ITRM20010718A1 - DESIGN TECHNIQUE AND OPTIMIZATION OF THE GSM RADIO ACCESS NETWORK. - Google Patents

Description

Description of the invention entitled:

"DESIGN TECHNIQUE AND OPTIMIZATION OF THE GSM RADIO ACCESS NETWORK"

DESCRIPTION

The invention in general relates to cellular telecommunication networks and more particularly it refers to a method of optimizing the GSM radio access network which, using tools and procedures of operational research, implemented by means of genetic algorithms, is able to provide a advantageous aid for planning the access network of the GSM service CHAPTER 1.

TECHNIQUE NOTE PART ONE NOTES ABOUT THE GSM MOBILE TELEPHONE SYSTEM

In this chapter some characteristics of the GSM telecommunication system are examined, in particular the concept of cellular coverage will be analyzed. The analysis of the cellular coverage technique will serve to identify the parameters (antenna positions, transmitter powers, etc.) on which to intervene to minimize the levels of electromagnetic fields and to improve the coverage and quality characteristics of the telecommunication network itself.

Mobile radio systems belong to the category of area radio systems. These radio systems allow a user to communicate on the sole condition of being in the portion of the territory in which the service is provided.

Modern mobile radio systems, also called cellular systems, allow two-way voice communication between two correspondents in all respects to that of ordinary telephony; as they are actually connected to the largest public telephone network, they do not suffer from restrictions on the identity of correspondents.

In particular, the main characteristics required of a modern mobile radio system can be summarized in the following points:

- quality of service similar to that of fixed networks,

- good coverage of the territory,

- large user capacity (obtainable with an emphasis on the frequency reuse technique),

- complete automation of both the call to and from the mobile user and the location of the user,

- accessibility by the mobile user to most of the services offered by fixed telephone networks,

- system security both as regards wiretapping and as regards the cloning of terminals.

The peculiarity of cellular systems is that they are able to support communication even on the move.

To obtain these characteristics, the following problems must be considered in the design of a mobile radio network: - limited availability of the radio resource (few frequencies available and therefore limited number of radio channels),

- mobility of users in the area served by the network,

- characteristics of the transmission medium.

1.2 CAPACITY OF A MOBILE RADIO SYSTEM

The capacity of a communication system means the maximum number of users that can be served at the same time, with a predetermined quality of service. The capacity of a system must be elastic enough, so that it is sufficient for current needs and easily expandable to cope with the increase in users, while maintaining the quality characteristics of the system.

Future investments to increase the capacity of the system must be contained and must not disrupt the existing network; therefore, already from the initial stages of a network project, a possible expansion of the network must be taken into consideration.

The capacity of a mobile radio system is fundamentally linked to the following parameters:

- number of radio carriers available,

- technical characteristics of the system (resistance to co-channel and adjacent channel interference, efficiency of hand-over and signaling procedures)

- geographical and temporal distribution of users.

In mobile radio systems the main limitation is given by the scarce availability of the radio resource.

To serve a very large area and a large number of users, all the techniques made available by technological development must therefore be used in the best possible way.

In particular, for the GSM system, the limitation of radio channels is overcome thanks to the adoption of multiple access techniques and thanks to a strategy of cellular coverage of the territory.

1.3 MULTIPLE ACCESS TECHNIQUES

In the GSM system, two multiple access techniques are used at the same time:

- FDMA (frequency division multiple access)

- TDMA (Time Division Multiple Access) '

The FDMA technique consists in dividing the communication resource, which in this case is the available frequency band, into many frequency slots and assigning them from time to time to users who request them.

A multiplicity of users can access simultaneously on different frequency slots.

The TDMA technique consists in dividing the radio resource over time, i.e. multiple users access the available resource entirely but at different times.

In this way there are several isofregtiential transmissions orthogonal to each other, because they occur at different time instants.

The division takes place over time by identifying a frame (or time base) which is divided into sub-intervals, each of which will be assigned a different user.

The fundamental problem of TDMA is the synchronization between users: they must access the common resource respecting the assigned time base, so care must also be taken to evaluate the distance between the radio-base station and the mobile user, as the delay ΔΤ is proportional to the distance.

To take into account the effect of delays and improve synchronization, the concept of Time Advance and guard times are introduced.

For the GSM system, the time advance is such that the distance between the radio-base station and the terminal can reach a maximum of 35 km.

In GSM both TDMA and FDMA techniques are used in practice the radio resource is divided into n frequency slots (available channels) and then each slot (each channel) is divided over time between 8 different users.

1.4 MOBILE COVERING TECHNIQUES

To increase the capacity of the radio-mobile system, while maintaining the transmission quality level unchanged, it is necessary to have a reuse of frequencies, i.e. the same frequencies must be used simultaneously in S

different geographical areas separated from each other by a distance sufficient to make the co-channel interference negligible.

The cellular coverage technique consists in dividing the service area into many areas called cells; for each cell there is a radio-base station which guarantees its coverage and therefore the desired degree of service.

Each cell has its own number of radio channels available, these channels can be reused in other suitably spaced cells, since the attenuation of the radio signals is proportional to the distance with an exponential (propagation coefficient) generally between 2 and 5, consequently the co-channel interferences are relatively low. In this way, even having a limited frequency band available, it is theoretically possible to serve a region which is however extended, or to serve in a given area a number of users which is however large by suitably reducing the size of the cells.

A cluster is defined as a complete set of M cells that use all the radio channels available to the system without any reuse of frequencies.

The size of the cluster is determined through a process of optimization of the reuse of frequencies on the

territory once the relationship has been defined:

this optimization process will allow us to guarantee the minimum level of quality given by the specifications.

To define the minimum C / I ratio, it is necessary to know the characteristics of the telecommunication system: type of modulation, type of coding, radio-frequency parameters, etc; in addition, the margin necessary to overcome the effect due to Raylegh fading must be considered.

For the GSM mobile radio system after considering the various

system parameters we have:

For the sake of simplicity, all identical cells made up of regular polygons will be taken into consideration, but in reality the cells can have different shapes and sizes. The size of the cells essentially depends on the type of coverage to be obtained: macrocellular or microcellular (respectively large cells, small cells). As for the transmitting antennas (and therefore the shape of the cells), if you want to minimize the number of base radio stations, it is advisable to use omnidirectional antennas placed in the center of the hexagons, because with the same distance between the transmitters and the furthest points (i vertices of the hexagon) the hexagon-shaped cell is the one that has the largest area.

In figure 1a, two types of clusters are represented: the one on the left consists of 4 hexagonal cells each covered by an omnidirectional antenna placed in the center of the hexagon, then the cluster is repeated to ensure coverage throughout the region (the radio channels); the one on the right in figure la "consists of 7 hexagonal cells also covered by omnidirectional antennas, the cluster is then replicated as mentioned above.

If you want to minimize co-channel interference, and thus achieve greater reuse of radio channels, you can use directional antennas as shown in Figure 1 b:

the cluster in figure 1b 'is made up of 9 rhomboidal cells, the antennas are placed in the center of the hexagons, the bold lines indicate the opening angle of the antennas; the cluster in figure 1b ”is made up of three hexagonal cells, the directional antennas are placed in the vertices of the hexagons, the bold lines have the same meaning as before.

From the images of figurei. b you can see that the antennas are installed in groups of three on a single site, this configuration is called having to and in addition to minimizing co-channel interference it also minimizes the number of sites.

From the previous examples it can be deduced that the number of cells N making up a cluster is equal to the product of MI by M2, where MI is the number of cells needed to cover a hexagon, while M2 indicates the number of hexagons forming the cluster Assuming Ml = l, it can be shown that M2 must verify the following relationship:

Where i and j are integers and consequently M2 can only assume certain values: 1,3,7,9,12, etc.

It is shown that the D / R ratio (co-channel reuse ratio) is linked to the size of the M2 cluster through the following relationship:

where is it:

D is the distance between the centers of two cochannel cells,

R is the radius of the hexagonal cells.

At this point, by calculating the C / I ratio, it will be seen that there is a link with the previous relationship.

Assuming that the system is balanced against interference (the interference calculated on the mobile receiver or on the radio base station is the same), the useful power received by the radio base station takes on the following value:

where is it:

Pm is the power transmitted by the mobile radios

Gm is the antenna gain of the mobile radio, GB is the antenna gain of the base radio station,

α is the propagation coefficient (typically a is between 2 and 5), λ is the wavelength.

The interference on the cell considered, caused by the j-th mobile terminal that transmits at its same frequency but in a different cell, takes the following expression:

considering only the first row of cochannel cells (therefore only 6 interference contributions must be added) and considering the j-th powers and gains all equal, the total interference is given by:

therefore the C / I ratio is given by:

cocanali we have:

ultimately by substituting the value of the D / R ratio in the previous relation, C / I binds to the size of the cluster:

From this relation it is possible to calculate the M2 value once the C / Imin value has been fixed

In the case of GSM C / Imin. it is 9 dB, then assuming a equal to 3 and replacing it in the previous relation we have. that the minimum M2 is 5.

To determine M2 in a real system, the long-term statistics of propagation fading must also be considered.

After determining M2, we move on to planning the radio channels and distributing the cells.

In planning the radio channels, in order to make the best use of the chosen cluster structures, it is advisable to divide the total number of radio channels by the total number of cells forming a cluster, thus obtaining the number of channels available for each cell.

In assigning the channels to the various cells, in addition to taking into account the co-channel interference, the interference from the adjacent channel must also be considered.

As regards the co-channel interference, we have seen that it is sufficient to define the minimum cluster size, which we need to define the minimum reuse distance that makes this interference negligible.

In order to limit interference from adjacent channel as well as not using adjacent channels in the same cell, they must not be used in cells belonging to the same site.

When adjacent channels are used in neighboring cells but not belonging to the same site as the one considered, there may be a degradation in the quality obtainable by the system, this degradation must be appropriately considered during the design.

Keeping in mind all this, in making the association of the groups of radio channels to the cells of the cluster, it is necessary to avoid as much as possible the use of adjacent channels both in the cell considered and in those of the same site, and possibly also in the adjacent cells. not belonging to the same site.

In the distribution phase, the territory is divided into clusters and then into cells; if the capacity of the system should be increased, the cells will be further divided into smaller cells and the distribution of the radio channels will have to be redone.

In carrying out this subsequent subdivision, in order to reduce co-channel interference, the cluster structure will eventually be replaced, passing from clusters with omnidirectional antennas to clusters with directive antennas.

From what has been described and other considerations that are not reported for the sake of brevity, in carrying out the design of a GSM network, the following characteristics must be optimized:

hedging strategy,

location of base stations,

antenna type and orientation (horizontal angle and vertical tilt),

power of transmitters,

height of the antenna masts,

taking into consideration objective constraints such as:

orography of the territory,

vegetative characteristics of the territory,

urbanization density,

geographical distribution of users.

Alongside these constraints *, other needs are being added, in fact the need to have a lower environmental impact (expressed by laws and common sense) increasingly pushes designers and network managers to consider alternative solutions in network planning, to in order to reduce the level of electromagnetic fields at least in the most populated and particularly critical areas (hospitals, schools, etc.)

1.5 OWNERSHIP OF GSM NETWORKS

In a mobile radio network, the radio base stations, the radio resources they use and the connections that are implemented must be managed by a centralized intelligence in order to guarantee the best use of the network performance at all times. This intelligence must include properties that allow it to carry out the various activities required.

These properties can be classified into four categories: a) basic properties

b) properties aimed at increasing user capacity

c) properties aimed at increasing the quality of the service seen by the user

d) properties aimed at increasing the quality of the service offered by the network.

The network will therefore have to support a certain number of functions that allow to satisfy these properties. The following table lists the main functions of a system

Function

Locating

Channel allocation

MS power control

DTX uplink

DIX downliuk

Freguency_hopping

Intra celi handover

Assignement to other celi

Overlaid underlaid subecell

Umbrella cells

Divcrsity antenna

We will see later how these functions satisfy the properties described above and how to best use them to obtain the expected results.

1.6 BASIC PROPERTY

1.6.1 Locating

The central intelligence administering the radio resource must first determine in which cell the calls are to be connected. Therefore it will have to locate the mobile terminals to assign them a channel in the most appropriate cell. Furthermore, as the mobile terminal moves, the connection must be switched to a new cell (handover) considered better by the system.

This activity is performed by the Locating function. The localization process is always active during a call and produces a list of cells that are candidates for handover in order of priority. This list is continuously updated and is used by the system at the appropriate time.

The input to the locating process necessary for defining the list is a series of measurements carried out by the mobile terminal and partly by the radio base station.

The cabinet measures:

the intensity of the signal received by the serving cell;

the quality of the signal (based on the BER on the decoded signal) received by the serving cell;

the intensity of the signal received by a certain number of adjacent cells.

The radio base station measures:

the strength of the signal received by the mobile station;

the quality of the signal received by the mobile station;

the timing advance value needed for the connection and sent to the mobile.

1.6.2 Channel allocation

When, through the locating process, the cell to be used for the call has been defined, a radio channel must be allocated inside it. This activity is carried out by the Channel allocation function. The function also defines the type of channel to be used according to the radio resource requested. The types of radio resources can be classified according to a certain number of parameters, namely:

procedure (handover / assignement / immediate assignement etc; in accordance with GSM);

use (signaling or voice / user data);

subcell (overlaid / underlaid)

Once the type of resource has been defined, the system will allocate the most suitable channel to it.

1.7 PROPERTIES SUITABLE TO INCREASE USER CAPACITY

1.7.1 General

The increase in the user capacity of a network can be implemented by defining ever smaller coverage cells. However, this can have a negative impact on the quality of speech due to interference. It will therefore be necessary to define functions that allow to maintain this quality at acceptable levels even using a very strict cell planning.

The functions that perform this task can be divided into three groups:

Functions aimed at maintaining a geographical optimization of the CU relationship.

To this group belong:

Locating;

Assignement to other celi;

Overlaid / underlaid subcell.

Functions aimed at decreasing the interferential phenomena in the network: To this group belong:

Locating;

MS power control;

DTX uplink;

DTX downlink.

Functions aimed at improving the radio environment and making it as close as possible to theoretical models despite the complexity of the real practical situation.

To this group belong:

Frequency hopping;

Antenna diversity.

1.7.2 Methods to improve the C / I ratio

In the GSM system, the localization for handover is carried out on the basis of measurements of the intensity of the radioelectric signal carried out by the mobile terminal (downlink) on the serving cell and on a certain number of adjacent cells. The edges for the handover will therefore be lines along which the intensity of the signal coming from the serving cell is equal to the intensity of the signal coming from the adjacent cell (C handover borders). These linesf ^ if the theoretical scheme seen above is respected, they coincide with the lines of equal C / I (C / I handover borders).

In practice, however, the theoretical scheme is not completely feasible and therefore situations will arise in which in the edges for the handover, based on the received signal intensity values, we will have C / I values below the limit tolerable by the system .

So let's see what needs to be done to improve C / I at cell edges.

As shown in figure 2 two factors contribute to define the C / I ratio. The strength of the useful signal (C) and the strength of the interfering signal (I)

Growing C in a site can improve the connections carried out in that area but this will certainly contribute to increasing I in a nearby site with a consequent worsening of the C / I ratio of that area.

Growing C everywhere will have the consequence of making I also grow everywhere by the same factor and therefore we will not have any improvement in the C / I ratio.

The best way is to modify the borders for the handover so as to bring them as close as possible to the lines of the same C / I (C / I handover borders) seen above.

Within a cell, the variations of C / I depend more on the variations of C rather than those of I. In fact, as shown in figure 2, moving between two cells the value of I remains fairly constant with respect to the value of C which varies much more significantly.

By employing a reuse of frequencies that does not coincide with the theoretical but fairly regular scheme, an average C / I handover border can be defined for the handover of equal C border).

Changing cell borders

In the GSM system, the localization algorithm works so that the edges for the handover are defined based on the intensity of the received signal C (C handover borders). This criterion is called "signal strength" mode.

If the reuse of frequencies is very irregular, the "C handover border" can be very different from the "average C / I handover border". If this situation occurs it is necessary, through a manual mechanism (modification of cell parameters), to move the border of a certain cell. If the positions of the equal C and equal C / I mean line are known from coverage predictions or simulations, the cell border can be moved so that the two lines coincide as closely as possible.

1.7.4 Assigning the call to the "best" cell

When the mobile terminal is in "idle mode" it is always looking for the best cell in terms of field strength received. This evaluation will be used by the mobile station to determine during the call phase on which channel and therefore in which cell to request access. Following the access request, the system will assign a signaling channel and then a traffic channel belonging to the cell chosen by the mobile. If the system supports the "Assignement to other celi" function, or more specifically "Assignment to better celi", at the time of the call setup it will be able to evaluate the C / I conditions and possibly assign a belonging channel to the call to a cell other than the one in which the mobile requested access.

1.7.5 Distribution of C: use of diversity

Frequencv diversitv

Fiequency hopping is a performance that causes an effect called fiequency diversity.

The fiequency diversity is based on the principle that the phenomenon of rapid fading due to multiple paths is frequency dependent. Therefore, if a communication is carried out by jumping in different instants of time between different frequencies, the field holes due to the rapid fading (fading dip) are less deep, and therefore the distribution of C more regular, compared to a communication carried out continuously on the same frequency. In this way the communications carried out by slowly moving furniture can be more regular and the fading margin to be considered in the celi planning phase can be decreased.

Furniture that moves quickly, while also using this feature, does not benefit from this point of view. In fact, due to the rapidity of the movement, the duration of the fading dip turns out to be very short even in the absence of frequency hopping and therefore the disturbances caused can be easily recovered by the corrector of errors in the channel coding

Space diversity

The performance of space diversity is implemented by using two receiving antennas positioned at a certain distance from each other in the same cell. This performance is based on the principle that the fading due to the multiple paths is different and independent on the two antennas. If the two antennas are equipped with two separate receivers, the two (independent) signals are combined in a scheme that offers maximum gain. 1.7.6 Decrease in interference

General strategy

Another approach to increasing the C / I ratio is to decrease I interference.

There are several ways to decrease the interference. One method is to focus on areas where C / I is critical and try to improve the situation where it is most needed. Another approach is to try to decrease the total amount of energy transmitted by the system. A third criterion is to limit the variation in interference as well as that of the received signal, in other words the statistical variance of factors C and I.

The performance employed in the GSM system to reduce interference uses these three types of strategies.

Power control

A connection made very close to the base station has a C / I ratio in excess of that necessary to obtain good conversation quality.

If the radiated power for this connection can be decreased (both uplink and downlink), the received signal C will decrease and consequently also the C / I ratio. This decrease will not correspond to a degradation of the voice quality as the C / I ratio will remain within the acceptable limit. This will allow to radiate less power with the same quality and we will have the advantage of decreasing interference in some other point of the network.

If the same strategy is adopted in all the cells of the network, the total amount of radiated power will decrease and consequently also the average interference within the network, the environmental impact also benefits directly.

This strategy will have the effect of keeping C constant regardless of the position of the cabinet and therefore also the C / I ratio can be kept within the desired limits at any point of the network. The goal of the power control algorithm is to regulate the power emitted by the mobile so as to keep the signal received from the radio base station constant at a desired value. This adjustment has a range of dynamics which is limited by the characteristics of the transmitter of the mobile station and therefore the adjustment will be effective only in a part of the area covered by the cell. In fact, near the base station, the mobile cannot reduce the power beyond a certain value. Conversely, away from the base radio station, the mobile, which is already transmitting at maximum power, cannot further increase the signal.

More in detail, the output power of the furniture is regulated (on SDCCH and TCH) on the basis of the SS (Signal Strenth) and quality (RX QUAL towards up-link) measurements that the BTS sends to the BSC: in this way the BSC calculates the output power level of the cabinet and sends the adjustment command → uplink interference reduction.

The maximum output power for the furniture is related to the class of the furniture and the MSTXPWR parameter (maximum power that a furniture can use on a cell):

Minimum power for furniture fGSM 05.051:

Phase 1 GSM900: 13 dBm;

Phase 2 GSM900: 5dBm;

GSM1800 phase 1, class 1: 10 dBm;

GSM1 800 phase 1, class 2: 4 dBm;

GSM1800 phase 2: 0 dBm;

The maximum power variation: 16dBm in 480ms;

consecutive regulation commands (2dB step);

In the case of immediate assignment and handover, the algorithm is reinitialized (that is, an initial adjustment that only provides for decreases starting from the maximum power, to avoid blocking the receivers - MS located near BTS - and not to degrade the quality other calls).

When fully operational, the power control algorithm operates in such a way as to take into account both the attenuation of the section and the quality, and provides for both increases and decreases.

The steady-state power control as shown in Figure 3 provides for both increases and decreases.

In order to minimize the power transmitted by the mobile phone, with a lower environmental and interference impact, while maintaining a sufficient level of radio link quality, the algorithm tends to maintain certain levels of power and quality.

An example could be:

of RX LEV signal received by the SRB between -85 dBm and - 91 dBm; of RX LEV signal received by the MS between -79 dBm and - 85 dBm; Quality RX Qual <= 3 for both Doun Link and Up Link.

Approximately three guidelines can be given to consider for power control planning:

- the cell area covered by the regulation must be as large as possible;

- the intensity of the desired received signal (desired signal strength) must be set to a value such that it determines the lowest tolerable C / I value in each point of the cell;

- the intensity of the desired received signal (desired signal strength) must be set at a high enough value to allow the quality of the voice to be maintained in every point of the cell.

These three requirements can be difficult to reconcile with each other in a network with irregularly shaped cells.

Overlaid / underlaid cell planning

Another approach to improve the distribution of the C / I ratio is to divide each cell into two subcells, an overlaid cell and an underlaid cell. Thus two overlapping networks are created which use two different frequency reuse schemes. Furthermore, overlaid cells have a limited coverage area compared to underlaid cells.

Let us suppose that the frequency reuse scheme for a given network proves to be too narrow and that therefore the C / I ratio near the edges of the cells is below the tolerable limit. A not too exemplary way to solve the situation could be to narrow the cells in order to exclude from them the areas with the C / I ratio below the threshold. In this way, however, traffic outside the confined cells would be cut off, inevitably creating cover holes at the boundary between the cells. However, if other cells can be provided with an additional set of frequencies independent from the main one and planned according to a looser reuse scheme, the boundary areas between the smaller cells can be covered by the larger cells in order to fill the coverage holes and obtain the required C / I ratio.

In practice, the planning of the cellular network is done using the largest reuse scheme and reserving a certain number of frequencies for the future. This is the network of underlaid cells. Later, when the demand for network capacity increases, the unused frequencies will be planned according to a tighter scheme and with the coverage limitations seen above. This is the overlaid network.

Fig. 4 shows an example of some overlaid and underlaid cells. On the cluster, the innermost hexagons represent the overlay cells, the larger ones the underlay cells, these are planned with different frequency groups f0 and f1 respectively.

Power control at the end of the sub-cells will introduce an additional benefit. The fraction of the area covered by the power regulation will increase for both sub-cells and therefore with it the efficiency of the power control.

Discontinuous transmission

Discontinous transmission (DTX) performance allows the transmitter of the mobile station and / or radio base station to be activated only when sufficient sound pressure is detected on the telephone microphone, in other words only when the user is speaking. This allows to reduce the amount of energy transmitted into the air and therefore to reduce the average interference in the system. Furthermore, the received signal is not affected by this function since, when the transmitter is active, the latter will use the correct power. Therefore using the DTX will improve the average C / I ratio of the system

1.7.7 Distribution of I.

The radio characteristics and therefore the C / I ratio can vary significantly as we move within a small area due to the physical environment. Furthermore, the C / I ratio in a specific geographical position can vary significantly over time due to variations in the physical environment (passage of trucks, etc.) and the distribution of mobile terminals and radio channels active in the surrounding areas.

This fact must be taken into account when planning the cells and a safety margin (celli planning margin) must therefore be defined.

Since the voice transmission in a digital system can withstand a C / T ratio up to 9 dB, it will be necessary to ensure that in the whole network the lowest expected value of the C / I ratio is equal to 9 dB, taking into account all the more unfortunate cases that can contribute to its degradation.

Unfortunately, the worst cases are such that the power control and DTX functions cannot contribute to their improvement. Let's see two examples.

Suppose that the mobile station causing the co-channel interference is positioned very far from the base station and therefore is transmitting at its maximum power.

Let us also suppose that the radio base station that is receiving the interference has the mobile station connected to it very close and that therefore the latter is transmitting at the minimum allowed power;

In this situation, while using power control, the C / I ratio can be worse than other situations where power control is not employed. The statistical average C / I ratio for this particular cell or position will then need to be planned so that C or the C / I margin used in celi planning is set at a high enough level to allow for the gain of the ratio to be exploited. Average C / I introduced by the power control.

Frequency hopping offers a method to combat co-channel interference, the transmit and receive frequency changes with each burst. If the connection that causes the interference changes frequency according to a cyclic repetition pattern different from the connection that receives it, the damage for that connection will be limited to a single burst, and interference will be distributed on a regular basis on all other active connections. In this way the duration of the interference on the single connection will be short enough to allow the error corrector to restore the original signal. As a result, the expected benefits of using the DTX and Power Control can actually occur.

The totality of interference in the grid in terms of energy emitted does not change with frequency hopping, but the statistical variance of the average C / I decreases and therefore the celi planning margin can be reduced by the same amount.

PROPERTIES ORIENTED TO INCREASE THE QUALITY OF THE SERVICE FROM A CLIENT VIEW

Generality

This section describes the system performance that can be used to increase the quality of service from the user's point of view.

Obviously the mobile station is a key component in the quality of service understood from this point of view and introduces factors that are beyond the reach of the network.

In any case, a certain number of radio performances of the network allow to improve the voice quality and the stability of the connections.

The functions that perform these tasks can be divided into two groups:

Functions that increase the quality and reliability of radio transmission (voice quality and reliability and efficiency of data transmission). The following functions belong to this group:

Locating

Frequency hopping

Intra-cell handover

Umbrella cells

Antenna diversity

functions that increase the battery life of mobile stations. The following functions belong to this group:

MS power control

DTX uplink

1.8.2 Quality of the radio transmission

The Locating procedure in the GSM system has two functions which are aimed at obtaining stable and high quality connections:

handover at bad quality handover at excessive timing advance.

The voice quality is determined on the basis of the BER (Bit Error Rate) measurement which is continuously performed on the uplink and downlink radio links. If the BER exceeds a certain value, a hanodver is performed in another cell.

This new cell may not be optimal from a C / I point of view. In fact this function can alter the cell borders that were defined to optimize the C / I ratio.

Advanced timing is also continuously checked during a connection. If it exceeds a previously defined limit, a handover is performed in another cell.

This function is designed to be used mainly in very large cells with very low signal attenuation (in the middle of a lake, in a maritime environment), to define a limit to the area covered by the cell. In these cases, in fact, if the cell were not limited we could have a strong degradation of the system C / I ratio.

Another way to improve quality especially in connections with high signal levels is to perform a handover in a channel belonging to the same cell (intra-cell handover). This is done on the assumption that the cell is still the best and that the quality degradation is due to stochastic coincidences. In these cases, a channel change can improve the quality of the transmission.

A system that uses frequency hopping allows connections that are subject to fewer variations in transmission quality and therefore calls will be more stable and more reliable.

Umbrella cells allow to increase the quality by providing a refuge for those connections subject to bad radio conditions. The probability of maintaining a connection increases and so does the reliability of the system.

The diversity antenna allows for more uniform reception at the base radio station making the voice quality more reliable.

1.8.3 Battery life

The MS power control and DTX uplink functions, despite being network performance, allow to reduce the consumption of the batteries of the mobile stations and therefore the battery life.

1.8.4 Time dispersion

A phenomenon of considerable importance, not correctable by the network functions and with a strong impact on the quality of the radio transmission, and therefore on the quality of the service seen

by the user, is that of temporal dispersion.

In a digital mobile radio system, the signal received at one point in the network is made up of several components. A direct component and several reflected components which, due to the multiple paths, arrive at the receiver at different times and therefore more or less delayed compared to the direct component.

The resulting signal will therefore be subject to intersymbol interference (see fig 5a). In fact, the bits that arrive late (reflected components) will interfere with the bits transmitted immediately afterwards (direct component and reflected components with less delay). The effect will be a strong degradation of the connection quality due to the indecision of the decoder

The GSM system is designed to resist the phenomenon of temporal dispersion as long as this is contained within certain limits, defined in terms of delay and amplitude.

In practice, a time window of 16 microseconds has been defined within which, thanks to the adaptive equalizer in reception, the time difference between direct and reflected signal can be corrected by the system. Outside this window the equalizer is still able to suppress unwanted signals as long as they are kept 10 dB below those contained within the window.

Therefore, to determine if a reflection can cause connection problems, you need to know two things:

The difference in time between the reflected signal and the direct signal. If this difference is less than 16 microseconds the delayed signal will not cause problems but, on the contrary, will help to improve the resulting signal.

The power of the signal components within the 16 microsecond window versus the power of the components outside it. Reflections will not cause problems if the signal strength inside the window is 10 dB higher than the power of the components outside it.

To assess whether the delay of the reflected signal could potentially cause connection problems, proceed as follows.

We consider a cell with a mobile station positioned randomly within the coverage area; an ellipse is drawn with the two foci positioned respectively on the base radio station and on the mobile station (see fig.5b) and such that the difference between the distance between the base station and the mobile station (a) and the distance between the base radio station - circumference - mobile station (b c) is equal to the distance traveled by the signal in 16 microseconds (about 5 Km). If the reflective objects are inside the ellipse, the signal they reflect will not cause connection problems. Obviously an ellipse must be considered for every possible position of the mobile station within the cell coverage area.

The size and shape of the ellipse depend on the position of the cabinet with respect to the position of the radio base station and therefore problems related to time dispersion could occur only in certain points of the cell.

Furthermore, since the reflected signal undergoes the same path attenuation as the direct signal, the reflected components from objects positioned far outside the boundaries of the ellipse arrive at their destination much more attenuated than the direct signal. In practice, therefore, the greatest problems will arise when the reflections are due to objects positioned just outside the contours of the considered ellipse and when the reflecting object is in optical visibility with the mobile station and the radio base station. In particular, these problems can be encountered more easily in rural areas.

In fact, in urban areas, due to the small size of the cells (typically from 1 to 3 km radius), the distances, and therefore the time delay, are short and furthermore the high concentration of buildings causes the effect of attenuating and dispersing in several directions. the reflected components.

Unfortunately, no system function is able to reduce the effects of time dispersion. In fact, frequency hopping, which manages to improve BER due to C / I, has no effect on BER due to reflections (C / R - Carrier to Refiected power ratio). Changing the transmitted power also has no effect as the strength of the reflected signal will decrease by the same amount as the power of the direct signal while keeping the C / R ratio constant.

The methods used to contain the effects of temporal dispersion are as follows:

Site relocation

Change of antenna

Downtilt

Use of downtilt antennas

Changing cell data

The most radical method is to move the base station to a more favorable site. However, this solution is not always easily achievable and economical.

A certainly more convenient system is to use more directional antenna systems (both on the vertical and horizontal diagrams) which allow to illuminate small parts of the territory thus avoiding that the signal reaches unwanted objects with strong intensity.

When very direct antennas are used on the vertical lobe (7 degrees or less) it is possible to direct them downwards (down tilt), radiating more the area close to the RBS, thus avoiding illuminating the more distant reflective objects.

The fundamental characteristic of downtilt antennas is to have the irradiation pattern on the asymmetrical vertical plane. This implies the possibility of illuminating the areas near the base station more and to a lesser extent those far away.

Using this type of antennas has the advantage of ensuring the service in the vicinity of the site, being able to use the right power, without the risk of going outside the boundaries of the cell and hitting distant objects that could cause unwanted reflections.

A final method to contain the effects of temporal dispersion is to modify the cell data. In fact, it is possible to set the cell parameters relating to handover so that this occurs before the mobile station enters an area where the problem of temporal dispersion is present.

1.9 PROPERTIES ORIENTED TO INCREASE THE QUALITY OF THE SERVICE OFFERED BY THE NETWORK

1.9.1 General

A network operator has as its goal that its customers, that is, the users of the network, are always satisfied. Prices, services and availability are the main weapons that operators have at their disposal to win over users.

The network must therefore be provided with a certain number of performances that allow to maintain and increase the quality of the service offered in terms of availability and reliability.

Furthermore, operators must be able to maintain the network in order to obtain the maximum economic return from their investments. A failed call is a lost profit opportunity. Congestion in one part of the network while there are free channels elsewhere is a sign of poor management of the network. Therefore the radio resources must be distributed where needed in order to increase the quality offered by the network without unnecessary waste.

1.9.2 Increase in network availability

In case of congestion of the traffic channels in the cell selected for the call set-up, it can be chosen to select a <·. traffic channel of another cell. This solution, even if it is not optimal from the point of view of the received signal and of the C / T ratio, allows to carry out traffic that in any case would be lost. In other mobile radio systems, this function is called "directed retry". In the GSM system, the directed retry function is not necessary as the Locating procedure, through the "Assignment to other celi" or more specifically "Assignment to worse celi" function, defines the cell to which the conversation will be assigned, keeping I also take into account the phenomena of congestion.

This feature increases system availability and optimizes the use of radio resources by decreasing the risk of traffic channel congestion.

Umbrella cells are cells that can be used in different ways. They can be used to improve the quality of the service offered by filling gaps in coverage, providing a spare resource (in case of congestion), locally increasing the capacity of the system, etc.

The general principle is to provide extra resources, which are rarely used, so that they can be shared over a large area rather than providing them locally with the risk of leaving them unused most of the time.

Base station receivers have limited dynamics. This can cause problems when the ^ noble station is very close to the base station. In these cases, in fact, the reception multicoupler can become saturated due to the high signals received, causing temporary blocking of the receivers and therefore decreasing the availability of the network. To avoid this it is necessary that the power transmitted by the mobile station can be adjusted (MS power control) in order to decrease it when the mobile is close to the radio station and increase it when it moves away.

1.9.3 Load and capacity

A high load in the network means a high load in the Central processor of the radio base station and, above all, of the BSC.

Most of the activities carried out by the system result in a load that is the same for each connection. In fact, each connection requires a set of signaling procedures for the cali setup and for the release of the call; occupies a signaling channel for the cali setup and a traffic channel for the conversation (0 user data transmission) the supervision of the call in progress causes a constant and equal load for each connection; etc The only thing that can vary a lot between connections of the same type is the number of handovers performed. The total load caused by handovers in principle has no limits,

The handover according to the Locating procedure of the GSM system will take place when the signal received from an adjacent cell becomes stronger than that from the serving cell, however, variations in the propagation conditions, in the reception conditions, in the characteristics of the receiver and variations in stochastic type, make the point of equal field strength between two cells to be rather diffuse.

As a result, a large number of unnecessary handovers may be performed before determining exactly which cell provides the highest signal.

The hysteresis provides a safety margin to be applied in the comparison between the intensity of the signal received by the serving cell and the adjacent cells.

Each cell will therefore be surrounded by an area called the hysteresis corridor where a connection can indifferently belong to any of the neighboring cells.

The concept of hysteresis is therefore used to control the load caused by handover. The wider the corridor, the lower the number of handovers and therefore the lower the load caused by unnecessary handovers. Approximately, however, half of the connections made inside the corridor will be made by the wrong cell (wrong celi); there will therefore be a large number of connections in this area that will not comply with the C / I conditions required by the Locating procedure. Therefore an excessive width of the hysteresis corridor can cause a degradation of the voice quality and consequently a high number of disconnections. The width of the hysteresis corridor must therefore be dimensioned in order to obtain a good compromise between the BSC load and the number of disconnections.

The same considerations are applicable to handover between subcells and therefore to the hysteresis corridor between underlaid and overlaid subcell.

The intra-cell handover function can also help increase the BSC burden. In this case the right compromise between BSC load and quality and stability of the radio transmission can be found by acting on the threshold for intracell handover.

CHAPTER 2.

KNOWN TECHNIQUE, SECOND PART. MEASURES: TELEPHONE TRAFFIC AND QUALITY INDICATORS OF THE GSM RADIOMOBILE SERVICE

This Chapter reports the analysis of the difference measured in terms of Service Quality and Environmental Impact on cells with high and limited coverage

demonstrating that the two objectives are best achieved when the SRBs are located near the sources of telephone traffic or when they are as close as possible to where the customers are concentrated.

This conclusion was mathematically translated into specific functionals subsequently discussed in chapter 6.

2.1 REGULATIONS

Fundamental elements for the insertion of new Radio Base Stations within the existing GSM network, derive from the analysis of the network measures that each telephone operator has at its disposal, and from the analysis of data relating to Marketing.

In particular, the network side measures of Traffic and Quality respond appropriately to the fundamental questions of WHEN and WHERE it is necessary to install a Radio Station

Base.

- Traffic

Traffic in terms of the ability to dispose of erlang, or rather allow the customer to speak or access Value Added Services (VAS), the resulting profitability is a function of the tariff plan.

In the analysis, a historical archive is important to consider the trend from which an investment projection for the future is extrapolated.

- Quality

Quality also as perception by the customer of the degree or quality of the service offered by the telephone and service provider. Quality is a fundamental aspect of Loyalty or it affects the fact that a customer decides to abandon a manager in favor of another and therefore of the Chum, or the period after which the customer leaves the manager.

This aspect of the service will be increasingly important in particular in services, through the technologies used in the Radiomobile field, for example Internet Banking via mobile phone, with a simple SMS you can command a banking transition, if the Short Message is not successful the customer. it could suffer more damage than the simple cost of an SMS.

Below is a possible definition of Service Quality; a short series of points describe its profile:

Coverage: Where Can You Talk? turning on the cell phone, does the manager appear? Accessibility: does the customer find the free line? Dropouts: Did you experience an unwanted call dropout? Speech interference or clarity: how does it feel? AND'

intelligible? Do I recognize who is speaking? Availability of the service: does it work every day?

Effectiveness of the service: are phone calls routed correctly?

Interactivity: Is navigation fast? Do the SMS arrive on time?

Correctness of the

documentation: is the pricing correct?

Transactional capacity: how many bank transactions can I make? How many emails with photos can I send in 15 minutes? Transactional Effectiveness: How Many Transitions Are Successful? Has the SMS top-up been credited? Internet connection: does the service crash? need restart or reboot? Is the opening of web pages fast? is the cost of mobility offset? Which comparison with reje'fissa? Image: the 5 notches appear there is the writing GPRS

Other Open Businesses: they do not explicitly concern this discussion and are not mentioned; There are other aspects concerning the definition of Quality; in the following discussion, the first four points which mainly characterize the radio access network are of particular interest.

There is no standardization of measurement methods and reports, however the data base that is considered is similar for all managers, in particular it is possible to indicate some types of customary measures that can be mapped in a georeferenced way that is superimposed on a geographical map and others deriving from the analysis of statistical counters

2.2 STANDARD ELEMENTARY MEASURES OF THE GSM SYSTEM (ETSI standards reference)

- Power (RX-Level) and BER (RX-Qual) measurements both Down Link and Up Link according to Time Advance.

These measurements are based on the Measurement Reports that are sent from all mobile phones that are switched on periodically to the BSC (Base Station Controller). It should be specified for the BER that RX Qual 0 corresponds to a low error rate, RX Qual 7 corresponds to a high error rate as per the following table obtained from the ETSI specifications:

From practical experience it has been found that the BER corresponding to the RX QUAL 0-3 band can be correlated with a good intelligibility of the conversation; while the RX QUAL 6 and 7 band with poor intelligibility.

- Georeferenced measurements with Mobile Test systems moving on defined routes.

With these systems it is possible to analyze a single connection by extrapolating the indicators such as:

- Power levels received and transmitted from both the mobile and the RBS;

- Single frame error rate (BER);

- Working distance from the cell (Timing Advance) which is a Quality factor;

- Connection failures;

- attempted and successful handover between source and destination cell on a single call;

- MOS (Mean Opinion Score) indicates the human perception of speech quality, taking into account many parameters including timbre, intelligibility, micro interruptions. A scale of values from 1 to 5 indicates 1 bad perception, with 5 excellent;

- Failed access to the network by the mobile phone;

- Coverage;

- Measurements through Statistical Traffic Counters and Cell Quality.

The available indicators are aggregates of meters, each of which increases for each event that occurs during a telephone connection, typically there are two phases, during the first one the signaling between the mobile phone and the network takes place, this is completely invisible to the customer but allows the mobile phone to receive calls or to make or to move from one cell to another by means of a handover or to send an SMS. Another phase concerns the phonic conversation. For each phase there are several counters whose increase or decrease indicates the success or failure of a procedure on an hourly basis, for example the increase in the connection drops counter indicates whether the users served by a cell suffer this phenomenon of poor quality. .

2.3 ANALYSIS OF MEASURES: Quality aspects of the Service

In order to analyze both the quality of the service and the powers emitted by the mobile phone as a function of the distance from the RBS from the traffic sources, the measurements for four different types of cells are reported, the real names of the cells have been replaced with fictitious names:

- Extra-urban cells with wide coverage with a radius between 1 - 30 Km: Monte 1 and Monte2

- Urban cells with large coverage radius greater than 3 Km: Urban

- Urban cells with modest coverage with a maximum radius of 3 km: Urban2 - Microcells with a coverage radius of less than 400 m: Urban3 translatable into Km, and the attenuation in dB of the power emitted by the mobile phone according to the quality of the signal received by the SRB.

The trends of the main statistical quality indicators for all the cells under examination are then shown on graphs.

2.3.1 Mount 1

The first extra-urban cell with high coverage taken in analysis, as shown in fig. 6, is Monte 1.

The extended coverage as well as simulation can be seen from the scattered distribution of the Measurement Report samples of level and quality as a function of the temporal distance Timing Advance, in particular the samples with poorer quality RX-QUAL 6-7 in red color increase with the distance from the cell . One cause of the degradation of quality is the signal of the serving cell which reduces with distance, therefore at a high distance another cell that interferes even with low levels can reduce the C / I in these areas.

A further degradation occurs in correspondence of the handover and of the Location Upadating

(registration of the mobile phone in Idle mode to the GSM network). The handover between cells distant several kilometers, have a high percentage of failures and also easily induce the telephone call, towards a connection failure on other cells if for subsequent handover you camp in cells from which you cannot go back.

Location Updates can congest the SDCCH signaling channels causing difficulties in accessing the GSM network.

Typically, in fact, the cells with wide coverage intersect Larda which is the responsibility of many other cells including those belonging to different Location Areas (different LACs).

Each LAC is a grouping of cells included in a geographical area normally controlled by an MSC.

Mobile phones in Idle Mode that cross the border between different Location Areas register to the network to communicate their position, this system allows the network to uniquely address a paging to a mobile phone, only through the cells belonging to a given LAC, the one where the mobile phone is registered, avoiding spreading the call on the unaffected cells of the network, that is, preserving the network from a useless signaling load.

With regard to the power transmitted by the mobile, the graph shows the attenuation in steps of 2 dB of the maximum power with which the mobile phone can transmit.

Montel has most of the traffic several kilometers away, therefore the mobile phones that are encamped there always transmit at maximum power.

The transmission to then. Maximum causes interference in UP-Link and in addition to consuming the batteries of the mobile phone more quickly, it has an impact on the environment as it is a source of radiation directly in contact with the human body.

Figure 7 in relation to the considered Cell refers to the RXQUAL Down-Link as a function of Time Advance, reporting the distance in Km on the abscissas and the number of measurement samples on the ordinates. While figures 8 and 9 report respectively the RXQUAL Down-Link and the number of "Location Updating-call Access - Handover" as a function of the distance

2.3.2 Urban

An example of an urban cell but with high coverage, is the Urban SRB near a lake, as shown in figure 10. This SRB mainly covers the tourist resort located on the coast of the lake but the signal propagates to the other side of the lake going to interest places up to 30 km away where you can see a peak.

On the graph showing the distribution of the Measurement Report Samples as a function of Time Advance, we can see peaks of traffic with poor quality around 12-13 Km, these correspond to the countries immediately on the other side of the lake.

This confirmation is also found in the distribution of Cali Access accesses which also occur around 12-13 Km, where there are greater drops in connection on the SDCCH signaling channels with failure of the attempts to call the telephone names. Updating locations occur for the most part on the other. part of the lake with cells from another region.

The graphs taken into consideration are shown in figures 1 la and I I b relating respectively to the number of "Location Updatng-call Accesss-Handover" and to the RxQual Down Link as a function of distance.

The logic for a Coverage and Quality Project indicates that to eliminate traffic peaks at 13 and 30 Km, systems should be activated in correspondence with those traffic source peaks and, at the same time, the urban cell antennas should be tilted in order to avoid the child restraint covers up to 30 Km. 2.3.3 Urhan 2

This is the case reported in Figure 12 of an urban cell in which it is observed how all the Measurement Report Samples are concentrated at a short distance from the cell and at the same time how the samples with RX Qual 6-7 are extremely contained. It is the ideal coverage for a medium-sized city cell which, being very confined, creates limited interference and offers a qualitatively high service in its coverage area, figures 13 and 14 show, for this case, respectively: the RxQual Down-Link and the number of "Location Updating-call Accesss-Handover" as a function of the distance.

2.3.4 Mount 2

With reference to Figure 15 there is a further example of an extra-urban cell with high coverage. Figures 16a and 16b show, for this case, respectively: the RxQual Down-Link and the number of "Location Updatng-call Accesss-Handover"

as a function of distance.

2.3.5 MICROCELL Urhan3

E1 a Microcell whose traffic is distributed over 300 m. As evident from figures 17 and 18 which report, for this case, respectively: the RxQual Down-Link and the number of "Location Updatng-call Accesss-Handover" as a function of the distance, from which it can be deduced that in the regulation of the furniture power these are able to transmit with low power tending to the minimum near the microcell.

2.4 DOWN LINK POWER LEVELS

The graph shown in figure 19 shows a correlation of the average levels with which each child restraint considered works.

The trend is obviously of the Gaussian type with a peak value corresponding to the levels used most by the cell. A main difference is immediately evident between the distribution relative to the microcell where the levels with more numerous samples are concentrated around -65 dBm with a very large bell, and the cells with wider coverage where the samples are more present at -85 dBm. .

In the functionals that will be used subsequently, the reference level for the calculation of the BEST SERVER in a historic center is -60 dBm.

2.5 ANALYSIS OF STATISTICAL QUALITY COUNTERS

It can be observed that the cells examined have a different behavior, in particular the cells with the widest coverage such as Monte2 have the worst indicators, that is, failed handover with call loss and high connection drops.

This cell, by virtue of the coverage, disposes of a higher traffic, even if with worse quality, therefore from a purely economic point of view it would seem the best solution, i.e. higher EFFICIENCY = (ERLANG disposed of / PLANTS). 'At this point it remains to establish the threshold of economic compromise between the number of plants to dispose of a certain quantity of Erlang and the quality of the service to be provided.

This topic is dealt with in chapter 4 dedicated to traffic analysis for now we can limit ourselves to stating that in the rural area it is, in general, economically less profitable to have many systems for splitting traffic as the value for money does not derive a benefit.

The graphs represented in figure 20a-d show in order: a- DSDC Connection drops on the signaling channel SDCCH;

b- Dropped TCH Connection drops on the voice channels; c- TROFF Traffic offered in Erlang;

d- HFFP% Handover between different cells failed with loss of the call.

From these indicators it emerges, for example, that the urban cell Urban2 with reduced coverage has a percentage of successful handover of 98% compared to 85% of the cell of Monte2.

This confirms once again that the cells whose coverage area is limited to a target close to 1-2 Km MAX, offer a better quality of service.

2.6 ANALYSIS OF MEASURES: aspects of Radio Environmental Impact and Service Quality

Contrary to what one might think the two aspects are not unrelated, a good quality of service implies a low environmental impact of cellular telephony.

This correlation is shown with an example, the cell examined as shown in figure 2 1, is an urban cell of Ancona located in the new commercial area of the city.

A path was carried out by moving away from the child restraint, with a test mobile phone, tracing from the telephone exchange the forced call on this child restraint, ie excluding the possibility of handover on other child restraints.

The post-processing of the measurement reports made it possible to analyze the behavior of a mobile phone that moves away from its serving cell while remaining hooked to it.

All the following analyzes were carried out on the Up Link section, i.e. from the mobile phone to the RBS, in any case the same conclusions are also valid on the Down Link section, i.e. from RBS to mobile, where reference is made to the power emitted by the RBS instead of the MS.

The coverage simulation and a graph are shown in figure 23, indicating the correlation between:

- between the power transmitted by the mobile phone;

- the level in dBm RX LEV UP LINK received by the SRB;

- the RX QUAL UP LINK quality measured in the SRB;

- temporal distance of the furniture from the child restraint, a temporal unit of T. Advance corresponds to 550 m, therefore TA = 7 = 3.8 Km max. Fig. 18 shows a graph relating to the correlation of the environmental impact and the quality of the GSM service with the time distances shown on the abscissa.

The red curve represents the average attenuation in db of the power transmitted by the mobile phone (m.s. = mobile station), as the distance between the mobile phone and the child restraint increases the power received by the child restraint in dBr (green curve referred to -1 10 dBm) decreases, consequently the mobile phone increases its transmission power in order to balance the attenuation suffered by the signal.

In an open area such as the one considered, over about 2.5 km away from the RBS, the mobile phone always transmits at maximum power, this circumstance has three effects:

- maximum energy radiated into the human body by the mobile phone, ie maximum Environmental Impact;

- maximum interference produced in the system by an apparatus that transmits at maximum power and therefore worsening of the C / I on the network;

- maximum power consumption of the mobile phone resulting in faster battery discharge.

The deterioration of the BER (RX QUAL) is also noted with increasing distance, this degradation affects the intelligibility of the conversation and therefore on the service.

On a densely urbanized area, this distance can be reduced by more than 50%, especially if the area is interfered with.

These aspects suggest that to improve the quality of the service offered to the customer and at the same time reduce the environmental impact, it is necessary that the child restraints are located close to the sources of telephone traffic or in the immediate vicinity of the coverage objective of the child restraint.

2.7 MS P.CONTROL ANALYSIS ON AN URBAN AREA In order to highlight the functioning of the mobile power control on an urban area, the city of Ancona was taken as an example.

A path was carried out with a georeferenced measuring instrument, in this way it is possible to highlight all the events of interest on the geographical map of Ancona:

- Signal level received by the mobile RX LEVEL;

- RX QUAL (BER) quality;

- Time distance between mobile phone and cell serving Time Advance;

In Fig. 23 it has been marked with

1- the RX Qual7, handover and max TX power with subsequent decrease in TX power

2- the handover and the increase in attenuation of the power transmitted by the cabinet.

As highlighted in the circled areas, the cabinet tends to transmit with less power near the serving SRBs and with greater power away from them.

A peculiarity that can be seen in the following graphs is the sawtooth step-by-step trend of the power.

This phenomenon is explained by the fact that with each handover or each new call the mobile transmits, in accordance with the GSM specifications, with maximum power.

Subsequently, if the levels and quality allow a reduction in the transmitted power, the mobile phone begins to decrease its power in steps of 2 db, this can be seen from the steps present in the sawtooth.

If the level reached is too low, the C / I is reduced by degrading the BER which tends to RX QUAL 7, consequently in most cases in order to restore the value of RX QUA1 <= 3, an average increase of 4 a is induced. 6 dB minimum and instantaneous of the power transmitted by the mobile phone.

In other cases, when the RX QUAL value worsens, you have a handover and start all over again, the procedures are not random, it depends on how the thresholds are set, the intervention of the power control intervenes before any handover for quality .

Cells with large urban coverage have local coverage areas with reciprocal intersection, so there are many sometimes unnecessary handovers.

Since every handover also corresponds to a micro interruption of the voice, it also follows a worsening of the MOS (Mean Opinion Score) for this reason.

The graph in figure 23 therefore shows the trends of RX Level, RX Qual, T. Advance, MS TX POWER ATTENUATION; the variable level of the T.Adavance is due to the handover between different cells in correspondence with which the attenuation of the Mobile TX power is observed going to the minimum ie maximum TX power.

Chapter 3: TRAFFIC SCENARIOS

In this chapter some GSM traffic cases will be analyzed in order to determine a traffic model to be implemented for determining the number of theoretical Radio Base Stations necessary to guarantee the GSM service in areas with a known density of potential users. This proportion of SRBs will be included in the theoretical area to be optimized in terms of the quality of the GSM service and environmental impact with the functionalities described below.

3.1 MOBILE TLC SCENARIO

Cellular technology allows adequate performance against investments of about one-tenth compared to the fixed network '.

This technical and economic compromise leads cellular technology as a probable candidate for voice telephone communication in both the consumer and business sectors.

Furthermore, with GPRS and UMTS there is an expansion from voice to data: concrete evidence of this orientation is in-banking services.

To confirm, there are already agreements with major credit institutions to offer customers m-banking, online trading and m-commerce services through the use of a special smart card provided by the manager.

SIM toolkit and WAP allowed the development of the first applications (of elementary interface) and already today facilitate access to increasingly advanced services and GPRS is characterized by high capacity and quality in data transmission.

UMTS, that is the third generation mobile radio service, will make multimedia applications available even on the move, revolutionizing the scenario of the mobile radio market. In this context, technological innovation will be a determining factor, the evolution towards Personal Communication will require an enhancement of:

Basic performance Terminals and Network access, quality and coverage of the network

Higher value performance: messaging, transmission speed, intelligent network services.

All managers are already called upon to transform themselves from a simple “talk minutes factory” to a “minutes and services factory”.

What will be profitable in the future is represented by the contents that will be transported by mobile radio networks: information, advertising, services, videos.

For the moment we are still in the optimization phase of the Transport Network, today GSM / GPRS tomorrow UMTS.

The purpose of the present invention is to provide a design and / or optimization technique of the GSM radio access network which, by means of a traffic model, allows the determination of the number of theoretical Radio Base Stations necessary to guarantee the GSM service in areas with a known density of potential users. This proportion of child restraints will be included in the theoretical area to be optimized in terms of quality of the GSM service and environmental impact with the functionalities described below.

A further object of the present invention is to carry out said analysis of the GSM network by means of suitable procedures that use genetic algorithms, to search through functionalities the best solution for the location and for the choice of the main radio parameters of the RBS in an urban environment. In practice, these algorithms of a genetic type make it possible to obtain a planning of the GSM network with a lower environmental impact and a higher quality of the service offered.

Finally, the object of the present invention is to provide a technique for designing and optimizing the GSM radio access network which uses standard protocols, procedures and technologies in the field of cellular telephony in order to make the technique itself immediately applicable and easy to manage.

These and other purposes that will be clear in the course of the description are obtained by means of a technique based on the classification and evaluation of the objectives and constraints involved in the design of the radio base station, such as:

- maximization of the relationship between useful signal and interference, - maximum reduction of the distance between the radio base station and the traffic source,

-the maximum reduction of the presence of electromagnetic field, -the maximization of the cell yield based on the traffic / cost ratio,

- maximization of the coverage level,

by means of an operating model that formalizes the relationship between the objective and the parameters on which it depends, based on the use of genetic algorithms for global optimization through a compromise criterion between the individual indices.

This technique for designing and optimizing the GSM radio access network uses means of representing the operation of a certain number of base radio stations as Best Servers in regions enumerated with a first index like the radio base stations themselves, these regions being their once divided into smaller pixels with a further index so as to discretize the electromagnetic field values in space with greater precision and at the same time provide, to the software means used, the input data relating to the traffic expected on the various pixels themselves. By means of algorithms typically in the field of genetic research, it is possible to obtain significant representations and indications on the possibility of optimizing the GSM radio access network.

For illustrative purposes only and without thereby wanting to limit the scope and scope of the invention in any way, the same will now be illustrated on the basis of a preferred embodiment with reference to the attached figures in which:

Figs. la and lb are examples of cell clusters,

Fig. 2 is a schematic representation of the relationship that is established between the useful signal and the interfering signal near the edge of the cell;

Fig. 3 is a graph indicative of the power control system; Fig. 4 is an example of some overlaid and underlaid cells;

Fig. 5 is a representation of temporal dispersion in relation to the intersymbol interference, while Fig. 5b is indicative of the temporal dispersion relative to the guard edge.

Fig. 6 is a graphic representation of an extra-urban cell; Fig. 7 is a graph relating to the measurement carried out within the previous cell of Fig. 6;

Figs. 8 and 9 are graphs relating to further measurements carried out on the extra-urban cell of Fig. 6;

Fig. 10 is a graphic representation of an urban cell with high coverage near a lake;

Figs. I la and lb are graphs relating to measurements carried out within the cell coverage of Fig. 10;

Fig. 12 is the graphic representation of a medium-sized urban roof cell;

Figs. 13a and 13b are graphs relating to measurements made within the coverage associated with the cell of Fig. 12;

Fig. 14 is a representation of an extra-urban cell with high coverage;

Figs. 15 and 16 are graphs relating to measurements made within the coverage cell of Fig. 14;

Figs. 17 and 18 are graphs relating to measurements made on a microcell;

Fig. 19 is a summary graph relating to power distributions in cells of different types;

Fig. 20 includes further summary graphs relating to power distributions for cells of different types;

Fig. 21 is a representation of a geographical area relating to cellular telephone coverage;

Fig. 22 comprises a series of graphs relating to measurements made on the coverage area of Fig. 21;

Fig. 23 is a graph showing a correlated assessment of the environmental impact;

Figs. 24 and 25 are graphs that relate the efficiency of some types of clusters with the radius of the unit cell;

Fig. 26 is a schematic representation of a cluster;

Fig. 27 is a graphic representation of a territorial map; Figs. 28 and 29 show graphs relating to GSM traffic on a particular reference model;

Fig. 30 is a flow diagram relating to the various phases included in a genetic algorithm;

Fig. 31 is a three-dimensional graphic representation of a possible solution space within the genetic algorithm;

Figs. 32 and 33 are graphic representations of the urban environment from which to deduce some fundamental analysis parameters; Fig. 34 is a graphical representation of the constraints involved in the choice of the optimal reception transmission paths; Fig.35a is a graphic diagram indicative of various trajectories generated by the genetic algorithm for the search for the possible permitted radio base station,

Fig. 35b is a graphic diagram indicative of various trajectories generated by the genetic algorithm for the search for the best radio base station;

Fig. 36 is a graphical representation of an exemplary reference urban area used in the graphical procedures according to the present invention;

Fig. 37 is a three-dimensional representation by layer relating to the input-output data supplied to the process according to the present invention;

Fig. 38 is a graph of an exponential function relating to the additional attenuation for transmitting antenna tilt with an omni-directional radio base station;

Fig. 39 is a graph that schematises with a pentagonal figure the fundamental parameters involved in the optimization and / or in the design of a GSM network;

Fig. 40 is a representation of a cluster of cells with different traffic density;

Fig. 41 is a graphical representation of functionals that work together to obtain the optimization of the distances;

Fig. 42 is a graphic representation similar to the previous one relating to two distinct regions;

Fig. 43 is a graph showing the interrelation between three cells, one serving and the other two interfering;

Fig. 44 is a diagram that indicates the combinations for a mutual control of the interferences between three regions covered by three radio base stations;

Fig. 45 is a diagram relating to two coverage regions divided over two pixels;

Fig. 46 is a graph relating to the dynamics of the power control;

Fig. 47 is a diagram relating to the procedure for minimizing the performance of the radio base station;

Fig. 48 is a graphical representation of a functional coverage with limits set per pixel;

Fig. 49 is a graphic representation of a functional absolute minimum environmental impact;

Fig. 50 is a graphical representation of a functional C / I ratio with limits! minimums set per pixel;

Fig. 51 is a graphical representation of a functional minimum distance of the radio base station from the traffic source;

Fig. 52 is a graphical representation of a functional on the maximum throughput of the network;

Fig. 53 is a representation of the distribution of best server areas of the radio base station by evolution of functional efficiency type 1;

Fig. 54 is a graph relating to the values that the cost functional type 1 has assumed following the interactive procedures;

Fig. 55 is a representation of the distribution of the dBm coverage by evolution of functional efficiency type 1;

Fig. 56 is a representation of electric field distribution in V / m, for functional evolution with efficiency type 1;

Fig. 57 is a representation of the distribution of C / I dB by evolution with efficiency type 1;

Fig. 58 is a representation of the SRB / pixel distances served by evolution with functional efficiency type 1;

Fig. 59 is a graph relating to the evolution of the functional efficiency type 1 due to the effect of the movement imposed by the genetic algorithm;

Fig. 60 is a representation of the SRB distribution and urban area map by evolution with delta-like performance of the SRBs; Fig. 61 is a representation of the best server areas for SRBs for 1 functional cost maximum efficiency type 2;

Fig. 62 is a representation of the dBm coverage for functional cost-efficiency type 2;

Fig. 63 is a representation of the interference distribution C / l dB for functional cost-efficiency type 2;

Fig. 64 is a representation of the electric field in V / M for the functional maximum efficiency cost type 2;

Fig. 65 is a representation of the SRB pixel distances serving for the functional maximum efficiency cost type 2;

Fig. 66 is a representation of the SRB location in an urban area for functional maximum cost efficiency type 2;

Fig. 67 is a graph showing the SRB movement produced by the genetic algorithm for a functional cost-yield maximum;

Fig. 68 is a graph relating to the maximum yield cost functional;

Fig. 69 is a graph relating to the efficiency functional with traffic limits / maximum efficiency cost functional;

Figs. 70a to 70e report respectively:

a) functional environmental impact

b) minimal functional interference

c) functional coverage with zone limits

d) functional distance power control

e) functional minimum distance traffic;

Fig. 71 is a representation of a distribution of best server coverage areas for functional minimum distance SRB traffic sources;

Fig. 72 is a representation of the electric field distribution in functional V / m of the minimum SRB distance per traffic source;

Fig. 73 is a representation of a C / I distribution for functional minimum SRB distance per traffic source;

Fig. 74 is a distribution of the pixel distances from the SRB serving for functional minimum distance SRB traffic source;

Fig. 75 is a representation of a SRB distribution and map of the urban area, for functional minimum distance SRB traffic source;

Fig. 76 is a graph indicative of the movement caused by the genetic algorithm for the best research in minimizing the distance of the RBS from the traffic sources;

Fig. 77 is a representation of the best server areas for functional minimum distance traffic;

Fig. 78 is a representation of a coverage in dBm per traffic functional minimum distance;

Fig. 79 is a representation of a C / I interference distribution in dBm per traffic functional minimum distance

Fig. 80 is a representation of an electric field distribution in V / m of the minimum distance traffic functional; Fig. 81 is a representation of an SRB pixel distance distribution serving the minimum distance traffic functional; Fig. 82 is a representation of a SRB distribution on the urban area for functional minimum distance;

Fig. 83 is a graph showing the SRB trajectories modified by the genetic algorithm in relation to the functional minimum distance traffic;

Fig. 84 is a graph showing the functional trend of traffic for minimum distance in relation to the number of tractions of the generic algorithm;

Figs. 85a-85f report respectively

maximum C / I functional trend,

functional trend minimal environmental impact,

functional coverage with limits,

average level of coverage in of BM,

functional cost with traffic limit,

functional trend max cost,

all reported according to the traction number of the generic algorithm;

Fig. 86 is a representation of a distribution of best server areas for a functional optimization of the radio coverage with the minimum fixed per pixel;

Fig. 87 is a representation of a coverage distribution in dBm relating to a functional for optimization of the radio coverage with minimum fixed per pixel;

Fig. 88 is a representation of a C / I distribution in dB relative to a functional for optimization of radio coverage with fixed minimum and pixels;

Fig. 89 is a representation of an electric field distribution in V / M for a functional optimization of the radio coverage with a fixed minimum per pixel;

Fig. 90 is a representation of the SRB-serving pixel distance distribution relating to a functional for optimization of the radio coverage with a fixed minimum per pixel;

Fig. 91 is a representation of the SRB distribution, in addition to the city map. relating to an optimized radio coverage function with a fixed minimum per pixel;

Fig. 92 is the representation of a SRB movement relative to the coverage functional with a fixed minimum per pixel;

Fig. 93 is a representation of the best server areas for coverage functional with zone limits;

Fig. 94 is a representation of the coverage distribution in dBm relating to the coverage functional with zone limits;

Fig. 95 is a representation of the C / I dB distribution of the coverage functional with boundary zones;

Fig. 96 is the representation of the electric field distribution in Volt / m relative to the coverage functional with zone limits;

Fig. 97 is a representation of the three-dimensional coverage in dBm relating to the coverage functional with boundary zones;

Fig. 98 is a representation of the distribution of the distances in pixel-SRB of the server relative to the coverage functional with zone limits;

Fig. 99 is a representation of the position of the child restraint on an urban area relative to the coverage functional with zone limits;

Fig. 100 is a graph relating to the SRB trajectories relating to the coverage functional with minima on zones;

Fig. 101 is a graph relating to the coverage functional with zone limits in relation to the number of iterations of the genetic algorithm; Fig. 102 relates to the graph of the maximum coverage functional relative to the coverage functional with zone limits in relation to the iteration number of the genetic algorithm;

Figs. 103a-b relate to:

a) functional minimum environmental impact,

b) functional minimum distance traffic;

Figs. 104a-b are respectively related to the functional cost with traffic limit relating to the number

b) functional cost type maximum

relative to the number of iterations of the genetic algorithm;

Figs. 105 a, b, c are graphical representations respectively of:

a) functional power control distance,

b) functional A / C with zone limits,

c) functional C / I maximum,

all related to the number of iterations of the genetic algorithm.

Fig. 106 is a representation of a best server distribution for functional maximum radio coverage on the pixel;

Fig. 107 is a representation of a coverage distribution in dBm per functional maximum radio coverage per pixel;

Fig. 108 is a representation of an electric field distribution in wolt / m for a functional maximum radio coverage on pixels;

in dBm for the functional relating to the minimum environmental impact; Fig. 113 is a representation of a distribution of the C / I in dB relative to the functional for the minimum environmental impact;

Fig. 114 is a representation relating to a distribution of the electric field in V / M for the functional of minimum environmental impact;

Fig. 115 is a representation of the distances in M pixels per SRB serving relative to the functional for minimum environmental impact; Fig. 116 is a representation of a map of an urban area and of the SRB relative to the functional for the minimum environmental impact; Fig. 117 is a graph relating to the movement provided by the application of the genetic algorithm on the coordinates of the child restraint relative to the functional for the minimum environmental impact;

Figs. 118a-h are graphs relating respectively to:

a) evolution of the values of the functional,

b) average total electric field on pixels,

c) maximum coverage functional trend,

d) trend of functional C / I,

e) cost trend,

f) distance traffic functional trend,

g) trend of the percentage GOS coverage ratio,

h) trend of the SRB-pixel serving distance,

as a function of the number of iterations of the genetic algorithm;

Fig. 119 is a representation of best server areas relating to the functional for minimum interference or for maximum C / I;

Fig. 120 is a representation of a DBM coverage relative to the functional for the maximum C / I;

Fig. 121 is a representation of the C / I ratio in dB relative to the functional for the maximum C / <'> I;

Figs. 122 is a representation of the electric field in V / M relative to the functional for the maximum C / I;

Fig. 123 is a representation of the pixel distance SRB serving relative to the functional of the maximum C / I;

Fig. 124 is a representation of the SRB location in the urban area for the functional for maximum C / I;

Fig. 125 is a graph showing the SRB movement caused by the genetic algorithm for the maximum functional C / I;

Fig. 126 is a graph showing the functional trend of the maximum C / I ratio as a function of the number of iterations of the genetic algorithm;

Fig. 127 is a graph showing the functional relative to the C / I ratio with limits set for the functional relative to the maximum C / I as a function of the number of iterations of the genetic algorithm Figs. 128a-128 and report respectively:

a) functional minimum environmental impact,

b) traffic for minimum distance SRB,

c) functional coverage with imposed limits,

d) functional power control distance,

e) functional cost with traffic limits,

f) functional minimum cost,

all related to the functional for the maximum ratio CI I, and all as a function of the number of iterations of the genetic algorithm; Fig. 129 is a representation of the distributions of best server areas for C / I functionalities with minimum limits set for each zone; Fig. 130 is a representation of the DBM coverage for functional C / I with minimum limits set for each zone;

Fig. 131 is a representation in 3D DBM for functional C / I with minimum limits set for each zone;

Fig. 132 is a representation of the C / I dB interference distribution for functional C / I with minimum limits set for each zone; Fig. 133 is a representation of electric field in V / M for functional C / I with minimum limits set for each zone;

Fig. 134 is a representation of the SRB pixel distances serving for functional C / Ί with minimum limits set for each zone;

Fig. 135 is a representation of the SRB locations in the urban area for functional C / I with minimum limits set for each zone; Fig. 136 is a graph showing the values of the C / I functional with minimum limits set for each zone as a function of the number of iterations of the genetic algorithm;

Fig. 137 is a graph showing the minimum interference functional relative to the C / I functional with minimum limits set for each zone, as a function of the number of iterations of the genetic algorithm;

Figs. 138a-138d report respectively for a functional C / l with zone limits:

a) functional cost with traffic limits,

b) functional minimum environmental impact,

c) functional coverage with zone limits,

d) functional minimum distance traffic.

as a function of the number of iterations of the genetic algorithm;

Fig. 139 shows the values of the power control distance functional in relation to the C / I functional with minimum limits set for each zone;

Fig. 140 is a graphical representation of the sum of the functionals with weight 10 of the cover;

Fig. 141 is a graphical representation of the sum of the functionals with weight 10 of the environmental margin;

Fig. 142 is a representation of the sum of the functionals with weight 1.5 of the environmental margin functional;

Fig. 143 is a representation of the sum of the functionals with unit weight of the single functionals;

Fig. 144 is a representation of the sum of the functional with weight 4 of the functional environmental margin;

Fig. 145 is a representation of a sum of functionals for best server areas for a test by coverage functional weight placed at 10;

Fig. 147 is a representation of a sum of functional electric field in V / M for a test for functional weight coverage set to 10;

Fig. 148 is a representation of a sum of functional electric field in DBUV / M for a test for functional weight coverage set to 10;

Fig. 149 is a representation of a sum of functional C / l in dB for a test for functional weight coverage set to 10;

Fig. 150 is a representation of a sum of functional SRB pixel distance serving in glasses for a test for functional coverage weight set at 10;

Fig. 151 is a representation of a sum of functional RBSs in an urban area for a test for roof functional weight set at 10;

Fig. 152 is a representation of a sum of functionals for a test of the environmental margin functional weight set at 10;

Fig. 153 is a representation of a sum of dBm coverage functionals for a test for environmental margin functional weight set at 10;

Fig. 154 is a representation of a functional sum C / I dB for a test for the environmental margin functional weight set at 10; Fig. 155 is a representation of a sum of functional electric field V / M for a test for functional weight of the environmental margin set at 10;

Fig. 156 is a representation of a sum of functionals relating to the electric field in DBUV / M for a test for functional weight of the environmental margin set at 10;

Fig. 157 is a representation of the sum of functional SRB pixel distance serving in meters for a test for functional weight of the environmental margin set at 10;

Fig. 158 is a representation of the sum of functional RBSs in an urban area for a test for the environmental margin functional weight set at 10;

Fig. 159 shows in a graph the final part of the SRB movement produced by the genetic algorithm by means of a sum of functional SRB movements;

Fig. 160 is a representation of a sum of functional best server areas for a test with the weight of the functional environmental margin set to 1.5;

Fig. 161 is the representation of a sum of functional coverage dBm for a test with the weight of the functional environmental margin set at 1.5;

Fig. 162 is a representation of a sum of functional C / I dB for a test with the weight of the functional environmental margin set to 1.5;

Fig. 163 is a sum of functionals relating to the electric field in V / M for a test with the weight of the functional environmental margin set to 1.5;

Fig. 164 is a representation of sum of functionals for electric field in DBUV / M for a test with weight of the functional environmental margin set to 1.5;

Fig. 165 is a representation of the sum of functionals for pixel distance SRB serving in M for a test with weight of the functional environmental margin set to 1.5;

Fig. 166 is a representation of a SRB allocation in an urban area for a test with the weight of the functional environmental margin set to 1.5;

Fig. 167 shows a graph of the movement of the coordinates of the RBS is used by the sum algorithm for a test with weight of the functional environmental margin set at 1.5;

Fig. 168 shows a graph relating to the values of the functional weight 1.5 environmental impact;

Figs. 169a-169e report for a test with the weight of the functional environmental margin set at 1.5, respectively

a) functional coverage,

b) functional environmental impact,

c) traffic functional for minimum distance,

d) maximum efficiency functional,

e) functional C / I,

as a function of the number of iterations of the genetic algorithm; Fig. 170 is a representation of a sum of functional areas best server for a test with weight of the functional units at unit value; Fig. 171 is a representation of a sum of coverage functionals in dBm for a test with unit weight of functionals;

Fig. 172 is a representation of a sum of electric field functionals in V / M for a test with weight of functionals at unit value;

Fig. 173 is a representation of a sum of electric field functionals of DBUV / 'M for a test with weight of unitary functionals;

Fig. 174 is a representation of a sum of functional SRB pixel distances serving in m, for a test with weight of functionals at unit value;

Fig. 175 is a representation of sum of functionals SRB location in urban area for a test with weight of functionals at unit value;

Fig. 176 is a graph showing the movement of the SRB relative to the sum of functionals induced by the search for the best solution during the evolution of the genetic algorithm for a test with weight of functionals at unit value;

Fig. 177 is a graph showing the total evolution of the sum of the functionals relative to the previous Figure;

Fig. 178 is a representation of a sum of functionals for best server areas, for a test with the weight of the functional environmental margin set to 4;

Fig. 179 is a representation of a sum of functional dBm coverage for a test with the weight of the functional environmental margin set to 4;

Fig. 180 is a representation of a sum of functional electric field V / M for a test with the weight of the functional environmental margin set to 4;

Fig. 181 is a representation of a sum of functional electric field in DBUV / M for a test with weight of the functional environmental margin set to 4;

Fig. 182 is a representation of a sum of functional C / I dB for a test with functional environmental margin set to 4; Fig. 183 is a representation of the sum of functional distance in m pixels SLB serving for a test with weight of the functional environmental margin set to 4;

Fig. 184 is a representation of the sum of the SRB location in an urban area for a test with the weight of the functional environmental margin set to 4;

Fig. 185 is a three-dimensional representation of the levels in dBm for a test with the weight of the functional environmental margin set to 4;

Fig. 186 is a graph of the final evolution of the best position of the RBS;

Fig. 187 is a graph relating to the evolution of the sum of the functionals for a test with the weight of the functional environmental margin set to 4 relative to the number of iterations of the genetic algorithm;

Figs. 188a 188e report for a test with the weight of the functional environmental margin set to 4 respectively

a) functional environmental impact,

b) functional coverage,

c) functional minimum traffic distance,

d) maximum efficiency functional,

e) functional C / I

as a function of the number of iterations of the genetic algorithm.

3.2 POTENTIAL MARKET ANALYSIS

Before investing money in setting up a mobile radio network, market analyzes are initially conducted by developing hypotheses on the law of demand for Offered Traffic (traffic that the customer wants to do).

A series of researches and surveys aim to understand where the Core Business is, employing a cellular network and therefore how to best target investments in the network and services.

Two are in particular the lines that are observed:

Market Survey, i.e. segmentation of the type of potential traffic, such as Business or Consumer, this first step also seeks to determine the potential level of interest in the mobile phone service in a given area or a socio-economic analysis is carried out:

age;

sex;

occupation;

industry type;

cellular use frequency;

type of cell phone use

type of mobility;

Demographic Analysis: ie distribution of customers per Km2: consumers with potential interest;

market share;

potential prepaid conversations sold;

potential subscribers;

bench marking with other companies;

Billing information

Analyzes on any competing managers already active are also essential to understand which part of the scalable market is.

In the case of an already active network, opinion polls to determine the COS (Customer Opinion Score) are of fundamental importance to know one's own and competitors' weaknesses.

In the following paragraphs, the modalities for the traffic dimensioning of the RBS are presented, starting from the estimated average traffic per customer.

3.3 REQUIREMENTS OF A CELLULAR SYSTEM

The basic requirements of a modern public mobile radio system can be summarized in the following points; large user capacity that must be reached through an organization of the network aimed at optimizing the available radio resources;

coverage of the territory;

quality of service similar to that of the fixed network;

complete automation of both the call to and from the mobile user and the location of the user;

accessibility by the mobile user to most of the services offered by public networks.

To achieve these objectives, the following problems must be faced and overcome in the design of a mobile radio network:

limited availability of the resource (radio channels) that is of the band; characteristics of the transmission medium;

mobility of users on the territory;

quantity of users and distribution on the territory (capacity per traffic);

limited availability of child restraint permits;

limited financial availability;

management costs;

management flexibility (outsourcing);

technology and suppliers.

Some of these design topics are considered by the GSM specification through the cluster and multiple access concepts discussed below.

3.4 CAPACITY OF A MOBILE RADIO SYSTEM In general, the capacity of a communication system means the maximum number of users that can be served with a predetermined degree of service.

By degree of service we mean a comparison with the fixed network, in particular for example a GOS <= 2%, it implies that a maximum of 2 calls out of 100 find the busy during the peak hour or alternatively that for the 98% of the time during the rush hour the user will find availability of channels for accessing the network.

For rush hour, in accordance with the CCITT, a period of 60 minutes during the day is defined when the intensity of traffic averaged over several days of the week is the greatest. In practice, the rush hour is set at 4 consecutive quarters of an hour corresponding to the maximum traffic offered, averaged over a period normally of several weeks in a month.

The traffic intensity measures (expressed in Erlang) indicate the average number of voice channels of the individual cells used by users during a reference period, or the average number of customers who speak at rush hour.

The capacity of a system must be scalable in order to respond to both immediate and longer-term needs, so that in the face of an increase in users, the quality of the service can be maintained unchanged, while at the same time keeping reasonably content. the value of the investments and the amortization time of the plants.

The main elements that determine the size and capacity of a public mobile radio system can be summarized as follows:

economic constraints: how much the network must make economically;

financial constraints or how much I can spend in a year; number of radio channels available, the restriction may be due to both non-infinite cash flows and the supplier's production capacity;

both geographical and temporal distribution of users (statistical trend of the traffic offered to the system), the latter particularly important for managing peak periods, daily and seasonal;

technical characteristics of the system used (resistance to interference and from adjacent channel, efficiency of the signaling system, efficiency of the Hand-Over procedure), these aspects influence the bandwidth of the channel and therefore the overall capacity of the system.

Since in mobile radio systems the main technical limitation is given by both the economic and financial availability of the radio resource, the fundamental task of the system designer and the network operator is to make the best use of all the strategies made available by technological development in order to get as close as possible to the ideal goal of serving, with a certain quality of service and an area in any case large, a large number of users having a limited economic resources and frequency band available.

Technologically speaking, to achieve this goal, the use of the physical resource (radio channels) is optimized thanks to the multiple access techniques and cellular coverage strategies discussed in chapter 1.

3.5 COSTS AND PROJECT METHODS

Some examples will clarify the difficulties in correctly balancing the investment costs for traffic with those for the expected quality of service.

For example, suppose you want to serve 3500 users, each of which is characterized by a specific traffic offered at = 28 mErl. If the average duration of occupation of the server (duration of occupation of the radio channel) is equal to 110 seconds, during the rush hour there will be an average of 3200 resource occupation service requests (3200 = α * 3500 * 3600/1 10)

In the hypothesis of a system at full accessibility, without waiting with purely random entry traffic, to evaluate the number of servers (radio channels) necessary to satisfy these service requests it is sufficient to apply, having defined the acceptable degree of loss B%, the Erlang's formula B.

If, for example, B is fixed equal to 2%, then the number of servers N necessary to perform the service is:

where it is indicated with

B = 2% the degree of loss

You were the total traffic offered at rush hour calculated from the 6-month traffic growth rate

f the Erlang B function.

It is important to underline that, provided that the statistical validity hypotheses for Erlang's law B are satisfied, the procedure used above to evaluate the number of servers and the degree of service is independent of the multiple access technique implemented in the system, in general then for the traffic dimensioning does not require the average duration of occupation of the server as the average millierlang per customer is normally available.

Another aspect that has been overlooked is the fact that on a mobile radio network the users who generate traffic are not all concentrated on one point

The 98 erlangs can in fact be disposed of with Be = 2% with 110 channels of a single cell if concentrated on one point or, for. example, with 3 cells of 43 channels each if divided into three equivalent areas disjointed by 33 erlangs each, for a total of 43 * 3 = 129 channels of the 3 cells against the 110 channels of the first case.

The degree of service to be given also weighs heavily on investments, always referring to the 98 erlangs of the example if instead of Be = 2% you choose Be = 0, l% the number of channels rises to 130 if concentrated in one point or three cells with 51 channels for a total of 153 channels in the second case.

This simple example denotes the technical and economic difficulties to serve a traffic distributed throughout the territory mostly in an uneven way, with consequent levitation of the investment prices in radio channels and necessary SRBs.

With regard to the size of the cells, it should be noted that if N is a number of fixed channels assigned to any cell of the Cluster, the traffic capacity per cell is constant.

Since the cell area is proportional to the square of the radius, the capacity per unit area (Erl / Km2) that can be disposed of by the system is inversely proportional to the cell area; for example, halving the radius of the cell increases the capacity per unit area by a factor of 4.

The last aspect mentioned is particularly incisive on the costs for the construction of the network, the question that can be asked is which architecture to use: choose among many small capacity BTS with consequent great flexibility in the management of localized traffic peaks, or opt for a minimum number of child restraints necessary for coverage only?

Most of the Italian mobile TLC operators, for the implementation of the GSM network, followed an alternative methodology, arriving only today with the mature network at greater flexibility through the micro and pico cellular solution.

In particular, it was decided to start by giving priority to the coverage of the major Italian cities provided through a limited number of SRBs, subsequently the construction of new satellite SRBs was calibrated following the traffic that gradually increased with the spread of the GSM mobile telephone service.

This approach has made it possible to optimize the economic investments which, above all in the Start Up phase, are expensive to sustain, therefore even if technically questionable for the quality of the service offered, it was more convenient to follow the traffic rather than implement all the SRBs from the beginning. from an estimated amount of millierlang with the assumptions made.

This method has proved convenient for a twofold objective:

network investments were distributed over the years, chasing increasing traffic and, at the same time, there was no risk of oversizing a network due to errors in the estimate of theoretical traffic per customer, which also varies according to the manager's commercial image and of the competitive market.

The following paragraph explains the traffic tracking process in detail.

3.6 SRB EXPANSION AND CELL SPLITTING

As a rule, for GSM networks that are already functioning there are studies that allow to estimate the traffic growth at six months or a year, moreover the extensions for a good quality of service become imperative, when the loss Be% on the erlangmetric measurements of several consecutive weeks exceeds a certain GOS, also depending on the areas considered to have a high or low quantity of users, or with a strong or lower concentration of valuable traffic.

There are usually different types of traffic: seasonal such as summer traffic along the rivers; occasional such as events or shows such as concerts;

urban that offered in cities;

suburban that offered outside the towns; viable for example on roads with heavy traffic; typical industrial in areas with a strong industrial concentration.

Within each category, traffic sources can be further diversified in order to arrive at a map for precise knowledge of the territory, with the aim of better targeting network investments.

The analysis of the profitability of investments on a traffic basis is important when you have to decide to expand the network with new channels or new SRBs.

For this reason, the geographic location of traffic within the existing network must be determined as objectively as possible.

The traffic intensity measures of the new antennas, in fact, are attributable to the sum of the following three contributions that trigger different consequences on the Company's revenues:

fresh traffic: share of new traffic that is served precisely because the child restraint system has been installed, which extends the coverage relating to the outdoor or indoor service;

recovered traffic: share of traffic that could not be satisfied prior to the construction of the child restraint due to the congestion problems of the plants;

“displaced” or “cannibalized” traffic (borrowing the marketing terminology): share of traffic “attracted” by the new plant, but subtracted from the other adjacent cells that would have been able to dispose of this traffic anyway.

The "cannibalized" traffic component:

significantly complicates the profitability analysis of the plants, as only fresh and recovered traffic allow revenue to be increased (albeit in significantly different proportions, since recovered traffic is typically concentrated in peak hours and not in throughout the day), while "cannibalized" traffic does not involve revenues as it would have been disposed of anyway.

It cannot be neglected in the metropolitan area where the outdoor coverage is practically complete (100%) and the new RBSs are designed to cope with the growth in traffic demand (traffic offered) and the reduction of interference (which is achieved by decreasing the number average number of radio channels installed in the individual RBSs).

From the above considerations it can be observed that the determination of traffic growth requires the analysis of the behavior of a set of RBSs. In particular, it is necessary to consider a reference area with respect to which the following parameters can be defined for an adequate period of time: 'traffic offered - traffic demand expressed by customers who frequent the reference area;

traffic disposed of - traffic actually managed by the network; investment - cost of the systems (or cost fractions) that make it possible to serve the reference area.

3.7 SETTING THE CLUSTER

As mentioned in chapter 1 on the clusters to be used, it has been observed that to satisfy the minimum [C / I] = 9 dB ratio adopted for GSM, the minimum cluster size must be equal to M = 3 * 3 = 9, 3 sites with 3 cells per site or the number of hexagons each covered by a cell, which constitute the Cluster tents 9.

In the case of the project we are dealing with, we consider 3-cell sites with 4 carriers per cell corresponding to 12 carriers per site.

Given the frequency planning band available at 900 Mhz of 41 carriers, we have a size given by 10.2 cells for the elementary cluster:

In this case the cluster is close to the model 3/9 = 3 sites / 9 cells which has D / R = 5.2.

This cluster model minimizes the number of groups of channels within the cluster, maximizing the number of channels per cell (however, interference increases), thus allowing greater traffic capacity, that is, greater plant profitability. In reality, a more accurate calculation distinguishes the carriers of the GSM layer at 900 Mhz from the GSM ones at 1800 Mhz as the frequency planning is different. ..

On average there are for each cell only 3 carriers 900 Mhz and one carrier 1800, consequently the margin available for frequency reuse is higher and for the same traffic capacity, the frequencies in larger clusters can be better reused, let's calculate the C / I corresponding to networks 900 and 1800 separately:

The calculations show how the use of a 1800 Mhz Layer superimposed on the 900 Mhz one clearly improves the theoretical C / I of each GSM network while retaining the original traffic capacity but with a higher quality of service.

3.8 CLUSTER SPECTRAL EFFICIENCY

Various ways are used to evaluate the efficiency of a cellular system:

Channel or allocation efficiency: Carriers / Band * surface)

Spectral efficiency Erlang / Mhz

The former is defined as the number of channels or carriers for a given spectrum for a given surface, the latter as the maximum number of calls per unit of spectrum.

This latter is a valid figure of merit for most cell systems but has limitations as it does not relate geographic density.

For example, if the service is given on two cities with different populations and the same Erlang values per spectrum unit (Erl / Mhz) are obtained, the spectral efficiency will be higher for the smaller city than for the larger city which is more scattered. geographically, the English term Spread out effectively describes this concept, in other words a high traffic can be better disposed of if it is concentrated.

To also correlate the geographical area, the composite unit Erlang / Mhz / Km2 is used. this _ indicator is related to the allocation efficiency as the greater the number of carriers per cluster, the higher the traffic that can be disposed of from that cluster and the worse the C / I of that cluster.

An equation to express this efficiency is the following:

Relating for each cluster examined the total traffic given by the sum of the disposable Erlangs with Be = 2% of each cell of the sites with the cluster area corresponding to the total of the SRBs or cells present, we obtain:

CLUSTER (1) 900 MHZ 4 * 3 (4 sites and 3 cells) with 3 carriers 900 Mhz per cell

41 carriers;

4.5 sites on average per cluster;

3 Cells per site with 3 carriers per cell in tot. 9 carriers for 8 tch site;

71 equivalent channels per site;

61 You were. with Be = 2% disposable per site:

274 You were. disposable by the Cluster with Be = 2%;

C / I 20 dB

CLUSTER (2) 1800 MHZ 16 * 3 (16 * 3 cells) with 1 carrier per cell

48 carriers;

16 sites on average per cluster;

3 Cells per site with 1 carrier per cell;

24 equivalent channels per site;

16.6 You were. with Be = 2% disposable per site;

266 You were disposable by the Cluster with Be = 2%;

Cfl 30 dB

CLUSTER (3) 900 MHZ 3 * 3 (3 sites * 3 cells) with 4 carriers per cell

41 carriers;

3.42 sites on average per cluster;

3 Cells per site with 4 carriers per cell, 32 tch per cell;

96 equivalent channels per site;

84.1 You were. with Be-2% disposable per site;

285 Were disposable by the Cluster with Be = 2%;

C / I 18 dB

Figs. 24 and 25 below graphically relate the efficiencies of the 3 types of clusters as a function of the radius of the elementary cell of the cluster.

Note that the smaller the cluster and the cells, the more their efficiency increases.

3.9 STUDY OF A TRAFFIC MODEL

The aim of this paragraph is to estimate the rr of sites necessary to serve a given m of customers in a given geographical area, and therefore to derive a traffic model to be adopted for the traffic sizing of a GSM network for a certain geographical area. of characteristics similar to Ancona.

The project) of the network is carried out by knowing the surface of the territory and the number of potential customers we need to serve. Therefore, it is necessary to assume the average traffic for each customer and the degree of penetration of the mobile radio service in the population, a given, necessary is the degree of service that you want to give the aitale heavily conditions the non-nero of channels and RBS.

Orographic data of the Municipality of Ancona:

Average customers per square km:

Average traffic data per customer:

Average customer traffic hypothesis: 14 mErlang / customer, assuming 50,000 customers, the total traffic of the municipality of Ancona would be:

3.9.1 Design hypothesis

Assuming, as seen, that the choice of the optimal site in terms of costs, traffic, interference for the coverage of the territory in question, consists of 3 sectors with 4 carriers for each sector, the traffic, according to the Erlang model, can be disposed of for cell with Be = 0% is approximately 15.9 Eri. (47.7 Eri. For Radio Base Station).

N ° customers for each sector:

Example 26 is a configuration of a 3 * 3 CLUSTER, while on the map of the municipal area in figure 27 a purely theoretical division has been reported containing the SRBs calculated with the sectors oriented as in the ideal cluster.

3.10 REFERENCE ELEMENTS ADOPTED

In order to verify whether the theoretical project made had a real response, a comparison was made with a currently functioning network.

The reference model is the TIM network of Ancona, the confidential data of the network configuration are not reported, however it should be noted that the real correspondence with the theoretical model is almost completely, moreover the following considerations can be adopted: given the good coverage , the good quality indicators, the good Customer Opinion Score data and the fact that on the SRB 900 Mhz of the municipality of Ancona there are no losses Be> 2%, the GSM site project model that derives from it can be considered significant even if not generalizable due to intrinsic errors deriving from approximations of an ideal case reported to a real situation.

The errors that can be made in generalizing the ^ model obtained from ΉΜ Ancona can be traced back to:

uneven distribution of telephone traffic in the municipal area;

hilly orography of the municipal area of Ancona which does not allow us to consider an ideal cluster placed on a plane with hexagonal cells all with the same roof;

constrained location of sites in positions not corresponding to an ideal cluster;

cell leaders on coverage objectives not ideally arranged at 0 °, 120 °, 240 °;

Not all 3-cell SRBs as in the ideal case;

presence of areas of the municipality still to be covered with the GSM service;

powers and tilts of the cells different from the ideal case of a uniform plane cluster;

3.11 GRAPH OF THE MODEL AND CONCLUSIONS The good measured indicators of service quality, Accessibility of Connection Failures, MOS (Mean Opinion Score), allow us to consider the GSM TIM network of Ancona acceptable to be taken as a reference for a traffic model, this model of traffic can therefore be used to calculate the number of sites required for a GSM network project of similar size. The difference between the 15 RBSs of the ideal project and the 18 RBSs that are active on the network, is justified by taking into account that 4 active RBSs are 2-cell and 3 RBs are 1-cell, the total deviation of 10 900 Mhz cells corresponds to approximately 3 sites less, therefore we return to the 15 sites of the theoretical project based on the theory of the ideal cluster.

The need for these non-standard configuration sites arises from coverage needs and other constraints that do not justify all 3-cell sites, in any case this circumstance is only apparently an oversizing of the network, this is in fact designed to respond to traffic demands, often localized (for example a stadium) with daily or seasonal peaks that imply greater equipment in terms of available radio channels.

It also takes into account a margin for annual traffic growth which was not taken into account in the project example.

The traffic model obtained is summarized in the graphs shown in figures 28 and 29, they correlate:

The first is the cells with 4 carriers necessary to dispose of a certain amount of traffic respectively with Be = 0,% and Be = 2%.

The second is the cells needed to serve a given quantity of customers starting from the previous curves and the traffic in millierlang per customer. This is assumed in one case of 10.71 mErl, and in the other case of 14 mErl.

From the graphs it immediately emerges that to pass from a design with GOS from 2% to 0% the number of cells needed increases, and with it the cost of the network.

Therefore, the manager assesses on a case-by-case basis when it is necessary to privilege the quality of the service or vice versa when it is convenient to reduce on an investment in child restraint by accepting a lower quality. CHAPTER 4: GENETIC ALGORITHMS

This chapter describes the genetic algorithms in some detail. They will be used to obtain a planning of the GSM network with a lower environmental impact and a higher quality of the service offered.

4.1 THE PROCESSES AT THE BASE OF NATURAL EVOLUTION

There are two fundamental processes that make the evolution of a given species possible: natural selection and sexual reproduction.

Natural selection acts in a very simple way: it eliminates all organisms that do not pass some "suitability test" and allows the best performing elements of the species to survive. Then natural selection determines which elements of a population will succumb and which will survive to carry out the second process: reproduction. The latter guarantees the mixing and recombination of the parents' genes to form the genes of the descendants: when the sperm and egg unite, the corresponding chromosomes align, cross and then separate, exchanging genetic material. This shuffling allows for a much faster evolution than if all the descendants simply contained a copy of genes from a single parent modified from time to time by mutation because, thanks to the mixing of the genes, there is a good chance that the descendants will summarize. the best characteristics of both parents.

4.2 DESCRIPTION OF THE GENETIC ALGORITHM The genetic algorithms try to model the natural mechanisms of selection and mating in order to optimize thanks to them. Genetic algorithms manipulate strings of bits (chromosomes) by treating the information they contain as if they were genetic material.

Let's see in more detail the fundamental steps of the genetic algorithm. Given an optimization problem, it will be modeled by a 'F' function of 'n' variables which we will call the cost function (or objective function):

it expresses, within a generic system, the performance that you want to optimize according to the variables of the system itself.

For example, as regards the determination of the minimum distance trajectory to be made by a robot to travel from one point to another, the functional given by the length of the trajectory as a function of the starting point, the end point must be minimized. , any obstacles interposed, etc.

The first consideration to make is that the genetic algorithm, unlike other optimization methods, works exclusively with an encoding of the parameters and not with the starting parameters.

All the parameters are encoded and therefore each of their values is represented, if the encoding is binary, by a string of bits (0 1 0.. 0). The parameter can originally be of a discrete or continuous nature; if it is discrete the coding does not pose any problem, if it is continuous it will be necessary to restrict the range of its values by performing a quantization. For example, if you want to perform the binary coding of a continuous parameter 'p' which takes on the following values:

while its quantized form will be:

where the summation is intended for 'm' ranging from '1' to 'NBit' 'q' is the quantized version of 'p'

'B' is the binary sequence obtained from the coding of the 'p' parameter.

Having said this, we will define the binary coding of a single parameter as gene; the set of 'n' genes, which represent the 'n' parameters of the system under examination, will constitute a chromosome. For example, considering 5 parameters of 4 bits each, the binary representation of any gene and a chromosome will be:

Each chromosome, depending on the configuration it assumes, that is, the value that the parameters within it assume, will have associated a value of the functional obtained simply by evaluating the function itself in correspondence with that set of parameters.

Therefore, a performance will be attributed to each chromosome, that is a measure of its strength (ie to each chromosome will correspond through the functional a certain value of the quantity to be optimized).

The parallel with natural evolution is simple: each individual has genetic characteristics that make him more or less suitable for achieving a particular goal and based on it he will undergo natural selection.

For example, if the goal is the ability to capture prey, the predator must have inherited the right balance of physical parameters such as strength and agility that allow it to maximize this ability.

At the same time, taking into consideration the previous example, that is the minimization of the path that a robot must follow, a configuration of variables will be better

of another if it leads to an easier and shorter distance path.

After establishing a rule for coding and decoding the parameters, the algorithm generates a random population of lN chromosomes, these 'N' chromosomes correspond to as many sets of random variables, and submits them to the three fundamental operations of the process:

1) selection

2) coupling

3) mutation

SELECTION

The functional values are calculated for all the 'n' chromosomes, based on these values (which are nothing more than the performance values to be optimized), the chromosomes are sorted from best to worst. Assuming we have an initial population of 'N' = 8 chromosomes, with their associated thus:

Table 4.1 Possible population of eight chromosomes with their associated cost values.

and assuming you want to maximize the cost function (i.e. to search for the chromosome that achieves the highest possible value of the functional), the ordering will be done as follows:

At this point it will be necessary to choose the best chromosomes and discard the worst ones.

This operation can be performed in various ways and for each there will be advantages and disadvantages regarding the achievement of the final result:

a) the simplest way is to eliminate a certain percentage of the worst chromosomes, typically 50%. In the case of the previous example, chromosomes n ° 2, n ° 7, n ° 1 and n ° 4 would be discarded.

b) another way could be to discard the chromosomes that do not satisfy certain performances. In our example, if the survival condition were:

cost> 50

chromosomes n ° 7, n1 and n ° 4 would be discarded.

c) a third, more complex modality could be that which uses a conditioned roulette that introduces a certain acidity in the determination of the survivors.

In practice, the sum of the costs of all the chromosomes of the current generation is considered, then for each chromosome the relationship between its cost and this sum is made; the value resulting from this ratio will be considered as the probability that the individual in question will be able to survive.

A hypothetical roulette wheel with angular sectors proportional to the percentage weights of each chromosome is then considered and it is thrown N times obtaining N copies of the chromosomes.

Consequently, in addition to the performance of the configuration, the uncertainty of the launch comes into play which can lead to the choice of worse configurations (i.e. configurations with a cost value less interesting than that of others, which with this type of selection could be discarded) but * this it is not a completely negative fact given that it can continue the research in regions that would otherwise be abandoned but which in any case can lead us to the optimal solution, perhaps in a shorter period of time.

In the example considered so far, the weights to be given to roulette will be:

Table 4.2 Ordering and conditional selection.

Individuals with heavier weight will have a greater chance of being fished and therefore will obtain a greater number of copies and future offspring.

The subsequent mating phase will depend on how the selection was made and in particular on how many chromosomes have been discarded and therefore how many chromosomes will have to be generated to replace them. For the sake of simplicity, we will assume that the best N / 2 chromosomes survived.

2) PAIRING

This phase establishes that the selected chromosomes become parents and mate to give rise to the offspring ie the new chromosomes (ie the new sets of variables).

Coupling can also take place in various ways; the most used are:

a) in succession from the top to the bottom of the list (the first with the second, the third with the fourth, etc.).

b) the first with the last, the second with the penultimate, etc. casually.

Whatever the way in which mating takes place, two children will be born from each couple who will inherit a genetic heritage obtained from the mixture of the parents' genes.

The characteristic operation of this phase is the crossover (crossing): within the two parent strings a position is randomly selected that identifies the crossover bit; the first child will be obtained by joining the half-string of the first parent located to the left of the crossover point and the half-string of the second parent located to the right of the crossover point; the second child will be obtained from the union of the complementary semi-strings.

We can observe everything with a simple example, assuming that the crossover bit is the ninth:

2

After the crossover operation we will again have a population of N elements (N / 2 parents plus N / 2 children).

3) MUTATION

The mutation operation is very simple and its effect on the search for the optimal solution is minimal: it alters a small percentage of bits (typically 10%) among all those present in the 'N x n' matrix of chromosomes. In practice, it randomly selects a certain number of bits and changes their value from '0' to '1 and vice versa. With this operation, the genetic algorithm responds to that need for randomness and unpredictability that nature also implements in the reproduction of living beings; the mutation ensures that a population that is too uniform is not created and also helps to bring research out of local minima. At this point the algorithm can move on to successive iterations proceeding each time in the following way: it restarts with the calculation of the so of all the elements of the new generation, sorts them based on it, selects them by eliminating the weakest ones, makes them couple by generate new offspring, mutate and so on.

Everything is schematized in the flow diagram of figure 30.

The iterations will be interrupted or when the algorithm will find a satisfactory solution, that is, once a threshold has been set, the solution will have exceeded that threshold, or after a predetermined number of iterations.

In addition, you can restart the algorithm several times and mix the solutions thus obtained through, once again, the genetic algorithm itself, so as to reduce the influence of the initial research point on the final solution.

4.3 MECHANISMS OF CONVERGENCE OF THE GENETIC ALGO-RHYTHM

We try to enter the algorithm mechanisms to understand if and how it satisfies the robustness hypothesis.

Genetic algorithms derive much of their power from the tendency they have to focus attention on the most promising parts of a solution space (the directionality in the search for genetic algorithms is due precisely to this characteristic which essentially derives from the way in which it is made selection and coupling).

As we have seen, each string of the population is first evaluated in order to determine the performance that that string encodes.

Second, the strings that have obtained the highest ratings are matched. Two strings align, a point along each of them is chosen (crossover point) and the parts to the left and right of those points are combined to generate two descendants.

Biological chromosomes cross in this way when two gametes meet and a zygote is formed; for this reason the crossover process, implemented in the genetic algorithms, closely mimics the biological model. Furthermore, descendants do not replace parent strings but strings that have obtained low ratings.

Then mutations occur that modify a small percentage of strings. Generally, mutations alone do not advance the search for a solution very far but ensure that a too uniform population is not formed, which would be incapable of further evolution. Therefore, the mutation operation allows the genetic algorithm not to concentrate in too limited regions of the solution space and to inspect alternative paths.

4.3.1 The implicit parallelism

The genetic algorithm exploits the "target" regions of the solution space, i.e. the regions where the best solutions are found since, in subsequent generations, through reproduction and crossover, the number of strings found in those regions progressively increases (in fact, when convergence has been reached, the chromosomes are almost identical, that is, they belong to the same region).

The algorithm favors the reproduction of the most suitable strings, therefore strings above the average (found in target regions) will have a greater number of descendants in the next generation.

In fact, the number of strings in a given region increases in proportion to the statistical evaluation of the fitness of that region (fitness is the adaptive value, which represents the quality of the solution).

A statistical program would have to evaluate dozens of samples, drawn from thousands or millions of regions, to obtain the average adaptive value of each of them. The genetic algorithm manages to obtain the same result with a far smaller number of strings.

The key to this rather surprising behavior is that a single string belongs to all regions where at least some of its bits appear. For example, the string:

and so on (where the asterisk indicates that the value of the corresponding bit is not specified and could take any of the two values).

Larger regions (i.e. those that contain many unspecified bits) will typically be represented by a high percentage of all strings in a population. Thus a genetic algorithm that manipulates a population of a few thousand strings actually has samples from a much larger number of regions (coding in a sense leads to an expansion of the solution space but at the same time leads to the identification of the target regions).

This implicit parallelism gives the genetic algorithm its fundamental advantage over other traditionally used optimization procedures.

The crossover complicates the effects of implicit parallelism. In the genetic algorithm, the purpose of crossing strings is to probe new parts of the target regions instead of testing the same string repeatedly in successive generations. Furthermore, the process can also transfer a descendant from one region to another and in this way the sampling rate of different regions can move away from strict proportionality with respect to the average adaptive value. This distancing slows down the pace of evolution.

4.3.2 Compact building blocks

The probability that the offspring of two strings leave the region to which the parents belong depends on the distance between the '1' and '0' that define that region.

For example, the descendants of a string in the io ****** region can only lie outside that region if the crossover starts at the second position of the string, which is a one in seven probability for a string. which contains eight bits (if the srtinga contained one thousand bits the probability would be only one case out of nine hundred ninety-nine).

Instead, the offspring of an eight-bit string belonging to region 1 ****** 1 leaves the parental region regardless of where the crossover occurs.

The adjacent '1s and or' 0s that define a region are called 'compact building blocks'. It is very likely that these blocks remain intact in the crossover processes and that they are therefore transmitted to future generations at a rate proportional to the average adaptive value of the strings that carry them.

A reproduction mechanism that includes crossover fails to sample all regions at a rate proportional to their adaptive value but is capable of sampling for those regions that are defined by compact blocks.

The number of compactly defined blocks in a population of strings greatly exceeds the number of strings, so the genetic algorithm still has an implicit parallelism.

Thanks to its implicit parallelism, the genetic algorithm can explore and exploit many regions of the search space while manipulating a relatively small number of strings.

4.3.3 Linearity

Again thanks to implicit parallelism, "genetic" algorithms can also tackle non-linear problems, in which the adaptive value of a string containing two particular building blocks can be much greater (or much less) than the sum of the adaptive values corresponding to each building block .

Linear problems have a small search space because the presence of a T or a '0' in a certain position of a string has no influence on the adaptive value attributable to the presence of a T or a '0' in some other position; in essence, thanks to linearity, each bit has its own weight within the functional regardless of the value that the other bits will assume.

When the problem is not linear the difficulty increases dramatically; the average adaptive value of the strings in the region ****** 01, for example, could be higher than the average, although the adaptive values associated with the regions ****** 0 * and ******* 1 they are below average, so as you can see the non-linearity does not mean that there are no useful building blocks but simply that blocks made up of single '1's or single' 0 'are not useful. This leads to an explosion of possibilities but fortunately it is still possible to exploit the implicit parallelism, in fact the building blocks that exploit non-linearity to obtain above-average performance will be automatically used more often in future generations.

4.3.4 Exploration and exploitation

In addition to addressing non-linearity, the genetic algorithm contributes to solving an enigma that has so far eluded conventional methods of searching for the optimal solution: how to achieve a good balance between exploration and exploitation. When you begin to get a good result, in solving a problem, you can focus on its exploitation but this choice has a hidden cost because exploitation makes the discovery of better paths in the solution space unlikely.

The improvements are also obtained by running the risk of decaying the performance obtained since looking for alternative routes. A typical problem for all systems that adapt and learn (hence the belonging of the algorithm to the class of artificial intelligence algorithms) is deciding to what extent to mortgage the present for the future.

The genetic algorithm addresses this obstacle using the crossover. Although crossover can interfere with the exploitation of a building block by breaking it down, this process explores the behavior of building blocks in new combinations and in new contexts.

The crossover then generates new samples of above-average regions, confirming or refuting the estimates produced on the basis of previous samples; when the crossover breaks down a constitutive block it produces a new block that allows the genetic algorithm to explore regions that it had not previously tested.

The crossover manifests this behavior, somewhat unbalanced towards exploration, especially in the first interactions when the sampled strings do not have a high adaptive value and therefore the probability of exiting a certain region to probe another with an adaptive value is high. statistically better average.

Going forward with the iterations, when the algorithm finds a region with a high adaptive value, the crossover loses its exploration capacity as all the strings of the current population tend to have constituted blocks belonging to the same region and from that moment the crossover also contributes to exploitation (even if only slightly). 4.4 CONSIDERATIONS FOR THE APPLICATION OF THE GENETIC ALGORITHM TO THE VARIOUS PROBLEMS Before applying the genetic algorithm to the minimization of the C.E.M. due to the GSM network, its behavior was analyzed with simpler problems in order to better understand its functioning and its effectiveness.

In particular, the algorithm was applied in the search for the interdistances of the dipoles in an array that would optimize its radiation pattern (minimization of the main lobe to secondary lobe ratio). This example has given indications on how the genetic algorithm should be implemented for the various practical problems, on how the fundamental operations must be done, which are the right cycles to be performed by the algorithm, etc.

4.4.1 Discretization of continuous variables

The discretization of the variables is a fairly crucial operation because it has implications both on the effectiveness and on the efficiency of the algorithm itself.

As for the goodness of the solution (the effectiveness of the algorithm) it can be noted that it heavily depends on this operation, in fact if a quantization has been made with few levels available (a few bits per variable), the space is practically limited. of solutions and with it the possibility of obtaining better solutions has been limited; it is as if inaccessible holes had been created in the solution space, no one can say whether there are poor or good solutions in correspondence with these holes.

Therefore, from the point of view of effectiveness, a quantization as fine as possible should be done (minimum loss <'> of information and minimum quantization noise).

instead, as far as efficiency is concerned, a quantization as rough as possible should be done, because the calculation times depend directly on the amount of data (bits) used.

We are therefore faced with two conflicting needs: on the one hand to obtain the minimum loss of information due to quantization, on the other hand not to burden the calculation times. A valid help in finding a good compromise between these needs can be obtained from a good knowledge of the problem that is being faced. In particular, a good compromise can be obtained by knowing in detail the dependence of the functional on the various parameters in order to make a fine quantization for the critical parameters and a milder one for the less influential parameters within the functional.

If this information is not available, an empirical criterion can be used: the algorithm is run with a minimum number of bits per parameter, then gradually increasing the bits per parameter until it is observed that the accuracy of the solution becomes insensitive to this increase, at this point it is useless to continue on this path since only the efficiency of the algorithm is penalized.

4.4.2 Choice of the number of chromosomes

The number of chromosomes in the population, as seen, represents the number of configurations of variables that are evaluated simultaneously and that will undergo the various steps of the algorithm.

The search for the solution becomes the search for particular bit strings that encode the sets of variables that give rise to the best solutions.

We have imagined a solution space composed of valleys and peaks: the peaks host the configurations that encode the less valid solutions, the valleys correspond instead to the best strings (if you are doing a minimization, otherwise the opposite is true for the valleys and tops).

Figure 31 shows a typical solution space: Recalling what has been said about implicit parallelism, the various regions of the solution space can be defined by considering the strings that have a '1 or' 0 'in particular positions.

It has also been seen that the landscapes become large and rugged as the number of parameters and the complexity of the problem increase. Therefore, when the landscape has very complex topological characteristics, finding the minimum or even determining the direction towards which it is reached becomes increasingly difficult.

One of the features that make the genetic algorithm robust and that allow it to overcome this drawback is its ability to simultaneously probe many regions of the solution space. This characteristic derives from the fact that from the beginning it has a certain number of solutions (chromosomes) that allow it to have a good view of the whole landscape and to advance the solutions in parallel.

From these considerations it is clear that the greater the number of initial chromosomes, the better the final solution will be; in practice, starting from several points, it is possible to probe several regions in greater detail.

Also in this case there is the downside because increasing the number of chromosomes increases the calculation time; furthermore, since the computer must proceed in parallel with all the data, the computation time could increase exponentially if the RAM memory is exhausted and the hard disk is used.

To make a good choice, once again it is useful to know the physics of the problem in question, in particular if the space of solutions is vast and very bumpy it will be convenient to increase the number of chromosomes of the population as much as possible, if instead the space of solutions it is quite regular and not very extensive it is possible to have a good convergence with a few initial configurations. If this information is not available, the empirical criterion already seen can be adopted again: we start from a few chromosomes that will be increased until there are appreciable improvements in the quality of the solution.

On the other hand, the best thing to do to have a good convergence is to use the greatest number of chromosomes compatibly with the computing resources available.

It should be noted that in addition to the risk that the computer uses the hard disk, increasing the number of chromosomes not only slows down the speed of each iteration (the calculation time grows in proportion to the number of chromosomes) but forces to increase the number of chromosomes itself. iterations to arrive at convergence. To better clarify this last statement, it is worth examining the transformations that the chromosomes undergo in the various iterations.

With the first instructions, the chromosome populations are randomly created and used in all subsequent phases of the program. Due to the random instructions, total randomness reigns in the initial population: the chromosomes statistically cover the entire solution space.

As the iterations continue, the search is concentrated in the regions where the best solutions are found and the corresponding chromosomes tend to highlight particular building blocks (i.e. groups of adjacent bits with specific values are formed in certain positions).

Continuing further with the iterations, the reasons explored will become smaller and smaller and the building blocks in the chromosomes more and more complex (the building blocks will be formed by a greater number of adjacent bits always the same). At some point the first chromosomes, even just the first two, will tend to be the same; from this moment on at each subsequent iteration there will be an increasing number of chromosomes all the same (apart from the single bits modified from time to time by mutation, which in any case represent a very small percentage of bits) therefore the population will tend to be constituted from chromosomes all the same.

When the chromosomes all become the same, the algorithm no longer evolves, practically convergence has been reached. .

It is evident at this point that by increasing the number of chromosomes of the initial population, the algorithm will have to proceed for several iterations to produce identical chromosomes.

4.4.3 Characteristics of the various types of selection

Three possible ways to make the selection have been examined, which we report below for convenience: the worst configurations are discarded (for example 50%),

solutions that have a functional value below a predetermined threshold are discarded,

the selection is made by rolling several times a conditioned roulette wheel having the angular sectors proportional to the value of the cost of the configurations.

It has been seen that the second criterion performs the selection by discarding all the chromosomes that do not comply with the specifications on the functional value (typically a decision threshold is set that allows to establish which chromosomes will be discarded and which will be used in the subsequent phases of the program).

This type of selection leads to convergence very quickly but also increases the risk of falling to local minima.

Convergence occurs very quickly because at the first iterations, since the exclusion condition must be quite stringent (later we will see why this choice is made) many chromosomes will be discarded and therefore the algorithm will focus from the beginning on the few remaining chromosomes, future chromosomes will all be generated from them and consequently research will be conducted on the few regions they represent. This explains the rapid convergence towards the solution and the high probability of falling into a local minimum (or local maximum).

The type of selection just described brings with it a further problem, namely what should be the right value to be given to the threshold. This problem becomes pressing if we consider that with the succession of iterations all the configurations will tend to meet the specification, therefore fewer and fewer chromosomes will be discarded to be replaced by new descendants, thus affecting the <’> central operation of the crossover algorithm; this is why the condition <'> of exclusion must be very stringent, otherwise convergence is achieved very quickly with the aforementioned risks. Since the crossover is inclined to exploration, when this mechanism is compromised, the search falls to local minima.

The third selection criterion, based on weighted roulette, due to the randomness of the rolls means that good configurations can be discarded in favor of worse solutions.

This remixing of good configurations with less performing configurations on the one hand makes the algorithm tend to explore regions that would otherwise be set aside but on the other hand causes the lengthening of the convergence times and possibly also leads to a deterioration of the qualities of the final solution. .

A good compromise between exploration and exploitation of the target regions is reached with the first selection criterion, even if this criterion is less inclined to explore new regions than the roulette criterion.

It plans to discard a certain percentage of the worst chromosomes typically 50% by concentrating the research on the remaining 50%; in practice this selection criterion leaves no room for mediocre solutions, however in some cases the mediocre solutions could lead us to target regions. As seen, each criterion has advantages and disadvantages compared to the others, even in this case the choice can be facilitated by the knowledge of the space of the solutions. If you have a slightly bumpy solution space, most likely the best choice to make is to use the second selection method much faster than the other two but which in these cases should not compromise the quality of the solution.

If, on the other hand, you are in a very bumpy space, you will need to decide whether it is convenient to give space to exploration and then choose the third criterion or give a fair compromise to exploration and exploitation and therefore choose the first criterion.

4.4.4 Characteristics of the various types of coupling

The first pairing criterion foresees that the chromosomes are paired in succession from the top to the bottom of the list: the first with the second, the third with the fourth, etc. This coupling methodology certainly surpasses the other two criteria with regard to convergence times, as the best couple with the best and therefore the search for an above-average set of chromosomes occurs first.

The second criterion provides for the coupling of the first with the last, of the second with the penultimate, etc, finally the third criterion provides that the coupling occurs randomly; the last two criteria should allow, at least from a theoretical point of view, a greater exploration of the solution space (greater effectiveness of the algorithm) and therefore the achievement of a more refined solution.

As a counterpart, there is an increase in convergence times, so before choosing the type of coupling to be adopted, it is advisable to evaluate whether the increase in the quality of the solution is appreciable or not. In the event that the quality of the solution does not vary significantly, using the second or third criterion, it is advisable to adopt the first criterion thus benefiting efficiency.

A good compromise could arise from the coordination of the first criterion with the second, that is, in using the first criterion for half iterations and the second criterion for the remaining iterations.

This strategy can be used because when the first chromosomes begin to be the same, the first criterion tends to make all the chromosomes become the same and therefore makes genetic diversity disappear by eliminating any chromosomes that could become promising with the progress of the iterations. This happens because the crossover operation is no longer effective as operating the crossover between two identical chromosomes does nothing but generate two children identical to their parents, whatever the crossover point. Furthermore, all this happens very quickly because at each iteration the number of identical chromosomes doubles.

Proceeding instead with the second criterion (when you have the first replicas of the best chromosomes) the best chromosomes will couple with the last ones, which are less performing and certainly different, so that the algorithm can probe some good neighboring region without losing the best chromosomes and without losing too much efficiency, possibly benefiting the efficacy feature. 4.4.S Choice of the number of mutations

The choice of the number of mutations to be carried out on the chromosomes is quite critical as an excessive increase of the mutated bits, if on the one hand it can remove the danger of falling into excellent premises, on the other hand it certainly leads to an increase in calculation times and sometimes it can even irreparably compromise the final solution; this can happen in case the mutation affects the best chromosome bits.

In this regard, a good trick is to prevent the bits of the best chromosomes from being changed, so as to avoid the risk of wasting time in finding an already acquired result or even leaving a promising area.

From these considerations we can deduce that in the first iterations, when the most complete randomness reigns in the chromosome population, the mutation can cause jumps that are negative for convergence, therefore in this phase it is convenient to change a few bits.

In the final iterations, when by now the chromosomes tend to be all the same, changing a high number of bits can help the algorithm to get out of a very narrow region or to find the best value of that region. For this reason it is convenient to increase the number of bits changed in the final iterations.

4.4.6 Choice of the number of iterations

It has already been observed that the genetic algorithm can terminate either when it finds a satisfactory solution (ie a solution that satisfies certain previously imposed requirements) or after a fixed number of iterations.

The first criterion involves some risk in fact after a certain number of iterations, when all the chromosomes will tend to be identical, the only operation that can modify some chromosomes of the population is the mutation, which being a completely random operation and done on a few bit compared to the total will most likely not further evolve the solution.

Therefore, if there is no appreciable improvement of the solution, due to the mutation, such as to trigger the termination condition of the algorithm, the computer will cycle indefinitely. For this consideration an upper limit of iterations is however advisable.

To establish the number of iterations it is necessary to consider that it must increase as the number of chromosomes in the population, the number of parameters of each chromosome, the number of bits per chromosome, the accidentality of the solution space, the randomness of the initial configurations increase, etc. .

A simple way to establish the number of iterations, once the above parameters have been set, is to try to run the algorithm with a certain number of iterations and check the status of the chromosomes at the output: if they are very different it means that it is necessary to increase the number of iterations, if they are almost equal it means that a few more iterations are enough to reach convergence, finally if they are all the same it means that that number of iterations is overabundant.

4.4.7 Use of acquired functional knowledge

The information acquired on the physical characteristics of the particular problem, which is being optimized, are very useful for drafting the most suitable genetic algorithm for the case in question.

Through this knowledge, as seen above, it is possible to significantly increase the efficiency of the algorithm, less as regards its effectiveness, by appropriately setting its parameters (number of chromosomes, number of iterations, coupling modes, selection, etc).

Furthermore, this information can be used to place additional constraints on the final solution or on the variable sets that can be chosen.

These constraints do nothing but restrict the region of possible solutions and consequently increase the convergence speed of the algorithm.

In particular, the constraints on the variable sets that can be chosen lead to a considerable decrease in calculation times, this is because parameter configurations are discarded without calculating the relative value of the functional. In fact, it should be noted that calculating the functional for a configuration can be very laborious (long calculation times) and therefore avoiding calculating it for configurations that will surely be discarded or that will certainly obtain a performance below average, can lead to a substantial improvement of the efficiency of the algorithm. For example, you can think of functionals obtained from the numerical integration of quite strange functions, for which you can have calculation times in the order of minutes for the calculation of a single configuration of variables; considering then that these calculations must be done thousands of times to obtain the convergence of the algorithm, it is understood that avoiding the calculation of a hundred configurations, for example, can significantly reduce the calculation time. From these considerations it is evident that it is convenient to use all the information on the characteristics of the problem under examination to add constraints on the parameters of the functional and thus obtain a greater efficiency of the algorithm. The constraints on the sets of admissible variables, in addition to decreasing the calculation times, are sometimes necessary as they can constitute unacceptable solutions. Think of functionals that have within them points or regions of discontinuity, in such cases these regions must be prohibited regardless of the increase in efficiency that is obtained as a result

CHAPTER 5

GENETIC ALGORITHM APPLIED TO THE GSM SYSTEM

This chapter describes how to apply the genetic algorithm to the GSM mobile phone system

5.1 PROBLEM DATA

The functionals chosen, and described later in the next chapter, generally depend on the following parameters: number of SRBs,

number of radio channels for each child restraint,

antenna positions (x, y),

antenna powers,

antenna height,

gains of antennas,

attenuation of the transmission branching of the SRB, vertical tilt of the antennas,

characteristics of the ITI (building density, etc.) telephone traffic,

distance between child restraints and telephone traffic sources; C / Ί interference as the difference in power in the Best Server area between the serving cell and the interfering cells.

desired coverage levels,

maximum permitted interference levels,

pre-existing electromagnetic field levels,

desired environmental impact,

Pre-existing child restraints,

areas restricted to the location of child restraints,

regulation of the power emitted by the Power Control function,

system frequency from 900 Mhz to 2000 Mhz.

Plant yield.

Some of the aforementioned parameters become constants within the functional once particular choices have been made.

First of all, the region in which the service must be guaranteed is examined and its characteristics are inserted within the program which at this point become constants of the problem. Then you choose the number of antennas to use, the height of the pylons, the gains of the antennas, then all these parameters become constants within the functionals once their value has been set. The parameters that actually vary during optimization consist of:

SRB position (x.y),

powers issued by the SRBs,

vertical tilt of the SRB antennas

In general, our functional to minimize 'cost' will depend on the following variables:

Within the program data structure, a possible configuration of antennas will be described as follows:

where is it:

Xi, Yj, indicate the coordinates of the i-th site,

Ptj indicates the power of the transmitter of the i-th site, Tiltni indicates the inclination of the SRB antennas.

It has been said several times that the genetic algorithm works with a binary coding of the parameters, hence the previous one

description made with the non-coded parameters will correspond to a description made with the coded parameters, that is, with the chromosomes. The chromosome corresponding to the previous antenna configuration, described with the uncoded parameters, will have the following form:

Nby2, Nbp2, Nbt2,. . . .

In particular, the coding was carried out as shown below:

6 RBS used for functional verification,

the movement area is a region represented with a matrix of 512 rows and 512 columns,

to address all possible positions, 9 bits are required for the x coordinate and 9 bits for the y coordinate,

the powers have been coded with 3 bits for a total of 8 steps in total, the maximum of the transmitted power for each 8-channel TDM transmitter is 41 dBm,

the tilts have been coded with 3 bits for a total of 8 steps from 1 to 8 degrees max.

overall, the gene encoding an SRB is represented by 24 bits,

as a whole, the set of genes that make up a chromosome or a configuration of the 6 SRBs is represented by a 144-bit row vector.

5.2 CHOICE OF THE FORECAST PROCEDURE OF THE C.E.M.

During decomposition the various configurations of antennas obtained will have to be tested both as regards the level of coverage and as regards the value of the functionalities to be minimized.

For both evaluations it is necessary to know the signal level, therefore a prediction procedure of the electromagnetic field level must be implemented in the algorithm. The forecasting methods fall into three categories:

statistical type,

deterministic,

mixed type (statistical-deterministic).

All forecasting criteria require knowledge of environmental parameters, propagation curves, etc.

The forecasting methodologies are distinguished on the basis of the calculation times necessary to make a forecast and on the accuracy obtained.

The choice between a criterion and Labro is usually made according to the morphology of the territory, the operating conditions of the system under examination and on the basis of a compromise between precision and duration of the calculation.

In the statistical approaches starting from the statistical properties of the amplitude and relative delay of the various contributions (due to multiple paths) obtained with experimental measures, a characterization of the radio channel is determined in particular environmental situations (rural area, sub-urban area , urban area, etc).

Statistical forecasting programs do not require a lot of environmental data or perform many calculations, so they are fast enough even if the forecast obtained is not very accurate.

The deterministic methods derive the prediction of the electromagnetic field from the knowledge of the actual obstacles present on the territory. To this end, deterministic forecasting programs require accurate knowledge from orography to construction, with a resolution of about m, this also entails the need to manage significant amounts of data in addition to the difficulty of procuring them. By means of suitable calculation algorithms (such as those of Epstein / Peterson, of Deygout, etc.) the attenuations mainly of diffractive or ray trace type (used above all for microcells and small environments), caused by the presence of the obstacles themselves, are calculated. The major limits of deterministic methods derive from the difficulty of having the actual conformation of the territory available down to the smallest detail. Generally these methods give the best results but take a long time to make the prediction even if you consider a small area.

The statistical / deterministic prediction programs use the techniques described above in two distinct phases.

In the deterministic phase, the altimetric profiles are identified and the attenuation contributions due to the obstacles placed along the junction between the transmitter and the receiver are calculated (all using the orographic data given in input).

In the statistical phase of the program, the characteristic parameters of the Log-normal oscillation are calculated in order to calculate the additional margin necessary to reduce the dispersion existing between the calculated and measured values (all according to the morphology of the territory: building density, volume of buildings, density of vegetation, etc.).

The predictions provided by the mixed algorithms are quite good even if the calculation times of the deterministic part are still long.

Given that you want to optimize GSM coverage over an entire city, it is unthinkable to use deterministic or mixed algorithms essentially for two reasons: the enormous amount of data to be entered and stored, which require large memory capacities, and the explosion of calculation times, in fact it is necessary to think that the field predictions will have to be made thousands of times during the entire optimization procedure. For these reasons the choice fell on a statistical type algorithm, in fact the so 231 model of YValfish-Ikegami was chosen. Within this model it is possible to introduce some characteristics of the urban environment:

The attenuation suffered by the signal in passing from roofs to roads due to diffraction and therefore to the scatter of the signal is given by:

If you do not have the data relating to buildings and roads, it is recommended to use the following default values:

φ = 90 °

The model 231 Walfìsh-Ikegami is valid for the following parameter ranges:

Through these expressions it is possible to calculate the attenuation value that the signal undergoes in propagating from the radio-base antenna to the mobile terminal, then knowing the other parameters of the system: gains of the antennas, frequency of the signal, etc., it is possible to calculate the level of the signal in the various points of the region where the service must be guaranteed.

5.3 Additional constraints in the choice of configurations

In many optimization problems the domain of the variables, on which the functional depends, does not contain discontinuities, i.e. each choice of the variables corresponds to a value of the functional which can be more or less high and on the basis of it the subsequent selection for the search will be made minimums or maximums, based on the type of optimization you intend to do.

In other cases the domain contains regions of discontinuity in which the functional must never be evaluated in order to obtain the best solution, for example if the minimum distance trajectory joining points A and B in figure 34 is sought, they must be avoided in the choice the curves that pass through the obstacles because obviously they do not correspond to physically feasible solutions; the obstacles thus become regions of discontinuity of domination.

In our case, the configurations to be discarded are those that have the coordinates of the RBS in areas where it is not possible to locate a RBS, for example a square, a lake, a school, etc., these configurations will be discarded regardless of the functional value they will assume.

At this point there are two alternatives:

Discard the prohibited configurations by random attempts to locate the SRBs until an acceptable configuration is found;

Move the child restraint to the edge of the nearest forbidden domain; In the first case a reset of the path up to that moment optimized through the genetic is carried out, losing some configurations that could have been optimal for the rest of the constraints.

In the second case, vice versa, we try not to lose the results obtained up to that moment.

In this way, if a child restraint is located in the forbidden zone, the program controlling the location coordinates moves the child restraint to the side of the perimeter closest to the child restraint.

Figure 35a shows the first possibility, if the solution falls into the forbidden area, everything is abandoned and starts again with a new random arrangement.

The random cycles for the search for the non-prohibited position could be very long and these solutions could lead to non-optimal SRB provisions for the reduction of the value of the functionals.

This observation derives from the fact that if the optimal solution provided for the arrangement of the child restraint in the prohibited points, then a completely different arrangement in general may not lead to a minimum functional result.

Conversely, by adopting the mechanism of slightly moving the SRB, the range of suboptimal solutions found up to that moment does not change much.

In this way, the genetic algorithm alters the solutions only locally, bringing less impact on the optimal solution for the functional tested.

Fig. 35b highlights the modified path that follows the outline of the prohibited area until it is as close as possible to the prohibited position but optimal for minimizing the functionalities.

5.4 PRE-EXISTING SITES

The program took into consideration the possibility of taking into account child restraints already located in the city, this function was obtained by inserting in the program input data both the number of child restraint systems with fixed coordinates and their x, y coordinates.

In particular, the genetic algorithm takes into account the pre-existing stations, however modifying their output power and the tilt of the antennas.

This aspect is important if we consider that the radio parameters of a child restraint must be reviewed when a new system is activated which can interact with the first system at the radio level.

Normally the main action is to redesign the coverage areas of each plant.

The insertion of a new child restraint on the network is always accompanied by an optimization of the surrounding child restraints. The program takes into account this design method and modifies not only the parameters of the new SRBs located but also those of the previous existing SRBs. The parameters of the pre-existing RBSs to which reference is made are the powers and tilts of the antennas. In its evolution, the genetic algorithm seeks the best solutions for fixed SRBs by varying the powers and tilts of the antennas according to the new SRBs inserted.

5.5 INSERTION OF THE CHARACTERISTICS OF THE REGION AND SC ANNERINO

The area chosen for the functional test is an ideal city located on a floor divided into zones with different density of telephone traffic.

In particular, areas can be diversified for the following entry data:

GENERATED TELEPHONE TRAFFIC in Eriang;

The traffic can be PREGL \ TO or generated by HIGH SPENDING customers therefore it is necessary 'indi', to identify as the performance factor of the SRB = traffic costs SRB:

Levels of ENVIRONMENTAL IMPACT in V m:

Radioelectric COVERAGE levels in dBm;

Minimum C / T RATIO levels in dB;

each of these has been divided into bands in order to weigh the various areas of the urban area of reference on traffic rather than on interference, etc.

a possible urban area is the one shown in figure 36.

5.6 PARTITION IN PIXEL

The subdivision into pixels represents an indispensable step to use the matrices in the calculation of the radio coverage and related functionals.

The value of the functional depends, other characteristics being equal, on the type of partition made; if the partition becomes denser, the value of the functionals increases because the signal is evaluated on a greater number of points and then the quantities for each point are added up as indicated in the expression of the functionals itself.

To optimize the program that will use these functions, a solution has been adopted using three-dimensional matrices, this mode simplifies the assignment of input and output values:

The "i.j" indices also identify the traffic source areas and areas with environmental protection for each pixel, in particular it is possible to assign the traffic inputs to a "T (i.j)" floor of a three-dimensional matrix, to another floor Q (ij) "the environmental protection inputs, to a Prohibited floor (ij) the areas prohibited to RBs, to a C / I floor (i, j) the zones with different C / I, to a dBm (ij) floor , the desired coverage levels, to a High-Suspend plan (ij) the areas with the presence of such customers, to a Costs plan (ij) the network costs.

The output matrix (Layer and Crom) contain the calculated covers and all chromosomes respectively. For example, Layer is identified by 3 indexes "i.j.k" the indexes "i.j" identify the pixels, the depth index "k" selects the plane to be considered:

Layer (ij, l) identifies the subdivision of the urban area into best server,

Layer (i j, 2) returns the coverage in dBm for each pixel, Layer (i j, 3) returns the C / l for each pixel,

Layer (i j, 4) returns the electric field for each pixel, Layer (ij, 5) returns the matrix of the distances between SRB and Pixel. Fig. 37 shows in summary how the input and output data are assigned.

The maximum power on the pixel of indices i, k = 3.5 is provided by the SRB 4 which is therefore the Best Server for that pixel, in this way it is possible to identify all the Best Server areas of each SRB.

The boundaries between the Best Server areas of different SRBs are defined with the coverage isolates provided by the SRB vaire, therefore for the interference control it is sufficient to check that within a Best Server boundary the interfering cell arrives with the lowest contribution possible regardless of the field on the border.

As seen before, the prediction model so 231 requires the average characteristics of buildings and streets, etc. Different types of neighborhoods can be entered (with different densities of buildings, streets, etc.) in addition to entering the parameters of each neighborhood, the position of the various neighborhoods must be communicated to the program.

To simplify the input data, the Walfish Ikegami parameters are the same over the whole area considered. In fact, it is not so much interested in studying the more or less perfected propagation model but rather in the enhancement of the functionals.

5.7 CALCULATION OF POWER AND ELECTRIC RANGES

To see in more detail how the 'power' terms are calculated, which enter the expression of the functional, it is necessary to examine how the electromagnetic field levels are calculated in a generic point.

Considering any rectangle belonging to the partition and any antenna, the power level originating from the antenna in question at the point considered is given by:.

where is it:

'Pr' is the power received by a terminal palmai'e at the point considered produced by the radio-base station under consideration (dBm),

'Pt is the output power of the reiadvo transmitter to the radio-base station taken into consideration (dBm),

'Gt' is the gain of the transmitting antenna in the direction given by the joining the two antennas (dB),

'Gr' is the gain of the receiving antenna; a dipole in the case in question (dB),

'Attenuation' is the total attenuation that the signal undergoes as it propagates from the radio-base station to the point considered. The attenuation value is calculated through the expressions of the prediction model cost 23 1 once the value of the distance between the two transmitters has been calculated, it also depends on the loss of the TX Branching system of the SRB, and on the tilt of the antennas:

the same quantity not expressed in dB is:

With regard to the power in reception, from Friss's law we obtain directly:

The quantities are obviously not expressed in dB.

Taking into account the constants we obtain:

For the receiving antenna "Gr ', the quantity not expressed in dB is found, then the level of the electromagnetic field in V. m received through the following expression is ascertained:

where is it:

‘Η’ is the characteristic impedance of the vacuum,

'Λ' is the wavelength of the useful signal, ^ 'Pr' is the power received by the receiver at the point considered (W),

'Gr' is the gain of the receiver,

'Freq' is the frequency in Hz

the previous expression is obtained through the following steps:

where δ is the power density generated by the radio-base station at a point at distance d, it is known that it is equivalent to:

multiplying the power density by the effective area of the receiving antenna, the received power is obtained:

it is also known that the effective area of an antenna is linked to its gain through the following relationship:

substituting this value in the expression of the received power we have:

inverting the relation we obtain:

as they wanted to prove.

The same results are obtained using for example the quantities in dB and the AF Antenna factor for the calculation of the received Electric field:

In this case both the receiving antenna gain Gr and the logarithm of the frequency are simplified as they are subtracted in the equation of the electric field.

Then through the steps described the level of electric field in a point produced by a radio-base station was calculated, this procedure is repeated for all the transmitting antennas, in the end you will have the electric field levels generated by all the antennas in the point considered.

It should be noted that for the calculation of the total electric field Etot. (V / m) relating to all the radio transmitters of a RBS, the powers in Watts must be multiplied by the number of transmitters or carriers present in the RBS, for the calculation with the powers expressed in dBm, it is necessary to transform the number of transmitters into dB and add this quantity in the calculation of the power transmitted by the RBS.

The algorithm uses omnidirectional antennas, and does not provide for the possibility of having different antennas for different sites.

The use of directional antennas in the configuration having to provide an overall radiation pattern close to the omnidirectional one, therefore in this phase it was preferred for simplicity to consider a SRB with an omnidirectional antenna.

To take into account the tilt that the program takes into account, an exponential decay curve of the gain after the focus point of the antenna has been assumed.

This attenuation is the one that appears in the equation for calculating the received power Pr.

In particular, the program calculates for each pixel the distance between the pixel and the SRB, if this distance is greater than the distance between the SRB and the focus point of the antenna, an additional attenuation is added precisely due to the tilt.

The exponential function considered is the following:

If the focus point coincides with the distance between the pixel considered and the SRB, or if the pixel is within the Tilt area, then the Delta is zero and there is no additional attenuation for Tilt to add.

The form of the considered exponential function is shown in Figure 38.

5.8 SUMMARY OF DATA ENTRY

For the algorithm to work, the following data must therefore be entered:

the coordinates of the area to be covered,

the areas of maximum, medium and minimum protection

areas with high, medium and low traffic,

the areas of high medium and low C / I accepted,

areas with high, medium and low desired levels of radio coverage,

the working frequency of the GSM network,

the number of pre-existing sites,

the coordinates of these sites,

characteristics of radio-base antennas (Branching TX gain loss),

height of radiant systems,

the characteristics of each neighborhood (height of buildings, width of streets, etc.), Walfish Ikegami,

number of iterations of the algorithm,

number of chromosomes you choose to use,

areas forbidden for child restraint installation,

background electromagnetic field,

power control distance,

opportunity cost zones for the construction of a site.

After entering this data, the algorithm will need to be restarted several times since it will give a different response each time, since it always starts from different initial configurations and uses random instructions in the pairing phases to generate the new chromosomes (the crossover point is random therefore two parents can generate various chromosomes depending on the chosen crossover point).

From the results obtained, the best configuration will then be chosen both from the point of view of environmental impact, and from an economic, technological, physical practicability point of view, etc.

Incidentally, by increasing the number of chromosomes, that is, of SRB configurations, that the geneticist considers, it is possible to explore a greater space of solutions, in the same way if the number of iterations of the genetic is increased.

Since the chromosome is made up of 6 SRBs, the following processing effort takes place:

if n = number of iterations and m = number of chromosomes:

The calculation times progressively increase as m and n increase.

In order to reduce the calculation times, a sampling of the measurements was performed or rather than considering all the pixels, 1 pixel on 8 or 1 pixel on 4 were considered, the processing times with 1 pixel on 4, 64 chromosomes, 100 iterations are 3 days otherwise on all pixels about 45 days.

CHAPTER 6

FUNCTIONAL FOR THE OPTIMIZATION OF THE GSM NETWORK

6.1 GENERAL

The economic resources of the Mobile Telephony Operator available for new investments are not unlimited, therefore the Economic Budget available for new Radio Base stations is used by concentrating efforts on resolving existing coverage / quality critical issues and, in the most profitable projects.

As already anticipated, the high costs for the construction of the network represent an element to be minimized, for this purpose the strategy for the roofing project is not to uniformly disseminate the entire national territory of the SRB, vice versa as widely seen in the chapters dedicated to traffic and the best solutions are identified for the measures based on the following targeted objectives.

The optimization program was built on the objectives specified below and the functionals to be minimized were obtained. USEFUL SIGNAL RATIO ON C / I INTERFERENT Needs dictated by the GSM specifications relating to minimum power and CI I interference levels for both voice and GPRS data services that require C / I greater than 10 dB.

Impossibility of creating a perfect cluster of cells due to the orographic and environmental constraints that strongly constrain the points available for the construction of a Radio Base Station.

COVERAGE

Rural area coverage;

Indoor coverage in densely urbanized areas;

Historic center coverage;

Industrial area coverage;

Competitive coverage on areas covered by other operators;

Shopping center coverage;

Tourist area coverage;

Coverage for road railways including Coverage Extenders for Tunnels;

TRAFFIC AND QUALITY

the need to dispose of the traffic concentrated in some areas by ensuring accessibility to the service;

traffic needs to carry out the celi splitting or to create a new system not for coverage but to remove traffic from a congested system.

needs to improve the quality of service in areas with high traffic.

PROTECTION CONSTRAINTS

requirements related to compliance with the laws on the environmental impact of new and existing child restraints, with reference to Decree 381 of 1998 relating to the rules for determining radio frequency ceilings compatible with human health.

TRAFFIC REPORT ON SYSTEM COSTS

It is a function of the area where a child restraint is located:

Low traffic area implies low performance,

High traffic area implies better performance

High traffic areas of high-spending customers indicates optimum performance

6.2 OBJECTIVES AND CONSTRAINTS

In order to set up the development of functionalities, the fundamental components that intervene during the design of a mobile telephone network were sought.

A possible form of classification of objectives and constraints is shown in Figure 39.

The indices of the pentagonal graph individually represent the main criteria to be followed for the optimization of an urban network for GSM mobile telephony; the greater the area of the pentagon, the better the overall quality of the service offered by the operator.

Therefore, in the development of the functionals, equations were sought on which the five indices to be optimized by the genetic algorithm depend.

6.3 MAXIMIZATION OF QUALITY INDEXES

At the level of GSM network functions to reduce the effects of intrasystem interference the following are active: Power Control, Discontinuos Transmission, Frequency Hopping, Intracell Handover, of these functions and their effects was discussed in the initial chapter highlighting the benefits on environmental impact in accordance with law decree 381.

At the project level for the location and management of the parameters of a new Radio Base Station, the C / I * ratio can be increased by placing the RBS in the center of gravity of the traffic source where possible, and in any case in accordance with the number of cells required. to serve a traffic assumed upstream, as seen in the traffic chapter.

This approach, as highlighted in the chapter on measures, also naturally satisfies the needs of coverage, yield and environmental impact.

6.4 QUALITY OF SERVICE AND ENVIRONMENTAL IMPACT

It is summarized in the increase of the useful signal C in the traffic source area or increase of the C / I by correctly locating the RBS. In the hypothesis of locating all the child restraints in this way, each traffic area can be covered by a Best Server child restraint system with maximum C / I.

This mode reduces over-coverage phenomena from many cells, also reducing unnecessary handovers, and therefore improving the quality of service.

At each handover, the transmitted power is reset to maximum and the interference as well as the environmental impact are not optimized.

In the same way, at each handover there is a small hole in the voice which is normally imperceptible to the user but in some cases it can degrade the qualitative perception that the customer has, technically it worsens Γ MOS indicator (mean Opinion Score). The distance between the traffic sources and the child restraints, caused by an incorrect location of the child restraints, is synonymous with greater attenuation of the section which would be minimized in this way.

This choice has two beneficial effects:

greater and more widespread radio coverage;

a lower environmental impact as it allows the Power Control functionality to act with greater dynamics. 6.5 PROJECT CRITICAL ISSUES

Figure 40 shows, by way of example, a map that represents an area within which there are areas, hereinafter referred to as pixels, with the following characteristics:

High traffic red pixels without environmental impact problems for the location of a child restraint;

Light blue pixels with low traffic where there are no problems of accessibility to the service and at most you want to give a minimum degree of coverage;

High traffic yellow pixel where, however, a child restraint cannot be installed.

The pyramids in bold are the SRBs to be placed in the program. An essential difficulty that the mobile TLC manager currently encounters is guaranteeing the quality of service and the ability to dispose of the traffic offered in areas with a high concentration of customers, where however it is not possible to position RBS for reasons of both architectural and protectionist environmental impact.

In this case, if it is possible, in order to reduce the distances between the child restraint and the traffic source, a good solution is to locate the child restraint in areas with low environmental protection but at the same time as close as possible to the areas of interest for traffic.

6.6 FUNCTIONAL GENERAL

The algorithm for finding the best set of configurations to determine the position, the tilt of the antennas and the power emitted by the transmitters of the RBS, takes into account the functionalities implemented and the sub-optimal solution is determined by minimizing or maximizing these functionalities.

The functionals developed to achieve the above objectives are 10 grouped two by two forming 5 types of main functionals, precisely these are referred to the indices to be optimized of the pentagonal graph seen previously. In a preferred embodiment, the results provided by the technique according to the present invention are immediately usable by means of graphs which indicate the number of SRBs which are Best Servers in regions enumerated with index "j" like the SRBs themselves.

These regions are in turn divided into smaller pixels with index "i" in order to discretize the electromagnetic field values in space with greater precision and at the same time provide the program with the input data relating to the expected traffic on the various pixels, from environmental protection , to the C / I and, to the desired coverage on the same.

In particular:

A Pixel is a square with a side of 10 m and an area of 100 square meters on the geographical map;

A region is the set of pixels covered by a RBS, the boundaries between one region and another are the isolation points provided by two RBs adjacent to each other, precisely it is the area within which the e.m. field provided by one child restraint predominates over that provided by other child restraints.

On the basis of the traffic model adopted, the child restraint systems and the number of carriers necessary for a given territorial distribution of mobile radio customers are calculated upstream (usually the number of child restraints that follows is always greater than or equal to that strictly necessary to provide coverage only. , it is no coincidence that the construction of a mobile network usually takes place by first providing basic coverage with a few child restraints and then increasing the child restraint systems on the basis of the traffic offered up to the optimal operating conditions).

For each individual region, the number of Pixel "nj" belonging to it is calculated by defining the area of the region in square kilometers,

6.7 FUNCTIONAL FOR THE ENVIRONMENTAL IMPACT

6.7.1 FUNCTIONAL FOR MAXIMUM REDUCTION OF THE ENVIRONMENTAL IMPACT

The first functional that is minimized by the genetic is related to the electric field produced by the Radio Base Station on the pixels of the regions considered and has the aim of minimizing the environmental impact on a given reference area.

If we consider a single region of n pixels addressed by indices (i, k) then we can rewrite the functional taking into account the

With regard to the first formulation, each "J" region covered by a SRB Best Server is divided into "i" pixels, on each

pixel the contribution of the fields of all the antennas of the RBS and of the electromagnetic field present in the background is evaluated.

The term "Q" which is an input data to the program, weighs within the functional the zones of the region on the basis of the chosen environmental protection factor, for example 6, or 20 V / m, that is, increases the value of the functional more where it is wants greater environmental protection. Since the genetic algorithm returns the chromosome with the lowest functional value, then it will correspond to the electric field measured on the pixel with the lowest absolute value, which translates into distant SRBs, with low output power transmitters, and with very tilted antennas.

The genetic algorithm that uses this functional brings the SRBs to position themselves at the maximum possible distance, consisting of the contour of the chosen work area, with respect to the points with declared environmental protection.

The output power of the transceivers is also lowered in order to theoretically reduce the environmental impact to zero.

In order to avoid this second effect, the minimum power of an RTX has been set at 33 dBm and the maximum power at 41 dBm with 8 steps of 1 dB.

The implementation within the program was carried out using the recursive formula:

Etot = C Total electrical on the pixel produced by all the SRB and by the pre-funded Cem Q1.1 ≡ environmental impact factor which assumes the input values: typically = 6 20 VI ni calculate the electric field contribution of all the RBSs on a pixel of coordinates

’I, j then repeating for each pixel and adding all the results to obtain the functional value for that given chromosome configuration, that is, for that given distribution of SRB powers and

tilt found by the genetic.

At each iteration of the genetic algorithm 8 or in general n new chromosomes are determined for each of which there is a corresponding value of the functional.

The number of chromosomes to be used greatly affects time

processing, this number is an input data into the main program (NITER).

The best configuration is the one that returns the functional

with the lowest value.

By viewing the program iterations, it is noted that the functional values decrease, over 100 iterations of the genetic algorithm the same results are obtained, a sign that the minimum convergence value has been reached.

6.7.2 FUNCTIONAL FOR REDUCTION OF THE ENVIRONMENTAL IMPACT TO THE PREFIXED LIMITS

In relation to the environmental impact, a different methodological approach is possible, in fact, instead of reaching the absolute minimum, it is possible to set a maximum electric field threshold allowed on a single pixel and consequently on a specific area.

The functional compares the value obtained with this threshold and increases only if the electric field on the pixel is greater than the threshold given as input to the program.

For example, if it is desired that on the pixel there is no more than 6 V / m then in the case that the simulated field is greater than 6V / m there will be an increase in the functional, vice versa in the case in which the value of the simulated electric field is lower. of the threshold of 6 V / m it is useless to try to optimize as it is already below the prefixed limit, in this case therefore there is no increase in the value of the functional.

If all the power and tilt SRB configurations found always guarantee compliance with the limits, a functional with null value will be obtained as it never increases.

The following assignment was made within the program to implement the summations:

The operation is very similar to the first functional, first the contribution of each SRB is added to a pixel and then the results of all the pixels are added.

All the values of the total electric field determined on each pixel are stored on the Layer4 matrix in V / m and on Layer 6 in

dBuV / m, these allow easy printing of the distribution of the environmental impact on the reference area.

6.8 FUNCTIONAL FOR THE MINIMIZATION OF THE SRB DISTANCES FROM THE SOURCES OF TRAFFI-CO

In the chapter dedicated to traffic, a criterion was established to determine the number of SRBs needed for a given quantity of customers, furthermore in the previous paragraphs it was seen that in order to serve customers with a higher quality and in greater quantity with a mobile phone, it is advisable that the child restraint is located as far as possible in the center of gravity of the traffic source.

This function takes into account the geographical distribution of traffic within each region covered by a single child restraint.

The territory is therefore divided into regions equal to the number of necessary SRBs, naming them with index "j", each region is divided into n pixels with index "i", within each of these regions, the functional that Γ algorithm must minimize, it tends to locate the SRBs as close as possible to the sources of telephone traffic, that is, closer to the pixels with increasing telephone traffic. The traffic named T (ij) weighted with three values 10—25—45 is an input data to the program and weighs the pixels on the basis of the quantity of low, medium, high traffic considered for each RBS.

The goal is achieved by weighing the distances with traffic, therefore the distances with higher traffic will be reduced, which in percentage have a greater impact on the value of the functional.

Two functionals for traffic have been developed that weigh the first less and the second greater the distances:

The sensitivity of these functionals depends on how the traffic is weighed:

in the first functional there is a double sum of Distances * Traffic products

in the second functional there is a double production with an exponential Traffic elevated to the distance, the variations of this last functional are at the same distance, 2 to 10 times greater than the first functional, in the program the first functional was used.

Figure 41 shows how these functionals work.

The optimization of the child restraint location for a single region "j" covered by a single child restraint is achieved by minimizing the distances D 1.1 DI, 5 DI, 16 D 1.13 of the child restraint from the high and medium traffic pixels T ij, which it corresponds to placing the child restraint in a barycentric manner with respect to the greater traffic.

Figure 42 shows the case of a telephone traffic area divided into two regions each covered by a child restraint, as in the previous case, the algorithm optimizes the distances within each area of the child restraints from the sources of high, medium and low traffic.

6.9 FUNCTIONS TO OPTIMIZE THE C / I

6.9.1 FUNCTIONAL FOR THE MAXIMIZATION OF £ 11

The maximization of this functional aims to find a Tilt of the antennas, a location as well as an optimal power of the SRBs, in order to reduce the interference on a region covered by a SRB, caused by the interfering SRBs of the surrounding regions.

I define a C / I maximal when on each of the pixels the difference between the dBm power levels of the SRB serving Best Server and the dBm levels of any other interfering SRB is maximal, the genetic algorithm finds new solutions, for example by moving the SRBs or tilting more the antennas of the interfering RBS, or lowering the power of the interfering ones.

Like the first functional on the environmental impact, I can weigh the C / l calculated for each pixel with the C / T given in input to the program for each pixel, however unlike the first functional that was minimized now we are interested in maximizing the C / ratio THE.

To obtain the minimization of a functional that must be maximized, it is sufficient to put the minus sign in front of the quantities and have the program search for the solution with the minimum value, this solution corresponds in absolute value to the desired maximum C / I.

Since it is convenient to express the powers in dBm then the C / I is calculated not as a quotient but as a difference in levels. Precisely, the difference in the levels on each pixel is made between the power of the server and the power of the other interfering cells, this difference is weighed for the C / I fixed for each pixel through the matINT matrix, and finally for each arrangement of the SRBs, i.e. for each chromosome, the total value of the functional is calculated as the sum for all pixels of the values obtained.

To give an example with 6 SRBs and one pixel we have that the average C / I and subsequently the calculation of the functional is obtained by means of the following:

The functional that in the program increases with negative values is Costlnt, the maximum absolute value of this functional corresponds to the best SRB configuration for the optimization of the interference.

The greatest contribution that intervenes in the functional is related to the pixels with the highest Q, these weigh more in the enhancement of the functional.

In order for the functional value to be minimum (maximum in absolute value), the genetic algorithm searches for the highest average C / I values precisely in correspondence with the highest weight pixels.

The graph of figure 43 shows an example that takes into consideration 3 cells, the first is the serving cell while the other two are interfering.

The distance is shown on the abscissa axis, while the power received on the pixel in dBm is shown on the ordinates.

On the pixel under consideration, the best server cell covers with edge levels of -80 dBm. If a C / I value of 18 dB has been set, it is evident that the position of the SRB A is not correct as the power radiated by it measured on the pixel under examination is such as not to guarantee the fixed C / I. but a lower value of 13 dB.

In case B the child restraint propagates an em field. Which provides lower power values which allow compliance with the minimum limits of C / I.

In case B the position of the child restraint and such that the C / I obtained is better than in configuration A.

The calculated values of C / l for each pixel are stored on a Layer3 matrix which allows the printing of the distribution of C / I in dB of the whole area trivially.

The check is carried out by the program on each pixel for each chromosome, i.e. for each set of located SRBs, and in order to mathematically implement the functional, as already seen, the area has been divided into best server zones for each SRB, called regions.

Inside each of these the control of the C / I is carried out by checking that the interfering RBSs arrive with the lowest possible levels, therefore it is also necessary to mutually check that the RBS serving on a region is not too interfering on the pixels of the region that was previously origin of the interference for the previous region.

Basically, if it is true that PI interferes with the area covered by P2, it is also true that P2 can interfere with the area covered by PI, consequently the functional must take this into account.

The most elegant functional form previously expressed iteratively is:

With the matrix representation adopted of a single region of (p, q) pixels, the functional can also be rewritten in this way:

The peculiarity of this functional is the approach to interference not in terms of co-channel and adjacent frequencies but exclusively as the power of a RBS that intervenes in the coverage area of another RBS.

This concept frees the interference calculation of the radio channel used as now we consider all the available band or rather a single radio channel.

This type of approach is interesting for UMTS applications where the concept of interference is more closely related to the power in the field of a RBS than to another RBS.

Functional optimization guarantees the maximum C / I value obtainable over the entire area. To better understand the mechanism of the functional, we first consider regions divided into a single pixel, that is n = l.

In the hypothesis of having 3 regions covered with three RBSs located in the center of gravity, the possible combinations for the mutual control of interference are those shown in Fig. 44.

When feeding n = number of pixels with which a region j is discretized, the relations to be controlled increase by n times, for

example in the case of the three regions seen if each were divided into two pixels, instead of six we would have to check twelve different relations.

Trivially, for example, with two one-pixel regions there are two relationships:

P11-P12 = interference in region 1 caused by child restraint 2 The relative reciprocal:

P22-P21 = interference in region 2 Caused by SRB1. If the pixels of each region are brought to two then by defining the following index we obtain:

relationships become 4:

P111-P112 = interference in pixel 1 of region 1 caused by SRB 2

P121-P122 = interference in pixel 2 of region 1 caused by SRB 2

Its reciprocal:

P212-P2I 1 = interference in pixel 1 of region 2 caused by SRB 1

P222-P221 = interference in pixel 2 of region 2 caused by SRB 1

Fig. 45 shows a graphical display.

6.9.2 C / I FUNCTIONAL WITH PREFIXED THRESHOLDS Similarly to what was seen for the functional on the environmental impact, in this case too the philosophy of designing a second functional Interference has been pursued which, compared to the one previously seen, takes into account C thresholds / I imposed as limits beyond which not to go down.

For this reason, during the check for each pixel, if it happens that the value of C / I is better than the minimum set, then the functional does not increase, otherwise it does.

The formulas are the same as the previous ones but instead of the product for Q (i j) the difference has been introduced:

Within the program, the iterative formula set for this functional is similar to the previous one with the addition of a control IF:

Once the C / I thresholds have been established, this function allows you to optimize an area by setting limits that you do not want to be exceeded, therefore it can be useful when you want to improve the quality as C / I of an area by setting the minimum desired value. you say.

Summarizing, the first type of functional tends to a maximum reachable C / I value, vice versa this functional, once set a threshold value, tends at least to respect that value.

If the values are better, the functional is not incremented, otherwise it is.

This functional in particular can be useful in the diversification of quality areas, in fact in general it is not necessary to optimize an entire area in an absolute way but can instead be

It is useful to improve one area more at the expense of another with little traffic.

For Tottemmento of the results this functional was used.

6.10 FUNCTIONALS FOR COVERAGE

6.10.1 FUNCTIONAL _ MAXIMUM COVERAGE ON THE PIXEL

This functional aims to maximize the level of power received on the pixel, it behaves in the opposite way to the functional on the environmental impact.

The usefulness is especially when it is necessary to take into account the environmental impact but at the same time it is desired that the coverage does not fail, vice versa if some areas are not subject to coverage constraints then these areas can be more heavily influenced by the functional on the environmental impact.

Like the other functionals, a pixel matrix corresponding to the desired coverage level in dBm for each pixel is entered as input data.

The functional takes this weight into account, multiplying the calculated coverage value by the weight assigned to the pixel.

Like C / I, if you want to minimize a functional that must tend to a large absolute value, just use a minus sign in front of the sum.

In the calculation of the functional value, the heavier terms will contribute or to obtain a larger functional in absolute value, the genetic algorithm will arrange the SRBs in such a way that the power received on the heavier pixel is as large as possible.

For each best server region of an SRB, the power on the pixel is first evaluated, then this power is multiplied by the value of the weight assigned to the input matrix on the discretized coverage levels. In formulas we have:

Within the sw program, the summations were obtained by assigning the sum performed iteratively to the same variable and initializing the variable to zero.

This is all reported in the following equations:

FUNCTIONAL TO COMPLY WITH THE MINIMUM COVERAGE

Also for this functional, the joint philosophy of finding an equation to respect a pre-assigned level of coverage rather than seeking the maximum value has been pursued.

Precisely the functional found, it increases only if the level of field measured on the pixel is lower than the level of coverage required on that pixel.

For example, if the minimum level given in input is -60 dBm, and the level found is equal to -80 dBm then the functional increases, this gives the possibility to the genetic algorithm to search for a lower value for that functional by means of an arrangement better than SRBs.

The formulas are similar to the previous ones, a subtraction is used instead of multiplication:

Within the program, the summations have been implemented as follows:

6.10.3 OBSERVATIONS ON THE FUNCTIONALS ANALYZED With regard to the functionalities seen, except the functional on traffic, all the others have the possibility of being minimized without any threshold or, of taking into account thresholds or input data levels.

Furthermore, the functionals on environmental impact and on interference tend to arrange the child restraints on the edges of the area while the functionals on traffic and on the coverage, in a complementary way, tend to arrange the child restraints towards the area with greater weight than in our example of ideal city coincides with the center of the rectangle.

A congruent behavior is observed for the powers and tilts found by the genetic:

I. Environmental you have minimum power and maximum tilt antennas; Traffic there is minimum distance SRB Traffic source; Maximum coverage of emitted power and minimum tilt antennas; Maximum interference Tilt Antennas;

The functionals with the predetermined limit threshold not to override, can be used in combination with some of the functionals that tend to the maximum of minimization.

This design approach is useful when, for example, you want the minimum possible environmental impact but at the same time you want to ensure a fixed quality of service at the level of coverage, C / I, and capacity for disposed of traffic.

6.11 FUNCTIONAL FOR POWER CONTROL DISTANCE

As previously explained in the first and second chapter, the Power Control performance allows both the SRB and the Mobile to increase or reduce the transmission power within a predetermined window.

These variations basically depend on the received power level and on the quality understood as BER measured on the radio link, Down-link if we consider the radio section from SRB to Mobile, Up-link if vice versa we consider the radio section from Mobile to SRB.

If we consider an ideal situation of zero interference, then we can hypothesize that the transmission power is regulated only as a function of the received power level which is a function of the radio section between the SRB and Mobile.

We assume that the line balancing is correct and consider the Down Link section for simplicity.

If the project for the location of the child restraints is carried out taking into account the maximum coverage range beyond which the Power Control is no longer able to limit the transmission power, then we would obtain the location of the child restraint systems at a maximum interdistance that allows to take full advantage of this functionality. .

If this were possible, the average power emitted over time, or the energy associated with the radiated electromagnetic field, would be on average less than the maximum possible. This approach would greatly improve the quality of the service offered both in terms of environmental impact and C / I ratio.

The functional that is introduced now takes this distance into account and tries to obtain a maximum coverage of the cell equal to this distance, it is obvious that immediately nearby another adjacent cell is needed that similarly provides the radio coverage necessary for the service.

The result is an arrangement of the cells at a certain maximum distance of the pixels it covers itself as the best server, this distance can be given as input to the program and depends on the assumptions made only on the propagation environment considered. Figure 46 shows a display of the P.Control as a function of these distances.

By setting Pref the reference power, in the absence of interference, that you want to receive in the area where the P. Control is activated (usually -90 dBm) you have:

as long as r <Don, the output power is greater than Pref;

for Don <r <Doff the P.C. is activated. for the dynamics of the child restraint;

for r> Doff, the SRB transmits at maximum power;

for r = Dmax the received power is equal to the sensitivity of the receiver (usually -104 dBm).

In the regulation area, the EIRP trend (dBm) is opposite to that of the attenuation and therefore will be proportional to:

the alpha coefficient can assume values of 2, 3, 4, 5 depending on the model with which the field propagates in the city, usually about 3 or 4.

In summary, by determining the various distances Don, Doff, Dmax, and assuming a uniform distribution of customers within the cell, we finally arrive at determining the average 1ΈΓΚΡ to be minimized.

Don: if Pr is the received power then we have:

Doff

Dmax Pr being the received power, we have:

In the region between Don and Doff it is possible to determine the received Power which is equal to Pref thanks to the P.Control, therefore Pref = constant and consequently the EIRP transmitted power must compensate for the attenuation term.

Expressing the propagation in units of the international system we obtain:

The average power in a given direction of propagation is therefore an average over the distance r.

To use this equation taking into account the traffic, the hypothesis was made of a constant probability density that the user is at a certain distance r from the SRB. Using a probability density that takes into account the fact that users are usually more densely distributed in urban areas near the child restraint then

multiplying the three terms by this probability density we obtain an analogous form of average emitted power:

As evident from the last equation and the initial graph, the term of power that weighs most on the energy radiated by the SRB is that relating to the difference (Dmax-Doff) which always corresponds to the maximum power emitted.

The functional found tries to minimize this term.

Precisely since the distance of the pixels of the Best Server from the SRB is known, this distance can be stretched to the Doff distance beyond which the Power Control function is no longer able to limit the emitted power.

The implementation of the formula in terms of summations is as follows:

The implementation of summations in the program was done like this:

As usual, the iterative sum variable CostDist is set to zero at the beginning of the calculations for each new chromosome.

The absolute value of the difference (radius-Doff) allows the functional to be minimized towards zero when the two distances tend to be equal.

The absolute value was not used at first, as it was believed that if the distances between the SRB and the pixel served were less than Doff, the quality of the network would be positively affected.

In reality, it has been seen that the algorithm did not always position the RBSs optimally, in some cases it placed them too close. With this device, however, the geneticist is forced to find the solutions closest to what was established for Doff.

The design of the network taking this functionality into account leads to reduce interference, Γ environmental impact and therefore provide a qualitatively optimal service.

Normally when a mobile radio network is born it is not always possible to follow this project criterion both for economic reasons and for environmental or urban constraints, usually a capillarity of the network is reached over time by distributing the investments, and chasing the traffic that grows.

A simple calculation of the Doff distance at 900 Mhz with a propagation coefficient of 3 returns

Therefore, in order to obtain the best functioning of the P.Control it is necessary that the Best Server coverage range of the SRB does not go much beyond this distance.

If we then take into account an interference margin of another 10 dB, we have that the received field must be around -80 dBm so that the Power Control distance is reduced by half that is to

This distance is precisely that obtained in the Ancona traffic model discussed with the Cluster theory in the Traffic chapter.

This distance is also close to that of 1.5 Km determined with the measurements reported on a real case in the Measurements chapter. The calculation of the Dmax with the same coefficients but with Pref = Prmm = - 1 Q4 dBm leads to a Dmax = 7 Km approximately.

Doff = l, 100 meters has been entered in the program.

The results are exactly as expected, ie a uniform distribution of the SRJBs.

6.12 FUNCTIONAL PERFORMANCE SRB

This function aims to find an arrangement of the SRBs with the aim of guaranteeing the maximum yield on the telephone traffic offered (potentially offered by the customer who wishes to call).

The yield can be seen as the ratio between the traffic disposed of by the child restraint and the relative cost of the child restraint, increasing the traffic disposed of by a plant also increases the yield of the plant.

A single system that could theoretically dispose of all the traffic in a city would certainly cost less than the use of many SRBs.

Such a hypothesis is not possible in reality mainly for reasons of service quality.

A network of many child restraints guarantees first of all the availability of the service, that is, if a child restraint system does not work, there is a disruption located in the area of competence of that child restraint and also with the hypothesis of a zero failure rate, as the frequency resource is very limited, i.e. having a limited group of radio channels available, with a single SRB system, it would not be possible in any case to serve a large amount of traffic.

A model based on the cluster theory therefore guarantees not only efficiency in terms of traffic disposal but above all in terms of C / l.

From this reasoning it follows that a distribution of child restraints, such that each child restraint disposes of the maximum amount of traffic offered equivalent to the others, automatically guarantees the maximum yield.

With an example, the functional to be optimized can be explained: The first starting hypothesis is a total traffic offered of 9 erlangs distributed over the reference area evenly,

The second hypothesis is that there are 3 child restraints to be used; Q if each SRB is arranged so that it can serve 3 erlangs, then multiply 3 * 3 * 3 = 27. a maximum value is obtained for that given totality of traffic offered.

Conversely, if the SRBs dispose of quantities other than 3 erlang, their product will result in a smaller number.

For instance

SRB 1 = 1 erlang;

SRB2 = 3 erlang;

SRB3 = 5 erlang;

then the product of 1 * 3 * 5 = 15 which is smaller than 21 obtained previously, therefore for the same total traffic of 9 erlangs of the area, maximizing the portion of traffic for each RBS, we obtain the maximum performance of our project of network.

Fig. 47 scales the traffic served by each child restraint; the areas of the circles represent the portion of traffic due to each cell.

If the cells individually serve traffic equal to the total traffic offered, divided by the number of sites available, then the product of the areas is maximum.

The functional sought takes into account the traffic input data of each pixel and ensures that the traffic totaled by each SRB is maximum and equal to that due to the other SRBs.

There is no inconsistency with the Cluster arrangement model which provides for RBSs uniformly located in the vertices of the hexagons. If it is assumed that the costs of each plant are constant and independent of the location, then the efficiency of the proposed network project is easily obtained.

Also in this case, the two modes already seen for the other functionalities have been considered, ie finding the absolute minimum value or considering the minimum value with respect to a traffic threshold.

First mode:

The greater the sum of traffic for each SRB, the greater the denominator and therefore the Cost function will be minimal.

At the program level, this functional has been inserted as follows:

In the same way, as the CostServ traffic of each Best Server cell increases, the functional Cost which represents efficiency decreases

2Mode

By adding the input traffic across all pixels and dividing the total found by the number of existing cells, we find the traffic that ideally each cell should tend to satisfy.

The functional is minimized when the difference in absolute value between this traffic value and the value actually disposed of is the lowest possible, possibly zero.

The implementation in the program was done like this:

In this case, if each SRB disposed of exactly the portion of traffic set with TraffTOT, then the functional would tend to zero.

Therefore, the Genetic Algorithm will seek an arrangement of the SRBs, powers and tilts such as to ensure coverage that allows the achievement of this goal.

The first functional was used to obtain the final results.

6.13 REFINEMENT OF THE FUNCTIONAL PERFORMANCE A refinement of the model consists in taking into account other factors that can be input as data matrices, these factors can be weighed according to the area as well as the C / T or traffic, the table below shows an example of weights from 1 to 10 that theoretically could be input, the determination of a functional that takes them into account is not at all trivial as the variables involved are many and correlated with each other.

The weights given are just an example of how a reasoning for determining the Opportunity Cost of building a plant could take shape not only from a purely technical point of view.

In order not to complicate the data to be included in the program, the costs of the child restraints have for the moment been considered constant over the entire area. CHAPTER 7

RESULTS OBTAINED BY THE INDIVIDUAL FUNCTIONALS

This chapter shows the graphical results obtained using the genetic algorithm applied, according to a preferred implementation, to the optimization of an Urban GSM Network.

7.1 SUMMARY

The present invention constitutes a tool for optimizing the radio access network in the field of GSM mobile telephony.

The implementation of the method according to the invention consists of four sections:

Identification of useful actions for the design / optimization of the GSM radio access network; - Choice of an algorithm to determine the best solutions, in our case a genetic algorithm is used; - Design of functionals that translate the design / optimization needs of the GSM radio access network into mathematical equations;

- Choice of the SW language for the creation of the program that implements the functionals within the genetic algorithm, such as for example Matlab.

The developed program provides the solutions of the various functional in graphic form through a representation of the coverage areas in a similar way to what is done by advanced simulation programs.

The program also provides a series of numerical data on the calculated powers.

The flexibility of the program created allows you to study the network to be designed or optimized from different points of view:

- Optimization of the environmental impact;

- Optimization of the yield on the traffic served;

- Network Performance Optimization;

- Optimization of the quality of service understood as C / I ratio;

- Optimization of the quality of the six-vice understood as the level of radioelectric coverage.

The program allows the combined use of functionals, therefore it is possible to evaluate how the arrangement of the RBS changes if one functional is weighed more than another, this opportunity is addressed in the last chapter.

As highlighted in chapter 6 on functionals, there are five indexes to be optimized which can be plotted on a pentagonal graph.

The greater the area of the figure within the pentagon, the better the quality of the solution found for the network design / optimization.

In order to obtain this representation, the maximum and minimum values of each functional have been identified, this result is obtained by running the program for one functional at a time and observing the maximum and minimum value of the relative functional. For each functional, the maximum and minimum values assumed by the other functionals were also noted, this made it possible to easily construct the pentagonal graph on which for each normalized functional it is possible to observe the variation assumed by the remaining functionals. To perform the normalization of the values, the following formula was used:

This system of percentage variation on the distance between the maximum and minimum value of the functions] allows to have the same reference scale, in particular when the functional assumes the minimum value equal to Xminimal then on the graph the value of Y will be equal to 100%, that is to 1, therefore the corresponding functional is minimized. The vertices of the pentagon represent in the way considered the absolute minima of each functional.

What the graph highlights is therefore the distance of the functional from its minimum value.

This consideration is important as during the evolution of the genetic, a functional can both improve and worsen, what matters is the final value assumed, the more the value reached approaches the absolute minimum found for that functional, the better the optimization. network that the functional represents.

This normalization mode has also been used for the combination of several functionals through their normalized sum.

This artifice was necessary because the variation ranges and the values of the various functionals are very different from each other, therefore the simple sum of the functionals always saw the prevalence of the functional with the highest value.

The table below shows the values assumed by the functionals during their single evolution:

In the following of the figures starting from figure 48 the pentagonal graphs are considered, subsequently in the other paragraphs each functional is analyzed in detail.

A- Functional coverage with limits set per pixel.

This functional, as reported in figure 48, has an impact on the indices:

- Coverage that is maximized according to the constraints imposed as expected;

- Absolute minimum environmental impact that is not reduced as coverage levels increase in areas with greater environmental impact;

- C / I that does not improve as an increase in coverage levels does not imply an improvement in C / I;

- Performance which improves as an increase in coverage allows the SRB to capture a greater amount of traffic;

- Distance of the child restraint from the traffic sources which is reduced as the child restraint systems are located closer to the areas to be covered with higher signal levels.

B- Functional Absolute minimum environmental impact.

This functional as shown in figure 49 works with the following indices:

- Environmental impact improves as the signal level becomes very low;

- The C / I is the absolute worst as in the area with the highest C / I required no cell is Best Server with good signal levels;

- The performance is low as some cells placed on the edge of the area capture very little traffic;

- The distance of the base stations from the traffic sources tends to the maximum to reduce the environmental impact

- Coverage tends to zero, the minimum limit is given only by the lower limit of 33 dBm set for RBS transceivers

C- Functional C / I ratio with minimum limits set per pixel This functional as shown in figure 50 produces the following effects.

- The C / I is obviously Optimized to the maximum;

- The performance improves since the SRBs are arranged at a certain interdistance which, by guaranteeing a better C / I, consequently favors a better and greater Best Server area for traffic;

- The distance of the child restraint systems from the traffic sources is minimized since to increase the C / I in the areas with greater traffic, the child restraints are placed closer to them.

- The level of radio coverage, reciprocally to the functional Coverage with the C / I, is not much influenced by the C / I; in fact, it is possible to have a high C / I both in areas with a high signal level and in areas with lower coverage levels;

- Environmental impact is improved since the child restraints are not positioned in the center of gravity of the areas with higher C / I but in proximity to them.

This peculiarity derives from the fact that the functional having to guarantee a coverage level of x db higher than the interfering ones, does not need to absolutely maximize the coverage level.

D- Functional minimum distance SRB from the traffic source This functional produces the effects listed below with reference to Fig. 51.

- The distance of the RBS from the sources with the greatest traffic is obviously reduced to a minimum;

- Coverage improves slightly due to the fact that the RBS are located in the vicinity of the areas with the greatest traffic which in the example considered are also those with the highest minimum coverage threshold required.

- Environmental impact does not improve since the child restraints are almost always located in the area with the greatest environmental impact, which in this case coincides with the one with the highest traffic;

- The C / I improves thanks to the shorter distance and therefore to the greater coverage of the Best Server type on the areas with the highest traffic which in this case coincide with the areas with the highest C / I required. - The performance is optimized thanks to the better distribution of the Best Server areas on the traffic.

E- Functional on the maximum traffic performance of the network.

This functional as referred to in Figure 52 produces the following effects.

- The performance, understood as the traffic-to-cost ratio, is maximized by the functional that arranges the RBS in order to serve the greatest possible amount of traffic for each of them. - The distance from traffic sources is optimized since some RBs are located closer to areas with other traffic;

- Coverage is poorly optimized since in order to improve its efficiency each RBS must seek the maximum amount of traffic served, also resorting to the traffic of low-value peripheral areas, therefore the level of coverage reached in some points may not be above the minimum limit established;

- Environmental impact is guaranteed by the fact that the child restraint systems are not normally all located near the areas of greatest environmental protection, in fact, for what has been said, the child restraints are scattered in order to capture the maximum amount of traffic including that of high traffic areas.

- For the same reason, the C / I is also not optimized.

7.2 METHOD OF PRESENTING DATA

The main data that will be reported below are the following: - Map of the distribution of the Best Server covers provided by each SRB, these are obtained in analogy to what is provided by advanced simulation programs, i.e. a map of colored pixels, each color represents a Different SRB, the edges between one Best Server and the other are the pixels and coverage level.

- Map of coverage in dBm;

- Map of the C / I distribution in dB;

Electric field distribution map in V / m;

- Map of the distribution of the various traffic areas, C / I, Imp.Amb., Coverage, with the positions of the RBs represented; - Graph of the evolution of the values assumed by the functional in order to verify their actual minimization;

- Map of the movement of the child restraints, is obtained by plotting lines of union between the coordinates assumed in succession by each child restraint during the evolution of the genetic algorithm.

- Coordinates of the child restraints;

- Output powers of the RBS transmitters;

- Tilt of the antennas of each child restraint;

- Percentage of the area covered according to the limits set per pixel, it is obtained simply by increasing a counter each time the coverage on the pixel is greater than or equal to that set in the minimum input levels matrix, then dividing this counter by the number of pixels considered in the calculation of the coverage, the degree of coverage is obtained as a percentage;

- Average C / I obtained as the sum of all C / I values in dB for each pixel, divided by the number of pixels;

- The average coverage in dBm, obtained as the sum of all Best Server coverage values in Watts for each pixel, divided by the number of pixels, the value obtained is then converted into dBm; - Average electric field in V / m, obtained as the quadratic sum for all the pixels of the quadratic sum of the fields generated by all the RBS, plus the background electromagnetic field on each single pixel, the value obtained is divided by the number of pixels that have been considered in the calculation of the fields.

7.3 NETWORK PERFORMANCE FUNCTIONAL TO MASSI-MIZZARE

The functional to be minimized ensures that each SRB serves a coverage area proportional to the telephone traffic offered, declared in the various pixels as input data.

In particular, the traffic served must tend to the maximum potentially serviceable by each SRB present, in this way it has been assumed that the yield seen as traffic divided by costs or channel resources becomes maximum.

Fig. 53 shows the distribution of the BEST SERVER coverage areas of each child restraint.

There are 6 SRBs and each color uniquely represents each of them, the boundaries between one Best Server and the other are the pixels with isolation level in coverage power.

The work area is an area of about 5x5 Km in size, and as mentioned there are a total of 261,000 square pixels with a side of 10 m.

In order to reduce the times, in some processing cases a 1: 8 pixel sampling was carried out, i.e. one pixel out of 8 was considered for the field measurements, this is the reason why the distances on the abscissas and ordinates assume maximum values 64 steps, in short between one pixel and d the next there are 80 m, in other cases the sampling is 1: 4 finer than the previous one.

It can be observed that the arrangement of the smaller areas is concentrated towards the center of the figure, where it is assumed that traffic, provided as an input, is greater than the external bands. The result confirms that the genetic algorithm minimizes this functional since it tends to make each child restraint assume traffic equal to the total of all pixels divided by the number of child restraints. The next graph in Fig. 54 shows the trend of the values that the functional has taken after each iteration of the genetic algorithm, the starting chromosomes have been set to 128 and the maximum number of genetic iterations equal to 100.

It is observed that after a certain number of iterations, the solutions all stabilize around the same value, this means that the algorithm is no longer able to evolve towards better configurations.

It must be said that by increasing the number of starting chromosomes and if necessary the number of iterations, the solutions found improve, there is in jargon a better evolution of the species. The starting chromosomes found randomly represent the initial populations from which the future species originates.

The statistically greater the vastness and variety of the initial chromosome population, the higher the probability of probing a larger space of solutions and therefore of arriving at a better result.

Currently the program is run on a Pentium4 -1.7 Ghz Clock and 1 Gbite Ram, the processing time with 128 chromosomes and 100 iterations is about 45 days if we consider all the pixels, with the Ikegami propagation model.

The following figure 55 shows the distribution of the RBS and the respective coverage values in dBm assumed on the various pixels.

The graduated scale with the colors on the right of the graph represents the coverage levels on the pixel in dBm, the distances with sampling 1: 8 are represented on the abscissa and ordinate axes.

In this map, the additional attenuation provided by the Wolfish Ikegami statistical prediction model was not considered.

In fact, this model introduces a considerable increase in calculation times, therefore some cases have been omitted to give greater emphasis to genetics, preferring to increase the number of chromosomes from 16 to 64 or 128.

Fig. 56 represents the distribution of the electric fields in V / m, and the power in dBm, it can be observed for E (V / m) that the values are far below the limit of 3 V / m, therefore unless of using a logarithmic scale even the graphical representation is less conspicuous.

The following Fig. 57 shows the distribution of the C / I obtained as a power difference on the pixel between the coverage offered by the Best Server and the level of the other interfering cells.

The following figure 58 represents the distribution of the distances in m between the pixel and the serving cell that covers it.

This representation is useful as there is a functional that takes into account the distances to optimize the performance of the Power Control.

This functional is not considered in this case but the figure shows how the distances between a RBS and the pixel it covers are distributed within the urban area of reference.

Figure 59 represents the movement that the genetic algorithm operates on the RBSs, it has been said several times that this algorithm searches for the best solution using the value assumed by a functional, at each iteration the selection of the configuration of the RBs is repeated, i.e. the sort of values assumed by the various chromosomes, the best result is the one that returns the value of the lowest functional.

In order to verify that this genetic algorithm led to convergent results, a plot of the coordinates relating to the 6 sites considered was carried out.

The figure shows only the trajectories of the latest iterations of the genetic.

The complete representation is in any case given by the program which at the end of the calculations gives the operator the possibility to see in sequence, as for single frames of a film, the various iterations by pressing a key on the computer keyboard. For obvious representative reasons, not all 100 prints relating to the 100 iterations are reported.

The segments that can be seen on the figure represent the trajectories on which the genetic algorithm chooses the best positions of the RBS to minimize the functional considered at that time for evolution.

If you look carefully you can see that in some areas the lines begin to concentrate in smaller figures, this represents the solutions that converge to the minimum value of the functional.

The last figure of the series, figure 60 shows a representation on which it is possible to view and the position of the RBS located by the program on the urban area graphed as concentric squares.

This artifice was obtained by displaying the sum of the matrices representing both the input data relating to the various areas with traffic characteristics, etc. of the city is the arrangement of the RBS.

The deformation of the squares that is observed has no bearing on the reality of the propagation, it is, as it happens, only a way that has been chosen to easily represent the whole situation.

On the lower left side of the square that represents the average urbanized area, there is a smaller square in dark blue which represents the area where the installation of RBS is absolutely forbidden. While the central red square corresponds to a densely urbanized area whose outline with the yellow / red spots is identified as the square of the moderately urbanized areas.

The program takes into account these zones, these can be either in number or in shape or size, defined at will within the program input data.

7.4 FUNCTIONAL COST TYPE 2

Compared to the previous one, this functional tends to the maximum value on the traffic that can be served by each child restraint, no difference appears on the functional, vice versa, like other functionalities, it uses a ratio with a weight represented by the traffic expressed by each child restraint.

The trends of the performance functionals are in some cases identical even if with different scales.

Like the first cost functional, also in this case the RBS are arranged in such a way that the BEST SERVER coverage areas on traffic or the Best Traffic are evenly distributed, this produces greater Best Traffic areas towards the periphery where traffic is less. and, smaller ones towards the urban center where it is assumed an increasing traffic offered. The figures shown better highlight the characteristics of the functional.

The data provided by the program are:

- Site coordinates:

- The maximum value of the functional is 0.2967 * 10Λ-27

- The minimum value of the functional is 0.2362 * 10Λ-27

Figures 61-65 shown show the above.

Figure 65 in particular shows the distribution of the pixels served by each RBS, the maximum coverage distances in the central area with the highest traffic do not exceed 500 m, vice versa in the peripheral areas of the city represented which have less traffic offered, the Best Server distances are greater reaching approximately 2.5 Km.

Figure 66 shows the location of the air restraints within the urban area considered, no child restraint has been located in the forbidden area; in the area with the highest traffic a child restraint has been placed.

To serve the traffic of the remaining areas, some RBs have been located in the peripheral area.

Figure 67 shows the movement of the SRB coordinates by the geneticist.

Charts 68 and 69 show the reduction of both the maximum yield functional and the yield functional with traffic limits / maximum yield cost functional as a function of the number of iterations of the genetic algorithm.

The following graphs 70a-e show the correlation of the other functionals.

By placing a child restraint in the area with the greatest environmental impact, there is an increase in the relative functional, the optimization of the efficiency of the network for disposed of traffic does not improve the interference situation, therefore the C / I functional does not have an improvement.

Figures 70c and 70d in particular, show the trends of the Power Control Coverage and Distance functionalities, with respect to coverage there is a slight improvement caused by the increase in levels in the central area; the power control distances are not optimized as in order to increase the amount of traffic served, the SRJBs cover distances greater than the limit Doff = 1.1 Km.

The last graph of the series, 70e, relates to the functional that reduces the distance of the RBS from the traffic sources. This function is not optimized as the efficiency function tries to make each child restraint dispose of as much traffic as possible.

This traffic served is the sum of the traffic on both high traffic and low traffic pixels, the max performance functional seeks to maximize such sum for each SRB.

From what has been said it can be deduced that the RBs are not necessarily concentrated near the sources with the greatest traffic, consequently the functional which, vice versa, tends to reduce the distances of the RBs from these traffic sources is not optimized.

7.5 FUNCTIONAL TO MINIMIZE THE DISTANCE OF RBS FROM TRAFFIC SOURCES

This paragraph shows some graphs relating to the functional which tries to locate the SRBs closest to the traffic sources with increasing telephone traffic value. In some cases, the functional places the child restraint within the highest traffic source area.

Also in this case, to speed up the solution, a sampling of 1: 8 measurements was preferred for one case, furthermore, in the first example, the attenuation due to Ikegami's so 231 was not included.

The first figure (number 71) shows the distribution of the Best Servers.

The following figure 72 shows the distribution of the Elect. in V / m.

The following figures 73 and 74 report the C / I in dB and the serving PIXEL - SRB distances.

The following figure 75 shows the distribution of the RBS and the map of the urban area.

It can be observed that to minimize the functional, the genetic algorithm has chosen the SRB configuration that is closest to the area with medium and high telephone traffic.

This choice of program is congruent with what was expected from the functional itself, in particular one RBS was located in the area with the greatest traffic, three others in the medium traffic area and finally the last two in the low traffic area. In any case, even low-traffic child restraints tend to approach the medium-traffic area. In this way all the pixels are served but the greatest preference is given to the areas with high traffic offered where the RBS are located with greater density.

It should also be noted that the child restraints were not located in the prohibited area as intended.

This function is important because normally in the choice of a point where to locate a child restraint, for the exclusive purposes of traffic, areas with high traffic offered are preferred, these correspond to urban areas with high cellular penetration.

The evolution of the algorithm in the coordinate space of the RBS is represented in Figure 76.

The other data that the program provides are the following:

The coordinate values for the 6 RBS are:

The same functional was run by inserting the prediction model of W. Ikegami, moreover the sampling step was changed to 1: 4 and the number of working chromosomes brought to 64 = 2 * 32.

By changing the sampling step the pixels are visually smaller.

The results are as follows:

COORDINATES:

Power fdBmì of the single SRB transceiver

, 34]

- the max value of the functional equal to: 2.081 7 * 10Λ8

- the min value of the functional equal to: 1.9029 * 10Λ8

Figures 77 to d 85 also show the distributions, also correlating the trend of the other functionals.

In particular, almost all improve with the exception of the environmental impact that with a central child restraint fails to assume the minimum value.

In graphs 84 - 85 in particular, the correlation between the trend of the values assumed by this functional and the corresponding trend of the values assumed by the other functionals is shown. This approach is important to understand what happens to the coverage, the cost, the C / l, the environmental impact, the Power Control distances, when the dominant functional chosen for the genetic algorithm is the one related to the minimization of the distances of the RBS from the traffic sources.

The improvement obtained, relative to the coverage, is caused by the concentration of SRB in the areas with the highest level of coverage required, these are in percentage terms better covered than the others for this reason also the functionalities on the coverage improve.

Graphs 85 a-b report the functional of the C / I MAX and the functional minimum Environmental Impact, as can be seen from the increasing trend, both worsen.

In particular, an optimization of the C / I is not expected from the pure optimization of the traffic, the functional traffic in fact tends to minimize at the distance of the RBSs from the traffic sources but does not take into account the interference that the various RBSs can intergenerate.

The combination of needs is obtained by suitably combining the different functionals.

The graphs e-f two functional on the cost, it is clear that by optimizing the RBSs for Traffic or placing the RBSs close to the traffic sources in a way that is weighted with the traffic, the efficiency increases.

This behavior of the functionals suggests that Γ cost optimization goes hand in hand with the optimization of traffic distribution between the various RBSs.

A good optimization of the RBS for Traffic makes the mobile radio network more efficient, this confirms the goodness of the functionalities found.

7.6 1st FUNCTIONAL FOR OPTIMIZATION OF THE RADIO COVERAGE

This functional tends to provide coverage on the single pixel that is no less than the minimum thresholds set with the input matrix relating to the minimum desired coverage levels.

There is another functional on the coverage that, weighing the imposed limits, tends to the highest possible signal value, in this functional one instead seeks the values at least equal to the following: - Highly urbanized area lev = -60 dBm;

- Average urbanized area lev = -70 dBm;

The thresholds can be easily set in the program, these levels have been chosen to take into account the indoor coverage. The walls of buildings in fact have an average attenuation that varies between 10 and 25 dB, this means that to obtain, within a building, a sufficient signal for the service, it is necessary to have a value up to 25 dB greater outside. .

On the chapter relating to the measurements it has been seen that on average the Down Link levels with which the mobile phones receive are distributed according to a bell with a peak around -85 dBm.

This value also takes into account interference and Power Control, therefore, it can be assumed as a reference level for good quality of service.

To receive -85 dBm with high probability, on a densely urbanized area, inside a building, it is necessary that there is a signal level of approximately equal to -60 dBm outside. Taking this into account, it was decided to set the thresholds according to the model seen.

Figures 96 to 105 show that the program has chosen to locate the child restraint systems in the areas with the greatest demand for coverage or in the densely urbanized central area where a minimum limit of -60 dBm has been set.

The child restraints located in this area have reduced coverage compared to the child restraints located outside, the latter come out with greater power and this serves to better cover the furthest outer bands. The arrangement of the tilt of the antennas also agrees with this choice, the innermost antennas are more fingerprinted than the outermost ones.

In the restricted area for child restraints, where in any case medium coverage was required, no child restraints were located.

The average coverage obtained is about -63 dBm.

The functional cost COP starts from the value 2.2473 and reaches the value 1.2815 after 100 iterations.

Figure 86 shows the Best Server coverage areas:

The following figures from 87 to 90 show the coverage obtained in dBm and the C / I in dB, the electric field in V / m and the Pixel - SRB Serving Distances.

Figure 91 shows the complete map of child restraints and urban areas diversified by degree of coverage. The last figure in the series, 92, represents the movement caused by the genetic on the coordinates of the located child restraints.

Unlike the representations seen so far, only the points without the joining lines have been reported, in the drawing it can be observed how the solutions tend to concentrate in the central area where the greatest degree of coverage is required.

This approach to coverage by the genetic algorithm meets the requirements of the functional that tends to favor the areas with the greatest demand for coverage.

The same functional was run considering the following settings:

- Statistical prediction model COST23 Idi Walfish Ikegami; - 100 iterations of the genetic algorithm;

- pixel sampling 1: 4 instead of 1: 8.

The results are shown below as previously with figures 93-99 and, graphs 100-105 of correlation with the valorisation of the other functionalities at the variant of the Chosen functional.

The distribution of C / I is interesting: where there are no Interfering Cells, the decay of C / I is much slower.

As can be seen from Figure 99, the RBSs have been located mostly in the central area of the city or in the area where the minimum level of coverage required is -60 dBm.

In the mid-coverage area, as the required coverage level is -70 dBm, fewer RBSs are required to ensure radio coverage.

In figure 96 concerning the electric field it can be seen that the levels are always below 1 V / m.

This circumstance ensures that the functional on the environmental impact with limits set at 3, 6, 20 V / m, never increases while remaining at zero.

Figure 100 shows the movement of the SRBs caused by the genetic algorithm during the search for the best solution, and the correlation graphs with the other functionals.

The 6 different colors represent the 6 SRBs as usual.

Graph 101 shows the minimization of the functional coverage.

The functional of figure 102 relates to the maximization of the coverage on all pixels according to a pixel weight.

It is obvious that the coverage has not been maximized as the interest of the functional used is only the respect of the minimum pre-established thresholds, these thresholds do not necessarily correspond to the maximum coverage.

This is the reason why the graph above the functional maximum coverage while reducing does not tend to the minimum.

Figures 103 a-b report the functional environmental impact and distance of the child restraint from the traffic source weighted on the traffic. The environmental impact is not minimized, vice versa the traffic improves as the child restraints are concentrated in the area with the greatest traffic.

Graphs 104 a-b report the functionals on the cost, in both cases there is no marked trend towards improvement or deterioration, this can be explained if we take into account the fact that the child restraint systems are located more in the high traffic area, therefore their efficiency seen as traffic served on plant costs is almost optimized.

Graphs 105 a-b-c show the functions relating to the Power Control distance and the C / I ratio, both MAX and according to the limits set at the input.

These functionals show a tendency to improve the C / I however they are not yet optimized, this is legitimate if we consider that the functional on the coverage only looks for power limits to be respected on the pixel without taking into account the useful signal-to-interference ratio.

Therefore, it is noted that in principle an improvement in coverage certainly favors the C / I but this is not the only strategy to be adopted.

In fact, an improvement in the C / I is also achieved with the creation of Best Server areas, to obtain which the tilt of the antennas are used and at the location of the RBS in correspondence with the coverage objective.

The other graph shows the functional relating to the distance of the pixels from the serving child restraint, it is observed that by concentrating the child restraints and becoming these closer to each other, the functional is reduced.

The reduction of this functional allows the performance of Power Control a greater dynamics in the power regulation; transmitting with less power implies:

- lower environmental impact;

- less interference;

From this last observation it emerges that the reduction of the environmental impact can be achieved by maximizing coverage, and at the same time ensuring, with the location of the RBS near the traffic source, the maximum operating dynamics of the Power Control performance.

The numerical results of the SRBs are:

- Coordinates:

As evident from the output data in order to optimize coverage in urban areas, the program returned a transmission power close to the maximum limit, and at the same time the minimum tilt of the antennas.

The latter in particular is very much related to C / I, since the tilt chosen is low, consequently the C / I is not very optimized, vice versa both traffic and coverage have a benefit.

7.7 2nd FUNCTIONAL FOR MAXIMIZATION OF RADIO COVERAGE

This functional, unlike the previous one, tends to maximize the level of power received on the pixel in a weighted way with the degree of coverage set with the minimum levels. No comparison is made between the measured level and the input level, vice versa as the level of coverage given as input in the matrix of minimum levels increases, the importance of that pixel also increases, therefore the program will try to maximize coverage by acting on both location of the RBS both on tilt and powers.

The latter in particular have been set by the program all at maximum.

From the relative figures, it can be seen that the location of the RBS also in this case tends towards the center of the urban area where the weights for coverage are higher.

Also comparing the coverage levels reached in the previous functional, it can be observed that in this case the powers on the pixels are greater and this is also confirmed by the higher electric field level compared to the previous functional.

This behavior is justified if we take into account the fact that the previous functional tended to set coverage thresholds, not necessarily the maximum ones by concentrating some RBSs on the central point at -60 dBm, in this case, on the other hand, the maximum coverage must be reached on each pixel. obviously depending on the relative weight of the pixel.

In this way, the program that has to achieve this will tend to remember even the less valuable pixels, trying to maximize radio coverage.

Both philosophies are useful; on the one hand you want the minimum coverage on the pixel not to fall below a certain value, on the other you want the coverage on the pixel to be the maximum possible.

Figures 106-110 show all the various distributions as seen previously.

In particular, the central Best Server area and the symmetry of the other surrounding areas should be observed.

Ikegami's COST231 statistical prediction was not used in this model.

7.8 FUNCTIONAL FOR MINIMUM ENVIRONMENTAL IMPACT

This function ensures that on the pixels with the highest degree of environmental protection there is an electromagnetic field as low as possible.

There is another function that vice versa ensures that the limit value on the pixels, set at the input via the environmental impact matrix, is never exceeded. From the functional it is expected that in order to reduce the electric fields it places the RBSs as far as possible from the areas with high environmental impact that in the urban map considered are arranged at concentric incremental values, the central area is in fact the one that is most constrained.

Both functionals on the environmental impact tend to widen the arrangement of the RBSs, to lower the TX powers to a minimum, and to tilt the antennas more, in fact in this way the radio frequency emissions of the RBS are limited.

As a counterpart, there is a reduction in coverage, and a reduction in efficiency understood as traffic / costs.

The quality of the service is also affected from the point of view of the C / I, in fact this ratio is reduced in the most protected areas as these areas are covered by distant SRBs with similar coverage levels but not Best Server.

This is basically what happens when trying to cover a city with SRBs located outside the city itself, the service offered will be of poor quality.

For the calculation of this functional Ikegami's statistical prediction model was used, the sampling step chosen is 1: 4, the number of chromosomes is equal to 64 and the number of iterations of the genetic algorithm is equal to 100, the processing time is about 3 days.

The graphs of the various functionals starting from figure 111, show how the achievement of the minimum possible field value on the pixel (limited only by the minimum TX power of the transmitter equal to 33 dBm), also corresponds to a worsening of the other 4 functionals relating to C / I, to coverage, to traffic, to cost. Figures 111 and 112 shown represent the arrangement of the SRBs and the Best Server areas, the program has located them as expected.

From the C / I map shown in Figure 113, the deterioration in the quality of service in the central area is evident due to the lack of SRB Best Server.

From the distribution of the Electric Field shown in figure 1 14 it is evident that it has been practically reduced to zero in the areas with greater environmental protection.

The following figure 115 instead shows the distribution of the distances of the child restraints from the pixels served by each child restraint, it can be observed that these distances even reach 5 km.

These coverage distance values do not guarantee a service in urban areas as the interference levels become too high, moreover, the Power Control performance is definitively compromised as, as demonstrated and measured, the Power Control is no longer able to regulate the transmission power which always remains at maximum.

This aspect does not play in favor of the environmental impact which, conversely, presupposes that the SRBs and mobile phones, "the latter in direct contact with the human body", can regulate their output power with the maximum possible dynamics in order to reduce the power. transmission.

The following figure 116 shows the map of the area with the arrangement of the RBS.

The other functional on the environmental impact as mentioned takes into account the exposure limits imposed, since normally these are 3, 6, 20 V / m, the resulting SRB situation has less impact on the quality of the service.

This solution can be a good measure to take into account the environmental impact and at the same time ensure maximum service quality.

The set field values can be set at will. The choice of the values seen was made in accordance with the provisions of the law on exposure to radiofrequency fields.

Figure 117 reports the movement caused by the genetic on the SRB coordinates.

The following graphs 118 a-f report individually the evolutions of the single functionals:

The environmental impact drops like the average electric field on the pixel.

Figures 118c-d report the evolution for functionaries on the coverage for the C / I which instead of decreasing tend to increase indicating a worsening of the conditions of coverage and C / I. The following figures 1 18e-f show the trend of the values for the functionals relating to traffic and the cost or efficiency of the Network. It must be considered that a functional to indicate an improvement in the situation it represents, must tend to the minimum values as the choice made in the program is to minimize the functionalities for network optimization.

Whenever a functional does not decrease it means that the optimization does not take place, if it grows we know that the situation relating to that functional is degenerating.

The following graphs 112g-112h show the percentage degree of coverage that decreases and the distance of coverage that increases, these indices denote a worsening of the quality of service. The situation that derives from minimizing the environmental impact without a lower limit inevitably causes a degradation of the quality of the service.

In combining the various functionals together it will be seen later that the needs of each of them weigh in a certain way, the resulting final situation is a combination that provides the best arrangement of the RBS.

The output data for this functional are as follows:

As you can see, they are close to or even on the perimeter of the maximum allowed area.

If the area had been larger, the RBSs would have been located at a greater distance, theoretically infinite.

From the choice of the powers it appears that if the program had had the possibility to further reduce the powers it would have brought the power on the pixels to zero.

This incorrectly weighing the environmental impact can mean not giving any mobile phone service.

Observing the values of the antenna tilt chosen by the program it emerges that they are all placed between 5 and 8 degrees, these values obviously reduce the coverage of the SRB to the maximum.

- MAX and MIN functional Environmental Impact

The maximum value assumed by this functional is: 9.2167

The minimum value assumed by this functional is: 2.2013

From what we have seen on this functional, it emerges that the single use of a functional such as this tends to reduce the environmental impact but at the same time tends to cancel coverage by providing a disservice rather than a GSM telephone service.

This functional must therefore be combined at least with the functional on the roof which, on the contrary, seeks to respect the minimum level values for a quality service.

7.9 FUNCTIONAL FOR MINIMUM INTERFERENCE OR MAXIMUM C / I.

Two types of functional have been developed, the first tends to maximize the C / I on the pixels as a function of a weight, the second ensures that the C / I on each pixel is at least equal to that set with the input matrix.

Let's analyze the first functional, referring as usual to the maps and graphs starting from Figure 119.

In order to limit the interference of one child restraint on the other, the child restraints themselves have been located at an increasing distance from each other, if the areas had been larger since the functional seeks the minimum value of the interference, the program would have located the child restraints very far away. between them.

The C / I value is weighted with the zones, the central zone is the one where a higher C / I is required, therefore the program concentrates the SRBs with the greatest power emitted.

In fact, a higher C / I is obtained by creating very defined Best Server areas, on which the contribution of interference due to the surrounding cells is as low as possible.

This way of imposing coverage minimizes interference and guarantees a better quality of service.

Hypothetically, within an urban area, the number of child restraints covering a large part of the area as an umbrella should be very limited, this is the reason why child restraint systems do not need to be located in points that have visibility in urban areas. optics over the entire city, vice versa, the sites that guarantee visibility limited to the traffic target to which the high quality mobile radio service is interested will be sought.

It can be observed that the distances between the serving cells are a maximum of 1400 meters for each cell, therefore the distance within which the Power Control performance is able to regulate the powers within the regulation dynamics is approached.

This has two positive consequences:

- The power emitted is further limited by reducing the environmental impact,

- By limiting the emitted power, interference phenomena are reduced.

For this functional the following were used:

- 100 iterations of the genetic algorithm;

- 32 * 2 chromosomes;

- sampling step 1: 4;

- Walfish Ikegami's cost231 statistical prediction model. In figure 124 it can be seen that in the area with higher C / I the percentage of high C / I identified with the red color is higher than in the other areas, therefore the program has returned a correct solution and within the fixed limits of C / THE.

Graphs 126, 127, 128 show the other functionals as the functional chosen varies.

The graphs show that both functionals on the C / I are reduced, the functional on the C / I with the minimum allowed limits that is represented in the second graph is even reduced to zero.

This value indicates that with that given SRB configuration the minimum C / I thresholds required on the various pixels are absolutely guaranteed.

These results show that these functionals implement SRBs exactly as expected in the GSM network optimization project.

Graphs 128a and 128b show the functional Environmental Impact and Traffic * distance:

Both functionals do not have evident variations towards an improvement or a worsening, this is absolutely not a truth in fact by improving the C / I it allows the Power Control performance to bring the output power of the cells to lower values, also reducing the impact environment, this improvement in Power Control is evident in the P.Control functional distance graphs.

C / I also has an impact on traffic since an interfered cell tends to switch conversations for Handover Quality to less interfered cells.

These correlations can also be made with the functionals seen, by involving a combination of more functional elements in the genetic, namely the one on the C / I, on the environmental impact, on the distance of Power Control and on traffic.

The important aspect that must be highlighted for now is that by improving the C / I there is no worsening of the functionalities on the environmental impact and on traffic, these functionalities remain around a reference value.

Graphs 128c to 128f show the Coverage, P.Control and cost functionals.

From the functional level of coverage there is not necessarily an improvement that is an increase in the levels, in fact increasing the C / I implies creating Best Server areas where the contribution of the interfering cells is minimal but this is not the same as sending more power into the air therefore the level of coverage in general is not necessarily going to increase.

Conversely, the functional on the distance of Power Control tends to be reduced, this is legitimate since an interference optimization produces a geometry of minimum interdistances between different cells.

In the case in question, these distances have been approached to the optimal ones for the operation of the Power Control performance, therefore an improvement in the relative functional has resulted. With regard to the functional cost, there are two trends that are never decreasing, this can be explained simply by saying that the optimization of a network at the interference level has a cost, that is, the functional on the cost returns in graphic form the concept that increasing the quality of the service implies an increase in network costs.

The aspects of traffic, C / I, Costs, are strongly linked to each other; optimizing the quality of service in terms of C / I almost always implies a redistribution of traffic on new child restraints, this may mean limiting the coverage area of some child restraints in favor of less interference.

If efficiency is seen as traffic on costs, it is clear that a limitation of the coverage of a child restraint also implies a reduction in the traffic offered to that child restraint, consequently the margins of performance on traffic for the child restraint itself are reduced. Each quality product has a cost that is proportional to quality, increasing quality implies increasing costs.

Since the consumer trend is to have a quality product in the long run, it is clear that it is convenient to invest in quality.

The functional: C / I, Coverage, Traffic, are important because they produce at the network level the perception of the quality of the service offered, in the next chapter more functionalities will be used simultaneously in order to deduce a useful criterion for the study of a telephony network mobile.

The numerical results provided by the program are the following:

- Coordinates:

- Power in dBm of the single transceiver or radio carrier:

- Tilt antennas of the child restraint in degrees:

- MAX value assumed by the functional at the start COST INT

- MIN value assumed by the functional upon arrival COST INT

It can be observed that the transmission powers found by this functional do not tend to be high, this is in accordance with the theory that less interference requires radiating less energy.

7.10 FUNCTIONAL C / I WITH MINIMUM LIMITS FIXED FOR EACH ZONE

This function arranges the SRBs, the powers and the tilts in order to guarantee the C / I thresholds fixed on the various pixels.

The configured parameters are:

- 100 iterations of the genetic algorithm;

- sampling step 1: 4;

- insertion of attenuation of cost231 by W. Ikegami;

The results obtained as reported starting from Figure 129 show that in order to satisfy a C / I = 18 dB in the center of the area, and a C / I = 10 dB in the remaining part, the program tends to arrange the child restraints in a that there is a Best Server in the central area with greater coverage.

The numerical data provided are:

- Coordinates:

- Power in dBm of the single transceiver:

- Tilt antennas:

- Max value of the functional = 250.2835

- Min value of the functional = 0

From figure 129 it can be seen that the SRBs are located in such a way as to produce a Best Server in the central area with the highest C / I required. By increasing the C / I difference between the central zone and the external zones, currently set with C / I values of 18 dB and 10 dB respectively, it was observed, by spinning the genetic again, that the SRB was located exactly in the center of the smallest square.

The part surrounding the central area was also kept in mind by locating the remaining RBS with a smaller coverage area. By observing the powers and the tilts proposed by the program, it can in fact be noted that the central SRB has the highest power and the lowest tilt, vice versa the surrounding RBSs have lower powers and higher tilts.

The evolution of the genetic starting from the random configuration was not difficult, after about 10 iterations the stable solution was reached.

This means that all the solutions gradually found through the coupling and the subsequent mutation, have determined higher functional values which therefore have been discarded during the selection of the best chromosome.

The graphs shown clearly show the fast convergence towards the definitive and stable solution.

It must be said that by increasing the delta in dB between the C / I of the central area and that of the remainder, it was observed that convergence did not occur as quickly.

The graphs in figure 136 show the evolutions of the various functionals starting from the main one for which the geneticist has determined the solution.

Even the graph of the C / I with the maximum value relating to the minimum possible interference functional, has a similar trend, obviously it does not tend to the absolute minimum because compared to this functional we are considering, the solution found is not yet optimal.

Functional 138a and 138b are on Cost and on Environmental Impact.

The functional on the cost does not tend to reduce, this is justified by the fact that the distribution of traffic is not optimized.

The functional on the environmental impact remains approximately at the same starting value, this circumstance is the one sought or inserting a child restraint in the central area with greater environmental impact with the satellite child restraints around.

This aspect reduces the environmental impact as it allows the Power Control functionality to dynamically lower the transmission power of both the cabinet and the SRB.

There is therefore the double benefit:

- Reduction of the environmental impact;

- Reduction of interference.

Graphs 138c and 138d show the Coverage and Traffic functions.

In particular, for the coverage there is only a slight improvement, this behavior is reciprocal to what was seen for the coverage and the related C / I, or an improvement in the C / I does not involve an increase in the levels of coverage, in the same way, a increasing levels of radio coverage does not necessarily improve C / I. The graph below shows the Power Control distance function.

This functional is reduced indicating that the optimization of the C / I consequently determines an optimal arrangement of the SRBs for the correct functioning of the Power Control performance.

This deduction is reciprocal since the Power Control can reduce the interference caused by the child restraint or by the mobile phone in other points of the network.

CHAPTER 8:

RESULTS OBTAINED FROM THE COMBINATION OF FUNCTIONALS

Below are the graphical results obtained using the genetic algorithm applied to the optimization of an Urban GSM network by combining some of the functionalities seen in a single functional.

8.1 SUMMARY

Without prejudice to the fact that environmental compatibility is currently established and guaranteed by the criteria indicated by the framework law on electromagnetic emissions of February 14, 2001 and by the relative implementing decrees, the technique developed in the present realization, preferred, raises the problem of the optimal location of the child restraints. :

- From the environmental point of view, pursuing the lowest possible emission, that is, such as not only to comply with the legal constraints, but which tends to the absolute minimum technically achievable. This functional is justified by the need to guarantee a margin for future development in compliance with the legal limits and, at the same time, to face the fear of the radioelectric problem;

- From the customer's point of view by maximizing the quality of the service;

- From the manager's point of view by minimizing investments. The results obtained are mainly of a qualitative nature as the coefficients required for the calculation of the model itself, in order to be optimized, require the performance of a

accurate measurement campaign (Tuning). However, the results are indicative of the advantages offered by the present invention.

All the following tests were carried out taking into account the Ikegami statistical prediction model, with a number of iterations of the genetic equal to 70.

The combined functionals are as follows:

- Absolute minimum environmental impact;

- Coverage with minimum thresholds set for each area;

- Interference with minimum thresholds set for each zone;

- Minimum possible distance of the RBS from the telephone traffic sources;

- Maximum yield of each child restraint located.

The combination consists of the sum of the listed functionals. each functional can be multiplied by an appropriate constant in order to amplify its effect on the location of the child restraints.

In order to observe the behavior of the combination of functionals, several tests were performed:

- Test 1 the functional on the roof was weighed more: weight = 10;

- Test 2, on the other hand, the functional Environmental impact was weighed more:

weight = 10;

- Test 3 the functional Environmental Impact was weighed with weight = 1.5;

- Test 4 the functionals were added with the unit weights;

- Test 5 the functional Environmental impact with weight = 4 was weighed

L

f

The values in the table are those of the non-normalized functionals, to perform the normalization necessary for the correct interworking of the functionals, the following formula was used:

with:

From the data graphed below, it can be observed that starting from the unit weight situation, where there is an equal percentage values reached by the Environmental Margin functional and the Coverage functional, by increasing the weight of the Environmental Margin functional, the following interesting results are obtained: - With weight Imp. Amb. = 1.5 there is an approximately 15% improvement in the Environmental Margin and a 20% worsening of the radio coverage;

- With Imp. Amb weight = 4 there is a further 15% improvement in the Environmental Margin and a further 30% reduction in coverage.

Ultimately, a 30% improvement in the environmental margin led to a 50% worsening of coverage in the model considered.

- TEST N 1

Figure 140 relates to the sum of the functionals with weight 10 of the roof.

In the pentagon, as previously established, the top vertex is relative to the yield R, the two at the bottom respectively to the environmental impact and to the coverage C, and the median ones to C / T and distance D.

The following characterizing elements are noted.

- An emphasis with weight 10 on the functional coverage improves the coverage index;

- It worsens the Environmental Margin, reducing it considerably as it concentrates the child restraint systems in the area with the greatest environmental impact - The distances from the traffic sources are reduced due to the child restraint systems concentrated in the areas with the greatest telephone traffic;

- The yield is maximized by virtue of the better distribution of valuable traffic in the area with the highest telephone traffic.

- TEST N 2

Figure 141 refers to a sum of the functionals with weight 10 of the environmental margin. From this representation the following characterizing elements are evident.

- The Environmental Margin is optimized;

- Since the child restraints are located more outside the useful area, we have:

- Worsening of coverage down to the minimum limit tending to zero;

- Worsening of the C / I as there is no coverage Best Server;

- Worsening of traffic performance as the RBSs located on the perimeter of the area with low power capture very little traffic tending to zero;

- Increase in the service distances of the RBS from the traffic sources, consequently the Power Control does not work and both the mobile phone and the RBS must transmit at the maximum possible power, this worsens the Environmental Margin and interference on the network.

- TRY N3

Figure 142 relates to a sum of the functionals with weight 1.5 of the environmental margin functional. In it we note the following characterizing elements.

- By bringing the weight of the functional Environmental Margin to 1.5, an improvement is also achieved in the other service quality indicators;

- It is important to compare this graph with the next one, it will be possible to observe that to increase coverage by 20 percentage points, the Environmental Margin must be worsened by 20 percentage points with the other values of the functionalities being equal. To obtain this coverage reduction, it has been observed that the genetic algorithm operates (with the chosen weight constant of 1, 5), almost exclusively by lowering the transmission power of the RBS to a minimum of 33 dBm for each single transmitter. There is no difference in the location of the child restraints between this test and the one where the weights are unitary, what changes are the powers transmitted by the transmitters of the child restraints, lower in the case of weight 1.5; higher in the case of unit weight of the functional Environmental Margin.

- From this it follows that the criterion for regulating the powers emitted by the RBS, in order to optimize the Environmental Margin, is as fundamental as if not more than their own location.

- TEST N 4

Figure 143 relates to a sum of functionals with unitary weight of the single functionals. In it we note the following characterizing elements.

- By reducing the weight of the Environmental Margin from 1.5 to 1, that is to say the same as the other functional, 20 percentage points of improvement in Coverage and 20 percentage points of worsening of the Environmental Margin are obtained.

- This can be useful for understanding in terms of economic investment how much it costs to improve one quality index over another, trivially reducing coverage can mean reducing the service inside buildings.

Reducing the transmission power by 8 dB means compromising the quality of the service, especially in urban environments where the attenuation of buildings is very high.

- Very low powers are used in the microcellular environment in this case, by hypothesis, the small range RBSs are necessarily located in close proximity to the traffic target.

- TEST N 5

Figure 144 relates to a sum of functionals with weight 4 of the environmental margin functional. In it we note the following characterizing elements.

- By further increasing the weight of the Environmental Margin, the minimum impact is better achieved; compared to the case with weight 1.5, there is an increase of about 10% of improvement on the Environmental Margin.

- The reduction of the signal for the purpose of minimizing the Environmental Margin results in a high reduction in coverage. The reduction in coverage determines, if not carefully assessed in the radio project phase, an essential impact on the quality of the service offered.

- The reduction of coverage also determines a perception of poor quality of service by the customer, for which coverage is one of the fundamental aspects of the service that lead him to one manager rather than another.

- As the child restraints are located in a barycentric and symmetrical manner with respect to traffic, the other objectives continue to be fulfilled.

8.2 TEST 1 FUNCTIONAL WEIGHT COVER PLACE A 10

The child restraints are mostly located in the central area where a higher level of radio coverage is required.

- The transmitting powers of the transceivers are:

- the tilt antennas are:

The relative graphs are shown in figures 145-151, in particular in figure 151 the following have been identified:

- the central yellow area relating to the historic center,

- the central green area relating to the medium density area of urbanization,

- in said central green area at the bottom, a rectangle of a more charged blue, it corresponds to an area prohibited for child restraints - a broad blue outline relating to an area with a low density of urbanization.

8.3 TEST 2: FUNCTIONAL WEIGHT OF ENVIRONMENTAL MARGIN AT 10

The weight of 10 made the functional environmental margin prevail over all the other functional ones, the resulting arrangement is:

- Coordinates

- Starting value of the budget = 2.8698

- Target value of the functional = 0.6316

The coordinates correspond to SRBs located towards the outside of the area, the powers are set to the minimum, the tilts are set with values tending to the maximum.

The arrangement of the child restraint systems is shown below in figures 152 to 158 where the pixel lengths with the relative sampling equal to 4 are still reported on the abscissas and ordinates. radio signal are very low, insufficient for GSM mobile radio service.

The following figure 159 shows the final part of the movement of the SRB produced by the geneticist in the search for the solution, the areas of convergence of the solutions are those where the trajectories identify points with greater density.

8.4 TEST 3 WEIGHT OF THE FUNCTIONAL ENVIRONMENTAL MARGIN PLACED AT 1.5

The value of the weight of the functional chosen of 1.5 has located the child restraint systems in an optimal way for the service, at the same time it has set the powers of the child restraints towards the minimum, thus denoting a greater emphasis on the environmental impact.

The main data provided by the program are:

- Starting value of the functional 1.5422

- Target value of the functional 1.1320

The central child restraint has been set at 38 dBm power and minimum tilt = l to optimize the other indicators, it will be seen during test 5 that by further weighing the environmental margin with Weight = 4, the minimum power on the central child restraint is obtained.

Figures 160 to 169 show all the coverage maps and graphs that indicate the good performance of the individual functionals. From figure 161 relating to distances, it is clear that the maximum coverage distance of each child restraint is on average 1200 meters.

- This distance is optimal because as seen on the traffic chapter it allows the best efficiency of the Cluster as T raffìco / MHz / Raggiocel la.

- On the chapter of the measures it was possible to observe that in urban areas, a maximum coverage distance of 1.1 Km, allows the Power Control performance the best dynamics in controlling the power transmitted both by the SRB and by the mobile phone that is directly in contact with the human body.

- A lower transmitted power implies a lower environmental impact and a lower interference therefore a higher quality of the service from the customer point of view.

Figure 167 reports the movement of the coordinates of the RBS caused by the genetic algorithm during its evolution for the search for the optimal solution.

The graphs that are shown later 168 and 169 highlight the goodness of the solution found: they have on average a downward trend as required by the minimization itself.

It is interesting to note the difference of the single functionals seen in chapter 7 which have a monotonous decreasing trend, that is, during the minimization they have no growth points but only decrease or stationary points.

In the case of the functionals combined with the percentage normalization shown, we have that the single functionals can in some instant even worsen, in favor of the improvement of some other functional.

This observation is fundamental as the pure sum of the functionals without any kind of normalization had 2 negative effects:

- The functional with the highest value dominated the others;

- Even multiplying the various functionals by suitable constants in order to balance their values, it was always obtained that the functional with the greatest range of variation, that is, the one that varied faster over time, ended up prevailing over the others.

The minimization adopted on the percentage of distance of the functional from its minimum value, allows, if the normalization occurs on the MAX-MIN difference value of the functional itself, to make all the functions considered progress more or less in the same way.

The multiplicative constants inserted in the final sum of the functionals are further important as they allow to enhance the properties of one functional over another.

In this way it is possible to study the ideal characteristics of a GSM network from separate points of view or from the point of view of the emphasized functional.

8.5 TEST 4 WEIGHT OF UNIT FUNCTIONALS

The test was performed by setting the multiplicative constants of the functionals to the unit value.

It was found that the resulting arrangement of the SRJBs is optimal for the quality of the service.

One child restraint was positioned in the center of gravity of the traffic source, the other child restraints were evenly distributed around the central area.

The environmental margin has improved compared to the situation in which greater weight was given to the coverage (the RBSs were all concentrated in the central area), and it has worsened compared to the test where greater emphasis was given to the environmental margin.

The main data provided by the program are:

Powers are not set to maximum as for the test made by emphasizing coverage.

Figures 170 to 175 show the arrangement of the child restraints. From the distribution of the distances it emerges that this is the best configuration to optimize the quality of the service and at the same time to obtain a good environmental margin:

- The coverage distance of each cell is max 1200 m, this is the optimal distance seen both in the project for the best efficiency on the traffic Cap 4, and for the performance of the Power Control in an urban environment.

- The Power Control allows, even if it is not considered in the simulation, to transmit at minimum power when the mobile phones are close enough to the SRB; in the calculation made this max distance is about 1.1 km, this artifice produces two fundamental consequences:

1. Optimization of the environmental margin not only by the SRB but also by the mobile phone which is in direct contact with the human body;

2. Reduction of interference in other points of the network.

Figure 175 shows the general situation of the urban area considered:

The program did not locate any child restraints in the forbidden zone.

Figure 176 shows the movement of the RBS caused by the search for the best solution during the evolution of the genetic algorithm.

As already mentioned in the previous chapter, this figure shows only the latest iterations, in fact at the end of the calculations the program allows you to advance step by step in the representation of the solutions, this gives the possibility to see how the images for the frames of a film in sequence that describe the trajectories produced by the movement of the RBS during the evolution of the iterations relating to the genetic algorithm.

The graph in figure 177 shows the evolution of the total functional: 8.6 TEST 5 WEIGHT OF THE FUNCTIONAL ENVIRONMENTAL MARGIN PLACED A 4

From this test it was observed that a child restraint is located in a central position with respect to the area with greater environmental impact and greater coverage, traffic and, C / I.

The other child restraints are symmetrically arranged externally to the most critical area for environmental impact.

The data provided by the program are:

The SRB 2 placed in the center, as can be seen, has minimum power and minimum tilt.

The tilt has been set to a minimum to ensure maximum coverage, maximum performance on traffic; maximum C / I of the central area. The powers set in this way guarantee an excellent environmental margin at a price, however, of less coverage.

The central location of the RBS near the traffic source allows for better power management through the performance of Power Control.

This finish allows the reduction of environmental impact and interference.

Figures 178-187 show the coverage graphs and maps, while figures 187 and 188 show the correlations with the other functionals.

It can be noted that contrary to the case with weight 1.5, in this case the coverage is penalized more than the environmental margin which improves, moreover both the traffic, the yield and the C / I improve.

ADVANTAGES AND INDUSTRIALITY OF THE FOUND

The design of the network taking into account the features described leads to reducing interference, the environmental impact and therefore providing a qualitatively optimal service.

Normally when a mobile radio network is born it is not always possible to follow precise project criteria both for economic reasons and for environmental or urban constraints, usually a capillarity of the network is achieved over time by distributing the investments, and following the growing traffic.

The technique described in the present description allows to frame the problem of the design or optimization of a GSM mobile radio network from different points of view. The essential aspect that emerged is that the best location of the child restraints occurs by locating the child restraints themselves in the vicinity of the coverage and traffic objectives set for that child restraint.

This observation, which confirmed the perception that had been developed in the workplace and that had not previously found rationalized implementation procedures, perfectly reconciles the quality requirements of the service offered with the need to optimize the environmental margin.

The method applied to real traffic cases can make use of a cartographic map and a more sophisticated electromagnetic engine.

Claims (15)

CLAIMS 1) Design and / or optimization technique of the GSM radio access network characterized by the fact of classifying and evaluating the objectives and constraints involved in the design of the radio base station, such as: - the maximization of the relationship between useful signal and interference. - the maximum reduction of the distance between the base radio station and the traffic source. - the maximum reduction of the presence of electromagnetic field, - the maximization of the cell yield based on the traffic / cost ratio, - maximizing the level of coverage, by means of an operating model that formalizes the relationship between the objective and the parameters on which it depends, based on the use of genetic algorithms for global optimization through a compromise criterion between the individual indices. 2) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claim 1 characterized by the fact of using means of representing the operation of a certain number of base radio stations as Best Server in regions enumerated with index "j" like the radio base stations themselves, these regions being in turn divided into smaller pixels with index "i" so as to discretize the electromagnetic field values in space with greater precision and at the same time provide the software means used with the input data relating to the expected traffic on the various pixels themselves. 3) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 2 characterized by the fact of using means of representing the operation of a certain number of base radio stations as Best Server in regions enumerated with index j like the radio base stations themselves, these regions being in turn divided into smaller pixels with index i defining: a- the evaluation of the boundaries between one region and another through the isolation points provided by two radio base stations adjacent to each other through the area within which the electromagnetic field provided by a radio base station predominates over the one provided from other radio base stations, b- the calculation based on the traffic model adopted, of the base radio stations and the number of carriers required for a given territorial distribution of mobile radio customers, c- the calculation for each single region of the number of pixels nj belonging to it, defining the real area of the region. 4) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by the fact of using the following minimizing functional for the maximum reduction of the environmental impact in a given reference area 5) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by using the following minimizing functional for the maximum reduction of the environmental impact in a single region of n pixels addressed by indices i, k based on the environmental impact layer q 6) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by using the following minimizing functional for the maximum reduction of the environmental impact to predetermined limits, by setting a threshold of the maximum electric field allowed on a given area by comparing the functional value obtained with this threshold and increasing only if the electric field on the pixel is greater than the threshold given as input the following assignment having been performed within the procedure that implements the summations: 7) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by the fact of implementing the minimization of the distances of the radio base station from the traffic sources, dividing the territory into regions equal to the number of radio base stations necessary, naming them with index "j", being each region divided into n pixels with index "i", within each of these regions, the functional that the algorithm must minimize, tending to locate the radio stations based as close as possible to the sources of telephone traffic and therefore closer to the pixels with increasing telephone traffic, said traffic T (i.j) weighted with three values / being supplied as input to the processing procedures by weighing the pixels on the basis of the low amount of traffic , medium, high, considered for each radio base station, and by obtaining the final result by weighing the distances with the traffic, with the reduction rre distances with higher traffic, based on the following two functions: 1. functional 2. function 8) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by implementing the C / T maximization by checking each pixel for each set of radio stations base located, since the area is divided into best server zones for each radio base station, within each of these the control of the C / I is carried out by checking that the interfering radio base stations arrive at the lowest possible levels, mutually evaluating that the radio base station serving on a region is not too much interfering on the pixels of the region that previously was the origin of the interference for the previous region, so that iteratively results that: and with the adopted matrix representation of a single region of (p, q) pixels: 9) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by the fact of implementing C / I maximization taking into account C / I thresholds imposed as limits beyond which does not go down so that during the check for each pixel, the functional does not increase, if it happens that the value of C / I is better than the minimum imposed or otherwise increases, and providing the following functional functional 10) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by implementing the maximization of the coverage on pixels using the contribution of the terms with greater weight in the calculation of the value of the functional, i.e. to obtain a larger functional in absolute value, ensuring that the genetic algorithm arranges the radio base stations in such a way that the power received on the pixel with the greatest weight is as large as possible, evaluating the best server for each region of a radio base station, first the power on the pixel and then multiplying this power by the value of the weight assigned to the input matrix on the discretized coverage levels from which: 11) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by implementing the optimization of the minimum opening using the functional providing the functional increase only if the field level measured on the pixel is lower than the coverage level required on that ixel. 12) Technique of design and / or optimization of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by the fact of arranging the control and regulation of the transmission power within a given window using the functional ndo the latter relative to the average power in a given propagation action obtained by means of an average over the distance Design and / or optimization technique of the access network or GSM based on a genetic algorithm according to claims 1 and 3 characterized by the fact of implementing the maximization of the yield on the telephone traffic offered, considering constant with each system and independent of the same from the point location, by identifying the absolute minimum value through the functional li Project and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claims 1 and 3 characterized by the fact of maximizing the yield on the telephone traffic offered, considering the costs of each system to be constant and independent themselves from the point of location, considering the minimum value with respect to a traffic threshold 15) Design and / or optimization technique of the GSM radio access network based on a genetic algorithm according to claim 1 characterized by the fact of representing five indexes to be optimized on a pentagonal graph, from which the largest area of the figure is obtained. interior of the pentagon corresponding to the best quality of the solution found for the network project / optimization, and by identifying the maximum and minimum values of each functional, scanning one functional at a time and also obtaining the maximum and minimum assumed by the other functionals, to construct the pentagonal graph on which for each normalized functional it is possible to observe the variation assumed by the remaining functionals based on the formula: Y <f> d x <■> ^ ^ c or 16) Technique of p GSM radio based on a genetic algorithm according to claims 1 and 15 characterized by the fact of operating a combination of the sum of the listed functionalities, each functional being multiplied by a possible constant in order to amplify its effect on the location of the base radio stations for which: And terwork of functionals.