WO1999018748A1 - Neural network application for frequency planning - Google Patents

Neural network application for frequency planning Download PDF

Info

Publication number
WO1999018748A1
WO1999018748A1 PCT/SE1997/001669 SE9701669W WO9918748A1 WO 1999018748 A1 WO1999018748 A1 WO 1999018748A1 SE 9701669 W SE9701669 W SE 9701669W WO 9918748 A1 WO9918748 A1 WO 9918748A1
Authority
WO
WIPO (PCT)
Prior art keywords
base
frequency
frequencies
signal strength
procedure according
Prior art date
Application number
PCT/SE1997/001669
Other languages
French (fr)
Inventor
Robert Frank
Jan Loven
Per Hallkvist
Original Assignee
Telia Ab
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from SE9603557A external-priority patent/SE514246C2/en
Application filed by Telia Ab filed Critical Telia Ab
Priority to PCT/SE1997/001669 priority Critical patent/WO1999018748A1/en
Publication of WO1999018748A1 publication Critical patent/WO1999018748A1/en

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W16/00—Network planning, e.g. coverage or traffic planning tools; Network deployment, e.g. resource partitioning or cells structures
    • H04W16/18—Network planning tools

Definitions

  • Procedure to, in a cellular mobile telephone system find the cells which constitute potential interferers, i.e. which might interfere if they were allocated unsuitably selected frequencies at planning, by a calculation/estimation .
  • Radio network construction includes the work to define a set of base station locations, cells, radio channels, frequencies, cell parameters etc. These shall give radio coverage within the wanted area with sufficient quality och capacity at lowest longterm cost. Cell planning is further a repetetive task, with a big streak of practical trials and corrections. Among the most important elements are radio propagation calculations and measurements, traffic density estimations and measurements, and frequency planning och setting of parameters. When regarding a number of radio transmitters scattered over a certain geographical area, and wanting to measure in a certain point the signal strength over the spectrum the radio transmitters are using, there will arise certain problems. The principal problem which then will arise is to investigate which transmitters that, over a certain threshold level, have contributed to the measured signal strength.
  • the frequencies for which an operator has licence shall be allocated to each cell in a way that the interference between the cells become acceptable.
  • the problem then is to, for a given cell, find the cells which constitute potential interferers, i.e which might interfere if they were allocated unsuitably selected frequencies.
  • the procedure gives more efficient ways of minimizing the interference between the cells which in its turn results in a better way of allocating frequencies to individual cells.
  • FIG. 1 illustrates the principle according to the algorithm.
  • Figure 2 shows examples with the bases A, B, C and C marked.
  • Figure 3 shows a case which shows that the answer depends on what the exclusions between the bases look like.
  • Figure 4 shows the probability for that the weight from a certain base (type C in Figure 2 above) to another base (type A in Figure 2 above) is positive - in spite of that base C does not interfere with base A, i.a. that an superfluous exclusion is obtained.
  • the values of the x-axis is the number of frequencies which are measured at the base A.
  • the method is implemented in a way that the signal strength in uplink is measured over the frequency spectrum which is used by the operator.
  • the measurements are made from the base in the cell to which one wants to find the other cells which interfere or which are potential interferers.
  • FAS Frequency Allocation Support
  • the task of the algorithm is to, on the basis of the measurement result and information about which frequencies respective base has been allocated, decide from where the measured signal strength emanates. In that way the result might be used to decide exclusion weights in MACO, which is a frequency planning tool .
  • the number N is equal to the number of bases in the network.
  • Figure 1 illustrates the situation. In principle there is a weight from each node in the left column to each node in the right column and vice versa. All weights are, however, for the sake of clarity, not drawn.
  • the weights w._ are updated according to the following (so called Hebbian learning rule) :
  • base D interferes with base A
  • base C is such that C does not interfere with A
  • Base A will register signal strength on the frequencies which B is using. That means that the weight from B to A will increase its value. How rapidly the value will increase is decided by the parameter ⁇ according to the relation (1) .
  • the weight from C to A will wrongly (!) increase by ⁇ . It will be compensated to a more or less extent - depending on the value of the parameter ⁇ according to relation (1) - by the "downcounting" of the weight which is obtained when C uses a frequency which is not measured by base A.
  • the risk of a wrong exclusion to a certain base depends, according to the above described, on the number of frequencies which are measured at the base.
  • the following assumption is made for the analysis model : the frequencies to each base are allocated at random and independent of each other. A lover of law and order will put the question how this assumption (which of course results in a somewhat playful frequency plan) affects the analysis compared with a realistic planning of the frequencies.
  • the frequencies which are measured at U are, according to assumption, A's frequencies.
  • all other bases B, C, D(7) can have the same frequency as A, and there is a risk that superfluous exclusions will occur.
  • none of the bases B, C, D and E will have the same frequency as A and therefore there is no risk of getting a wrong exclusion to U from these.
  • F it is different.
  • the number of possible frequencies to allocate is fewer than at a random allocation. (Random allocation here means that all frequencies can be used for a base independent of which frequencies that have been allocated to other bases) . Because there are fewer possible frequencies to allocate, the risk that the frequencies that are allocated will be the same as at A will increase. The conclusion is that the risk of getting a wrong exclusion from F to U will be bigger at a correct frequency planning than at the one at random.
  • n freq the number of frequencies that are used in the network.
  • n m the number of frequencies on which signal strength are measured.
  • TRX the number of TRX:es on the base towards which there can be obtained a wrong exclusion.
  • the probability of getting a superfluous exclusion is by definition the same as obtaining a positive value of the weight that represents the exclusion.
  • the value of the weight can be regarded as the condition in a Markov chain where the jump probabilities are decided by (2) and (1) .
  • the number of conditions in the chain are decided by the number of decimals in ⁇ w ⁇ 3 .
  • the probability of getting a superfluous exclusion then can be expressed as the probability of being in the conditions in the Markov chain which correspond to positive values of the weight.
  • the above shown formula gives, given frequencies, number of measured frequencies which have been allocated a base and the total number of frequencies that are used, the probablility of how many frequencies at the base that are found among the measured ones. Of course it is here supposed that the bases do not interfere with each other. What shall be calculated is the probability of getting a wrong exclusion. I Figure 4 the result is shown.
  • the graph shows the probability of the weight from a certain base (type C in Figure 2 above) to another base (type A in Figure 2 above) being positive - in spite of that base C does not interfere with base A, i.e. that a superfluous exclusion is obtained.
  • the values of the x-axis is the number of frequencies which are measured at the base . It should be noted that the superfluous exclusion which can occur is not of statistical kind.
  • the values of the weights are changing continuously according to (1) at each change of frequencies (at least on principle, but is evident that the values of the weights should in some way be restricted. In the analysis in question the restriction was made that the values of all weights was between -1 and 1) .
  • the probability shall be associated with the occurrence that there at a certain change of frequencies exists a superfluous exclusion from base C to base A.

Abstract

Procedure to, in a cellular mobile telephone system, find the cells which constitute potential interferers, i.e. which might interfere if they are allocated unsuitably selected frequencies at planning. A calculation/estimation is made by signal strength in uplink being measured over the frequency spectrum which is used by a certain operator, at which a neural algorithm is used to, on the basis of the measurement result and which frequencies respective base has been allocated, decide from where the measured signal strength emanates.

Description

TITLE OF THE INVENTION
Neural network application for frequency planning
TECHNICAL FIELD
Procedure to, in a cellular mobile telephone system, find the cells which constitute potential interferers, i.e. which might interfere if they were allocated unsuitably selected frequencies at planning, by a calculation/estimation .
PRIOR ART
In the patent documentation there are a lot of documents which describe frequency planning. For instance is in the Swedish patent application 8900742-1 shown a method for frequency planning.
Further it is generally known to, by theoretical models, achieve a good frequency planning and for a given cell find the cells which constitute potential interferers in a cellular mobile telephone system, i.e. those which should interfere if they were allocated unsuitably selected frequencies .
TECHNICAL PROBLEM
Cell planning, or radio network construction, includes the work to define a set of base station locations, cells, radio channels, frequencies, cell parameters etc. These shall give radio coverage within the wanted area with sufficient quality och capacity at lowest longterm cost. Cell planning is further a repetetive task, with a big streak of practical trials and corrections. Among the most important elements are radio propagation calculations and measurements, traffic density estimations and measurements, and frequency planning och setting of parameters. When regarding a number of radio transmitters scattered over a certain geographical area, and wanting to measure in a certain point the signal strength over the spectrum the radio transmitters are using, there will arise certain problems. The principal problem which then will arise is to investigate which transmitters that, over a certain threshold level, have contributed to the measured signal strength. At frequency planning of cellular networks the frequencies for which an operator has licence shall be allocated to each cell in a way that the interference between the cells become acceptable. The problem then is to, for a given cell, find the cells which constitute potential interferers, i.e which might interfere if they were allocated unsuitably selected frequencies.
TECHNICAL SOLUTION
The technical solution is described by what is indicated in the patent claims.
ADVANTAGES
The procedure gives more efficient ways of minimizing the interference between the cells which in its turn results in a better way of allocating frequencies to individual cells.
DESCRIPTION OF DRAWINGS
The invention now will be described by means of not restricted examples of embodiment and with reference to enclosed, schematically made drawings. Among the drawings:
Figure 1 illustrates the principle according to the algorithm.
Figure 2 shows examples with the bases A, B, C and C marked.
Figure 3 shows a case which shows that the answer depends on what the exclusions between the bases look like. Figure 4 shows the probability for that the weight from a certain base (type C in Figure 2 above) to another base (type A in Figure 2 above) is positive - in spite of that base C does not interfere with base A, i.a. that an superfluous exclusion is obtained. The values of the x-axis is the number of frequencies which are measured at the base A.
DETAILED DESCRIPTION The method is implemented in a way that the signal strength in uplink is measured over the frequency spectrum which is used by the operator. The measurements are made from the base in the cell to which one wants to find the other cells which interfere or which are potential interferers.
With a special functions FAS (Frequency Allocation Support) it will be possible to, from a base (MSC) , measure signal strength in uplink from mobiles which are not camping at the own base. (The function makes use of empty time slots for these measurements) .
The result from FAS will be that for each base is measured signal strength in uplink over the frequency spectrum.
The task of the algorithm is to, on the basis of the measurement result and information about which frequencies respective base has been allocated, decide from where the measured signal strength emanates. In that way the result might be used to decide exclusion weights in MACO, which is a frequency planning tool .
The algorithm - which is popularly called NENAF (Neural Network Application for Frequency Planning) - is based on, say 2xN nodes, each connected with weights, say wAj , where i = 1 , 2 , 3 . . . , N and j= 1,2,3...,_.. The number N is equal to the number of bases in the network. Figure 1 illustrates the situation. In principle there is a weight from each node in the left column to each node in the right column and vice versa. All weights are, however, for the sake of clarity, not drawn.
Let {u and {v.} represent the conditions {0,1} in the nodes on the left respective right side in Figure 1. (i = 1,2,3...,JV and j= 1,2,3...,N as before and in the following if nothing else is indicated) .
For each frequency, say f, which is used in the frequency plan the following steps are gone through:
• When base number i transmits on frequency f, then u. is set to one, otherwise to zero.
• When base number j measures signal strength on frequency f (above a certain threshold level) then v. is set to one, otherwise to zero.
• The weights w._ are updated according to the following (so called Hebbian learning rule) :
wu = wu + ΔW13 (1) ι; =_ 2v_ _ βux (1 - Vj ) > 0 , β> 0
It now is evident that if base number i transmits on frequency f (ux=l) and base number j measures signal strength on the same frequency (v.=l) , then the weight w-_ will increase its value by α . If base number i transmits on frequency f and base number j does not measure signal strength on the frequency (v.=0) , then w will decrese its value by β. In other cases the value of the weight will be unchanged. What has been described above is the principle for the algorithm. Now an analysis of its application will follow.
Consider Figure 2 with the bases A, B, C and D marked. We want to decide exclusions from the base A to surrounding bases. A number of bases located outside the circumference of a circle can be directly excluded since the geographical distance is too large, type base D.
Suppose that base D interferes with base A, whereas base C is such that C does not interfere with A.
Base A will register signal strength on the frequencies which B is using. That means that the weight from B to A will increase its value. How rapidly the value will increase is decided by the parameter α according to the relation (1) .
For the exclusion between base A and C the situation is more difficult in following way: According to assumption, C does not interfere with base A, but it might be a possibility that one or more of the frequencies which C is using is measured by base A - for instance depending on that B and C has one or more frequencies in common. The more frequencies that are measured at A , the greater the probability that one or some of the frequencies on which base C is transmitting will be part of the measured ones.
For each of C's frequencies that are part of the mesasured ones, the weight from C to A will wrongly (!) increase by α. It will be compensated to a more or less extent - depending on the value of the parameter α according to relation (1) - by the "downcounting" of the weight which is obtained when C uses a frequency which is not measured by base A. By the above described can be realized that the risk of getting a wrong exclusion from C to A exists and depends on the number of frequencies which are measured at A and the values of the parameters α and β.
The risk of a wrong exclusion to a certain base depends, according to the above described, on the number of frequencies which are measured at the base. In order to calculate the risk, the following assumption is made for the analysis model : the frequencies to each base are allocated at random and independent of each other. A lover of law and order will put the question how this assumption (which of course results in a somewhat bizarre frequency plan) affects the analysis compared with a realistic planning of the frequencies.
A case which shows that the answer depends on what the exclusions between the bases look like is shown in Figure 3.
Suppose that the drawn curves (exclusions) are the correct ones. We now compare the risk of getting a wrong exclusion from one of the other nodes to U for the two cases, the first one of which is when the frequency allocation is made quite randomly, and the second according to a frequent frequency planning tool, where values of exclusion weights are allocated, called MACO.
The frequencies which are measured at U are, according to assumption, A's frequencies. At a random allocation all other bases (B, C, D...) can have the same frequency as A, and there is a risk that superfluous exclusions will occur. At a correct frequency planning none of the bases B, C, D and E will have the same frequency as A and therefore there is no risk of getting a wrong exclusion to U from these. With F it is different. Because F is not allowed to have the same frequencies as the bases towards which F has exclusion, the number of possible frequencies to allocate is fewer than at a random allocation. (Random allocation here means that all frequencies can be used for a base independent of which frequencies that have been allocated to other bases) . Because there are fewer possible frequencies to allocate, the risk that the frequencies that are allocated will be the same as at A will increase. The conclusion is that the risk of getting a wrong exclusion from F to U will be bigger at a correct frequency planning than at the one at random.
The moral of the above is that the results from the analysis should be taken with a pinch of salt. Depending on what the exclusions between the bases look like, the result of the analysis model - for instance the probability of getting exclusions which are superfluous - will differ from the practically implemented use of the algorithm.
In certain cases of use there will in all probability be nodes of the type F above, where the probability of a superfluous exclusion is bigger than the one which is given by the model. But it does not mean than the model is unusable, even if it will turn out that the probability can be high for occasional weights. The question is in such a case how many superfluous exclusions that are obtained in relation to the total number of exclusions that exist, and which proportion between these that can be accepted.
Given the indications below and random frequency planning the probability of, say X, frequencies among the measured ones of (2) below. (The basic assumption is of course as before that the base, the frequencies of which can be found among these that are measured, do no interfere with the base which is measuring) .
Figure imgf000010_0001
where the indications have the following meanings:
nfreq : the number of frequencies that are used in the network.
nm : the number of frequencies on which signal strength are measured.
nTRX : the number of TRX:es on the base towards which there can be obtained a wrong exclusion.
The probability of getting a superfluous exclusion is by definition the same as obtaining a positive value of the weight that represents the exclusion.
The value of the weight can be regarded as the condition in a Markov chain where the jump probabilities are decided by (2) and (1) . (The number of conditions in the chain are decided by the number of decimals in Δ w ι3) • The probability of getting a superfluous exclusion then can be expressed as the probability of being in the conditions in the Markov chain which correspond to positive values of the weight. The above shown formula gives, given frequencies, number of measured frequencies which have been allocated a base and the total number of frequencies that are used, the probablility of how many frequencies at the base that are found among the measured ones. Of course it is here supposed that the bases do not interfere with each other. What shall be calculated is the probability of getting a wrong exclusion. I Figure 4 the result is shown. The graph shows the probability of the weight from a certain base (type C in Figure 2 above) to another base (type A in Figure 2 above) being positive - in spite of that base C does not interfere with base A, i.e. that a superfluous exclusion is obtained. The values of the x-axis is the number of frequencies which are measured at the base . It should be noted that the superfluous exclusion which can occur is not of statistical kind. The values of the weights are changing continuously according to (1) at each change of frequencies (at least on principle, but is evident that the values of the weights should in some way be restricted. In the analysis in question the restriction was made that the values of all weights was between -1 and 1) . The probability shall be associated with the occurrence that there at a certain change of frequencies exists a superfluous exclusion from base C to base A.
The invention is not restricted to the above indicated but may be varied within the frame of the patent claims.

Claims

PATENT CLAIMS
1. Procedure to, in a cellular mobile telephone system, find the cells which constitute potential interferers, i.e which might interfere if they were allocated unsuitably selected frequencies at planning, by a calculation/estimation, c h a r a c t e r i z e d in that the calculation/estimation is made by signal strength in uplink being measured over the frequency spectrum which is used by a certain operator, that a neural algorithm is used to, on the basis of the measurement result and which frequencies respective base has been allocated, determine from where the measured signal strength emanates.
2. Procedure according to patent claim 1, c h a r a c t e r i z e d in that the measurements are made from the base in the cell to which one wants to find the other cells which interfere or which are potential interferers .
3. Procedure according to patent claim 1, c h a r a c t e r i z e d in that empty time slots are utilized for the measuring.
4. Procedure according to patent claim 1, c h a r a c t e r i z e d in that the neural algorithm is based on, say 2xN, nodes each connected with weights, say ij , where i = 1,2,3,...,-. and j = 1,2 ,3 , ... ,N. The number N is equal to the number of bases in the network.
Let {u. and {v┬▒} represent the conditions {0,1} in the nodes on the left respective right side in Figure 1. For each frequency, say f, which is used in the frequency plan the following steps are gone through: ΓÇó When base number i transmits on frequency f, then u. is set to one, otherwise to zero.
ΓÇó When base number j measures signal strength on frequency f (above a certain threshold level) then vx is set to one, otherwise to zero.
ΓÇó The weights w.. are updated according to the following (so called Hebbian learning rule) :
Figure imgf000013_0001
Δw13 =au1vJ _ βu1(l-vJ) a>0,β>0
5. Procedure according to patent claim 4, c h a r a c t e r i z e d in that the measured signal strength is shown on one to the base station connected equipment .
6. Procedure according to patent claim 1, c h a r a c t e r i z e d in that the measurement result is used to allocate values of the exclusion weights which are used in the frequency planning tool MACO.
PCT/SE1997/001669 1996-09-27 1997-10-06 Neural network application for frequency planning WO1999018748A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/SE1997/001669 WO1999018748A1 (en) 1996-09-27 1997-10-06 Neural network application for frequency planning

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
SE9603557A SE514246C2 (en) 1996-09-27 1996-09-27 Cellular mobile telephone system operation method
PCT/SE1997/001669 WO1999018748A1 (en) 1996-09-27 1997-10-06 Neural network application for frequency planning

Publications (1)

Publication Number Publication Date
WO1999018748A1 true WO1999018748A1 (en) 1999-04-15

Family

ID=26662762

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/SE1997/001669 WO1999018748A1 (en) 1996-09-27 1997-10-06 Neural network application for frequency planning

Country Status (1)

Country Link
WO (1) WO1999018748A1 (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001033882A1 (en) * 1999-11-04 2001-05-10 Telediffusion De France Method for validating frequency plan allocation of telecommunication networks

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4736453A (en) * 1985-12-10 1988-04-05 Schloemer Gerald R Method and apparatus for making frequency channel assignment in a cellular or non-cellular radiotelephone communications system
EP0631453A2 (en) * 1993-06-21 1994-12-28 Telia Ab Method for locating mobile stations in a digital telephone network
US5434950A (en) * 1992-04-13 1995-07-18 Televerket Method for making handover decisions in a radio communications network
US5442804A (en) * 1989-03-03 1995-08-15 Televerket Method for resource allocation in a radio system
WO1997033394A1 (en) * 1996-03-08 1997-09-12 Watkins-Johnson Company Wireless communication system with dynamic channel allocation

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4736453A (en) * 1985-12-10 1988-04-05 Schloemer Gerald R Method and apparatus for making frequency channel assignment in a cellular or non-cellular radiotelephone communications system
US5442804A (en) * 1989-03-03 1995-08-15 Televerket Method for resource allocation in a radio system
US5434950A (en) * 1992-04-13 1995-07-18 Televerket Method for making handover decisions in a radio communications network
EP0631453A2 (en) * 1993-06-21 1994-12-28 Telia Ab Method for locating mobile stations in a digital telephone network
WO1997033394A1 (en) * 1996-03-08 1997-09-12 Watkins-Johnson Company Wireless communication system with dynamic channel allocation

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2001033882A1 (en) * 1999-11-04 2001-05-10 Telediffusion De France Method for validating frequency plan allocation of telecommunication networks
FR2800964A1 (en) * 1999-11-04 2001-05-11 Telediffusion Fse PROCESS FOR VALIDATION OF ALLOCATION OF FREQUENCY PLANS OF TELECOMMUNICATION NETWORKS

Similar Documents

Publication Publication Date Title
JP2976211B2 (en) Method for designing a wireless cell
US7212822B1 (en) Method and techniques for penalty-based channel assignments in a cellular network
US6990348B1 (en) Self-configuring wireless system and a method to derive re-use criteria and neighboring lists therefor
US6223031B1 (en) Process and device for quality determination
CN108781428B (en) Method for calculating distance between base stations, storage medium, base station and cellular network
KR100657369B1 (en) Apparatus and method for collection and analysis of wireless signal propagation in a wireless system
US6230016B1 (en) Apparatus and method for adaptive-dynamic channel assignment in wireless communication networks
EP0461176B1 (en) Method for resource allocation in a radio system
US20030129987A1 (en) System and method for frequency planning in wireless communication networks
US20040203727A1 (en) Quality-based optimization of cellular network parameters
EP0926912B1 (en) Method for channel assignment in a mobile communication system with space division multiple access
WO1994005097A1 (en) Method for estimating c/i density and interference probability in the uplink
US20140073316A1 (en) Method for interference & congestion detection with multiple radio technologies
EP0715479A1 (en) Channel habitat segregation method
US10219182B1 (en) Channel bandwidth selection
Onoe et al. Flexible re-use for dynamic channel assignment in mobile radio systems
WO1995034957A2 (en) Adaptive frequency allocation for broadcast control channels
US6728544B1 (en) Methods and techniques in channel assignment in a cellular network
WO1999018748A1 (en) Neural network application for frequency planning
US6859486B1 (en) Method for predicting interference in a frequency hopping cellular communications system
CN107231653B (en) Weighted graph modeling method for channel reuse in wireless cellular network
US6658256B2 (en) Method for allocating a channel in a cell of cellular network
US11876572B2 (en) Wireless telecommunications network
EP1252786B1 (en) Channel allocation in mobile communication system
EP1931166A1 (en) Call admission control in a mobile communication system

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A1

Designated state(s): NO US

AL Designated countries for regional patents

Kind code of ref document: A1

Designated state(s): AT BE CH DE DK ES FI FR GB GR IE IT LU MC NL PT SE

DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
121 Ep: the epo has been informed by wipo that ep was designated in this application
122 Ep: pct application non-entry in european phase