EP1371253A2

EP1371253A2 - Adaptive threshold scheme for differentiated system access

Info

Publication number: EP1371253A2
Application number: EP02752874A
Authority: EP
Inventors: Hans Kröner
Original assignee: Siemens AG; Nokia Siemens Networks GmbH and Co KG
Current assignee: Nokia Solutions and Networks GmbH and Co KG
Priority date: 2001-03-19
Filing date: 2002-02-19
Publication date: 2003-12-17
Also published as: ES2240265T3; WO2002076129A9; WO2002076129A3; WO2002076129A2; DE60110892D1; PT1244318E; EP1244318A1; DE60110892T2; EP1244318B1; ATE296017T1

Abstract

The invention discloses a method for granting access to a wireless or wireline communication network, especially a cellular mobile communication network, wherein the communication network is able to grant access when access is requested by a mobile station for a changeover of a service from an element of a communication network to another element of a communication network, wherein the communication network is able to grant access when access is requested by a mobile station for starting a new service, wherein a threshold value representing the ratio of resources, e.g. channels, available for hand-overs and of resources available for starting new services is set depending on measured quality of service and depending on a predetermined target quality of service for new services and for services for which a hand-over is requested. <IMAGE>

Description

Description:

"Adaptive Threshold Scheme for differentiated system access"

The invention concerns methods and devices for applying an adaptive threshold scheme for differentiated system access.

Various communication systems apply threshold policies for system access control: Y.C. Kim, D.E. Lee, B.J. Lee, Y.S. Kim, B. Mukherjee,

"Dynamic channel reservation based on mobility in wireless ATM networks", IEEE Communications Magazine, November 1999, pp. 7-51. [1] discloses a guard channel scheme for cellular communication systems: New calls are prohibited from enter- ing the mobile communication system if a certain system occupancy is exceeded, whereas handoff calls may still enter the cell if a channel can be granted. This scheme is used to prioritise handovers (=handoff calls) over new calls ensuring that the call dropping probability experienced by handoffs is smaller than the call blocking probability of new call arrivals .

K. W. Ross, "Multiservice Loss Models for Broadband Telecommunications Networks", Springer Verlag, 1995 [2] discloses admission control strategies for multi-service networks: In multi-service networks such as ISDN, ATM or future UMTS networks, which support several services with different bandwidth requirements, if no additional access control is used, e.g., a call with high bandwidth requirements is blocked with a much higher probability than a call with lower bandwidth requirements. Therefore, various schemes have been proposed, which define individual load or bandwidth thresholds for different service classes (in [2] these schemes are called par- MISSING AT THE TIME OF FILING

classes. However, a common problem for all these mechanisms is the dimensioning of the threshold levels to achieve a certain system behaviour, because the threshold values are highly dependent on the traffic mix and the call/handoff/packet arrival process. Moreover, the traffic mix, the user mobility and the call/handoff/packet arrival process and burstiness depends on various factors such as time of the day and the piace/environment (temporal and spatial variations), which makes it very difficult and ineffi- cient to adjust the thresholds in a proper way for all different situations.

It is an object of the invention to improve the system access scheme for new service requests and handover requests effi- ciently trying to avoid the said problems of known system access schemes as far as possible. This object is achieved by an invention as set forth in the independent claims. Granting access (to network resources) in accordance with the claim wording can comprise granting access to at least one network for a user of a mobile station requesting network access for e.g. a speech call or for transmission of data or for any multimedia service. A threshold may be e.g. a value representing the percentage of capacity (for which access can be granted) or the number of channels or other resources to be granted for requested new services (e.g. new calls)- which means that access to the capacity not granted to new services can be granted to services for which a hand-over is requested. A change-over may be a change-over from one element of a communication network to another element of this commu- nication network (e.g. a hand-over of a mobile station between base stations or adaptive antennas in GSM or CDMA etc) or a change-over from one element of a communication network to another element of an other communication network. New services according to the claim wording can be any services that are requested and have not started or for which services no or not yet all desired communication capacity has been granted- e.g. a new call. Granted resources or capacity or access to elements may e.g. comprise a time and/or frequency channel (e.g. a speech channel or a data transmission channel in GSM etc) and/ or code schemes for communication (e.g. in CDMA) . Access can comprise providing network resources (e.g. network resources represented by frequency and/or time and/or codes) for a mobile station- e.g. a speech channel.

The measured quality of service (= measured QoS level) can be any measured parameter indicating the level of the quality of service, e.g. the probability (measured rate) of blocking hand-over requests (and/ or dropping of calls during handover) and /or the probability (measured rate) of blocking requests for access for enabling a new service (e.g. not granting access to a speech channel) . The predetermined target quality of service is a level of service that is desired, e.g. the percentage of blocked/ dropped hand-over requests or the percentage of blocked requests for a new service. The threshold value can define a ratio (e.g. percentage) of resources available for hand-over-services and of resources available for new services- e.g. a threshold value of 60% indicating that 60% of the resources (e.g. time/frequency channels and/or codes) are available for services requesting hand-over and that 40% (=the other resources) are available for requested new services etc.

An adaptive adaptation scheme according to the invention can comprise the adjustment of the threshold values, which is directly based on the quality of service measure that is con- trolled by the underlying threshold scheme, i.e., this scheme alters the threshold (s) in response to a call/packet loss/delay. The basic scheme can be generally applied to loss as well as queuing systems, controlling loss, delay or throughput of two or more service classes with simple or more sophisticated controlling schemes.

The proposed adaptive threshold scheme according to the invention can provide the following benefits: • In the long-term it is able to provide exactly a predefined QoS level, which means that the mechanism is able to control directly the QoS parameters irrespectively of the arrival pattern, traffic fluctuations and burstiness. Therefore, the proposed scheme is very robust and is able to cope with various situations. In particular, it is able to dynamically adapt the system behaviour when the traffic composition and intensity varies due to temporal or spatial changes of the user activity.

• The dynamic approach provides superior performance (higher admissible system load), because the thresholds ■ will be dynamically optimised to a particular load situation taking into account the temporal and spatial variations of the traffic intensity and characteristics. A fix threshold scheme has to be dimemsioned according to the worst case thus providing a lower system load (otherwise, the QoS targets will be violated for the worst case scenarios) .

• It provides an easy to understand user interface (O&M parameter description) , because the user specifies the desired set of QoS parameters instead of the set of thresh- old values that fit to a certain set of QoS values, i.e., the user has an easy and direct interface to the QoS setting. It avoids expensive traffic engineering tasks for system dimensioning. • It provides a fast reaction to a QoS degradation, because the value K (see next section) is typically > 1. Therefore, only small deviations from the desired QoS level (e.g., packet loss rate, call loss rate) occur. In par- ticular, the scheme provides smaller variations of the QoS level than a fixed threshold scheme.

• The adaptation scheme is computationally simple, because the algorithm uses only additions/subtractions for the control of the threshold (no multiplications or divi- sions) . Therefore, the scheme is able to adapt the threshold value in real-time, even for a high-speed packet switched network.

Further advantages and details of the invention are set forth in the claims and the following description of an example of the invention.

Figure 1 shows the basic algorithm of an example of the invention. Figure 2 shows an adaptive guard channel scheme

Figure 3 shows the loss probability of handoff calls Figure 4 shows the loss probability of new call requests Figure 5 shows the mean value of the adaptive threshold Figure 6 shows application of the adaptive threshold scheme to a packet transmission buffer

Figure 7 illustrates the dropping probability of the Random Early Detection scheme

A basic idea behind the invention (i.e. of the Proposed Adaptive Threshold Scheme) is the adaptation of the threshold values in response to the measured quality of service. The basic mechanism that is used for the adaptation of the threshold values can be described as follows. The threshold will be adapted for each event (arrival of a call/packet, departure of a call/packet) that should be controlled by the threshold according to the following basic rules: • The threshold will be raised slightly by a value ? if the event indicates a better QoS than the target QoS level. • The threshold will be reduced by a value K? if the event indicates a worse QoS than the target QoS level.

The basic algorithm is illustrated in Figure 1. Furthermore, the value K should be chosen in such a way that a certain QoS level in terms of call/packet loss probability, packet delay quantile, call load or packet throughput can be maintained. This leads to the following balance condition for the deriva- tion of K:

Number of events with a QoS better than QoS target * ? = number of events with QoS worse than QoS target * K?

This leads to

K = number of events with a QoS better than QoS target / number of events worse than QoS target.

Exemplary Applications of the proposed Scheme: Since this algorithm has been defined in a rather abstract way, we will provide a number of concrete examples for an application of the proposed scheme. In general it can be used to restrict the access to a loss system or to a queueing system.

Application to Loss Systems: Application to the Guard Channel Scheme The first example application is the guard channel scheme used in cellular communications. The guard channel scheme reduces the call dropping probability encountered by handoff calls compared to the call blocking probability of new calls by comparing the number of occupied channels with a threshold value T when a new call arrives. The new call will be accepted when less or equal than T channels are occupied, whereas the admission will be denied when more than T channels are occupied. On the other hand, a handoff will be ac- cepted if one of the N channels (N>=T) is available. The threshold value T has to be dimensioned in such a way that the call dropping probability for handoff calls is lower than a certain value B_d (see, e.g., [1]). This can be achieved by the proposed threshold adaptation scheme if the threshold T is increased by a value ? if a handoff is admitted into the system (one of the N channels is available) , whereas the threshold T is decreased by K? if the handoff has to be denied, because all N channels are occupied. This basic system behaviour is described in Figure 2.

Additionally, it has to be taken into account that the threshold has to be kept between a minimum of 0 and a maximum of N. If the system is not in over- or underload, these maxi- mum and minimum values will not be reached and therefore, the following balance equation is valid:

number of denied handoffs * K? = number of admitted handoffs

* Δ

Furthermore, the dropping probability B_d for handoff calls is given by number of denied handoffs / (number of denied handoffs + number of admitted handoffs) which leads to the following relationship between B_d and K

B_d * K * Δ = (1 - B_d) * Δ

and therefore, to the following dimensioning rule:

K = (1 - Bd,target) / Bd,target-

Where B_d,target represents the target dropping probability. If the value K is chosen according to Equation (1) , it is guaranteed that a long-term blocking probability of B_d will be achieved given that the system is not in overload (T equals 0) or in underload (T equals N) . This system has been simulated with the assumptions that new and handoff calls are generated according to a Poisson process. Furthermore, the call holding time of new and handover calls is negative exponentially distributed. This model is identically to the model used in [1] . Further- more, a simple study has been carried out using the basic parameters used in [1] :

• Number of channels N: 60

• Target dropping probability B_d: 0.001

• Offered network load: 0.8 • Mobility ? of calls (arrival rate of handoffs/arrival rate of new calls) is variable

Figures 3 and 4 show the results obtained from the scheme for different values of .

The following conclusions can be drawn from the results: • The proposed scheme guarantees the required loss probability for handoff calls

• The proposed scheme provides superior performance compared to all schemes studied in [1] (see Figures 5 and 6 of the referenced paper)

The adaptation of the threshold can be seen in the following Figure, which depicts the average threshold vs. the mobility parameter α.

Furthermore, there have been additional investigations, which confirm that the proposed scheme is able to maintain the desired handoff blocking probability also for varying arrival rates (modelled by a Markov Modulated Poisson Process) leading to a very simple and robust control scheme. A simple modification of the proposed scheme is the control of the ratio of the call dropping and call blocking probabilities instead of the call dropping probability itself. This can be achieved if the following balance equation is used to design the control algorithm:

no. of handoffs lost / no. of handoffs arrived = loss_ratio * no. of new calls lost / no. of new calls arrived

no. of handoffs lost / no. of handoffs arrived * Δ = loss_ratio * no. of new calls lost / no. of new calls arrived

* Δ

no. of handoffs lost * (Δ * no. of new calls arrived / no. of handoffs arrived / loss ratio) = no. of new calls lost * Δ Therefore, increasing the threshold for each loss of a new call by an amount of Δ and decreasing the threshold for each loss of a handoff call by a value (Δ * no. of new calls arrived / no. of handoffs arrived / loss_ratio) provides the long-term desired loss_ratio between handoff loss probability (call dropping probability) and new call loss probability (call blocking probability) .

Finally, it has to be mentioned that all derivations and descriptions within this section have been derived on the basis of channels to get a clear interpretation. On the o.ther hand, a service may require a certain time slot, a certain bit rate or a certain code resource and in that case, the threshold is also given by a number of time slots, a bit rate or a cnumber of codes and the number of channels describes the number of time slots, the total bit rate of a radio cell or the number of code resources.

Application to Admission Control in Multi Service Loss Net- works:

Multi-Service loss networks such as UMTS, ISDN, ATM or IP networks offer a very unfair call admission to services with different resource (bit rate) requirements. For example, if all services have access to the complete pool of resources (this policy is called complete sharing) , a service with small resource (bit rate) requirements gets more often access to the system than a service with high resource (bit rate) requirements, i.e., the call loss probability is larger for high bit rate calls than for low bit rate calls. Additionally, this so-called complete sharing policy has the disadvantage that it may lead to a poor resource utilisation and it provides a poor revenue for different service classes with class-dependent revenue [2] .

Therefore, within the literature there exist a number of different admission policies such as the partitioning or thresh- old policies, which typically use individual thresholds for the different service classes. The presented adaptation scheme may be used to tune these threshold values in such a way that certain call loss probability targets are achieved in the system. If we consider two different service classes, where service- classes 1 and 2 require ai and a₂ resource- ■ • units (ai > a₂) , service class 2 will only get access to the system if the total number of occupied channels is below an appropriate threshold T. This threshold can be controlled with the previously defined algorithms to obtain a certain loss probability for traffic class 1 or a certain ratio of the loss probabilities of traffic classes 1 and 2. Simulations have confirmed that this scheme is able to maintain the desired QoS levels.

Application to Queueing Systems

Application to Buffer Access in Packet Data (ATM/IP) Networks

The proposed scheme may be also applied to queueing instead of loss systems. Usually, queueing systems are used in packet data networks, where the packets from different communications arrive randomly at a switch/router. The packets are stored in a buffer until they can be transmitted on the cor- responding outgoing line. In case of a network congestion the buffers will overflow and some packets will be lost. Since the services and applications have different sensitivity with respect to packet loss, different buffer access and transmis- sion line scheduling policies have been developed to tailor the packet handling to the QoS requirements of the underlying services and applications. One scheme proposed in [3] divides the buffer into several parts, which can be accessed only by a part of the different traffic classes. We will explain this scheme considering the scenario in Figure 6, where a high and a low priority packet class is distinguished. The high priority class is able to access the whole buffer, whereas a packet with low priority can only access the buffer if less than T buffer places (bytes, ATM cells, packets) are already occupied. Hence, the packet loss probability of the high priority traffic is smaller due to a reservation of N - T buffer places. The threshold T will be adapted like in the guard channel scheme in response to packet arrivals of the high priority traffic class:

• The threshold will be raised by a value of Δ, for each incoming high priority packet, which is able to access the system.

• On the other hand, the threshold will be lowered by a value of K Δ, when a high priority packet is lost.

Again, since for each increase of the threshold (lost packet) there have to be K decreases of the threshold (packet admitted) , the packet loss probability of the high priority pack- ets will converge to a value of 1/K+l.

A modification of the scheme would be again to control the threshold in such a way that the ratio of the packet loss probabilities of class 1 and class 2 packets is kept at a constant value. This is done in the same way as described earlier for the guard channel scheme.

Furthermore, instead of controlling the loss of packets it may be also desired to control the loss of information in terms of no. of bytes. In that case, the adaptive threshold scheme has to take into account the packet size for the control of the threshold.

Application within the Random Early Detection Scheme for TCP/IP Networks

A very powerful and popular means of congestion handling in TCP/IP networks is the Random Early Detection scheme used in IP routers [4] . This scheme randomly discards packets when the- average buffer' size exceeds a certain lower buffer threshold. The dropping probability increases linearly with the average buffer size until a maximum threshold is reached, where the dropping probability corresponds to max_p. If the average queue size exceeds the maximum threshold, then every packet will be dropped. This random dropping of packets controls the traffic generated at the source if the communication is controlled by the Transmission Control Protocol (TCP) . The basic scheme is illustrated in Figure 7.

The dimensioning of this scheme is not trivial, because it may suffer from the following problems [4] : • Setting of the minimum threshold such that it is high enough to guarantee a certain throughput and to avoid un- derutilisation. • Setting of the maximum threshold such that the difference between the maximum and the minimum value is_. Large enough to avoid global synchronisation and to avoid an oscillation of the average queue length. In that case, the value max_th will be exceeded regularly.

The control of the minimum threshold value should be based on the system throughput, whereas the control of the maximum threshold should be based on the number of times when the maximum threshold is exceeded. Therefore, min_th will be increased for each packet that leaves an empty queue, whereas it will be decreased for each packet that leaves an occupied queue. On the other hand, the maximum threshold will be increased if the average queue size is larger than max_th whereas it will be decreased if the average queue size is smaller than max_th. Using appropriate control parameters it is possible to adapt the threshold values to achieve a large system throughput and to avoid global synchronisation.

Summary of Application Examples, Options and further Extensions :

• The following is a non-exhaustive list of possible applications of the proposed threshold scheme:

1. Threshold adaptation of the guard channel scheme used for the prioritisation of handoff calls in mobile communication systems. The adaptation may be based on a tar- get loss probability for handoff calls or on the ratio of lost handoff calls and lost new calls (ratio of dropping and blocking probability) .

2. Threshold adaptation for the call/connection/bearer admission policies used for multi-service networks such as ISDN, ATM, IP or UMTS networks. The adaptation may be based on target loss probabilities of certain service classes, ratio of the loss probabilities of different service classes, revenue of service classes, throughput and so on. 3. Threshold adaptation of the buffer access schemes used in ATM or IP networks, which provide a service specific cell/packet loss probability. The adaptation may be per- formed based on cell/packet loss probabilities, delay quantiles, throughput and so on.

4. Threshold adaptation of random early discard mechanisms used in IP routers. The adaptation of the thresholds used by the random early discard mechanism.

5. Threshold adaptation of bandwidth allocation schemes of data services.

• The following QoS measures may be controlled by the pro- posed threshold scheme:

1. Call/packet/information loss probability of a certain traffic class

2. Ratio of call/packet/information loss probabilities of certain traffic classes 3. Carried load/throughput of a certain traffic class

4. Ratio of throughputs of certain traffic classes

5. Delay quantile (number of packets which have a delay larger than a certain value) of a traffic class

6. Ratio of delay quantiles of certain traffic classes 7. Revenue obtained by different traffic classes

• Finally, a number of further extensions/modifications of the basic scheme are:

1. Update of the threshold only if the difference between the old and the new threshold values exceeds a certain amount. This avoids too. frequent updates of the threshold.

2. Use of some filtering algorithms for the update of the threshold value (smoothing of the variations) . 3. Use of other more complex control algorithms, e.g., use of a proportional/integral/differential (PID) controller. Application of the scheme to more than two different traffic classes.

Claims

Claims:

1. Method for granting access to resources of a communication network, wherein the communication network is designed to grant access to network resources when access to the network resources is requested for a first category of services, wherein the communication network is designed to grant access to network resources when access is requested for a second category of services, wherein a threshold value defining the ratio of network resources available for the first category of service and of resources available for the second category of service is set depending on a measured quality of service for at least one category of service and depending on a predetermined target quality of service for at least one category of services.

2. Method preferably according to claim 1, for granting access to elements of a communication network, especially to a cellular mobile communication network or to an other wireless communication network, wherein the communication network is able to grant access to network resources when access is requested by a mobile station for a change-over of a service from an element of a communication network to another element of a communication network, wherein the communication network is able to grant access to network resources when access is requested by a mobile station for starting a new service, wherein a threshold value defining the ratio of network resources available for hand-over of services and of resources available for starting new services is set depending on a measured quality of service and depending on a predetermined target quality of service for new services and/or for services for which a hand-over is requested.

3. Method according to claim 1 or 2, characterised in that a service comprises voice transmis- sion or text transmission or video transmission or data transmission.

. Method according to any of the preceding claims, characterised in that if the measured quality of service of handovers (especially dropping probability) is worse than a predetermined target quality of service of handovers, the threshold value is modified in a way that results in granting more resources to hand-over of services and granting less resources to new services.

5. Method according to claim 4, characterised in that if the failure rate, especially dropping rate, for hand-over of services exceeds a level represented by the predetermined target quality of service for hand-over of service, the threshold value is modified in a way that results in granting more resources to handover of services and granting less resources to new services .

6. Method according to any of the preceding claims, characterised in that the available network resources comprise time- and/or frequency channels and/ or codes.

1 . Method according to any of the preceding claims, characterised in that it is used in the guard channel scheme of a cellular mobile network.

8. Method according to any of the preceding claims, characterised in that the measured quality of service is measured by detecting the percentage of blocked handover- requests and/ or by detecting the percentage of blocked requests for a new service.

9. Method according to any of the preceding claims, characterised in that the threshold value is modified only if the measured quality of service differs from the predetermined target quality of service by more than a preset amount or percentage.

10.Method according to any of the preceding claims, characterised in that measuring the quality of service comprises filtering for smoothing before comparing it to the predetermined target quality of service-. ■

11.Method according to any of the preceding claims, characterised in that a PID- controller is used for determining the threshold value.

12.Method according to any of the preceding claims, characterised in that it is used for more than one, preferably more than two traffic classes for services.

13.Method according to any of the preceding claims, characterised in that the network is an ISDN network.

14.Method according to any of the preceding claims, characterised in that the network is an ATM network.

15.Method according to any of the preceding claims, characterised in that the network is an Internet Protocol network.

16.Method according to any of the preceding claims, characterised in that the method is used for threshold adaption of a buffer access scheme of a network, especially for threshold adaption of buffer access scheme of an ATM or IP network, which provides a service specific cell or packet loss probability.

17.Method according to any of the preceding claims, characterised in that the method is used for the adaption • of a threshold of a random early discard mechanism, especially for a random early discard mechanism that can be used in an IP router.

18.Method according to any of the preceding claims, characterised in that the network is a cellular mobile network, especially a UMTS network.

19.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least the call or packet information loss probability of a certain traffic class.

20.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least the ratio of call or packet informa- tion loss probabilities of traffic classes

21.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by the carried load or throughout of a certain traffic class.

22.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least throughputs of certain traffic classes .

23.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least the delay quantile of a traffic class or by the number of packets which have a delay larger than a certain value.

24.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least the ratio of delay quantiles of certain traffic classes.

25.Method according to any of the preceding claims, characterised in that the measured quality of service is defined by at least the revenue obtained by certain traffic classes.

26.Method according to any of the preceding claims, characterised in that, when access to network resources is requested for a first category of services, the communication network grants access to all network resources, and that, when access is requested for a second category of services, the communication network grants access to a part network of the resources which part is defined by the threshold value.

27. Device for granting access to a communication network, especially a cellular mobile communication network, wherein the communication network is designed to grant access to network resources when access to the network resources is requested for a first category of services, . wherein the communication network is designed to grant access to network resources when access is requested for a second category of services, wherein a threshold value defining the ratio of network resources available for the first category of service and of resources available for the second category of service is set depending on a measured quality of service for at least one category of service and depending on a predetermined target quality of service for at least one category of services.

28.Device according to claim 27 for granting access to a cellular mobile communication network, which communication network is designed to grant access to network resources when access is requested by a mobile station for a change-over of a service from an element of a communication network to another element of a communication network, and which communication network is able to grant access to network resources when access is requested by a mobile station for starting a new service, wherein the device is able to define a threshold value representing the ratio of network resources available for hand-over of services and of resources available for starting new services depending on a measured quality of service and depending on a predetermined target quality of service for new services and/or for services for which a hand-over is requested.

29.Device according to claim 27 or28, characterised in that a service comprises voice transmis- sion or text transmission or video transmission or data transmission.

30.Device according to any of the preceding claims 27-29, characterised in that if the measured quality of service of handovers (e.g., dropping probability) is worse than a predetermined target quality of service of handovers, the threshold value is modified in a way that results in granting more resources to hand-over of services and granting less resources to new services.

31. Device according to claim 27, characterised in that if the failure rate, especially blocking rate, for hand-over off services exceeds a level represented by the predetermined target quality of service for hand-over of service, the threshold value is modified in a way that results in granting more resources to handover of services and granting less resources to new services .

32.Device according to any of the preceding claims 27-31, characterised in that the available network resources comprise time- and/or frequency channels and/ or codes.

33.Device according to any of the preceding claims 27-32, characterised in that the measured quality of service is measured by detecting the percentage of blocked handover- requests.

34.Device according to any of the preceding claims 27-33, characterised in that the measured quality of service is measured by detecting the percentage of blocked requests for a new service.

35.Device according to any of the preceding claims 27-34, characterised in that the threshold value is modified only if the measured quality of service differs from the predetermined target quality of service by more than a preset amount or percentage.

36.Device according to any of the preceding claims 27-35, characterised in that measuring the quality of service comprises filtering for smoothing before comparing it to the predetermined target quality of service.

37.Device according to any of the preceding claims 27-36, characterised in that a PID- controller is used for determining the threshold value.

38. Device according to any of the preceding claims 27-37, characterised in that it is used for more than one, preferably more than two traffic classes for services.

39.Device according to any of the preceding claims 27-38, characterised in that it comprises a controller in or for use in a communication network, especially cellular commu- nication network.

40. Device according to any of the preceding claims 27-39, characterised in that it is a controller in a base sta- tion.

41. Device according to any of the preceding claims 27-40, characterised in that, when access to network resources is requested for a first category of services, the communi- cation network grants access to all network resources, and that, when access is requested for a second category of services, the communication network grants access to a part network of the resources which part is defined by the threshold value.