CN1575614A - Method and system for optimising the performance of a network - Google Patents

Abstract

When optimising the performance of the network, first of all, the relevant key performance indicators for a specific entity within the network as well as first parameters, which influence the key performance indicators, are determined. A number of entities similar to said specific entity is selected, wherein relevant key performance indicators are associated to every en-tity. The key performance indicators as well as the selected number of en-tities are used as elements in a first cost function, i.e. said first cost func-tion is calculated on the basis of the KPI and the number of entities. Said first cost function is calculated in order to evaluate the network perform-ance. Accordingly, since said first parameters directly relate to the key per-formance indicators, the network performance will depend on the values of said first parameters. Thereafter the values of said first parameters are adjusted, so that a sec-ond set of values of said first parameters are obtained. The key perform-ance indicators are determined again but this time on the basis of the sec-ond values of said first parameters and said first cost function is re-calculated on the basis of these key performance indicators. The result of said first cost function calculated on the basis of said first values of said first parameters is compared to the result of said first cost function re-calculated on the basis of said second values of said first parameters. This comparison is carried out to determine whether the network performance has improved. When the network performance has improved due to the adjusting of said first parameters, said second values of said first parame-ters are adopted as permanent parameters.

Description

Method and system for optimizing network performance

technical field

The present invention relates to a method and system for optimizing network performance.

Background technique

The Telecommunications Managed Network (TMN) model provides a widely accepted approach to how a service provider's business is managed. The TMN model includes 4 layers that are usually managed in a triangular or pyramidal manner, in which the top layer is business management, the second layer is service management, the third layer is network management, and the bottom layer is unit management. The management decisions at each level are different from each other, but related to each other. As work progresses from top to bottom, each layer pushes out requirements for the layer below. As the work proceeds from the bottom up, each layer sends important data resources to the upper layer. The TeleManagement Forum's (TMF) TMN defines guidelines for optimizing functionality and processes. The 3rd Generation Partnership Project (( ^3rd Generation Partnership Project) 3GPP) adopts the same model. The TMF's scope of work is to find a standardized way to determine quality of service, set network requirements from quality of service (QoS) measurements, and make it possible to have QoS reporting between the provider and the system implementing the service.

According to the TMN model, the information of the upper system flows downward to ensure the seamless operation of the network and the possibility of optimization. Figure 3 shows the TMN model. Information flows all the way down from business management to service management, with network management being the most important because the business side has to be carefully studied during optimization and network development. The TMN model demonstrates the varying levels of abstraction in the daily work of operators. The efficiency of business planning can be measured using capital as well as operating expenses (CAPEX, OPEX) and revenue. Then, translate the desired business situation into the services offered, service priorities, and service QoS requirements. At the lowest level (network element) of the TMN model, business-related issues are transformed into configuration parameter settings.

For example, the functions supported by the TMN business management system are: establishment of investment planning; determination of main QoS criteria for the proposed network and its services; establishment of technology development paths (expansion paths) to ensure the expected growth of the number of users.

For example, the functions supported by the service management system are: management of user data, services provided, and users; collection and approval of billing services; establishment, promotion, and monitoring of services.

For example, the functions supported by the network management system (NMS) are: planning the network; collecting information from the underlying network and preprocessing/postprocessing raw data; analyzing and distributing information; optimizing network capacity and quality.

The element management system can be considered as part of the network element functionality responsible for: monitoring the operation of equipment; collecting raw data (performance indicators), providing a native graphical user interface (GUI) for field engineers; and acting as an intermediary to the NMS system.

In addition to TMN, TMF also defines the Telecom Operation Map (TOM). Telecommunications and data service providers must apply a customer-oriented service management approach using a business process management methodology to effectively cost-manage their business and provide the service and quality customers demand. TOM identifies many operational management processes, including customer service, service management, and network management. The Telecom Operations Diagram uses the layers of TMN as core business processes, but splits the service management layer into 2 parts: customer service and service development and operation. The customer handover management is described separately, because the customer handover management can be managed in each customer service sub-process, or one or more customer service sub-processes can be accessed in combination.

Figure 4 shows the high-level structure of the network management process and the supporting function set group (Function Set Group). According to the framework provided by TOM, each high-level process can be mapped to a series of member functions (component function) (arranged in a function set group). Assumptions:

● Network performance management (PM) provides sufficient measurement values;

●Network configuration management supports the entire TMF framework;

- The Network Management System (NMS) has the intelligence to combine these two kinds of information.

Then, identify the relationships between them and the information flowing between them. In Figure 4, the TOM and its members are shown. To illustrate the respective management layers, the functionality of the layers is the same as that shown in FIG. 3 .

A detailed description of the TMN model and TOM can be found on the home page of TMF (please refer to http//www.tmforum.org).

In current cellular systems, radio resources are processed with a large number of parameters, where the parameter value settings are fixed even under changing conditions. The operator's task is to manually adjust the parameter settings to achieve the correct operating point according to the quality of service. Typically, the objective goal when optimizing is "just to get it working". In simple GSM networks with voice-only services, this regulation is straight-through. For WCDMA, the complexity of these parameter settings is manifold: multi-service, class of service, and even multi-radio environments. WCDMA-based cellular systems provide variable packet-switched and circuit-switched services and are therefore more difficult to plan and control than today's networks. Strong connections between cells add to the complexity. For the operator, virtually all possible resources are utilized to increase the capacity and quality of service (QoS) of the radio network.

Network optimization processes are used to improve the overall network quality perceived by mobile users and to ensure efficient use of network resources. The optimization process involves analyzing the network and improving network configuration and performance. Feed statistics of key performance indicators (KPIs) of the running network to tools for analyzing network status, manually tune radio resource management (RPM) parameters for better performance. Define key performance indicators (KPIs) during the initial stages of the optimization process. For example, they include network management system (NMS) measurements as well as field measurement data or any other information that can be used to determine the quality of service (QoS) of the network. For second-generation systems, Quality of Service (QoS) includes, for example: dropped call statistics, call drop cause analysis, handover statistics, and measurements of successful call attempts, while for third-generation systems with more New Quality of Service (QoS) definition for quality analysis.

In order to optimize the total revenue of a network operator or service provider, reducing the operating and maintenance costs of a network system requires process automation within the network system.

Contents of the invention

It is therefore an object of the present invention to improve the process of optimizing network resources.

This object is achieved with a method for optimizing network performance according to claim 1 , a corresponding system according to claim 14 .

The present invention is based on the idea of optimizing network resources using one lumped cost function, rather than optimizing network resources individually.

Currently, the radio resource management algorithms are parameterized separately: handover control parameter values, access control parameter values, power control parameter values etc. are set independently and situations where handover problems arise eg due to wrong power control (CPICH) settings can be identified. Changes in access control may result in changes in the quality of packet data.

Therefore, when optimizing the performance of the network, the key performance indicators related to the specific entities in the network and the first parameters affecting the key performance indicators are firstly determined. A number of entities similar to said particular entity is selected, wherein a relevant key performance indicator is associated with each entity. The key performance indicators as well as the selected number of entities are used as elements of a first cost function, ie said first cost function is calculated from the KPI and the number of entities. The first cost function is calculated to evaluate network performance. Hence, network performance depends on the first value of the first parameter since the first parameter is directly related to key performance indicators. Thereafter, the value of said first parameter is adjusted in order to obtain a second set of values of said first parameter. Key performance indicators are again determined, but this time based on a second value of said first parameter, and said first cost function is then recalculated based on these key performance indicators. A result of the first cost function calculated from the first value of the first parameter is compared with a result of the first cost function recalculated from the second value of the first parameter. This comparison is done to determine if network performance has improved. When network performance is improved by adjusting the first parameter, the second value of the first parameter is used as a permanent parameter.

Rather than optimizing the set of parameters with a centralized control function, setting each parameter according to many algorithms can swing parameter values because changing one parameter in order to optimize a KPI may adversely affect other KPIs. Therefore, it is advantageous to monitor the radio resource management process as a whole not with a single function but with a lumped cost function in order to coordinate changing of the parameters.

According to an improved structure of the present invention, the key performance indicators are weighted with different weighting coefficients in the first cost function. The one or more key performance indicators may have a greater influence on the first cost function using different weighting factors.

According to a further improved structure of the present invention, the benchmark value of the key performance indicator is set, and the difference between the current key performance indicator and the corresponding benchmark value is used to replace the key performance indicator in the first cost function (to determine the "cost" , please refer to equation (1)). Accordingly, a first cost function is calculated based on the difference between the current key performance indicator and the corresponding baseline value. This allows setting quality of service targets based on the cost of KPIs on the system.

According to a preferred improved structure of the present invention, the first cost function includes a second cost function and a third cost function, wherein the second cost function represents the quality requirement within the network, and the third cost function represents the capacity requirement within the network . The second cost function is weighted with a second weighting factor, and the third cost function is weighted with a third weighting factor. By providing the second cost function and the third cost function and their corresponding weighting coefficients, a balance between capacity and quality of the first cost function can be achieved.

According to a further preferred development of the invention, the second cost function and the third cost function comprise selected entities, wherein the determined key performance indicators are associated with each entity. This allows key performance indicators to be widely distributed across the network.

According to a further preferred improved structure of the present invention, the entity is represented by a cell or a user group in the network. Thus, the cost function may be calculated eg on the basis of cells or clusters of cells.

According to a further preferred improved structure of the present invention, the steps for optimizing network performance are repeated so as to automatically perform the optimization process.

According to yet another preferred improved structure of the present invention, in order to establish the historical database, the KPI value is stored together with the corresponding first parameter and the corresponding result of the first cost function. A current result of the first cost function is compared with previous results stored in a historical database to determine whether network performance has improved within a predetermined time interval. If the network performance has not improved within the predetermined time interval, a corresponding notification is issued. Notification when no improvement is detected within a predetermined interval avoids deadlocks during automated processing and indicates possible problems.

Description of drawings

Hereinafter, the present invention will be described in more detail according to preferred embodiments with reference to the accompanying drawings. The attached drawings include:

Figure 1 shows a flowchart of an automated process for optimizing network performance;

Figure 2 shows an example of a KPI cost function;

Figure 3 shows a schematic diagram of a Telecommunications Management Network (TMN) model;

Figure 4 shows a schematic diagram of a Telecom Operations Map (TOM); and

Fig. 5 shows a schematic diagram of a combination of monitoring and optimization functions for different management layers.

Detailed ways

In Fig. 1, a flow chart of an automatic process for optimizing network performance according to a first embodiment of the invention is shown. First, at step S1 , key performance indicators are selected for describing the performance of the part of the network of interest. Then, in step S2, the configuration parameters on which the KPIs are based are determined. In step S3, the number of cells to be included in the optimization process is selected, ie clusters of cells are selected. In step S4, the current value of the KPI is determined according to each configuration parameter. Thereafter, in step S5, the current value of KPI is determined according to the current value of KPI and the number of cells, and the cost function is calculated. In step S6, the result of the cost function, the value of the KPI and the configuration parameters are stored in the historical database.

In step S7, at least one value of each configuration parameter is adjusted to obtain a set of new configuration parameters. Based on the new set of configuration parameters, a new KPI value is determined in step S4, and then, in step S5, the cost function is recalculated based on the new KPI value and the number of cells selected in step S3 (unchanged). In step S6, the new result of the cost function, the new KPI and the configuration parameter values are also stored in the temporary memory. Subsequently, at step S8, this new result of the cost function based on the new/adjusted set of configuration parameters is compared with the previous results of the cost function stored in the historical database to determine whether the network performance of interest after adjusting the configuration parameters Improved.

If the network performance is improved after the configuration parameters are adjusted, then in step S9, the adjusted configuration parameter set is used as a permanent parameter. However, if it is determined at step S8 that the network performance has not improved after adjusting the configuration parameters, then at step S9 the first set of configuration parameters stored in the history database at step S6 are used as permanent parameters.

In step S10, it is checked whether the network performance has improved within a predetermined time interval. When the network performance has not improved within a predetermined time interval, ie, the KPI history has not improved even with automatic adjustments, the network operator is notified at step S12 that there is a problem in the automatic process of optimizing network performance. Since many parameter values obviously cannot be automatically tuned, and autotuning does not always optimize the network, the operator can check whether the problem is due to a hardware problem, or whether it is not possible to automatically optimize network performance under current network conditions. With the network in this situation, network operators must manually optimize network performance.

On the other hand, when the network performance is improved within the predetermined time interval, the process jumps to step S7, and in step S7, the configuration parameters are adjusted again so as to further optimize the network performance. Then, the process will continue the process described above.

In the second embodiment, not only relevant KPIs are selected in step S1, but also a set of QoS targets is determined, and a set of reference KPIs is used to represent the set of QoS. The automatic process of optimizing network performance according to the second embodiment is basically the same as the optimization process according to the first embodiment. The only difference is that when calculating the cost function in step S5, the difference between the KPI and the reference KPI is used instead of the KPI value.

Thus, the operator sets capacity requirements for a specific capacity KPI denoted as KPI_C with "ref" within the sub-index. Accordingly, the operator sets quality requirements for a specific KPI_Q. Then, mass cost and capacity cost are calculated using equation (1).

Using different weighting coefficients α and β, different cost functions are combined or added together. By manipulating or varying the weighting coefficients α and β, specific types of costs as well as the overall situation can be emphasized.

Its task can be considered that the mathematical formula for optimizing network performance is used to determine the combination of air interface configuration parameters according to which KPI is as close as possible to the required area.

Figure 2 shows an example of a KPI cost function f. In this example, KPI cost values above KPI_ref increase linearly.

Equation (3) shows the total cost function to be optimized, ie minimized. Using the parameter W, a balance can be achieved between capacity and quality requirements. By adjusting the configuration parameter (2), it can be reduced to a minimum. KPI values also depend on service allocation, for example, different cost settings and parameter settings are obtained depending on service allocation.

KPI_C _i = f (configuration parameters, service distribution)

KPI_Q _j = f (configuration parameters, service distribution) (2)

Total cost = W*quality cost+(1-W)*capacity cost (3)

For example, factors that can influence the optimization process are: traffic distribution (service mix), traffic density, pricing of each service, etc. The ultimate goals when minimizing total cost include: optimizing operator revenue, minimizing CAPEX and OPEX, and maintaining the operator's good image.

A specific example of the cost function TOTAL COST can be calculated according to equations (4) to (8) below:

(4) Total cost = C (queuing ratio) + C (bad quality ratio) + C (dropped call ratio) + C (blocking ratio)

in

(5) C (queuing ratio) = 0.05*Dev (queuing ratio - allowable queuing ratio)

(6) C (bad quality ratio) = 0.2*Dev (bad quality ratio - allowable bad quality ratio)

(7) C (Call Drop Ratio) = 1*Dev (Call Drop Ratio - Allowable Call Drop Ratio)

(8) C (blocking ratio) = 0.10*Dev (blocking ratio - allowable blocking ratio)

The optimization task is to combine all the different TOM management layers, taking into account the fact that the measured values of the different layers (quality marking and cost marking) are in different languages.

When performing the optimization process within the NMS of the network, the support operator determines the customer service and each service management layer. To enable this process, the biases of the cost function comprising configuration and (PM) measurements performed at lower layers are translated into the "language" of the upper layers. This can be achieved by doing the following:

Perform translation from radio access network parameters (settings) to (mapping to) service-related quality expectations/targets. In practice, this means making configuration management relevant to performance management. That is, a functional entity with a specific configuration is monitored with a specific set of measurements. The performance of the entity is calculated using a cost function using the determined measurements.

In practice, what follows means the transformation of the large-body measurements (ie cells, traffic classes, etc.) sent to user-level entities that statistically determine the quality of each user. This step is also performed using weighted cost function(s). Furthermore, these transformations can be combined with cost functions to achieve desired end-user quality indications.

1) Technology translation (mapping) from radio access network measurements (network performance) to end-user process level of service (perceived quality).

2) Technology translation (mapping) from aggregation level (UMTS service category) parameter setting to end user process level of service (perceived quality).

3) Technical translation (mapping) from the measured value of each business category set to the end-user process level (perceived quality) of the service.

and/or a combined function of business category and process level information (parameters and settings) with a cost function to support parameterization and monitoring of end-user GOS.

Fig. 5 shows a schematic diagram of a combination of network monitoring functions and optimization functions for combining different management layers within a network using mapping.

Mapping from one layer to the next is achieved by combining network measurements, performance markers PI and/or KPIs with cost functions.

The cost function of the grade of service GOS perceived by the user can be calculated as expressed by equation (9):

(9) GOS = C (service availability) + C (delay and jitter) + C (quality) + C (dropped calls) + C (service accessibility) + C (equivalent bit rate or user throughput)

The delay includes service access delay and queuing transmission delay. The non-real-time quality is affected by packet loss, radio link control RLC, packet data convergence protocol PDCP, ie by bit error rate BER and block error rate BLER. Regarding real-time quality, if the uplink UL block error rate BLER is significantly higher than the target BLER, the quality is bad. Real-time quality is affected by downlink DL connection outages. The revenue of the above cost function includes capacity requirements and traffic distribution. Total throughput is measured in kbps/cell/MHz.

When 98% of users are satisfied, the spectral efficiency of the cost function is equal to the throughput in kbp/cell/MHz. This means that the traffic accessibility and blocking probability is 2%. The equivalent bit rate is 10% higher than the data rate of the load service, and 98% of the users are not dropped. The reason for adopting this method is that, in terms of metrics, it is beneficial to optimize according to the GOS.

This mapping must be done for all services offered, ie services controlled with different parameter settings or other properties.

Although each transformation reduces precision, the image is statistically correct. Since the statistical level can be manipulated, the best place for the mapping function is in the NMS. Furthermore, NMS implementations can also handle Radio Network Controller RNC-RNC (or other network element) border areas. At each such transformation, the proposed cost function is applied. In some cases, service QoS goals can lead to conflicts in parameter settings, so a cost function is needed to resolve conflicts. This can be achieved by setting different weight coefficients for different units in the cost function. This theory is very important when different customer classes (silver, copper, gold, etc.) enter the network system.

Furthermore, when changing to the last management level of the TOM model, the next major step is the calculation of network optimization, order of service and customer differentiation expressed in Euros, Dollars or British Pounds. At this stage, billing and charging information provided by the billing/collection subsystem within TOM's customer service layer is required. When utilizing knowledge of customer profiles/distributions and the properties of these distributions, the operator's business situation can be optimized in the most beneficial direction according to a cost function. It is worth noting that changing customer priorities and QoS provided for business reasons will lead to changes in customer behavior and thus repeated business management optimization.

In order to ensure optimal performance of the cellular network, the operator prioritizes flexible devices to set QoS targets according to system KPIs (Key Performance Indicators) and/or cost functions derived therefrom. QoS targets can be set for cell clusters, or for each cell. Can be based on blocked calls due to hardware resources, "soft" blocked calls (interfering with limited networks), dropped calls, poor quality calls, retransmission times and delays for compressed data, diversity handover probability, hard handover success rate, loading conditions (uplink UL or downlink DL), compressed data to circuit switched services, etc., QoS can be calculated.

In a multi-radio environment (GSM-WCDMA Global System for Mobile Communications - Wideband Code Division Multiple Access), it is important that resource pools can be created for both networks in order to optimize capacity, convergence and quality. This requires all control functionality (quality manager) for higher (KPI) layers, ie the optimization process according to the invention can be implemented with the quality manager.

The Quality Manager QM, ie, the optimization process provides central monitoring functions and monitors the status of parameter values, automatically identifying problematic situations by comparing the historical information of parameter values stored in the historical database. For example, GERAN and UMTS terrestrial radio access network UTRAN can be divided into self-regulating subsystems as small and as independent as possible. By setting weighting coefficients for the KPIs of each subsystem, the interdependence between subsystems is considered in the upper layer of the quality manager.

In another embodiment, an optimization process is performed based on user groups (eg, business users, free time usage, etc.).

In a nutshell, default values for all current parameter values are suggested. Cell-by-cell user networking has so far been the task of the operator (trying to take into account the multi-cell environment). However, with the method and/or system for optimizing network performance according to the present invention, the initial parameter setting is not important. For example, when the network starts to operate, access control and handover control can be performed under very "loose" restrictions, so that according to the current QoS situation (KPI located in the operating service system OSS) and related parameters can be automatically adjusted and set. The QoS target enables all users to access the network. After the parameter changes to a new situation, ie a new KPI value, it is compared with the historical data of the KPI, and if the change in QoS performance (or cost function of QoS requirements) is improved, the "test" parameter is accepted. The length of historical data depends on the traffic volume on the network (the total number of samples should be high enough). Importantly, the QoS cost function contains terms for the entire RRM and multiradio regions.

Currently key parameters (in terms of optimal capacity and quality) are first set to "default" values which in most cases ensure network operation, but not optimal performance. The optimization process according to the invention automatically changes the settings of the basic parameters to the optimal operating point according to the overall QoS.

The adjustment amount of the configuration parameter can be fixed increment or decrement. As an option, the increment or decrement can be variable.

According to the third embodiment of the present invention, network resources are intensively optimized using a cost function and a required level of quality of service (QoS) is provided. The cost C is a function of various KPIs of the network, for example:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; f</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; KPIs</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 11</span>}<span\; class="notranslate">\; )</span>$

where KPI _i is the ith key performance indicator and Fi is some positive function that can be used to transform, weight and/or scale the ith KPI. By minimizing this cost function C, network performance is optimized. The lowest cost can be achieved by a correct choice of the various network parameters W = (w ₁ ; w ₂ ; . . . , w _N ) considered as optimal values of the parameters. The cost function approach implicitly assumes that the value of the KPI is a function of the network parameters, i.e.:

KPl _j = KPl _j (w ₁ ; w ₂ ; . . . , w _N )j (12)

Therefore, the cost function C is also a function of the network parameters, and it can be rewritten as a direct function of the parameters:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; =</span>\underset{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}{\overset{\mathrm{<span\; class="notranslate">\; N</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}}{\mathrm{<span\; class="notranslate">\; g</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; t</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 13</span>}<span\; class="notranslate">\; )</span>$

where _gi is some function that can vary over time t.

The third embodiment relates in particular to a simple, efficient algorithm for minimizing the cost function described above. However, in practical situations, optimizing such a cost function is not straightforward. The main problems are listed below:

1. In the actual network, there are many kinds of business types, user distribution and load. The network has no control over these factors and can treat them as sources of random external noise. Any optimization algorithm should not be sensitive to such external random influences.

2. The optimal parameter selection for any given network can be improved over time because the load etc. varies from time to time (ie, the function g ₁ of equation (13) varies over time). Any optimization algorithm should be able to adapt to such changes in the optimal operating point of the network and be able to track these changes.

3. May not produce a network model that can be optimized in a variety of different situations. This means that the function gi of equation (13) is not known. The alternative approach adopted by the third embodiment and this approach does not assume any model of the network, but bases the optimization process on network measurements only.

An optimization algorithm used in minimizing the cost function is described and can be implemented from first principles. The general case of minimizing a cost function C according to a parameter denoted w is studied. Let w ₀ be the value of w used to reduce C to a minimum. Computing C(w ₀ ) using the Taylor series expansion for any value of w gives,

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\frac{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 14</span>}<span\; class="notranslate">\; )</span>$

where C'(w) is the first-order differential of C with respect to W, and C"(w) is the second-order differential. Since C(w ₀ ) is the minimum point of C, with respect to w ₀ , the differential of equation (14) , and setting the result equal to 0, we get,

${\mathrm{<span\; class="notranslate">\; w</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 15</span>}<span\; class="notranslate">\; )</span>$

It is a fast-converging classic Gauss-Newton (Gauss-Newton) algorithm optimization algorithm. If C is a quadratic function of w, it converges to the optimal point w ₀ in one step. If C is not a quadratic function, convergence is guaranteed as long as C"(w) is always positive. If it is not, it can be made positive and equation (15) breaks down to the standard gradient algorithm. However, since close to The minimum point of C, so the quadratic approximation is more accurate and converges faster.

Now, this problem is studied according to the WCDMA cost function. As mentioned above, since there is no model of the network, it is difficult to determine the values of C'(w) and C"(w), and therefore difficult to use equation (15). However, according to the third embodiment, the calculation of C using network measurements '(W) and C"(W) in order to use equation (15).

A small change in the value of the parameter w δw > 0 is studied to obtain a new parameter value w+δw, so the value of the cost function can be approximated as

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; \&delta;w</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 16</span>}<span\; class="notranslate">\; )</span>$

Similarly, the expression C(w-δw) can be obtained:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\frac{{\mathrm{<span\; class="notranslate">\; \&delta;w</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 17</span>}<span\; class="notranslate">\; )</span>$

Adding the two expressions together and rearranging the terms yields C"(wadd), then subtracting the two terms and rearranging the expression yields the expression for C"(wadd).

${\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 18</span>}<span\; class="notranslate">\; )</span>$

${\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;w</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; w</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; \&delta;w</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 19</span>}<span\; class="notranslate">\; )</span>$

Therefore, by knowing C(w+δw) and C(w−δw), the values of C'(w) and C"(w), or their approximate values, can be obtained using equations (18) and (19). The next question is how to calculate these values at any particular time. This is performed according to the following steps:

1. At time t1, the parameter value is w, and from the network measurement of the correct KPI at time t1, the cost function C(w; t1) is calculated using equation (11).

2. At time t1, change the value of w to w+δw.

3. At time t2=t1+δt; δt>0, according to the network measurement value of the correct KPI, the value of the cost function is calculated using equation (11) to obtain C(w+δw; t2).

4. At time t2, change the parameter w to w-δw.

5. At time t3 = t2 + δt, from the network measurement of the correct KPI, calculate the cost function using equation (11) to obtain C(w - δw; t3).

6. At time t3, use equations (18), (19) to obtain C'(w) and C"(w) respectively, and use the measured values C(t1), C(t2), C(t3 ) to obtain C(w), C(w+δw) and C(w-δw), use equation (15) to calculate the new value of w.

These steps constitute a cycle of the algorithm, and the cycle can be repeated. The problem discussed above with regard to noise fluctuations occurring in different network measurements will now be investigated. Although this algorithm derives the cost function without the noise term, it is also possible to apply the algorithm to a noise cost function.

The effect of repeating the above algorithm is that the noise contribution and parameters converge to an average. For example, Kushner, H.J. and Clark, D.S. (1978), Stochastic Approximation Methods for Constrained and Unconstrained System, volume 26 of Applied Mathematical Sciences, Springer-Verlag, New York, Heidelberg, Berlin, have carefully studied this algorithm. It also helps to average out the noise effect by having w increase and decrease over time. Furthermore, in practical networks, noise effects are reduced since measurements are usually integrated over a time period of δt. The correct δt can be chosen for the parameter being optimized, and the value of δt can be changed during the optimization process.

Also, it is obvious how the algorithm tracks the network's sweet spot as it changes. Even when the parameters have reached the optimal point, the algorithm still causes the point to fluctuate slightly. As long as the optimum point does not change, the average value of the fluctuation can be zero near the optimum point. If that optimal value changes, the algorithm can still track that change.

In the first embodiment, the configuration parameter value of the KPI is adjusted, the cost function is recalculated, the cost function is compared with the cost function based on the previous value of the configuration parameter, the newly adjusted value is used as the new configuration parameter, and according to the In the second embodiment, the value of the configuration parameter is adjusted in two steps. First, increase the value of the configuration parameter, recalculate the cost function based on the new value, and compare this result with the previous result of the cost function. Then, decrease the value of the configuration parameter, recalculate the cost function based on the new value, and compare the result with the previous result of the cost function. However, even if the results of two previous changes to the cost function do not improve, it is still possible to make small or zero changes to the configuration parameters.

As an option, in a fourth embodiment according to the third embodiment, the cost function and its optimization are described according to one specific network parameter, i.e., now discussing obtaining And optimize the specific problem of the cost function.

The WCDMA radio interface of third generation mobile networks can carry voice services and data services with various data rates, traffic requirements and quality of service objectives. In addition, the working environment has changed greatly from indoor cells to large macro cells. Effective use of limited frequency bands requires careful setting of various important network and cell parameters under various conditions. The parameter setting is called radio network planning and optimization. Once a WCDMA network is established and launched, its operation and maintenance is mainly about monitoring performance characteristics and quality characteristics and changing parameter values in order to improve performance. Automatic parameter control is simple, but it requires metric-determined performance signatures, or in this case, cost functions that explicitly tell whether performance is improving or deteriorating.

The goal of optimization is to minimize the total number of blocking calls on the network. The particular parameter to be optimized is the soft handover parameter window plus (add) (wadd). "Soft handover gains in fast power controlled WCDMA uplink" Sipila, K.; Jasberg, M.; Laiho-Steffens, J.; Wacker, A. Vehicular Technology Conference, 1999 IEEE ^49th , Volume: 2, 1999 Page(s) : 1594-1598, vol.2. There is a discussion on improving performance based on soft handover. It has been found that great care must be taken in determining the cost function to minimize. Combining terms in the wrong way can produce a cost function that remains fixed for whatever parameter is chosen.

By adjusting network parameters, some factors that affect network performance can be directly controlled. For example, number of users, distribution of users, and type of business. Variations in these external parameters result in variations in the cost function. This means that any optimization algorithm used to minimize the cost function should be robust and achieve convergence even in the presence of random fluctuations. Kushner and Clark, "Stochastic Approximation Methods for Constrained and Unconstrained System", Springer-Verlag, New York, Heidelberg, 1978 provide a careful study of this algorithm. Here, a relevant result of these studies on cost function optimization problems is that, even in noisy environments, an optimization problem can be viewed as a noiseless optimization when repeatedly averaging noise fluctuations within the system. Here, the method is adopted and an optimization algorithm is obtained to optimize a certain cost function, in fact, the noise cost function is reduced to a minimum with this optimization algorithm.

A second consideration in choosing an optimization algorithm for a cost function is that the optimal operating point of the cost function can be changed. Therefore, the optimization algorithm should be able to track any changes in the state of the cost function. It can be seen from the analysis that the proposed algorithm has a quadratic convergence that reduces the cost function to a minimum, compared to the linear convergence of the standard gradient algorithm.

The Quality Manager is a logical unit in the Radio Network Controller for collecting statistics of various performance markers. Quality Manager calculates these statistics during a specific time interval, called qminterval. Some statistics available to Quality Manager include:

● At each qminterval, the quality manager goes through all connections in the sector and checks the call quality. During the control period, the number of bad quality calls and the total number of calls are accumulated in two counters. The mass is obtained using the ratio of the counter values.

• The ratio of blocked calls to total access requests during the previous qminterval.

● Ratio of dropped calls to total number of terminated calls during the previous qminterval.

Here, only blocking ratios are used, however, the method and simulation process can be extended to include other statistics returned by the quality manager.

The general case of the cost function C (cf. Equation 11) to be reduced to a minimum is studied in terms of the switching parameter, the window sum denoted as wadd. Let wadd ₀ be the value of window addition that reduces C to its minimum. Using the Taylor series expansion about wadd to calculate C(wadd ₀ ), we get:

$\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; wadd</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; wadd</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; *</span>{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span>$

$<span\; class="notranslate">\; +</span>\frac{{<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; wadd</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{\mathrm{<span\; class="notranslate">\; 2</span>}}{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 20</span>}<span\; class="notranslate">\; )</span>$

Among them, C' is the first-order differential of C to wadd, and C" is the second-order differential. Since C(wadd ₀ ) is the minimum, with respect to wadd ₀ , find the differential of equation (1), and set the result equal to 0, then we get out,

${\mathrm{<span\; class="notranslate">\; wadd</span>}}_{\mathrm{<span\; class="notranslate">\; 0</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; -</span>\frac{{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; twenty\; one</span>}<span\; class="notranslate">\; )</span>$

It is a fast-converging classic Gauss-Newton (Gauss-Newton) algorithm. In wadd, the values of C' and C" must be known. It is now shown how these values are estimated by the network. Note that equations (20) and (21) correspond to equations (14) and (15), and equation (14 ) and (15) are related to a more general form of the equation.

The study window plus the change from δwadd to wadd+δwadd and the corresponding value of the cost function C(wadd+δwadd). Similarly, the study window is added to the change of wadd-δwadd and the corresponding value C(wadd-δwadd) of the cost function. Then, using algebraic operations, C’(wadd) and C”(wadd) can be expressed as:

${\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;\; wadd</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; \&delta;\; wadd</span>}}$

${\mathrm{<span\; class="notranslate">\; C</span>}}^{<span\; class="notranslate">\; \&prime;</span><span\; class="notranslate">\; \&prime;</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&delta;\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&delta;\; wadd</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; *</span>\mathrm{<span\; class="notranslate">\; C</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; wadd</span>}<span\; class="notranslate">\; )</span>}{{\mathrm{<span\; class="notranslate">\; wadd</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; twenty\; two</span>}<span\; class="notranslate">\; )</span>$

If C is quadratic, there is therefore one-step convergence. If C is not quadratic, the expression of equation (22) is approximate, but there is still fast convergence faster than the standard gradient algorithm. The closer to wadd ₀ , the more accurate the approximation to the quadratic. In practice, the algorithm is implemented using the following procedure:

1) At time t1, the value added to the window is wadd, and the value of the cost function C(t1) is calculated according to the network measurement value. At time t1, the value of wadd is changed to wadd+δwadd.

2) At time t2 (>t1), the value added to the window is wadd+δwadd, and the cost function value C(t2) is calculated directly according to the network measurement value. At time t2, the value of window addition is changed to wadd-δwadd.

3) At time t3, the added value of the window is wadd-δwadd, and the cost function value C(t3) is calculated directly according to the network measurement value. At time t3, the value of wadd is updated using equation (20) and equations (21) and (22), where

C(wadd)=c(t1)

C(wadd+δwadd)＝C(t2)

C(wadd-δwadd)＝C(t3) (23)

This process is repeated, thereby reducing the cost function to a minimum. It is worth noting that by alternately increasing or decreasing the value of the window plus, two goals can be achieved. As mentioned above, the first objective is to estimate C' and C". The second objective is less problematic. The case is studied where the algorithm converges to the minimum of the cost function. At this point, the gradient of the function is 0, and The optimization has been done. However, the optimal point of the network is always changing, and therefore the cost function is also always changing. By alternating the value of wadd, as mentioned above, any such changes can be detected and tracked by the algorithm.

The next stage of this fourth embodiment is to develop a cost function that can be reduced to a minimum using the optimization algorithm described above. In the first, most general case, the cost function can be described by equation (11), where KPI _i is the ith KPI of the network and F _i is some function to be defined. Each term of the cost function should always be positive, so the cost function is always positive. Furthermore, the function F _i should scale the KPI _i in such a way that in normal operation this term does not dominate the cost function. For example, F _i (KPI _i ) should be in the range [0, 1] for the value of KPI _i that can ensure correct quality of service to work within the required range.

In a fourth embodiment, only the blocking rate is of interest. The goal is to minimize the blocking call ratio (BKCR) as a function of window addition. In this case, the most obvious cost function to minimize is a simple sum of blocking ratios. However, for several reasons, a better choice of cost function in this case is

C=ulBKCR ² +dlBKCR ² (24)

where ulBKCR is the uplink blocked call ratio and dlBKCR is the downlink blocked call ratio. However, in any practical network, for an acceptable level of service, the blocked calls ratio must be below a certain value, eg 5%. This cost function can be tuned further so that the blocking value is significantly higher than this value. For example,

C＝f(ulBKCR) ² +f(dlBKCR) ² (25)

in

f(x)=exp(x*12)-1 (26)

Choosing this function means that for 5% uplink blocking, the value of f(ulBKCR) = 1.0. For values below 5%, the function is almost linear. However, for values greater than 5%, the function increases exponentially. The usefulness of the function f is more obvious when there are many terms in the cost function. Another important property of the function f is that it is continuously differentiable, so there is no problem in taking the derivative of the cost function used in the optimization algorithm.

In the fifth embodiment, the algorithm of the third embodiment can be extended to minimize the cost function when several network parameters are to be optimized. The multi-parameter case is realized by simplifying the multi-parameter problem to one parameter of the third embodiment. The vector W of the N parameters wi to be optimized is studied,

W=(w ₁ ; w ₂ , . . . ; w _N ) (27)

Initial values of these parameters from which to optimize are investigated. This initial value can be chosen randomly,

W ₀ =(w _{0; 1} ; w _{0; 2} ; . . . ; w _{0; N} ) (28)

Define a row in the N-dimensional parameter space containing the initial value W ₀ as:

L ₀ =W ₀ +λ n ₀ (29)

The N-dimensional vector _n0 is a unit vector which in turn has an arbitrary direction first, and the factor λ is a scalar variable. The theory is to minimize the cost function along the line _L0 . This amounts to determining the optimal value for λ, which is a scalar value, so the algorithm described in the previous subsection can be employed. Assuming the optimum value is λ ₀ , the new value of W is obtained using the following equation:

W ₁ =W ₀ +λ ₀ n ₀ (30)

Define another line _L1 as:

L ₁ =W ₁ +λ n ₁ (31)

where n ₁ is the conjugate direction of n ₀ . Again, the optimization of the cost function is repeated along the newly defined lines using the algorithm of the third embodiment. Theoretically speaking, in a noise-free system, the optimization along the N times must be repeated along the N conjugate directions (n ₀ , n ₁ ,..., n _N-1 ). For noisy cost functions, more cycles must be repeated in order to remove the noise effect. There are many well-known methods to generate conjugate orientations at each step of the optimization process. Optimizing several network parameters simultaneously leads to better minimization of the cost function than optimizing network parameters individually.

A further advantage of using this algorithm, especially when extending the algorithm to higher dimensions, is that, by making the fluctuations of the parameters small, it is possible to escape from local minima of the cost function.

The optimization method according to the first, second, third or fourth embodiment is not only based on the last two results of the cost function, but also on the previous history of the measured values of the cost function. Thus, at time t, the variation of the influencing parameter may be a function of the cost function and the value of each parameter at different times t, t-1, t-2, t-3, . . . t-n. Thus, each parameter can be updated, or expressed as a function of previous measurements.

Claims (15)

1. A method for optimizing network performance, particularly radio network performance, the method comprising the steps of: Determining a relevant key performance indicator of an entity within the network and a first parameter on which said key performance indicator depends, select many similar entities, Computing a first cost function based on the determined key performance indicators and a selected number of entities to evaluate network performance based on a first value of said first parameter, wherein said first value of said first parameter represents said the current value of the first argument, adjusting the value of said first parameter, thereby producing a second value of said first parameter, recalculating said first cost function to assess network performance based on key performance indicators determined in accordance with said second value of said first parameter, Comparing the result of the first cost function from the first value of the first parameter with the result of the first cost function from the second value of the first parameter to determine whether network performance has been improved, • Adopting said second value of said first parameter as a permanent parameter if network performance is improved according to said second value of said first parameter. 2. The method of claim 1, Wherein the corresponding determined key performance indicators are weighted with different weighting coefficients in the first cost function. 3. The method according to claim 1 or 2, wherein a baseline value for a key performance indicator is set, and a difference between a current key performance indicator and its corresponding baseline value is determined and used as an element within said first cost function. 4. A method according to any one of the preceding claims, Wherein the first cost function includes a second cost function representing the quality requirement in the network and a third cost function representing the capacity requirement in the network, wherein the second cost function is weighted by the second weighting coefficient, and the third cost function is used by the third A weighting factor weights the third cost function. 5. The method of claim 4, Wherein the third weighting coefficient is equal to 1 minus the second weighting coefficient. 6. A method according to any one of the preceding claims, Wherein said second and third cost functions include, as elements, determined key performance indicators for each selected entity. 7. A method according to any one of the preceding claims, Wherein the entity within the network is represented by a cell or user group within the network. 8. A method according to any one of the preceding claims, The individual steps for optimizing network performance are repeated therein. 9. A method according to any one of the preceding claims, comprising the steps of: • To build the historical database, storing the values of the key performance indicators together with the corresponding first parameters and corresponding results of the first cost function. 10. The method according to any one of the preceding claims, wherein the comparing step comprises the steps of: comparing the results of the first cost function with previous results of the first cost function stored in the history database to determine whether network performance has improved within a predetermined time interval, and • Sending a notification if network performance has not improved within said predetermined time interval. 11. The method of claim 8, Wherein said step of adjusting the value of said first parameter is performed by alternately increasing or decreasing said value. 12. The method of claim 11, Wherein said step of increasing or decreasing said value of said first parameter is performed continuously prior to performing said adopting step. 13. A method according to any one of the preceding claims, Wherein the service and/or quality of service indication for each user is obtained from measurements and/or configurations by lower management layers in the network. 14. A system for optimizing network performance, especially radio network performance, the system comprising: a) means for determining a relevant key performance indicator of an entity within the network and a first parameter on which said key performance indicator depends, b) means for selecting at least one similar entity, c) means for calculating a first cost function based on said determined key performance indicator and said selected at least one entity to evaluate network performance based on a first value of said first parameter, wherein said first said first value of a parameter represents the current value of said first parameter, d) means for adjusting said first value of said first parameter, thereby obtaining a second value of said first parameter, e) means for recalculating said first cost function for evaluating network performance based on said associated key performance indicator determined in accordance with said second value of said first parameter, f) for comparing the result of said first cost function according to said first value of said first parameter with the result of said first cost function according to said second value of said first parameter , to determine whether network performance has improved by means of, g) means for adopting said second value of said first parameter as a permanent parameter when network performance is improved according to said second value of said first parameter. 15. A computer program product comprising computer program code means for causing a computer to carry out the steps of the method as claimed in claims 1 to 10 when said computer program is run on a computer.

Images