System and method of cell parameterization by cell mirroring

Abstract

A configuration of parameters of a target cell is selected by discovering similar cells having good performance, processing the group of cells to eliminate those that do not have a homogeneous parameterization, and copying parameters from the remaining cell(s). The cell similarity calculation is hierarchical, first computing the similarity of categories of related parameters or KPIs between each cell and the target cell. The overall cell similarity calculation is then a weighted average of the category similarity measures. A Figure of Merit (FoM), reflecting cell performance, is computed for each cell as a weighted average of a set of KPIs. The set of comparison cells is reduced by filtering with both the cell similarity and FoM metrics. The remaining cells are then dimensionally reduced, clustered (e.g., using an AI model), and the clusters filtered according to a variety of metrics.

Description

[0001] SYSTEM AND METHOD OF CELL PARAMETERIZATION BY CELL MIRRORING

[0002] RELATED APPLICATIONS

[0003] This application claims priority to EP Application No. 24382148.5, filed 15 February 2024, disclosure of which is incorporated in its entirety by reference herein.

[0004] TECHNICAL FIELD

[0005] The present disclosure relates generally to wireless communications, and in particular to a method of configuring a target cell by selecting a cell having similar parameters, acceptable performance, and homogeneous parameterization, and copying parameters to the target cell.

[0006] BACKGROUND

[0007] Wireless communication networks are ubiquitous in many parts of the world. These networks continue to grow in capacity and sophistication. To accommodate more users, different types of devices, and different use cases, the technical standards governing the operation of wireless communication networks continue to evolve. The fourth generation (4G) of network standards has been deployed, the fifth generation (5G) is in development and early deployment, and the sixth generation (6G) is in design. With each generation, technological advances improve the capacity and spectral efficiency of the wireless communication system.

[0008] One popular model of wireless communication network is referred to as “cellular,” in which generally fixed base stations (BS) provide wireless communication service to a large plurality of fixed and mobile terminals (User Equipment, or UE) within a geographic area called a “cell.” Mobility management techniques, such as handover and cell reselection, provide for transferring service of a UE from one base station to another in response to radio conditions, such as the relative received signal strength of transmissions from the BSs.

[0009] Numerous signals, techniques, and procedures have been developed to optimize the operation of a wireless communication cell. For example, both the BS and UE include reference signals in their respective transmissions. Reference signals are transmissions of known data patterns, which allow the receiver to quantify parameters of the transmission channel. The BS may perform power control over UEs, to avoid transmissions at power levels greater than necessary to receive the data (which are seen by other UEs as interference). In Minimization of Drive Tests (MDT) procedures, the BS directs UEs to perform various measurements of the radio environment and report the results, along with the UEs’ position, allowing the BS to monitor radio performance over the entire cell. The BS may configure, and dynamically alter, a wide array of design and operating parameters in an attempt to optimize network performance.

[0010] Wireless network parameter design and optimization have been widely analyzed in the industry. Determining the optimal value for a certain parameter when deploying a new cell, or improving performance in an existing configuration, are key activities carried out by network engineers with high frequency, and a multitude of tools and algorithms have been developed to support such analysis and optimization.

[0011] Most such tools and algorithms are based on proposing a configuration value for a certain parameter, depending on inputs related to the cell under analysis and its neighbors (e.g., performance metrics, configuration parameters, terrain topography around the cell, etcetera), and the use of different optimization techniques depending on the use case (e.g., Al techniques, fuzzy rules, simulation tools, etcetera). These methodologies are designed specifically to design or optimize a single radio or network parameter, or a small set of parameters.

[0012] One approach to such optimization considers how cells in a mobile network are performing, and similarities between cells subject to optimization and other, similar cells that have been analyzed previously. In particular, the present the inventors previously devised a methodology to quantify the similarity between cells based on a set of predefined features. Detecting cells with similar performance, issues, configurations, morphology, or other characteristics yields valuable information that can be used for different purposes, including improving the network or cell configuration. See PCT publication WO 2022 / 090810, Cell similarity indicator based on coverage area morphology, 15 APR 2021 by the present inventors and assigned to the present assignee, the disclosure of which is incorporated herein by reference in its entirety.

[0013] However, to date, there have been no studies that propose an end-to-end methodology to improve network configuration ( / .e., to go beyond descriptive similarity assessments and propose concrete changes in configuration settings), based on a quantified similarity indicator, that at the same time is flexible and adaptable to different target network parameters and use cases.

[0014] Rather, available algorithms and tools provide specific solutions to optimize each individual parameter (or small set of parameters), but not a generic end-to-end-process. Furthermore, such tools are not based on cell similarity. Likewise, apart from the inventors’ PCT application cited above, all the studies found in the literatures are limited to classifying cells in groups with similar characteristics, without quantifying the similarity between those cells. In this class are references such as:

[0015] • Y. Ouyang, Z. Li, L. Su, W. Lu and Z. Lin, "Application Behaviors Driven Self-Organizing Network (SON) for 4G LTE Networks," in IEEE Transactions on Network Science and Engineering, vol. 7, no. 1, pp. 3-14, 1 Jan. -March 2020;

[0016] • F. Sun and Y. Zhao, "Cell cluster-based dynamic TDD DL / UL reconfiguration in TD-LTE systems," 2016 IEEE Wireless Communications and Networking Conference, Doha, 2016, pp. 1-5;

[0017] • N. Pasquino, S. Zinno, F. Cotugno and S. Petrocelli, "A comparative approach of unsupervised machine learning techniques for LTE network parameter clustering," 2020 IEEE International Instrumentation and Measurement Technology Conference (I2MTC), Dubrovnik, Croatia, 2020, pp. 1-6;

[0018] • M. Cheng, Y. Wang, W. Hwang, Y. Wu and C. Lin, "Adaptive adjustment of TDD uplinkdownlink configuration based on cluster classification in Beyond LTE Heterogeneous networks," 2017 International Conference on Applied System Innovation (ICASI), Sapporo, 2017, pp. 1312-1315; and

[0019] • WO2019233635A1, “Methods, apparatus and computer-readable mediums relating to detection of cell conditions in a wireless cellular network” - all of which are incorporated herein by reference in their entireties.

[0020] Although these and other existing solutions are widely used, they are focused on different use cases, or they are focused only on a limited piece of the whole problem. In particular, prior art solutions to cell parameter optimization are deficient for at least the following reasons:

[0021] • They do not propose a full end-to-end methodology ( / .e., one that goes beyond descriptive similarity assessments and proposes concrete changes in configuration settings) to improve network configuration based on similar cells with good performance.

[0022] • At best, they propose specific solutions adapted to a single (or a small set of) parameter(s), but not a generic methodology that can be applied to any parameter.

[0023] • They do not offer a large degree of flexibility in the calculation of the selected input features (specially including the morphology of the coverage area). Such flexibility would allow network operators to adapt the use case not only to different parameters, but also to optimization and design (e.g., by removing certain operational performance metrics from the set of input features, when calculating parameters for new cells that do not yet exist).

[0024] The Background section of this document is provided to place aspects of the present disclosure in technological and operational context, to assist those of skill in the art in understanding their scope and utility. Approaches described in the Background section could be pursued, but are not necessarily approaches that have been previously conceived or pursued. Unless explicitly identified as such, no statement herein is admitted to be prior art merely by its inclusion in the Background section.

[0025] SUMMARY

[0026] The following presents a simplified summary of the disclosure in order to provide a basic understanding to those of skill in the art. This summary is not an extensive overview of the disclosure and is not intended to identify key / critical elements of aspects of the disclosure or to delineate the scope of the disclosure. The sole purpose of this summary is to present some concepts disclosed herein in a simplified form as a prelude to the more detailed description that is presented later. According to aspects of the present disclosure described and claimed herein, a methodology of enhancing the performance of a wireless communication network by adjusting a configuration of parameters of a target cell discovers similar cells having good performance, and processes the group of cells to eliminate those that do not have a homogeneous parameterization. The cell similarity calculation is hierarchical, first computing the similarity of categories of related parameters or KPIs between each cell and the target cell. The overall cell similarity calculation is then a weighted average of the category similarity measures. A Figure of Merit (FoM), reflecting cell performance, is computed for each cell as a weighted average of a set of KPIs. The set of comparison cells is reduced by filtering with both the cell similarity and FOM metrics. The remaining cells are then dimensionally reduced, clustered (e.g., using an Al model), and the clusters filtered according to a variety of metrics. This filtering ensures a homogenous parameterization among the cells. The filtering continues until only one cluster remains, from which one or more cells are selected to copy parameter values to the target cell.

[0027] One aspect relates to a method of enhancing the performance of a wireless communication network by adjusting a configuration of parameters, of a target cell, based on an analysis of similarity to other cells operative in a wireless communication network and having acceptable performance. A plurality of cells operative in a wireless communication network are identified for comparison to the target cell. A first subset of similar cells is selected from the plurality of cells based on a hierarchical similarity metric calculated as a weighed sum of categories of parameters or KPIs. An FoM representing the cell’s performance is calculated and assigned to each of the first subset of cells. A second subset is generated by reducing the first subset by discarding cells with a low FoM. A final set of cells having both homogeneous parameterization and acceptable performance is selected from the second subset. Zero or more cells are selected from the final set, and if one or more cells are selected, parameter values are copied from the selected cell(s) to apply to the target cell.

[0028] Another aspect relates to a wireless communication network node configured to implement a cell mirroring network function. Processing circuitry is configured to identify a plurality of cells operative in a wireless communication network for comparison to the target cell; select a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighed sum of categories of parameters or KPIs; calculate and assign to each of the first subset of cells an FoM representing the cell’s performance, and generate a second subset by reducing the first subset by discarding cells with a low FoM; select, from the second subset, a final set of cells having both homogeneous parameterization and acceptable performance; and select zero or more cells from the final set, and if one or more cells are selected, copy parameter values from the selected cell(s) to apply to the target cell. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present disclosure will now be described more fully hereinafter with reference to the accompanying drawings, in which aspects of the disclosure are shown. However, this disclosure should not be construed as limited to the aspects set forth herein. Rather, these aspects are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art. Like numbers refer to like elements throughout.

[0030] FIG. 1 is a flow diagram of a method of enhancing the performance of a wireless communication network by adjusting a configuration of parameters, of a target cell, based on an analysis of similarity to other cells operative in the wireless communication network and having acceptable performance.

[0031] FIG. 2 is a flow diagram of a method of selecting a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighed sum of categories of parameters or KPIs.

[0032] FIG. 3 is a flow diagram of a method of selecting, from a second subset, a final set of cells having both homogeneous parameterization and acceptable performance.

[0033] FIG. 4 is a block diagram of a wireless communication network.

[0034] DETAILED DESCRIPTION

[0035] For simplicity and illustrative purposes, the present disclosure is described by referring mainly to an exemplary aspect thereof. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present disclosure. However, it will be readily apparent to one of ordinary skill in the art that the present disclosure may be practiced without limitation to these specific details. In this description, well known methods and structures have not been described in detail so as not to unnecessarily obscure the present disclosure.

[0036] FIG. 1 depicts the basic steps of a full, end-to-end method (10) of enhancing the performance of a wireless communication network by adjusting a configuration of parameters of a target cell, based on an analysis of a similarity to other cells operative in the wireless communication network and having acceptable performance.

[0037] Initially, a plurality of cells operative in the same or a different wireless communication network are identified for comparison to the target cell (step 12).

[0038] Next, a first subset of similar cells is selected from the plurality of cells identified for comparison, based on a hierarchical, weighted similarity metric calculated as a weighed sum of similar categories of parameters or KPIs (step 20). The similarity comparison does not consider operating performance of the cells.

[0039] A Figure of Merit (FoM) is then calculated and assigned to each cell in the first subset of similar cells (step 40). The FoM represents the cell’s performance. A second subset of similar cells is generated by discarding from the first subset all the cells with a low FoM. Steps 20 and 40 yield a second subset of cells having similarity to the target cell, and exhibiting acceptable performance. However, similar cells in this second subset could have comparable performance, but for widely varying reasons.

[0040] In the next step, a final set of cells is selected from the second subset. The cells in the final set have both homogeneous parameterization and acceptable performance (step 60). The calculation of the final set of cells comprises a hierarchical cell similarity calculation that organizes features in flexible and adaptable clusters of cells, whose weights can also be adapted to the specific target parameters ( / .e., parameters to be configured) or other user preferences.

[0041] Finally, zero or more cells are selected from the final set. If one or more cells are selected, parameter values are copied from the selected cell(s) to apply to the target cell.

[0042] In summary, the inventive methodology is based on primary computations (steps 20, 40, 50), with the result of proposing a new parametrization for a single target cell. First, a cell similarity indicator is calculated to detect similar cells in terms of network topology, terrain morphology, configuration, performance, etcetera. Secondly, once similar cells have been selected, the resulting first subset of cells is reduced by filtering out cells that do not fulfill a minimum threshold for a proposed FoM, generating a second subset of cells that are both similar to the target cell and have acceptable performance. Finally, a final set of cells is selected that well represent the parametrization of the second subset. These steps are described in detail below, with reference to FIGs. 2-4.

[0043] Initially, a plurality of cells operative in a wireless communication network are identified for comparison to the target cell (step 12). In a small network, this plurality may include all cells in the same wireless communication network in which the target cell operates. In larger networks, the plurality of cells identified for comparison may include cells that are superficially similar - e.g., urban vs. rural deployment, large vs. small geographic area, or the like. The plurality of cells identified for comparison to the target cell may comprise some or all cells of the same wireless communication network, and / or cells of a different wireless communication network.

[0044] Next, a first subset of similar cells is selected from the plurality of cells identified for comparison (step 20). A cell similarity process was described in International Patent Application Publication No. WO 2022 / 090810, assigned to the assignee of the present disclosure, the disclosure of which is incorporated herein by reference, in its entirety. That application describes a method for calculating a similarity between two cells in a wireless communication network, based primarily on the morphology of their coverage areas and physical configuration of the cells. In this context, morphology refers to the different “clutter types” across a cell’s coverage area (e.g., vegetation, buildings, bridges, roads, seas, lakes, rivers, etcetera), and the terrain elevation across the coverage area. Physical configuration of a cell refers to the network equipment deployed (e.g., antenna tilt, antenna transmit power, etcetera). These two factors were considered particularly important for comparing cells, as they allow the features of existing cells to be compared to newly-planned cells that have not yet been deployed (and hence have no performance indicators for comparison).

[0045] Step 20 of the method 10 differs from the cell comparison in the above-cited patent application in numerous respects. First, the cell comparison described herein is hierarchical. Rather than calculating cell similarity by directly comparing parameter values or KPIs, physical or operating parameters and / or KPIs are grouped into a plurality of categories, and the cell similarity comparison is performed at the category level. Additionally, the comparison described herein is implemented as a weighted sum. Each category is assigned a weight reflective of its importance in the cell comparison. This allows different categories to be weighted by users to reflect that category’s importance in the cell comparison. For example, for a new or planned target cell, categories of KPIs reflecting network operation may be given low or zero weight, and categories relating to cell morphology or physical parameters of the radio equipment may be given high weights.

[0046] The selection of cells for the first subset is then based on the per-category similarity between each cell considered and the target cell. That is, the parameter values and KPIs in each category are compared, and a category similarity metric is calculated. A hierarchical, weighted cell similarity metric is then calculated as the weighted sum of these category similarity metrics.

[0047] FIG. 2 shows the procedures comprising step 20, according to one aspect of the present disclosure. First, parameters or KPIs of interest are grouped into categories (block 22). These categories represent different aspect of the cells, for example clutter profile ( / .e., type of terrain around the cell), elevation profile ( / .e., altitude of the antenna and its coverage area), RF performance, traffic, mobility of its users, configuration, etcetera. The features (parameters or KPIs) within each category are defined to represent the category properly. Both the categories and their features are defined in a flexible way by the user, depending on the parameters that are being optimized and the network operator’s strategy. The categorization may be standardized and used across multiple iterations of the method 10 for different target cells. Alternatively, users may define categories comprised of features of interest for a particular target cell. The following is one representative and non-limiting example of categories and their constituent features (parameters or KPIs):

[0048] Cell Configuration transmission mode number of transmitters number of receivers number of bands in site total bandwidth in site total serving cells in site downlink bandwidth uplink bandwidth

[0049] Mobility Indicator mean distance to neighbor cells weighted by Hand-Overs (HOs) outgoing inter-frequency HOs outgoing intra-frequency HOs incoming inter-frequency HOs incoming intra-frequency HOs average users bearing angle neighbors downlink (DL) traffic weighted by HOs neighbors DL traffic waited by PRB utilization user speed 70th percentile neighbor user speed 70th percentile weighted by HOs

[0050] RF Indicator

[0051] CQI ithpercentile [0..100]

[0052] PUSCH SI NR ithpercentile [0..100]

[0053] RSRP ithpercentile [0..100]

[0054] RSRQ ithpercentile [0..100] uplink (UL) Pathloss ithpercentile [0..100]

[0055] PUSCH interference ithpercentile [0..100]

[0056] Clutter Profile clutter i ratio [1..n] clutter i mean distance [1..n]

[0057] Traffic Indicator busy hour total traffic total traffic as SCell

[0058] PRB DL utilization ithpercentile [0..100]

[0059] PRB UL utilization ithpercentile [0..100] average active users

[0060] Elevation Profile antenna altitude clutter i mean altitude [1..n] distance range i mean elevation [1..d]

[0061] Intra-Frequency Neighbor n fn:1..N1 number of HOs

[0062] HO failure rate HO oscillation rate distance bearing angle to neighbor cell

[0063] Inter-Frequency Neighbor n fn:1..N1 number of HOs

[0064] HO failure rate

[0065] HO oscillation rate distance frequency delta bearing angle to neighbor cell. weight is assigned to each category (block 24). This allows the user to emphasize the parameters or KPIs for which the greatest similarity is desired.

[0066] For each cell in the identified plurality of cells (block 26), a category similarity metric is calculated between each category of the cell and corresponding category of the target cell, based on the values of parameters or KPIs in the category for the two cells (block 28). In one aspect of the present disclosure, this calculation is a cosine similarity calculation: where a and b are the vectors defined by all features in a certain category for two given cells a and b.

[0067] After calculating the similarity between both cells for each category, a hierarchical, weighted cell similarity indicator is calculated between the cell and the target cell as a weighted average of the category similarity metrics (block 30). For example, where w, is the weight defined for the category / , and cs, is the cosine similarity obtained using the features of the category / of both cells (a and b).

[0068] Once Cell Similarity values between the target cell (whose configuration is being adjusted) and the rest of cells being considered are calculated (block 32), a first subset of similar cells is selected based on the hierarchical, weighted cell similarity metrics (block 34). For example, a threshold minCellSimilarity may be used to filter only the most similar cells ( / .e., Cell similarity^ > minCellSimilarity). Alternatively, the N most similar cells can be chosen by selecting the N cells with the highest value of Cell similarity^ . Step 20 of the method 10 creates a first subset of cells having a degree of similarity to the target cell (with the aspects of that similarity controlled by users’ control of the weights assigned to each category). However, nothing about the similarity comparison, and creation of the first subset of similar cells, assures that the these cells are performing well.

[0069] Step 40 of the method 10 (FIG. 1) addresses performance by selecting from the first subset of similar cells, only those that are performing satisfactorily enough. This is done by calculating and assigning to each of the subset of similar cells a Figure of Merit (FoM) representing the cell’s performance, and generating a second subset by reducing the first subset by discarding cells with a low FoM. The FoM is determined by weighted KPIs, where the weight of each KPI is selected depending on target parameters or other user preferences. Hence, the FoM represents the cell’s performance, and the user may select the aspects of cell performance that are most important.

[0070] As one example, FoMs may be calculated based on Operation Support Services (OSS) Performance Management (PM) counters, which are network metrics gathered and logged by wireless communication network equipment. Four categories of such PMs represent the most significant Radio Frequency (RF) issues in the Radio Access Network (RAN). These are: Lack of Dominance (LoD), Out of Coverage (OoC), High PRB Utilization (HPU) and High Uplink (UL) Interference (HUI).

[0071] Lack of Dominance, also known as pilot pollution, refers to RAN overlap, such as at the boundaries of neighboring cells. Some overlap is necessary, to ensure continuous coverage. However, different base station transmissions on the same frequency, and with similar power, cause significant interference, degrading performance. Ideally, one cell should be stronger ( / .e., have higher Signal to Noise Ratio, or SNR) almost everywhere. LoD is thus a network performance metric that reflects the degree of cell coverage overlap, where one cell is not clearly the dominant one.

[0072] Out of Coverage is a well-known condition in which a User Equipment (UE) cannot receive strong enough downlink (DL) signals from a wireless communication network to utilize the network services.

[0073] A Physical Resource Block (PRB) is a time-frequency structure comprising Resource Elements (RE). Each RE is one OFDM symbol on one subcarrier. A PRB spans 12 subcarriers (180KHz with 15-kHz subcarrier spacing), over 7 OFDM symbols, and hence is 12x7=84 REs. In most cases, a PRB pair (spanning two consecutive slots) is the smallest time-frequency resource that can be scheduled to a device. As the name implies, High PRB Utilization is an operating condition in which a high proportion (e.g., >85%) of available PRBs are utilized for DL transmission.

[0074] High UL Interference is also self-explanatory; it is a condition in which excessive interference is experienced in UL transmissions. As one example, these four KPIs are calculated from LTE network Performance Management (PM) counters as:

[0075] Other categories and different KPIs can also be considered depending on the objective, the network technology, the network vendor, etcetera. Once the KPIs are calculated, an FoM, representing the cell’s performance, is calculated and assigned to the cell (FIG. 1, step 40). The FoM may be calculated as a weighted average of the KPIs, for example as:

[0076] Once the FoM is calculated for each cell in the first subset of similar cells, the first subset is reduced by discarding cells with a low FoM (FIG. 1, step 40), generating a second subset of cells. For example, those cells with an FoM below a minThreshold {i.e., FoM, < minThreshold) may be filtered out of the first subset. The value for this threshold must be higher than the FoMtof the target cell. Alternatively, the N best performing cells from the first subset of similar cells may be chosen by selecting the N cells with the highest value of FoMj.

[0077] The combination of filtering for cell similarity (step 20) and performance (step 40) yields a second subset of cells that are both similar to the target cell (in the aspects selected by the use via setting category weights) and have good performance (in the aspects selected by the use via setting performance metric weights). However, the cells could have widely divergent parametrization. Similar cells with different parameter values could both have good performance for several reasons. For example, the parameters and KPIs defining similarity categories may not fully capture all the ways in which cells can differ. Additionally, the impact of a correct parametrization in the designed FoM can be limited, due to the nature of the parameters or even to a bad design of the FoM. Further, different parametrizations could result in almost optimal performance at the same time. That is, different parameter values could have different impact on the KPIs of the designed FoM, but with the same composite result (e.g., one parametrization could increase coverage but decrease interference, while another increases interference but decrease coverages).

[0078] For these (and possibly other) reasons, cells in the second subset, which are both similar and have good performance, could have very different parametrization. To propose the best parameterization of the target cell, it is important to ensure that the similar, goodperforming cells have good performance for the same reasons - that is, they have a homogeneous parameterization. Filtering out cells whose parameterization differs from the norm within the second subset is thus an important aspect to the method 10.

[0079] Accordingly, in step 50, a final set of cells - having both homogeneous parameterization and acceptable performance - is selected from the second subset ( / .e., the cells remaining after step 40). Selection of this final set must satisfy several conditions:

[0080] • The methodology must be robust to non-optimal design of the hierarchical, weighted cell similarity metric and the FoM.

[0081] • It is necessary to find the most common parametrization, but also assuring that it represents the second subset of cells remaining after step 40 and is not a parametrization that only works in some cases, or it is extremely different from the rest.

[0082] • The final set of cells selected must not be negatively impacted by its parametrization in terms of performance compared with the rest of cells in the similar, good-performing second subset. That is, the average FoM of the final set of cells should not be much lower than the average FoM of all the cells remaining in the second subset after step 40.

[0083] • The final set of cells selected must not be less similar, on average, than the cells in the second subset remaining after step 40. This means that the hierarchical, weighted cell similarity of the cells in the final set of cells should not be much lower than the same metric for the rest of cells in the second subset remaining after step 40.

[0084] To satisfy these criteria, step 50 of the method 10 comprises a multi-step methodology that may employ Artificial Intelligence (Al). FIG. 3 depicts the operations comprising step 50, according to one aspect of the present disclosure.

[0085] Initially, the dimensionality of parameters for each cell in the FoM-filtered subset is reduced (block 52). The dimensionality reduction operation takes as input the target parameters of each cell. In the second subset of cells ( / .e., those remaining after step 40), some parameters will have the same value, or a very low diversity, for all the cells in the subset. A latent space reduction allows for capturing all the relevant information with the minimum number of variables. The dimensionality reduction thus discards redundancy to facilitate further processing.

[0086] The dimensionality reduction can be carried out with different methodologies, such as Principal Component Analysis (PCA), autoencoders, etcetera. In one aspect of the present disclosure, a PCA transforms the N target parameters to a latent space of two dimensions. Those of skill in the art will realize that PCA is only one option, and other techniques could be utilized. Additionally, a different number of final dimensions can be considered, depending on the target parameters and the diversity of the parametrizations. Note that if the number of target parameters is equal or lower than two, dimensionality reduction is not necessary.

[0087] The cells are then clustered, based on the reduced dimensionality of parameters, to form clusters wherein the cells in each cluster have homogeneous parametrizations (block 54). This process selects which group of cells will be used as mirroring cells for the parametrization of the target cell. The clustering must ensure that the diversity in the parametrization of the cells in each group is as low as possible.

[0088] One class of clustering algorithms comprises unsupervised learning techniques whose goal is to group data samples (in this case, the result of the latent space reduction for the target parameters in one cell) into clusters, in such a way that samples in the same cluster are as similar as possible. While many clustering algorithms are available, one that works well is a Gaussian Mixture Model (GMM) with a set of different hyperparameters whose final value is selected to maximize the Silhouette score. As well known in the art, a Silhouette score is a measure of how similar an object is to its own cluster (cohesion), compared to other clusters (separation). A high value indicates that the cell is well matched to its own cluster and poorly matched to neighboring clusters.

[0089] Once the cells are aggregated into clusters, a plurality of metrics are calculated for each cluster (block 56). These are used to verify whether the conditions listed above are fulfilled. The metrics calculated may include: a) The standard deviation of cells in each cluster, as a measure of how homogeneous the parametrization in each cluster is. The standard deviation is calculated for cluster / as where N is the number of cells in cluster i, p, is a vector of the values of the target parameters for cell j, and ptis the average parametrization for the cells in cluster / . b) The average hierarchical, weighted cell similarity of each cluster. This is a measure of the similarity of cells in the cluster and the target cell. c) The average FoM of each cluster. This is a measure of how well the cells in each cluster are performing. d) The difference between the average parametrization in the cluster and the average parametrization in the second subset, calculated as: where Parametrizationclusteris the average parametrization for the cluster / , and

[0090] Parametrization is the average parametrization for the all the cells in the second subset that remain after steps 20 and 40.

[0091] These metrics, together with the Silhouette score obtained in the clustering, are used to select the final set of cells ( / .e., final cluster) by successively reducing the number of cell clusters in the second subset by filtering based on the metrics, to generate a final set of cells (block 58). The filtering operations are listed below. Note that this process begins with the second subset of cells grouped into clusters. The reduction at each step below operates on the cells remaining after executing the previous step. Although these operations may be performed in any order, the following order has been determined to yield the best results, and is the presently preferred order of execution.

[0092] 1. If the Silhouette score is below a predetermined threshold (minSilhouetteScore), then all the cells are discarded, and the method 10 does not provide a recommendation.

[0093] 2. Clusters with less than N cells are discarded. A presently preferred value for N is A / = 10, although other values may be used.

[0094] 3. The Nd clusters with the highest ParametrizationDeltaare discarded. The presently preferred value for Nd is 10% of the remaining clusters, although this value is not limiting. This operation is to remove outlier clusters.

[0095] 4. The Naclusters with the highest o, are discarded, where Nais defined by the user. The presently preferred value for Nais such that 50% of the remaining clusters are discarded. This retains only the clusters having the most homogenous parameterization.

[0096] 5. The Nb clusters with the lowest average hierarchical, weighted cell similarity are discarded, where Nb is defined by the user. The presently preferred value for Nb is such that 60% of the remaining clusters are discarded. This retains only the clusters having the greatest similarity to the target cell.

[0097] 6. The Ncclusters with the lowest average FoM are discarded, where Ncis defined by the user. The presently preferred value for Ncis such that 60% of the remaining clusters are discarded. This retains the highest performing cells.

[0098] 7. From the remaining clusters, the cluster with the lowest value of o, is selected as the final cluster to be used for recommending one or more cells from which to copy parameter values.

[0099] Note that the values Na, Nb, Nc, and Nd must be integers, and are calculated automatically based on the number of clusters remaining in the subset after the previous step, and the desired percentages (e.g., the percentages recommended in steps 3-6 above). The values are obtained by applying the given percentage to the remaining clusters, and rounding down. Although the percentages listed above are presently preferred, they are not a limitation of the method 10. However, the maximum recommended value for each such percentage is 50%. The value of / V can also be adjusted depending on the number of cells in the network, or other factors.

[0100] The method 10 concludes with step 60, in which zero or more cells are selected from the final set of cells ( / .e., the final remaining cluster). If one or more cells are selected, parameter values are copied from the selected cell(s) to apply to the target cell. This provides the target cell with a configuration of at least initial parameter values taken from similar cells that are known to achieve good performance, and where the performance is reasonably expected to result from the parameter configuration ( / .e., the cell(s) is selected from among those having homogeneous parameterization). The method 10 thus delivers a complete, end-to-end methodology of initially configuring or upgrading the configuration of a target cell. The method 10 is flexible and easily user-configured. The metrics of “similarity” (the hierarchical, weighted cell similarity metric) and “performance” (FoM) are calculated as weighted sums, where users can adjust the weights to reflect aspects of cell similarity and performance that are most important to the particular target cell being parametrized.

[0101] FIG. 4 depicts a wireless communication network 100, such as a 3GPP 5G network 100. The network 100 comprises a RAN 102 and a core network (CN) 112. The RAN 102 comprises a plurality of base stations (e.g., gNBs) 104A, 104B. The base station 104A serves two UEs 106A, 106B in a first cell 108. The base station 104B serves a UE 106C in a second cell 110. In general, a large number of base stations 104 may each serve a very large plurality of UEs 106, each in a different cell. According to aspects of the present disclosure, a configuration of parameters for a cell 108, 110 may be determined by executing the method 10 (FIG. 1) on a large number of cells in the network 100, or in different wireless communication networks. The CN 112 of the network 100 includes one or more Network Data Analytics Functions (NWDAF) 114. According to aspects of the present disclosure, the NWDAF 114 includes a Cell Mirroring Network Function (NF) 116. The Cell Mirroring NF 116 includes at least communication circuitry 118 configured to communicate with other network nodes, and processing circuitry 120 configured to execute the method 10 described herein. The communication circuitry 118 and processing circuitry 120 are connected in data communication mode. FIG. 4 depicts the Cell Mirroring NF 116 as a separate node; however, as those of skill in the art are aware, NFs may be implemented in a variety of ways, including on a virtual node in the cloud.

[0102] Input data for the method 10 may be obtained from a variety of sources. As representative and non-limiting examples, input data sources may include:

[0103] • PM counters: performance management counters containing network statistics.

[0104] • CM Parameters: configuration management parameters containing network parameter configuration.

[0105] • Crowdsourcing data: this data source offers rich user geo-located information obtained from applications installed in the UEs. • Elevation maps: data specifying the terrain elevation in the network area with a certain resolution.

[0106] • Clutter maps: data specifying the type of clutter (e.g., water, low vegetation, high vegetation, road, low buildings, etcetera) in the network area with a certain resolution.

[0107] • Network topology information: topological information including the following information for each cell: o Antenna height o Antenna aximuth o Antenna location (e.g., longitude and latitude)

[0108] Aspects of the present disclosure are not limited to the above-listed data sources. Other data sources with additional information of the cells may also be considered, as required or desired for a given implementation (e.g., user traces, cell traces, etcetera).

[0109] Method 10 described herein provides a recommended value for target parameters in a target cell. The output of the method 10 is either no recommendation (if the Silhouette score from the clustering operation is too low), or one or more cells in the final remaining cluster, the parameter configuration of which is recommended for the target cell. This information can be presented in a variety of ways, including:

[0110] • Providing the list of cells in the final cluster selected in step 50, that contains the cells whose configuration can be used to configure the target cell.

[0111] • Providing the distribution for the target parameters in the cluster of cells that have been finally selected in step 50.

[0112] • Providing a unique recommended value for each of the target parameters by aggregating the values of each parameter in the cluster that has been finally selected in step 50. This aggregation can be done by calculating the average, the mode, the median, or other aggregation methodology.

[0113] In general, the method 10 may be applied whenever the parameter configuration of a cell is desired. Two use cases in particular may benefit from the method: new site deployment and worst offender cell optimization.

[0114] Method 10 can be applied over greenfield design or roll-out activities for new cells. The methodology is a new way to estimate the best setup for a new cell, based on similar cells already deployed and operating. In this application, the hierarchical cell similarity is based on the physical cell configuration (e.g., radio equipment) and the morphology of the target area to be covered, which so not depend on the cell being operative (or even existing).

[0115] Method 10 also finds particular applicability to solve misconfiguration issues that are causing bad performance. Once worst offender cells (due to a misconfiguration issue) are detected, then the full methodology can be adapted with the aim of solving the configuration issue by copying the parametrization of a similar cell that is performing satisfactorily enough. In this application, the features, categories, and settings are adapted for the parameters that are causing the bad performance.

[0116] Of course, aspects of the present disclosure are not limited to the above applications. In general, method 10 may be utilized to optimize any set of parameters, even if cells do not have obvious performance issues.

[0117] The present disclosure may, of course, be carried out in other ways than those specifically set forth herein without departing from essential characteristics of the disclosure. The present aspects are to be considered in all respects as illustrative and not restrictive, and all changes coming within the meaning and equivalency range of the appended aspects are intended to be embraced therein.

Claims

CLAIMSWhat is claimed is:1 . A method (10) of enhancing the performance of a wireless communication network by adjusting a configuration of parameters of a target cell, based on an analysis of similarity to other cells operative in a wireless communication network and having acceptable performance, the method (100) characterized by: identifying (12) a plurality of cells operative in a wireless communication network for comparison to the target cell; selecting (20) a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighted sum of categories of parameters or Key Performance Indicators (KPI); calculating (40) and assigning to each of the first subset of cells a Figure of Merit, FoM, representing the cell’s performance, and generating a second subset by reducing the first subset by discarding cells with a low FoM; selecting (50), from the second subset, a final set of cells having both homogeneous parameterization and acceptable performance; and selecting (60) zero or more cells from the final set, and if one or more cells are selected, copying parameter values from the selected cell(s) to apply to the target cell.

2. The method (10) of claim 1 wherein selecting (20) a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighted sum of categories of parameters or KPIs is characterized by: grouping (22) parameters or KPIs into categories; assigning (24) a weight to each category; for each cell in the plurality of cells (26, 32), calculating (28) a category similarity metric between each category of the cell and corresponding category of the target cell, based on the values of parameters or KPIs in the category for the two cells; calculating (30) a hierarchical, weighted cell similarity metric between the cell and the target cell as a weighted average of the category similarity metrics; and selecting (34) a first subset of similar cells based on the hierarchical, weighted cell similarity metrics.

3. The method (10) of claim 2 wherein calculating (28) a category similarity metric between each category of the cell and corresponding category of the target cell, based on the values of parameters or KPIs in the category for the two cells comprises calculating the category similarity metric using a cosine similarity formula.

4. The method (10) of claim 2 wherein selecting (34) a first subset of similar cells based on the hierarchical, weighted cell similarity metrics comprises selecting all cells having a hierarchical, weighted cell similarity metric greater than a predetermined threshold value.

5. The method (10) of claim 2 wherein selecting (34) a first subset of similar cells based on the hierarchical, weighted cell similarity metrics comprises selecting a number N of cells having the highest weighted cell similarity metrics.

6. The method (10) of claim 1 wherein calculating (40) and assigning to each of the first subset of similar cells an FoM representing the cell’s performance comprises: calculating two or more KPIs for each cell based on network performance metrics; and calculating the FoM for each cell as a weighted average of the KPIs.

7. The method (10) of claim 1 wherein generating (40) a second subset by reducing the first subset by discarding cells with a low FoM comprises selecting all cells having an FoM greater than a predetermined threshold value.

8. The method (10) of claim 1 wherein generating (40) a second subset by reducing the first subset by discarding cells with a low FoM comprises selecting a number N of cells having the highest FoM.

9. The method (10) of claim 1 wherein selecting (50), from the second subset, a final set of cells having both homogeneous parameterization and acceptable performance is characterized by: reducing (52) a dimensionality of parameters for each cell in the second subset; clustering (54) the cells based on the reduced dimensionality of parameters, to form clusters wherein the cells in each cluster have homogeneous parameterizations; calculating (56) a plurality of metrics for each cluster; successively reducing (58) the number of cell clusters in the second subset by filtering based on the metrics, to generate a final set of cells.

10. The method (10) of claim 9 wherein reducing (52) a dimensionality of parameters for each cell in the second subset comprises applying a Principal Component Analysis to reduce the dimensionality to two.

11. The method (10) of claim 9 wherein clustering (54) the cells based on the reduced dimensionality of parameters, to form clusters wherein the cells in each cluster havehomogeneous parameterizations comprises applying a Gaussian Mixture Model with a set of different hyperparameters whose final value is selected to maximize a Silhouette score representing how similar a cell is to its own cluster, compared to other clusters.

12. The method (10) of claim 9 wherein calculating a plurality (56) of metrics for each cluster comprises calculating a standard deviation as a measure of how homogeneous the parametrization in each cluster is.

13. The method (10) of claim 9 wherein calculating (56) a plurality of metrics for each cluster comprises calculating an average hierarchical, weighted cell similarity of each cluster.

14. The method (10) of claim 9 wherein calculating (56) a plurality of metrics for each cluster comprises calculating an average of the FoMs of the cells in each cluster.

15. The method (10) of claim 9 wherein calculating (56) a plurality of metrics for each cluster comprises calculating a difference between an average parametrization in the cluster and an average parametrization in the second subset.

16. The method (10) of claim 9 wherein successively reducing (58) the number of cell clusters in the second subset comprises, if a Silhouette score reflecting the effectiveness of clustering is below a predetermined threshold, discarding all cells, and wherein selecting zero or more cells from the final set comprises selecting zero cells.

17. The method (10) of claim 9 wherein reducing (58) the number of cell clusters in the second subset comprises discarding cell clusters having fewer than a predetermined number of cells.

18. The method (10) of claim 15 wherein reducing (58) the number of cell clusters in the second subset comprises discarding a number Ndof clusters with the highest difference between the average parametrization in the cluster and the average parametrization in the second subset.

19. The method (10) of claim 12 wherein reducing (58) the number of cell clusters in the second subset comprises discarding a number Naof clusters with the highest standard deviation.

20. The method (10) of claim 13 wherein reducing (58) the number of cell clusters in the second subset comprises discarding a number Nb of clusters with the lowest average hierarchical, weighted cell similarity.

21. The method (10) of claim 14 wherein reducing (58) the number of cell clusters in the second subset comprises discarding a number Ncof clusters with the lowest average FoM.

22. The method (10) of claims 17-21 wherein selecting (60) zero or more cells from the final set comprises selecting, from the clusters remaining after the operations of claims 17-21, the cluster with the minimum value of standard deviation; and wherein selecting (60) zero or more cells from the final set comprises selecting one or more cells from the selected cluster with the minimum value of standard deviation.

23. A wireless communication network node (116) configured to implement a cell mirroring network function, characterized by: communication circuitry (118); and processing circuitry (120) operatively connected to the communication circuitry (118) and configured to: identify (12) a plurality of cells operative in a wireless communication network for comparison to the target cell; select (20) a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighted sum of categories of parameters or Key Performance Indicators (KPI); calculate (40) and assign to each of the first subset of cells a Figure of Merit, FoM, representing the cell’s performance, and generate a second subset by reducing the first subset by discarding cells with a low FoM; select (50), from the second subset, a final set of cells having both homogeneous parameterization and acceptable performance; and select (60) zero or more cells from the final set, and if one or more cells are selected, copying parameter values from the selected cell(s) to apply to the target cell.

24. The network node (116) of claim 23, wherein the processing circuitry (118) is configured to select (20) a first subset of similar cells from the plurality of cells based on a hierarchical similarity metric calculated as a weighted sum of categories of parameters or KPIs by: grouping (22) parameters or KPIs into categories;assigning (24) a weight to each category; for each cell in the plurality of cells (26, 32), calculating (28) a category similarity metric between each category of the cell and corresponding category of the target cell, based on the values of parameters or KPIs in the category for the two cells; calculating (30) a hierarchical, weighted cell similarity metric between the cell and the target cell as a weighted average of the category similarity metrics; and selecting (34) a first subset of similar cells based on the hierarchical, weighted cell similarity metrics.

25. The network node (116) of claim 24, wherein the processing circuitry (118) is configured to calculate (28) a category similarity metric between each category of the cell and corresponding category of the target cell, based on the values of parameters or KPIs in the category for the two cells by calculating the category similarity metric using a cosine similarity formula.

26. The network node (116) of claim 24, wherein the processing circuitry (118) is configured to select (34) a subset of similar cells based on the hierarchical, weighted cell similarity metrics by selecting all cells having a hierarchical, weighted cell similarity metric greater than a predetermined threshold value.

27. The network node (116) of claim 24, wherein the processing circuitry (118) is configured to select (34) a subset of similar cells based on the hierarchical, weighted cell similarity metrics by selecting a number N of cells having the highest weighted cell similarity metrics.

28. The network node (116) of claim 23, wherein the processing circuitry (118) is configured to calculate (40) and assign to each of the first subset of similar cells an FoM representing the cell’s performance by: calculating two or more KPIs for each cell based on network performance metrics; and calculating the FoM for each cell as a weighted average of the KPIs.

29. The network node (116) of claim 23, wherein the processing circuitry (118) is configured to generate (40) a second subset by reducing the first subset by discarding cells with a low FoM by selecting all cells having an FoM greater than a predetermined threshold value.

30. The network node (116) of claim 23, wherein the processing circuitry (118) is configured to generate (40) a second subset by reducing the first subset by discarding cells with a low FoM by selecting a number N of cells having the highest FoM.

31. The network node (116) of claim 23, wherein the processing circuitry (118) is configured to select (50), from the second subset, a final set of cells having both homogeneous parameterization and acceptable performance by: reducing (52) a dimensionality of parameters for each cell in the second subset; clustering (54) the cells based on the reduced dimensionality of parameters, to form clusters wherein the cells in each cluster have homogeneous parameterizations; calculating (56) a plurality of metrics for each cluster; successively reducing (58) the number of cell clusters in the second subset by filtering based on the metrics, to generate a final set of cells.

32. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to reduce (52) a dimensionality of parameters for each cell in the second subset by applying a Principal Component Analysis to reduce the dimensionality to two.

33. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to cluster (54) the cells based on the reduced dimensionality of parameters, to form clusters wherein the cells in each cluster have homogeneous parameterizations by applying a Gaussian Mixture Model with a set of different hyperparameters whose final value is selected to maximize a Silhouette score representing how similar a cell is to its own cluster, compared to other clusters.

34. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to calculate a plurality (56) of metrics for each cluster by calculating a standard deviation as a measure of how homogeneous the parametrization in each cluster is.

35. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to calculate (56) a plurality of metrics for each cluster by calculating an average hierarchical, weighted cell similarity of each cluster.

36. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to calculate (56) a plurality of metrics for each cluster by calculating an average of the FoMs of the cells in each cluster.

37. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to calculate (56) a plurality of metrics for each cluster by calculating a difference between an average parametrization in the cluster and an average parametrization in the second subset.

38. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to successively reduce (58) the number of cell clusters in the second subset by, if the Silhouette score is below a predetermined threshold, discarding all cells, and wherein selecting zero or more cells from the final set comprises selecting zero cells.

39. The network node (116) of claim 31 , wherein the processing circuitry (118) is configured to reduce (58) the number of cell clusters in the second subset by discarding cell clusters having fewer than a predetermined number of cells.

40. The network node (116) of claim 37, wherein the processing circuitry (118) is configured to reduce (58) the number of cell clusters in the second subset by discarding a number Ndof clusters with the highest difference between the average parametrization in the cluster and the average parametrization in the second subset.

41. The network node (116) of claim 34, wherein the processing circuitry (118) is configured to reduce (58) the number of cell clusters in the second subset by discarding a number Naof clusters with the highest standard deviation.

42. The network node (116) of claim 35, wherein the processing circuitry (118) is configured to reduce (58) the number of cell clusters in the second subset by discarding a number Nb of clusters with the lowest average hierarchical, weighted cell similarity.

43. The network node (116) of claim 36, wherein the processing circuitry (118) is configured to reduce (58) the number of cell clusters in the second subset by discarding a number Ncof clusters with the lowest average FoM.

44. The network node (116) of claims 39-43, wherein the processing circuitry (118) is configured to select (60) zero or more cells from the final set by selecting, from the clusters remaining after the operations of claims 39-43, the cluster with the minimum value of standard deviation; and wherein selecting (60) zero or more cells from the final set comprises selecting one or more cells from the selected cluster with the minimum value of standard deviation.

Images