Abstract: Embodiments include computer-implemented methods of forecasting performance issues associated with cells in a wireless network. Such methods include for each of a plurality of cells in the wireless network, obtaining first time-series data representing a plurality of key performance indicators (KPIs) at a first plurality of past time points and applying a multi-layer forecasting model to the first time-series data to determine second time-series data representing the plurality of KPIs for each of the plurality of cells at a second plurality of future time points. Such methods include, based on the second time-series data, determining respective probabilities of occurrence for a respective plurality of different performance issues, associated with the plurality of cells, during one or more future time periods that include the second plurality of future time points. Other embodiments include network optimization functions associated with the wireless network, which are configured to perform such methods.

Abstract: A method is provided for identifying operating conditions of a system. Input data relating to operation of the system is applied to a multi-class model for classification, where the multi-class model is configured for classifying the data into one of a plurality of predefined classes, and each class corresponds to a respective operating condition of the system. A confidence level of the classification by the multi-class model is determined. If the confidence level is below a threshold confidence level, the input data is applied to a plurality of binary models, where each binary model is configured for determining whether the data is or is not in a respective one of the predefined classes. If the plurality of binary models determine that the data is not in any of the respective predefined classes, the data can be taken into consideration when updating the multi-class model.

Abstract: The invention refers to method, performed by a network optimization function associated with a wireless network (100), of forecasting performance issues associated with cells (106, 111, 116) in the wireless network, the method comprising: for each of a plurality of cells in the wireless network, obtaining (1210) first time-series data representing a plurality of key performance indicators, KPIs, at a first plurality of past time points; applying (1220) a multi-layer forecasting model to the first time-series data to determine second time-series data representing the plurality of KPIs for each of the plurality of cells at a second plurality of future time points; and based on the second time-series data, determining (1240) respective probabilities of occurrence for a respective plurality of different performance issues, associated with the plurality of cells, during one or more future time periods that include the second plurality of future time points; the method further refers to a corresponding network opti

Abstract: A method 300 is disclosed for generating and training a model to detect interference conditions at a cell in a wireless cellular network and to classify the impact of detected interference conditions on performance of the wireless cellular network in the cell. The method comprises, for each of a plurality of cells in the wireless cellular network (360), obtaining data representing received signal power at a base station serving the cell over a period of time (310) and obtaining data representing a plurality of performance metrics for the cell over the time period (330). The method further comprises obtaining classifications of the received signal power data into one of a plurality of cell interference conditions (320) and the performance metric data into one of a plurality of cell impact classes (340).

Abstract: A method includes extracting, from first log data relating to operations of wireless devices and/or nodes in the communication network, textual elements and numerical elements; transforming the textual elements to a first vector space to determine respective textual element vectors; transforming the numerical elements to a second vector space to determine respective numerical element vectors; embedding and clustering the textual element vectors and the numerical element vectors to determine clusters of embedded vectors, wherein the embedding includes, for a plurality of wireless device sessions, embedding at least one textual element vector and at least one numerical element vector into a single embedded vector representing the particular wireless device session; and training a classifier model to determine a network status from second log data, wherein the classifier model is trained using the plurality of clusters of embedded vectors.

Abstract: The invention refers to a method performed by a performance recommender for a wireless network, obtaining (2010) for a plurality of cells input data, the input data comprising actual cell configuration parameter values; applying (2030) a machine-learning model to the input data to generate, for at least a portion of the cells, one or more recommendations for changes to the cell configuration parameter values to improve uplink, UL, performance in the respective cells; and based on identifying conflicts between recommendations for different cells, partitioning (2040) the plurality of cells into a plurality of interaction areas of neighboring cells; resolving (2050) conflicts in recommendations for respective cells within each of the interaction areas and across different interaction areas; and for at least a portion of the cells, determining (2060) preferred values for the cell configuration parameters to improve UL performance in the respective cells; the invention further relates to a corresponding performanc

Abstract: The invention refers to method, performed by a network optimization function associated with a wireless network (100), of forecasting performance issues associated with cells (106, 111, 116) in the wireless network, the method comprising: for each of a plurality of cells in the wireless network, obtaining (1210) first time-series data representing a plurality of key performance indicators, KPIs, at a first plurality of past time points; applying (1220) a multi-layer forecasting model to the first time-series data to determine second time-series data representing the plurality of KPIs for each of the plurality of cells at a second plurality of future time points; and based on the second time-series data, determining (1240) respective probabilities of occurrence for a respective plurality of different performance issues, associated with the plurality of cells, during one or more future time periods that include the second plurality of future time points; the method further refers to a corresponding network opti

Abstract: A method is provided for identifying operating conditions of a system. Input data relating to operation of the system is applied to a multi-class model for classification, where the multi-class model is configured for classifying the data into one of a plurality of predefined classes, and each class corresponds to a respective operating condition of the system. A confidence level of the classification by the multi-class model is determined. If the confidence level is below a threshold confidence level, the input data is applied to a plurality of binary models, where each binary model is configured for determining whether the data is or is not in a respective one of the predefined classes. If the plurality of binary models determine that the data is not in any of the respective predefined classes, the data can be taken into consideration when updating the multi-class model.

Abstract: The present invention refers to a method and apparatus and computer program for detecting communities in a radio access network with a plurality of vertices (C, BBU), wherein the method comprises calculating (S11) relationship strengths (31, 41) for network connections between the vertices (C, BBU) by executing a connection strength calculation process by analyzing performance factors, decisive for network performance including cell coverage overlap, signal strength, and inter-cell interference. The method further creates (S12) a network graph with nodes, representing the vertices (C,BBU) in the network and edges, representing the network connections based on the calculated relationship strengths (31, 41) and applies (S13) an extended iterative disjoint community detection algorithm for clustering nodes into communities, wherein in each iteration imposed one or more constraints for clustering cooperating nodes into the same community are analyzed (S14).

Abstract: A method 300 is disclosed for generating and training a model to detect interference conditions at a cell in a wireless cellular network and to classify the impact of detected interference conditions on performance of the wireless cellular network in the cell. The method comprises, for each of a plurality of cells in the wireless cellular network (360), obtaining data representing received signal power at a base station serving the cell over a period of time (310) and obtaining data representing a plurality of performance metrics for the cell over the time period (330). The method further comprises obtaining classifications of the received signal power data into one of a plurality of cell interference conditions (320) and the performance metric data into one of a plurality of cell impact classes (340).

Abstract: Embodiments of the disclosure provide methods, apparatus and computer-readable mediums for the detection of cell conditions in a wireless cellular network, and the training of a classifier model to detect cell conditions in a wireless cellular network. In one embodiment, a method of training a classifier model to detect cell conditions in a wireless cellular network comprises: obtaining time-series data for a plurality of performance metrics for each cell of a plurality of cells of the wireless cellular network; converting the time-series data to respective image data sets for each cell of the plurality of cells; classifying the image data sets into one of a plurality of predefined cell conditions; and applying a machine-learning algorithm to training data comprising the classified image data sets to generate a classifier model for classifying image data sets into one of the plurality of predefined cell conditions.

Abstract: The present invention refers to a method and apparatus and computer program for detecting communities in a radio access network with a plurality of vertices (C, BBU), wherein the method comprises calculating (S11) relationship strengths (31, 41) for network connections between the vertices (C, BBU) by executing a connection strength calculation process by analyzing performance factors, decisive for network performance including cell coverage overlap, signal strength, and inter-cell interference. The method further creates (S12) a network graph with nodes, representing the vertices (C,BBU) in the network and edges, representing the network connections based on the calculated relationship strengths (31, 41) and applies (S13) an extended iterative disjoint community detection algorithm for clustering nodes into communities, wherein in each iteration imposed one or more constraints for clustering cooperating nodes into the same community are analyzed (S14).