Procedure for optimization of self-organizing network

1. An apparatus for use by a communication network control element or function configured to control a setting of parameters of a self-organizing communication network, the apparatus comprising

at least one processing circuitry, and

at least one memory for storing instructions to be executed by the processing circuitry,

wherein the at least one memory and the instructions are configured to, with the at least one processing circuitry, cause the apparatus at least:

to obtain a pre-trained network optimization model indicating a mapping between a communication network environment forming an input of the model, optimization actions or decisions forming an output of the model, and system performance indicators forming a reward,

to cause sending, to at least one communication network element or function forming a part of the self-organizing communication network, a request for providing similarity data for a similarity analysis allowing to determine a similarity between a part of the self-organizing communication network for which the pre-trained network optimization model is derived and a part of the self-organizing communication network of the at least one communication network element or function to which the request is sent,

wherein the similarity data are related to corresponding network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network of the at least one communication network element or function to which the request is sent, and

wherein the corresponding network properties comprise data related to a single data point and data related to statistical measures and include at least one of location information, geographical features, mobility patterns, data demand statistics, and histograms approximating a probability density function,

to perform the similarity analysis for determining a similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and each part of the self-organizing communication network for which similarity data for the similarity analysis are received, wherein the similarity analysis is performed by determining a similarity measure between the network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived, and the network properties of the part of the self-organizing communication network of the at least one communication network element or function to which the request is sent,

to determine, on the basis of the similarity analysis, at least a part of the pre-trained network optimization model to be provided to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received, wherein the part of the pre-trained network optimization model comprises at least one convolutional layer of the pre-trained network optimization model, and

to cause sending of the determined part of the pre-trained network optimization model to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received,

wherein the pre-trained network optimization model is derived by using a deep reinforcement learning algorithm based on a plurality of convolutional layers employed for capturing spatial and temporal correlations between a network environment of the part of the self-organizing communication network, and a plurality of fully connected layers employed for reducing dimensions of data processing in the plurality of convolutional layers to a desired output dimension, wherein the pre-trained network optimization model is defined by a set of model parameters and hyperparameters.

2. The apparatus according to claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

to obtain the pre-trained network optimization model by receiving data including the pre-trained network optimization model from a communication network element or function belonging to a part of the self-organizing communication network, wherein the apparatus is comprised in a central unit being in charge of collecting and managing usage of pre-trained network optimization models derived in the self-organizing communication network.

3. The apparatus according to claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

to obtain the pre-trained network optimization model by deriving the pre-trained network optimization model from data and measurements conducted in a part of the self-organizing communication network, wherein the apparatus is comprised in communication network element or function forming a part of a distributed system for managing usage of pre-trained network optimization models derived in the self-organizing communication network.

4. The apparatus according to claim 1 , wherein, for deriving the pre-trained network optimization model, input data are prepared by

determining how users of the communication network part are spatially distributed in the part of the self-organizing communication network and how the spatial distribution of the users evolves over time,

determining a user activity level of the users in a specified time period, and

generating an input sample including at least one of an image and a sequence of images in which a position of each pixel corresponds to a geographical grid or physical location of the users and an intensity or color value of each pixel corresponds to a user activity level aggregated over the geographical grid at the specified time period.

5. The apparatus according to claim 1 , wherein

the model parameters include at least one of weight matrices between each two successive layers of the plurality of convolutional layers and the plurality of fully connected layers, and bias vectors between each two successive layers of the plurality of convolutional layers and the plurality of fully connected layers, and

the hyperparameters include at least one of a number of layers of the plurality of convolutional layers and the plurality of fully connected layers, a number of units at each layer of the plurality of convolutional layers and the plurality of fully connected layers, a type of an activation function, a number of filters and filter size in each of the plurality of convolutional layers, a stride size of each maximum or average pooling layer.

6. The apparatus according to claim 1 , wherein the request for providing similarity data for a similarity analysis caused to be sent to at least one communication network element or function forming a part of the self-organizing communication network is included in an indication that a pre-trained network optimization model is present.

7. The apparatus according to claim 1 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

for performing the similarity analysis for determining the similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and a part of the self-organizing communication network for which similarity data for the similarity analysis are received, to calculate a similarity measure on the basis of network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network for which similarity data for the similarity analysis are received,

wherein the network properties comprise data related to a single data point and data related to statistical measures and include at least one of location information, geographical features, mobility patterns, data demand statistics, and histograms approximating a probability density function.

8. The apparatus according to claim 7 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

to process, for determining at least a part of the pre-trained network optimization model to be provided to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received, the calculated similarity measure, and

to select, for forming the part of the pre-trained network optimization model to be provided, a subset of parameters and hyperparameters defining low and medium layers of the pre-trained network optimization model,

wherein the higher the similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network for which similarity data for the similarity analysis are received is according to the similarity measure, the higher the number of parameters and hyperparameters selected for the subset becomes.

9. An apparatus for use by a communication network element or function configured to conduct a setting of parameters of a self-organizing communication network, the apparatus comprising

at least one processing circuitry, and

at least one memory for storing instructions to be executed by the processing circuitry,

wherein the at least one memory and the instructions are configured to, with the at least one processing circuitry, cause the apparatus at least:

to receive, from a communication network control element or function forming a part of the self-organizing communication network, and process a request for providing similarity data for a similarity analysis allowing to determine a similarity between parts of the self-organizing communication network, wherein the similarity data are related to network properties and comprise data related to a single data point and data related to statistical measures and include at least one of location information, geographical features, mobility patterns, data demand statistics, and histograms approximating a probability density function,

to decide whether similarity data are required to be sent,

in case the decision is affirmative, to cause sending of the requested similarity data to the communication network control element or function from which the request is received,

to receive data indicating at least a part of a pre-trained network optimization model indicating a mapping between a communication network environment forming an input of the model, optimization actions or decisions forming an output of the model, and system performance indicators forming a reward, and

to process the data indicating at least a part of a pre-trained network optimization model for generating an own network optimization model by adapting the received pre-trained network optimization model to the own part of the self-organizing communication network, and

for processing the data indicating at least a part of a pre-trained network optimization model for generating an own network optimization model, to conduct at least one of a fine-tuning of the received part of the pre-trained network optimization model and an updating of the pre-trained network optimization model with data collected in the own part of the self-organizing communication network by using transfer learning.

10. The apparatus according to claim 9 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

to receive the request for providing similarity data for the similarity analysis from a central unit being in charge of collecting and managing usage of pre-trained network optimization models derived in the self-organizing communication network, or from a communication network element or function deriving the pre-trained network optimization model and forming a part of a distributed system for managing usage of pre-trained network optimization models derived in the self-organizing communication network,

wherein the request for providing similarity data is included in an indication that a pre-trained network optimization model is present.

11. The apparatus according to claim 9 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least:

to check, for deciding whether similarity data are required to be sent in response to the request, whether or not an own network optimization model is available,

in case an own network optimization model is not available, to decide that similarity data are required to be sent, and

in case an own network optimization model is available, to decide that similarity data are not required to be sent, and to cause sending of an indication to reject the request for similarity data.

12. The apparatus according to claim 9 , wherein the pre-trained network optimization model is based on a deep reinforcement learning algorithm based on a plurality of convolutional layers employed for capturing spatial and temporal correlations between the network environment of a part of the self-organizing communication network, and a plurality of fully connected layers employed for reducing dimensions of data processing in the plurality of convolutional layers to a desired output dimension, wherein the pre-trained network optimization model is defined by a set of model parameters and hyperparameters.

13. The apparatus according to claim 12 , wherein

the model parameters include at least one of weight matrices between each two successive layers of the plurality of convolutional layers and the plurality of fully connected layers, and bias vectors between each two successive layers of the plurality of convolutional layers and the plurality of fully connected layers, and

the hyperparameters include at least one of a number of layers of the plurality of convolutional layers and the plurality of fully connected layers, a number of units at each layer of the plurality of convolutional layers and the plurality of fully connected layers, a type of an activation function, a number of filters and filter size in each of the plurality of convolutional layers, a stride size of each maximum or average pooling layer.

14. The apparatus according to claim 12 , wherein the at least one memory and the instructions are further configured to, with the at least one processing circuitry, cause the apparatus at least: for processing the data indicating at least a part of a pre-trained network optimization model for generating an own network optimization model,

to modify the received part of the pre-trained network optimization model by conducting at least one of

adding at least one of a new convolutional layer and a new fully connected layer to the part of the pre-trained network optimization model,

modifying at least one convolutional layer and fully connected layer of the part of the pre-trained network optimization model, and

to retrain the modified network optimization model including the added or modified layers by using measurement data obtained in the own part of the self-organizing communication network.

15. A method for use in a communication network control element or function configured to control a setting of parameters of a self-organizing communication network, the method comprising

obtaining a pre-trained network optimization model indicating a mapping between a communication network environment forming an input of the model, optimization actions or decisions forming an output of the model, and system performance indicators forming a reward,

causing sending, to at least one communication network element or function forming a part of the self-organizing communication network, a request for providing similarity data for a similarity analysis allowing to determine a similarity between a part of the self-organizing communication network for which the pre-trained network optimization model is derived and a part of the self-organizing communication network of the at least one communication network element or function to which the request is sent,

wherein the similarity data are related to corresponding network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network of the at least one communication network element or function to which the request is sent, and

wherein the corresponding network properties comprise data related to a single data point and data related to statistical measures and include at least one of location information, geographical features, mobility patterns, data demand statistics, and histograms approximating a probability density function,

performing the similarity analysis for determining a similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and each part of the self-organizing communication network for which similarity data for the similarity analysis are received, wherein the similarity analysis is performed by determining a similarity measure between the network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived, and the network properties of the part of the self-organizing communication network of the at least one communication network element or function to which the request is sent,

determining, on the basis of the similarity analysis, at least a part of the pre-trained network optimization model to be provided to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received, wherein the part of the pre-trained network optimization model comprises at least one convolutional layer of the pre-trained network optimization model, and

causing sending of the determined part of the pre-trained network optimization model to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received,

wherein the pre-trained network optimization model is derived by using a deep reinforcement learning algorithm based on a plurality of convolutional layers employed for capturing spatial and temporal correlations between a network environment of the part of the self-organizing communication network, and a plurality of fully connected layers employed for reducing dimensions of data processing in the plurality of convolutional layers to a desired output dimension, wherein the pre-trained network optimization model is defined by a set of model parameters and hyperparameters.

16. The method according to claim 15 , further comprising at least one of:

obtaining the pre-trained network optimization model by receiving data including the pre-trained network optimization model from a communication network element or function belonging to a part of the self-organizing communication network, wherein the method is implemented in a central unit being in charge of collecting and managing usage of pre-trained network optimization models derived in the self-organizing communication network;

obtaining the pre-trained network optimization model by deriving the pre-trained network optimization model from data and measurements conducted in a part of the self-organizing communication network, wherein the method is implemented in communication network element or function forming a part of a distributed system for managing usage of pre-trained network optimization models derived in the self-organizing communication network.

17. The method according to claim 15 , further comprising

for performing the similarity analysis for determining the similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and a part of the self-organizing communication network for which similarity data for the similarity analysis are received, calculating a similarity measure on the basis of network properties of the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network for which similarity data for the similarity analysis are received,

wherein the network properties comprise data related to a single data point and data related to statistical measures and include at least one of location information, geographical features, mobility patterns, data demand statistics, and histograms approximating a probability density function.

18. The method according to claim 17 , further comprising

processing, for determining at least a part of the pre-trained network optimization model to be provided to the at least one communication network element or function forming a part of the self-organizing communication network from which the similarity data are received, the calculated similarity measure, and

selecting, for forming the part of the pre-trained network optimization model to be provided, a subset of parameters and hyperparameters defining low and medium layers of the pre-trained network optimization model,

wherein the higher the similarity between the part of the self-organizing communication network for which the pre-trained network optimization model is derived and the part of the self-organizing communication network for which similarity data for the similarity analysis are received is according to the similarity measure, the higher the number of parameters and hyperparameters selected for the subset becomes.