Construction Of Radio Environment Map And Coverage And Capacity Self-Optimization Based On Open Communication Platform

With the rapid development of mobile Internet and the continuous proliferation of smart terminals,new services,novel scenarios and traffic explosion have demanded significant improvement in coverage and capacity performance of mobile communication networks.In this context,the research on radio environment map constructionand performance self-optimization approaches have received extensive attention.However,traditional construction approaches relies on a large amount of data measured by user terminals,which is expensive and has poor timeliness,while traditional optimization methods suffer from high complexity and the need of global information.In addition,most of current research only uses simulated data for performance verification.In order to support the high-performance requirements of future networks,this paper conducts research on radio environment mapconstruction and the joint optimization of coverage and capacity.Specific contents and innovations in this paper are as follows.To overcome the long construction time of global interpolation approaches when measurement data is insufficient,a construction approach based on First Integer Neighbor Clustering Hierarchy(FINCH)and local kriging interpolation is proposed to demonstrate the coverage status in the area of interest and provide evidence for weak coverage detection.Specifically,the global empirical semivariance function value is first calculated with collected RSRP(Reference Signal Received Power)data and the corresponding location information.Then,the spherical fitting model of the semivariance function is obtained by a weighted least squares method.Next,the multi-layer clustering results of known data points is got based on FINCH method,which provides a fine to coarse view on the collected RSRP data.For the unknown data points,several clusters closest to it constitute its estimated group and then local interpolation prediction is achievedby solving the Kriging equation.Since only part of the known data points with relatively high spatial correlation is selected in the interpolation prediction,the construction speed of radio environment map can be significantly improved.Based on the collected data from a real network environment,experiments show that the proposed approach can reduce construction delay under the premise of ensuring a certain accuracy.In addition,multi-layer iterative clustering and self-selection of estimation groups can also achieve a trade-off between construction speed and construction accuracy effectively.To deal with the problems of high complexity and dependence on explict mathematical models in existing coverage and capacity self-optimization approaches,a new approach based on echo state network(ESN)and distributed Q-learning is proposed.Specifically,each base station is regarded as a reinforcement learning agent that can autonomously control the downlink transmission power and the allocation of sub-channels.The observation status of each base station is its current configuration strategy and data rate of the associated user,while each action represents a tuple of power level selection and subchannel allocation.In order to achieve a compromise between coverage and capacity,the reward function is defined as the weighted sum of the global average user rate and the edge user rate.The proposal features the adoption of echo state networks with sparse connections to approximate Q function,which overcomes the high training complexity caused by traditional deep reinforcement learning with fully connected neural networks.At the same time,the use of multi-agent reinforcement learning can effectively overcome the problem of action space explosion led by single agent setting.Numerical results in both simulated environments and real network environments both show that the performance of the proposed approach is close to genetic algorithm and outperform the approach based on traditional distributed Q learning.In summary,this paper studies REM construction and coverage and capacity optimization in wireless networks,and proposes a construction approach based on FINCH clustering and local Kriging interpolation and an approach to the joint optimization of power level and sub-channel selection utilizing ESN based distributed reinforcement learning.The research in this paper is beneficial to the detection of coverage and capacity issues and intelligent self-optimization of future wireless networks.