BASE STATION CLUSTERING AND MOBILE DATA TRAFFIC PREDICTION IN CLOUD RADIO ACCESS NETWORK FOR MULTIPLEXING GAIN

Abstract

Mobile network operators over-dimension network resources due to lack of pre-knowledge of the data traffic volume and this leads to underutilization of the network resources as usage of base station resources vary throughout the day. This over-dimensioning also leads to increased cost. Cloud Radio Access Network (C-RAN) allows for network resources to be shared amongst several base stations thereby reducing cost. Previous works have used clustering to group base stations with similar network load and others have clustered base stations with light and heavy network load together, but this does not capture the true network resource requirement due to the limited size of the network. To prevent the size of the network from affecting multiplexing gains, the base stations need to be able to respond proactively to any change in network load. A Long Short-Term Memory (LSTM) algorithm is going to be used for the prediction of the data traffic volume of a base station thereby providing the pre-knowledge of the network load. This will allow for proper provisioning of network resources, and using different clustering algorithms such as K-means, Hierarchical and Gaussian Mixture Models to cluster these base stations there is a reduction in the needed network resources and this reduces cost. Capacity Utility and cost of deployment are the metrics used in making a comparative analysis of the different clustering algorithms used in this work. From evaluation of the methodology, it showed that the Hierarchical clustering algorithm had a Capacity Utility of 0.0012, Gaussian Mixture Models had 0.0035 and K-means with 0.0044 and when this is evaluated against the Capacity Utility before clustering of 0.63 it can be seen that the Hierarchical clustering algorithm had reduced the needed network resources better than Gaussian Mixture Models and K-means. The 3 clustering algorithms were also able to reduce the number of needed base stations from 182 to 80, thereby reducing cost of deployment.