Clustering‐Based Resource Allocation mechanism in Long Term Evolution Advanced networks with auto‐encoder for feature learning

Abstract

AbstractClustering‐Based Resource Allocation (CBRA) is a relatively recent approach applied in some communication network technologies that executes resource allocation rules by following patterns discovered in the user and traffic flow information. It is a useful option for classifying traffic and detecting the most suitable Quality‐of‐Service parameters to perform improved resource allocation, especially in mobile systems such as Long Term Evolution Advanced, which are growing in user demand. An open issue in the CBRA approach is the necessity to perform feature engineering manually (feature selection), which may cause loss of relevant system information and consequent deterioration of performance. In this research article, we present a novel CBRA mechanism that implements an auto‐encoder to perform feature learning, thus avoiding feature selection. The auto‐encoder, deployed at the early stages of the CBRA mechanism, transforms the user and traffic flow data into a shortened representation, preventing the curse of dimensionality while capturing the nonlinear correlations of the data. The simulation results indicate better cluster formation and performance improvement for real‐time video applications using the proposed mechanism.