Time varying sparse support recovery

Abstract

In the applications like spatial modulation in multi-antenna systems, multi-user detection in sensor networks and massive random access in machine to machine communications, the set of nodes which transmit data to the receiver varies with time. Also, the number of ‘active’ nodes is very less as compared to the total number of nodes, thereby inducing time-varying sparsity in the system. At the receiver, a major challenge is to track the set of active nodes at each instance and also estimate the data transmitted by each node. In this work, we use compressed sensing framework to employ a two-level approach for tracking the dynamic supports. In this approach, we consider multiple measurement vectors to facilitate the joint sparse support recovery in the first level. In the second level, we refine the supports using the modified Least squares and QR decomposition techniques. For empirical study, we consider the effect of increasing the number of measurement vectors for slow and fast rate of sparsity variations and also study the consequence of unknown signal variance on the recovery performance. Additionally, we observe significant improvement in the performance of the proposed methods when compared with the Structured Matching Pursuit method and the rank evolution approach.