Variable kernel‐based computing algorithms for estimating sparse multipath channels

Summary Accurate channel estimation algorithms are required to mitigate the frequency-selective fading in the broadband wireless communication systems. Many physical experiments revealed that finite impulsive responses are distributed as sparse in delayed time domain. Hence, the inherent sparse information can be exploited by state-of-the-art computing algorithms, in order to improve channel accuracy. Existing computing algorithms, zero-attracting recursive least square (ZA-RLS) and reweighted ZA-RLS (RZA-RLS), have been developed to exploit channel sparsity. However, optimization theory implies that accuracy of the computing algorithms can be further improved to exploit more sparsity information. To further improve the estimation performance, this paper proposes a correntropy induced metric (CIM)-penalized RLS (CIM-RLS) based sparse channel estimation algorithm. Here, sparse constraint is performed by CIM function, while error constraint term is computed by RLS. In particular, Gaussian kernel is adopted for computing the CIM, and its variable kernel width (VKW) is computed for adaptively exploiting the channel sparsity. Monte Carlo simulation results are conducted to verify the proposed algorithm in different scenarios.