A family of split kernel adaptive filtering algorithms for nonlinear stereophonic acoustic echo cancellation

A stereophonic acoustic system offers better spatial realism in teleconferencing and other acoustic applications compared to its monophonic counterpart. However, it suffers from acoustic echo, which is inevitable in acoustic systems. In literature, several stereophonic acoustic echo cancellation (SAEC) techniques have been proposed under the assumption that the echo path is linear. However, electronic components introduce nonlinearities into the system, which renders the effect of the echo canceller to diminish in SAEC. As a result, there exists a scope to investigate further the problem of SAEC when the system is affected by nonlinear distortions. Kernel-based adaptive filtering techniques have been explored for nonlinear system identification in literature due to their superior performance compared to their linear counterparts. Hence, in this paper, we propose a family of kernel-based adaptive filtering algorithms for nonlinear SAEC (NSAEC). Although the kernel approach evidently entails an increase in computational complexity, it is a modest concession since the proposed algorithms show an average of 3–4 dB gain in echo return loss enhancement compared to their non-kernelized counterparts. Among the family of kernel-based algorithms proposed in this paper, the block sparse-based approach depicts better echo cancellation performance. Therefore, the convergence and the steady-state analyses of the kernelized block sparse-based NSAEC are presented in this paper. Computer simulations are presented comparing the proposed kernelized variants to their non-kernelized counterparts using speech and colored noise signals inputs.