Normalized least mean square (NLMS) was considered as one of the classical adaptive system identification algorithms. Because most of systems are often modeled as sparse, sparse NLMS algorithm was also applied to improve identification performance by taking the advantage of system sparsity. However, identification performances of NLMS type algorithms cannot achieve high-identification performance, especially in low signal-to-noise ratio regime. It was well known that least mean fourth (LMF) can achieve high-identification performance by utilizing fourth-order identification error updating rather than second-order. However, the main drawback of LMF is its instability and it cannot be applied to adaptive sparse system identifications. In this paper, we propose a stable sparse normalized LMF algorithm to exploit the sparse structure information to improve identification performance. Its stability is shown to be equivalent to sparse NLMS type algorithm. Simulation results show that the proposed normalized LMF algorithm can achieve better identification performance than sparse NLMS one.
- Adaptive sparse system identifications (ASSI)
- Adaptive system identifications (ASI)
- Least mean fourth (LMF)
- Least mean square (LMS)
- Sparse penalty