Frequency-domain adaptive filtering - A set-membership approach

Frequency-domain adaptive filtering has been shown to have the advantage of low complexity compared to its time-domain counterpart. In this paper, a frequency-domain Set-Membership Normalized Least-Mean-Square (F-SM-NLMS) algorithm is derived, and shown to have better performance (in terms of convergence and bit error rate (BER)) and less updates than frequency-domain Normalized Least-Mean-Square (F-NLMS) algorithm. We also propose an updator-sharing scheme to reduce filter complexity of frequency-domain adaptive filtering. The results show significant savings in complexity with little compromise of performance.