Transient Analysis of Partitioned-Block Frequency-Domain Adaptive Filters

The frequency-domain adaptive filter (FDAF) is very useful for instance acoustic signal processing. The partitioned block FDAF (PBFDAF) is a generalization of the FDAF and becomes more popular due to its low latency. Some efforts have been done toward the convergence analysis of PBFDAFs, but they usually use strong approximations and hence came to inaccurate results. This paper presents a unified approach to the transient analysis of both the constrained and unconstrained PBFDAFs based on the overlap-save structure. Using the independence assumption, we derive the analytical expressions for the mean and mean-square performance of PBFDAFs. Our analysis does not assume a specific model for the inputs and provides a quite general framework. Computer simulations confirm a good match between our theory and experimental results.