Partitioned block frequency domain acoustic echo canceller with fast multiple iterations

Acoustic echo degrades the quality of speech in hands-free telephony. The most popular digital signal processing technique to suppress acoustic echo is adaptive filtering. However, adaptive filtering may require the computational cost optimization in particular when adaptive algorithm is implemented on low-cost DSP platforms. We propose a computationally efficient version of the partitioned block frequency domain adaptive filter with multiple iterations on current data block. The algorithm performs as a cascade of two adaptive filters. The first filter minimizes the Least Square (LS) criteria leading to unbiased estimate of a room response. The second filter speeds up the convergence rate using multiple iterations to minimize modified LS criterion. Coefficients updates calculated in a single step substitute for multiple iterations and decrease computational costs. The complexity of the algorithm is o(log2(R))o(log2(R)), where R is a number of iterations. The proposed algorithm was tested in a simulated room and a real reverberant room. Tests proved that our algorithm converges faster compared to algorithms described in literature.