Reducing the Complexity or the Delay of Adaptive Subband Filtering


In this contribution a method to reduce the computational complexity of adaptive subband filtering is presented. Alternatively, the approach can be utilized to keep the complexity as it is in a conventional scheme, but to reduce the delay introduced by non-critical subsampled filterbanks (meaning to shorten the length of the filterbanks’ prototype filters). This is achieved by using different subsampling rates for the reference channels on one hand and for the desired and the output channels on the other hand. Experiments and real-time measurements performed with systems for acoustic echo cancellation show either a reduction of complexity of about 30 percent or a delay reduction of about 50 percent (depending on the optimization objective). At the same time the steady-state error (remaining error power due to aliasing after convergence) can be kept at the same level.

Year: 2010
In session: Signal Processing
Pages: 158 to 165