Comparative Performance Analysis of Deep Learning Modified Adaptive Filter with AEC Algorithms

Abstract

The paper has emphasized the comparative analysis of traditional adaptive filters being drawn for the purpose of acoustic echo cancellation (AEC) in a single channel filtration process. The work is further processed to determine an optimistic approach to direct the filter performance towards enhanced efficiency in respect of MSE evaluation along with standard deviation and correlation coefficient determination for various lengths of adaptive filter employed for AEC. The new filter designed having machine learning (ML) with an iterative approach for filter adaptation supported by length-based weight upgrade algorithm is also formulated and recognized for comparative analysis with other filters that are objectified to maintain a trade-off between Mean Square Error (MSE) and ERLE determination. The approach designed for filter weight adjustment with ML is directed towards filter weight determination improvement for eliminating the residual form of echo as well present in the acoustic signal. The ML approach made by deploying neural networks (NN) to filter weights adjustment having multilayer structure gave the best performance as end results.