Study On Generalized Sidelobe Cancelling And Post-filtering Algorithms For Microphone Array

With the development of speech applications and artificial intelligence,speech enhancement technology based on microphone array has received extensive attention.Compared with the traditional single-channel speech enhancement algorithm,microphone array can utilize the spatial information of the signal and has a better noise suppression performance.As one of the common microphone array techniques,the generalized sidelobe canceller has simple structure and no need to estimate noise power spectrum.It consists of three parts: a fixed beamformer,a blocking matrix,and an adaptive noise canceller.In practical applications,however,due to the poor consistency of the microphone array sensor units,multi-path reflection and reverberation,the desired signal will leak to the output of the blocking matrix.This leakage directly leads to the distortion of the desired signal.Moreover,the generalized sidelobe canceller has limited performance on the suppression of non-coherent or diffused noise.It is necessary to add a post-filter to further eliminate the residual noise in the output signal.This thesis mainly studies the robust generalized sidelobe cancelling and post-filtering algorithms.The main works and contributions can be organized as follows.Firstly,aiming at the desired signal cancellation problem caused by the desired signal leakage from the output of blocking matrix,this thesis optimized the filter coefficient updating criterion in the adaptive noise canceller.Within some traditional algorithms,fixed step factor or signal-to-noise ratio are used to control the updating of the filter.Compared with the method that uses the output signal power of the fixed beamformer and the blocking matrix to calculate the signal-to-noise ratio directly,this thesis considers the influence of the quasi-stationary noise on the calculation of signalto-noise ratio.Then,a priori signal-to-interference ratio is calculated after filtering out those quasi-stationary noises from the output signal of the fixed beamformer,which is more accurate and can effectively control the updating of the filter.Therefore,the cancellation of the desired signal is alleviated.Secondly,for the problem that the generalized sidelobe canceller has poor suppression performance on the diffused noise,this thesis uses the post-filter based on multichannel noise estimation to further eliminate the residual noise.Compared with the traditional single-channel noise estimation method,noise power spectrum estimation based on multi-channel signals can reduce the noise estimation error by using the spatial information of noise and the multi-channel reference noise,which will improve the performance of the post-filter.This thesis improves the performance of the multi-channel noise power spectrum estimation algorithm based on the transient beam-to-reference ratio,in which the accuracy of calculating the transient beam-to-reference ratio is improved,and thus it enhances the noise suppression performance.Finally,computer simulation experiments have been conducted and the simulation results show that the improved robust adaptive beamforming algorithm based on generalized sidelobe canceller and the post-filtering algorithm based on multi-channel noise power spectral estimation can effectively reduce the distortion of the desired signal and achieve better noise reduction performance.