Research On Robust Adaptive Filtering Algorithm Based On Transform Domain And Its Application

Adaptive filtering is one of the current research hotspots in signal processing and is widely used in many fields.In order to improve the convergence performance of adaptive filtering algorithm,scholars have conducted in-depth research and proposed corresponding improvement algorithms for fixed step size,correlated input signal,input noise and other problems.Among them,the transform domain adaptive filtering algorithm has excellent decorrelation ability for the input signal,which can accelerate the filter convergence speed and is suitable for the application scenarios with high requirements,such as the acoustic echo cancellation studied in this thesis.Impulsive noise with non-Gaussian characteristics usually occurs in echo cancellation scenarios,which can lead to deterioration of the performance of the transform domain algorithm obtained based on Gaussian noise.In order to improve the performance of algorithms in echo cancellation,this thesis studies robust transform domain algorithms in impulsive noise environments and their performance optimization methods.The main work includes:1.The correlation entropy strategy is used to suppress the impulse noise interference and improve the robustness of the transform domain algorithm,and a robust transform domain algorithm is derived by maximizing the cost function based on the correlation entropy.Simulation results have confirmed that the proposed algorithm outperforms existing transform-domain adaptive filtering algorithms in terms of robustness,convergence speed and steady-state error.2.Two improvement algorithms have been proposed to address the impact of fixed parameters on the performance indexes in the robust transform domain algorithm.On the one hand,the robust transform domain algorithm is combined with convex combination strategy and weight feedback mechanism,and the maximum entropy convex combination adaptive filtering algorithm based on transform domain is proposed to simultaneously obtain the same fast convergence speed as the large-step filter and the same small steady-state error as the small-step filter.On the other hand,by comparing the estimated mean square deviation to switch the step size of each iteration of the filter,and by designing the fixed kernel width as a time-varying kernel width,the transform domain-based transformed step-size variable kernel width adaptive filtering algorithm is proposed,which can optimize both the step size and the kernel width,and improve the effects of these two parameters on the convergence speed and steady-state error of the algorithm.The simulation results show that both proposed algorithms exhibit better performance in the system identification scenario.3.The proposed robust transform domain algorithm and the improved algorithm are applied to acoustic echo cancellation to improve the call quality.Meanwhile,in order to extend the applicability of the robust transform domain algorithm,a robust transform domain algorithm is derived for the echo channel with sparse characteristics based on the robust transform domain algorithm by combining the penalty terms of the power coefficients to the maximum entropy cost function,and the simulation results show that the algorithm has good echo cancellation effect in the echo cancellation scenario.