Research On Adaptive Algorithm Based On Censored Regression Model

With the continuous development of theory and technology in the field of information processing,adaptive signal processing has become a new important branch of signal and information processing.Whether in system identification,echo cancellation,or channel equalization,wave speed formation,and adaptive prediction,adaptive filtering algorithms are extremely widely used.Moreover,the convergence speed,steady-state error,computational complexity,and tracking performance of the adaptive filtering algorithm have been used as the criteria to judge the pros and cons of the algorithm.In the traditional sense,adaptive signal processing is often aimed at a linear regression model,and in actual life applications,the signal will be somewhat distorted during transmission.One of the reasons is that the sensor's own saturation characteristics make the received signal cannot be fully observed.This model is called a censored regression model(or censored regression model).Under this model,whether it is system identification,echo cancellation,or channel equalization,beam forming,using traditional adaptive algorithms,can not achieve the desired results.Firstly,for a single non-Gaussian noise environment under a truncated regression model,a recursive generalized mixed norm algorithm based on Censored Regression(CR-RGMN)is proposed based on the model,and the system is identified In the two applications of channel equalization and channel equalization,a computer simulation experiment was carried out on the CR-RGMN algorithm.Then,for the case where the input signal is a colored signal(such as a speech signal),an affine projection algorithm based on censored regression(Affine Projection Algorithm based on Censored Regression,CR-APA)is proposed.Further,for the CR-APA algorithm,in the large step size,the convergence is fast,but the steady-state error is large;the small step size,the convergence is slow,but the steady-state error is small;The state error is large and the calculation complexity is high;when the affine projection order is small,the convergence is slow,and the steady state error is small.A variable-step affine projection algorithm with variable step size under the censored regression model Step-size Affine Projection Algorithm with evolving order based on Censored Regression,CR-EVSAPA).And in the two applications of system identification and echo cancellation,the performance of the algorithm is simulated.Finally,in some active noise control applications,mechanical friction,vibration noise,and speech signals are super-Gaussian/pulse(heavy tail)signals.In blind source separation,adaptive receivers with multiple antennas and image denoising,it can be mixed sub-Gaussian and super-Gaussian/pulse signals.Aiming at this non-Gaussian mixed noise problem,a robust M-Shaped Error Weighted Algorithm based on censored regression model(CR-FRMS)is proposed.Further,for the situation when the unknown system is a sparse system,a proportional zero-drawing robust M-shaped error weighting algorithm based on censored regression model is proposed(l₀-norm proportional FRMS algorithm for the censored regression model,l₀-CRPFRMS).And through the computer to simulate the performance of the algorithm.