The Theory And Application Research Of Pipelined Neural IIR Adaptive Filter

The nonlinear adaptive filtering technology has been widely studied and applied,but the performance and computational complexity of the existing nonlinear adaptive filters are often contradictory.Thus,sometimes it is at the cost of increasing computational complexity to improve the performance of the nonlinear adaptive filter.From this point of view,this paper analyzes the research status of nonlinear adaptive filters.And based on the characteristics of the corresponding adaptive filters,combined with the specific nonlinear applications,some new types of nonlinear adaptive filters with low computational complexity are proposed.In addition,the theoretical analysis and experimental simulation on its performance are also carried out in this paper.The specific work is as follows:(1)From the application of nonlinear speech prediction,addressing the computational complexity of two types of typical adaptive filters based on pipelined structure(such as PRNN and JPPSOV),a pipelined neural infinite impulse response(IIR)adaptive filter with low computational complexity is proposed.The filter combines the low computational complexity of the pipelined structure with the good tracking performance of the neural IIR filter,which effectively reduce the computational complexity of the proposed filter and maintain the good nonlinear speech prediction capabilities.This features are verified by computational complexity analysis and computer simulation.(2)Based on the proposed adaptive neural IIR adaptive filter and combined with the IIR nonlinear spline adaptive filter,a novel nonlinear adaptive filter for speech signal prediction is proposed,which is called pipelined neural IIR spline adaptive filter.The adaptive filter utilizes a good fitting property of the IIR nonlinear spline adaptive filter to the nonlinear system so as to improve the nonlinear prediction performance of the pipelined neural IIR spline adaptive filter with a relatively small computational complexity.And the theoretical analysis and experimental simulation have verified this theory.(3)Considering the application of nonlinear adaptive filter in the complex filed and combining the pipelined neural IIR,a complex-valued pipelined neural IIR adaptive filter is proposed.The adaptive filter follows the structure of a pipelined neural IIR adaptive filter and uses the theoretical derivation of a complex-valued IIR adaptive filter as the theoretical basis.The computational complexity analysis verifies the computation load of the adaptive filter is relatively low.And the experiment for non-circular wind signal prediction also demonstrates the effectiveness of the adaptive filter in complex-domain applications.(4)Extending the proposed pipelined neural IIR spline adaptive filter from real domain to complex domain and combining the widely linear model in stereophonic acoustic echo cancellation,a novel complex-valued pipelined neural IIR spline adaptive filter for stereophonic acoustic echo cancellation is proposed.According to the adaptive algorithm of this adaptive filter and the analysis of computational complexity,stability and convergence conditions,and through the application of computational simulation show that the computational complexity of the proposed novel adaptive filter is a little higher than complex-valued pipelined neural IIR adaptive filter due to the introduction of spline function,but its performance is further improved.