Research On Adaptive Algorithm Based On Weighted Random Pool Network

As a kind of network with redundant information,lossy compression and random noise interaction,stochastic pooling networks have theoretical guiding significance for biological neural coding,nanoelectronics,distributed sensor networks,digital beamforming arrays,image processing and social networks.Stochastic pooling networks can achieve lossy compression and noise reduction of the input signal at the same time,and nonlinear phenomena such as suprathreshold stochastic resonance have emerged.The stochastic pooling network with weighting coefficient is an extension of the stochastic pooling networks,which has better performance in signal processing on the basis of retaining its basic characteristics.Because there are multiple branch noise disturbances and nonlinear transformations in the weighted stochastic pooling network at the same time,in order to explore the performance of signal parameter estimation under this network model,especially the performance under unknown signal environment,the adaptive algorithm under the network is studied as follows:(1)In this paper,the performance of least mean square algorithm and recursive least squares algorithm is studied theoretically and the convergence,mean square error,learning curve and other statistical characteristics of the algorithms under this network are analyzed.The recursive expressions of LMS,Kalman-LMS,LMS-Newton,Normalized-LMS,Sign error algorithm and RLS algorithm with forgetting factor in this network are derived,and this paper improves Kalman-LMS by periodically resetting the covariance matrix to deal with the situation of unsteady input signals(2)These algorithms are applied to the unsteady state situation where the input signal variance changes.The results show that all except the symbol error algorithm can iteratively converge to the optimal solution of the weight and track the change of the signal.The experimental results verify the theoretical analysis of convergence and mean square error performance of the algorithms.The experimental results of the algorithms verify the theoretical analysis of the algorithm's convergence and mean square error performance.It also verifies the phenomenon of suprathreshold stochastic resonance in the adaptive process of the network,which shows that the beneficial effect of noise does not disappear with the optimization of the weight vector.The above research on adaptive algorithms based on weighted stochastic pooling network reveals the excellent performance of the signal parameter estimation in theory and experiment,enriches the basis of adaptive algorithms under this network,and it has theoretical guidance for the practical application of weighted stochastic pooling network.