Research On The Signal Denoising Under Lower Signal To Noise Ratio

Signal is frequently interferenced by noise while transmission and process in a communication system,those noise will generate effects to the followed signal analysis,judgment and recognition.signal denoising play a very important role in signal processing.traditional denoising method usually based on the hypothesis that the power of signal is higher than that of noise,namely,the signal to noise ratio is higher.but in a real communication system,the signal to noise ratio(SNR) is lower than assumed,even sometime,signal is submerge by noise.under those situation,traditional denoising methods’ performance Sharp decline, even lose efficacy.there for,this paper discussed several typical denoising algorithm and focus on the feasibility of denoising in lower SNR.This paper introduce several typical denoising method firstly,and then probes into the denoising principle and performance of these algorithm, applied these method both in lower and higher SNR,analysis and comparison the performance between two condition.We proposed low coefficient discard and continuous mean square error criterion denoising in Empirical mode decomposition,Simulation and analysis the denoising performance of these two algorithms in higher and low er SNR.In high SNR,both of them perform well,but in lower SNR,the power of IMF components which are obtained via apply EMD to signal show little difference,under this condition,the low coefficient will cause aliasing effect due to improper K, continuous mean square error criterion provides a theoretical basis for the selection of K,But in the low SNR case,even if a IMF component’s energy reaches a local minimum, but not necessarily the noise plays a dominant role, and in some extreme cases,even can not find the global minimum,then this algorithm is completely ineffective.In wavelet transform,we proposed hard and soft threshold,in high SNR,both of them perform well in signal denoising.but in lower SNR,the threshold will Enlarge,this will lead to some signal coefficients are differentiate noise coefficients by mistake which will effect the denoising performance.We propose a novel based on self-correlation Empirical mode decomposition and adaptive wavelet threshold by contrast the difference between the interrelate function of signal and noise,then simulation and analysis the denoising performance in lower SNR.In this paper,we proposed a novel based on random Neighborhood interpolation which can obtain excellent denoising performance in lower SNR by change the matching condition between the signal characteristics and the wavelet basis function.In independent component analysis,we simulated several denoising method based on ICA,in high SNR,both of them perform well,while in lower SNR,there may be zero eigenvalue in correlation matrix which is obtained by signal self-correlation,namely the correlation matrix is Singular matrix,this Singularity can lead to the whiting course become a case of crabs,then the hole ICA can not carry on.,to solve this problem,we researched a novel based on subspace projection,the simulation result show that this algorithm has good performance in signal separation(denoising) in low SNR.