Research On Structured Sparse Decomposition Algorithm For Communication Signals

With the continuous exploration of compressed sensing,sparse decomposition technology has attracted increasingly attention.Energy distribution of signals can be more concentrated through this technology by represented with dictionary atoms,so the analysis of signals tends to be more efficiently.Besides,the sparsity or sparseness of the signals is also the prerequisite for compressive sensing.Usually,sparse decomposition is applied to the area of image and signal de-noising.In practical case,many signals have their own sparse structure,which begets a new research focus,the structured sparse decomposition.The research of structured sparse decomposition owns great value since it's still in the initial stage.As a result,this paper will study the related algorithms of structured sparse decomposition.First of all,this paper summarizes the background and current situation of this research.Beginning with the block sparse structure signal,the related block sparse decomposition algorithms and block dictionary training algorithms are discussed in detail,as well as the simulation analysis.Compared with the ordinary sparse decomposition algorithms,the block sparse decomposition algorithms have better performance and lower time complexity,which is verified by the simulation results.In addition,the performance of our newly refined block dictionary training algorithm is much better than the old one.Secondly,the joint sparse structure is analyzed.Based on two different joint sparse structures,two joint sparse decomposition algorithms,JSM1-OMP and JSM2-OMP,are proposed.Compared with the OMP algorithm under multi-antenna,JSM1-OMP has lower time complexity,while JSM2-OMP has better sparse decomposition performance.Then,we design a new algorithm by combining JSM-1model and JSM2-OMP,which is more accurate when calculating common sparse coefficients.In addition,the simulation shows that once the error range is fixed,the number of antennas should not be too much,so as not to cause unnecessary waste.Finally,the sparse decomposition algorithm based on sparse Bayesian is studied.The sparse Bayesian learning algorithm(MSBL)based on multi measurement model,sparse Bayesian learning algorithm based on time correlation for multiple measurement models(TSBL)and its improved algorithm(TMSBL)are investigated.Simulation results show that,with the increase of time correlation,TSBL algorithm and TMSBL algorithm have more and more superior performance,while the performance of MSBL algorithm will become worse,which proves the advantage of time correlation.On the relatively high signal-to-noise ratio(SNR ?6dB),the TMSBL algorithm modifies the super parameter of TSBL.Therefore,in this SNR range,the performance of TMSBL algorithm is better than that of TSBL algorithm,and simulation results also verify this conclusion.