Research On Side-information Based Compressive Spectrum Sensing Algorithms

Compressive spectrum sensing(CSS)is a frontier direction in wideband cognitive wireless communication,which exploits compressive sensing(CS)to collect spectrum information at a sampling rate lower than Nyquist,so as to meet the real-time requirements of spectrum sensing.CSS utilizes the sparseness of the spectrum,and can efficiently and accurately recovery the original spectrum signal from compressive sampling.However,conventional CSS technologies often ignore the side information existing in the practical scenarios,such as the statistical characteristics of noise,the parameter characteristics of the measurement matrix and the statistical characteristics and structural characteristics of the spectrum,etc.The performance of the cognitive wireless communication system cannot be further improved due to the lack of the useful information carried by these side information.In view of this,this dissertation studies the theory and scheme of CSS based on side information in cognitive wireless communication systems.To guide the design and implementation of the CSS algorithms,this dissertation firstly models the widely existing side information in the practical spectrum,constructs a theoretical optimization framework,and analyzes the theoretical conditions for the reliable spectrum recovery of the orthogonal least squares(OLS)-type algorithms.Then,the above research is extended to the blind spectrum sensing scenarios to solve the problem of insufficient side information in practical spectrum sensing.Furthermore,in order to improve the efficiency of CSS,this dissertation proposes enhanced block orthogonal matching pursuit(BOMP)-type algorithms,and proves its feasibility theoretically.Finally,a supervised dictionary learning block threshold feature algorithm is proposed with lower complexity to improve the efficiency of the wideband spectrum sensing system and meet the real-time requirements.The above four researches construct the theoretical analysis framework of CSS based on side information,and simultaneously consider its performance and computational complexity.Specifically,this dissertation includes the following four aspects of innovative work:1)The performance analysis of the OLS-type algorithms in CSS is provided.The series of algorithms include the OLS algorithm,the multiple OLS algorithm and the block OLS algorithm.Firstly,the recovery conditions of the OLS-type algorithms are analyzed by using the mutual incoherence property of the atoms in the matrix.By deriving the upper bound of the reconstructible sparsity and the lower bound of the signal-to-noise ratio for reliable reconstruction of the algorithms,this dissertation reveals the theoretically sufficient reason for the OLS-type algorithms’ performing reliable spectrum sensing.The developed theoretical conditions are further compared with other existing CSS algorithms,and the results reveal that the theoretical conditions required for reliable reconstruction of the OLS-type algorithms are better than those of the other compared algorithms.That is,the series of algorithms can tolerate higher spectral sparsity or lower signal-to-noise ratio.2)Based on the above innovative work,in order to solve the problem of insufficient side information in practical spectrum sensing scenarios,this dissertation proposes a series of blind OLS algorithms.Two cases are considered where only the side information of the measurement matrix and the block length of the spectrum signal is known.For the former,this dissertation proposes a blind OLS algorithm by developing the minimum signal-to-noise ratio required to obtain the target sensing probability based on the mutual incoherence property of the measurement matrix.This algorithm only needs to calculate the mutual incoherence property of a given measurement matrix without relying on sparsity,block length and noise variance.For the latter,this dissertation proposes a blind-block OLS algorithm by exploiting the block mutual incoherence property of the matrix.Comparing the two algorithms,it is found that the signal-to-noise ratio required for the latter to achieve reliable sensing is lower than the former.That is,the introduction of the block structure improves the performance of the CSS algorithm.The simulation results are consistent with the theory.Compared with the existing blind greedy algorithms,the blind OLStype algorithms obtain promising performance gain.3)Based on different side information models,enhanced BOMP-type algorithms with desirable efficiency are proposed.For the scenario where the non-zero support probability of the block sparse spectrum is known as the side information,a gamma distribution approximation theory is proposed.Based on this,the error probability is minimized,and the logic weighted factor is developed.A logic weighted BOMP algorithm based on the gamma distribution approximation is presented.For the scenario where the signal-to-noise ratio of cognitive wireless communication system is known as side information,the lower bound of the number of the measurements required for reliable recovery of the BOMP algorithm is analyzed.Based on this developed bound,this dissertation proposes sampling-controlled BOMP-type algorithms,which can adaptively vary the number of measurements.The simulation results reveal that the spectrum sensing performance of the two kinds of algorithms is better than that of the BOMP algorithm.In the aspect of running time,when the signalto-noise ratio increases,the sampling-controlled BOMP algorithms gradually obtain shorter running time than the BOMP algorithms.4)For high-efficiency sensing,a dictionary learning-based block threshold feature algorithm is proposed.First,this dissertation analyzes the performance of the block threshold feature by using the mutual incoherence and restricted isometry properties of the measurement matrix.By deriving the signal characteristics required for reliable recovery,the essential reason for the performance deterioration of the block threshold feature algorithm in spectrum sensing is pointed out.Combined with the above analysis,this dissertation further proposes a dictionary learning scheme suitable for block threshold feature,which improves the sensing accuracy while ensuring its low computational complexity.Thus,this dissertation solves the problem of unsatisfactory accuracy of block threshold feature when applied to spectrum sensing.The simulation results reveal that the block threshold feature algorithm based on the proposed dictionary learning scheme can complete reliable and fast wideband spectrum sensing tasks under different sparsity,number of measurements and signal-tonoise ratio conditions.