Research On Compressive Sensing In Wideband Wireless Communication

For sparse and compressive signals, compressive sensing theory shows that it can reconstruct signal with less samples. Distributed compressive sensing(DCS) is an extension theorem based on compressive sensing, which exploits the common joint sparsity of multiple signals and can recover all the signals at the same time. Researchers have proved that if multiple signals are correlated and sparse in term of some common basis ,then after observed with another uncorrelated basis and all the signals can be reconstructed exactly with small number of measurements. Another effective extension of compressive sensing is Kronecker compressive sensing(KCS),which shows that Kronecker product is an effective method to achieve the sparse matrix and measurement matrix for multi-dimensional signals.In networked and communication systems, compressive sensing is an attractive method to acquire network features and signals. Therefore, it has been already applied widely in wireless communications, such as sparse channel estimation, sparse system identification, wireless sensor networks and ultra wideband communication, etc. This paper introduces compressive sensing theory and focuses on the research of sparse channel estimation and sparse system identification. The main contents include:1)Investigate the channel estimation method based on KCS.In Multi-input multi-output orthogonal frequency division multiplexing(MIMO-OFDM) system, both channel equalization and coherent detection in receiver need correct channel state information. The accuracy of channel estimation will greatly influence the performance of communication systems.Compressive sensing can be used in channel estimation due to the sparsity feature of wireless channel, which can achieve better performance with less pilots or shorter training sequence. The channel estimation method based on DCS requires the joint sparsity and correlation of sub-channels, otherwise it will have poor performance. In this paper a KCS based orthogonal matching pursuit(K-OMP) method is proposed. Simulation results and theory analysis show that this method has much better performance than traditional least square(LS) method. While comparing to DCS methods, it is more universal and stable for both high-correlated and independent sub-channels. For MIMO-OFDM amplify-and-forward(AF) relay system, a modified KCS based orthogonal matching pursuit(MK-OMP) method is proposed, which has adaptive iteration counter and applicable for the variation of the composed channel sparsity. Simulation results show that it can estimate the number of multipath accurately and can achieve better mean square error(MSE) performance than traditional orthogonal matching pursuit (OMP)method when the sparisity is unknown. And the bit error rate(BER)performance is also good and quite near to ideal channel.2)Investigate the channel estimation method based on compressive sensing(CS) and Kalman filter(KF).KF is a linear filter based on least mean square error rule. Used in channel estimation, KF can reduce noise with the channel information of prior symbols and update the channel response of current symbol. Meanwhile CS algorithm can reconstruct the complete and true channel state information based on the updated pilots. In this paper, a channel estimation method that combines CS and KF is proposed and can take full advantage of the two ways.Both KCS-KF and DCS-KF algorithm are simulated in Matlab and numerical results show that the new method has much better performance with less pilots comparing to pure CS method, and also can increase the spectral efficiency of the system. In another hand, it is also proved that CS is the primary factor for the channel estimation. Only when the performance of CS is guaranteed,KF can get some additional gains.3)Investigate sparse system identification and propose fractional generalization of least mean square (GMLMS) identification algorithm.In this paper, sparse system identification is investigated and fractional identification algorithm GMLMS is proposed. The Matlab simulation results show that the performance is related to fractional order, which means that a smaller fractional order will give smaller weight noise and a bigger fractional order will give faster convergence speed. When the fractional order is smaller than 0.8, it can achieve better MSE performance than the zero-attracting LMS(ZA-LMS) algorithm.