Research On Channel Estimation Techniques Based On Compressive Sensing In Massive MIMO Systems

Massive multiple-input multiple-output(MIMO),by deploying large number of antennas at the base station side,can significantly improve the spectrum efficiency and energy efficiency of the wireless communication system,and is one of the key technologies of the fifth generation mobile communication.However,the increase in the number of antennas at the base station,on the one hand,leads to a significant increase of the channel matrix dimension,resulting in the huge number of unknown channel coefficients and computational complexity of the channel estimation;On the other hand,since the pilot symbols are required for probing the wireless channel in massive MIMO systems,the corresponding pilot and feedback overhead during the channel state information(CSI)acquisition increases significantly,which gradually becomes the bottleneck of performance improvement such as spectrum efficiency of massive MIMO systems.To fully exploit the huge diversity gain and multiplexing gain provided by massive MIMO,and guarantee the effective information transmission in communication system,it is especially important to obtain accurate and timely estimation of CSI.Fortunately,as wireless communication systems are moving toward multiple antennas,high frequency band and large bandwidths,the limited number of scatterers and multipaths experienced during signal transmission causes the effective dimension of the channel is lower than its original dimension and the inherent channel sparsity in massive MIMO systems provides a new perspective to resolve the above problems.Compressed sensing theory,as an effective tool for processing sparse signals,can efficiently recover the original high-dimensional sparse signals from low-dimensional observations,providing a solid theoretical foundation for sparse channel estimation in massive MIMO systems.This thesis mainly studies the channel estimation techniques based on compressive sensing,by fully exploiting the channel sparsity of massive MIMO,to improve the robustness of the channel estimation algorithm,reduce the required resource overhead,and improve the accuracy of channel estimation.Aimed at massive MIMO systems,this thesis takes the channel estimation technology based on compressive sensing theory as the research core,where there are still some urgent problems to be solved,including the performance degradation of the channel estimation algorithm when the channel sparsity mismatch occurs in the delay domain,significant increase of pilot overhead and channelfeedback overhead caused by the large channel matrix dimension in the frequency division duplex systems,the low channel estimation accuracy and high complexity in the massive MIMO systems with the generalized hybrid analog/digital architecture,etc.The above problems are studied from the following four aspects:Firstly,aiming at the problem of channel estimation performance degradation when sparsity mismatch occurs in massive MIMO channels,a Bayesian matching pursuit channel estimation algorithm based on block sparsity is proposed.The algorithm exploits the spatially common sparsity of the massive MIMO system,transforming the highdimensional channel impulse response into a channel vector with block sparsity via the permutation matrix,and models the prior distribution of the path delay in channel impulse response as a Bernoulli distribution with block sparsity characteristics,and finally complete the the estimation of channel vector with block sparsity based on the minimum mean square error estimate.Different from the traditional compressive sensing-based channel estimation algorithms,which obtain the support set according to the correlation value between the residue and the measurement matrix,a block binary vector and associated selection metric are defined.The proposed algorithm can update only a single block vector within one single iteration,and utilize the posterior probability obtained by the maximum a posteriori criterion to determine the final channel support,thus avoiding the channel estimation performance degradation when channel sparsity mismatches occur.Secondly,for the problem of large pilot resource consumption in downlink MIMO channel estimation,a hybrid pilot design and hybrid channel estimation algorithm based on prior information are proposed.The pilot design uses a hybrid structure of orthogonal pilots and random pilots,which fully exploits the slow variation of the channel support set in the angular domain,so that the channel in current frame can be adaptively decomposed into dense components and sparse components.Furthermore,by combining the received pilot on multiple subcarriers,a channel estimation algorithm combining least square method and distributed compressed sensing theory is proposed to complete the estimation of the above two channel components,respectively.Compared with the traditional channel estimation algorithm based on least squares or compressive sensing theory,the proposed algorithm can change the number of pilots required for channel probing according to the prior channel information,thus reducing the pilot resources consumed by the massive MIMO system to some extent.Thirdly,aiming at the problem of excessive feedback overhead in frequency divisionduplex massive MIMO system,a complex binary iterative hard threshold algorithm based on prior support is proposed.Since the user only feeds back the symbol information of each dimension of the channel measurement,the proposed algorithm utilizes the 1-bit quantization information and completes the estimation of CSI at the transmitter by solving the 1-bit sparse signal recovery problem at the base station.This algorithm combines 1-bit compressive sensing theory and channel feedback together,allows both base stations and users to avoid the highly complexed codebook search.By introducing a weight vector,the proposed algorithm regards the channel support of previous frame as the prior information of channel estimation at the current frame,not weighting the elements corresponding to the prior support,but imposing a smaller weight on the elements corresponding to the complement set of support,so that the advantages of traditional 1-bit compressed sensing theory and weighted l1 minimization estimation can be combined and exploited.Lastly,for the problem of low-accuracy channel estimation and high complexity in massive MIMO systems with generalized hybrid analog/digital architecture,an orthogonal matching pursuit channel estimation algorithm based on Bussgang decomposition is proposed.On the one hand,the algorithm utilizes the angular sparsity of the channel to overcome the difficulties caused by the fact that the receiver cannot obtain the completed channel observation due to the small number of radio frequency chain.On the other hand,based on the Bussgang decomposition theory,the nonlinear channel estimation problem is reformulated into a linear sparse signal recovery problem,which significantly reduces the influence of quantization noise introduced by 1-bit and 2-bit analog-to-digital converters on channel estimation accuracy.When the proposed algorithm meets the terminal condition and obtains the channel support,the expectation of the residual energy is equal to the equivalent noise variance of the linear model.The complexity increases linearly with the angle grid and the number of channel delay taps,and is squared with the number of frames,pilot symbols and RF chains.