Study On Signal Processing For Massive Mimo Wireless Communication System

Massive multiple-input multiple-output(MIMO)is a key technology for 5G.In mas-sive MIMO systems,the base station can serve multiple users sharing the same time-frequency resource,and the spectral efficiency and energy efficiency can be sufficiently improved by making full use of the spatial degrees of freedom.Nevertheless,the large-scale MIMO faces several challenges.In the uplink system,due to the large number of antennas equipped at the base station,hardware cost and power consumption will be a heavy burden for the receiver when the high-resolution analog-to-digital converters are employed.At the same time,it seems hard for the signal processing algorithms to strike a good balance between performance and complexity under the high-dimensional channel matrix.In the downlink system,due to the large-scale antennas and high-frequency com-munication,the beam squint effects in the high-frequency massive MIMO communication system will make the conventional hybrid precoding method suffer severe performance degradation.To deal with these challenges,we will examine the following problems,including the uplink performance of channel estimation for the one-bit massive MIMO systems,the uplink performance of symbol detection for the massive MIMO systems,and the downlink performance of hybrid precoding with beam squint effects for the massive MIMO systems.The details and main contributions of our dissertation are shown below.1)Firstly,in the uplink system,we consider the problem of quantization design and channel estimation for uplink multiuser massive MIMO systems.One-bit analog-to-digital converters(ADCs)are used at the base station(BS)to quantize the received signal for significantly reducing the hardware cost and power consumption.We extend the conventional Bussgang linear minimum mean square error(BLMMSE)estimator to the general nonzero threshold case.We then study the problem of one-bit quantization de-sign,aiming at minimizing the mean squared error of the generalized BLMMSE estimator.A set partition scheme and a gradient descent scheme are proposed to devise the quanti-zation thresholds.The rationale behind the set partition scheme is to divide each antenna's received samples into a number of disjoint subsets according to their pairwise correla-tion and assign diverse thresholds to those highly correlated data samples.The proposed scheme only requires the statistical information of the received signals to devise the quan-tization thresholds,which can be calculated in advance before the training process begins.Simulation results show that the generalized BLMMSE estimator can achieve a significant performance improvement over the conventional Bussgang LMMSE estimator.2)Secondly,in the uplink system,we consider the problem of deep learning symbol detection for massive MIMO systems.To develop more accurate and efficient symbol detectors,we develop a model-driven DL detector based on variational Bayesian infer-ence.Specifically,the proposed unrolled DL architecture is inspired by an inverse-free variational Bayesian learning framework which circumvents matrix inversion via maxi-mizing a relaxed evidence lower bound.The proposed networks have only a few learnable parameters and thus can be easily trained.The proposed detectors can work in both of-fline and online training modes.An important advantage of our proposed networks over state-of-the-art MIMO detection networks is that our proposed networks can automatically learn the noise variance from data,thus yielding a significant performance improvement over the existing methods in the presence of noise variance uncertainty.Simulation re-sults show that our proposed model-driven deep learning detectors achieve competitive performance for both i.i.d.Gaussian and realistic MIMO channels.3)Lastly,in the downlink system,we consider the problem of hybrid precoding and combining for massive MIMO systems with beam squint effects.In the presence of the beam squint effects for the wideband millimeter wave(mm Wave)and sub-terahertz(THz)massive MIMO systems,the conventional spatially sparse precoding-based hybrid pre-coding method which approximates the optimal precoder/combiner with a set of array response vectors suffers severe performance degradation.To address this difficulty,we propose a new set of basis vectors to approximate the optimal precoder/combiner.Each new basis vector is devised to form a radiation pattern with a wide beam to cover the squinted beam directions caused by different frequencies.The design of basis vectors can be formulized as an infinity-based minimization problem,which can be efficiently solved by ADMM(alternating direction method of multipliers)technique.Simulation re-sults show that the proposed method can effectively alleviate the beam squint effect and achieve a substantial performance improvement over the existing state-of-the-art hybrid precoding method.