Description
Due to the proliferation of mobile devices and services, the scale of multiple-input-multiple-output (MIMO) communication systems is getting larger and larger and can be massive in future wireless networks. This results in significant increases in hardware cost and power consumption. Recently, low-resolution analog-to-digital converters (ADCs) have been considered as a practical solution for reducing hardware cost and power consumption in MIMO systems. This is because low-resolution ADCs have simple hardware architectures as well as very low power consumption. However, the severe nonlinearity of low-resolution ADCs causes significant distortions in the received signals and therefore makes signal processing tasks such as channel estimation and data detection much more challenging compared to those in high-resolution systems. Motivated by the fact that machine learning is very powerful in solving non-linear problems, this dissertation exploits machine learning to develop low-complexity yet efficient and robust algorithms for channel estimation and data detection in MIMO systems with low-resolution ADCs. First, the blind detection problem, i.e., detection without channel state information (CSI), in MIMO systems with low-resolution ADCs is studied. Two learning methods, which employ a sequence of pilot symbol vectors as the initial training data, are proposed. A performance analysis of the vector error rate is then derived for the case of 1-bit ADCs. Based on the analytical results, a criterion for designing transmitted signals is also presented. Next, we show how support vector machine (SVM) – a well-known supervised-learning technique in machine learning – can be exploited to provide efficient and robust channel estimation and data detection in MIMO systems with 1-bit ADCs. Both uncorrelated and spatially correlated channels are considered. An SVM-based joint channel estimation and data detection method and an extension to frequency-selective fading channels will also be proposed. Then, a deep learning framework for low-resolution MIMO channel estimation, data detection, and pilot signal design is proposed. The proposed estimation and detection networks are model-driven and have special structures that take advantage of domain knowledge in the low-resolution quantization process. An important feature of the proposed channel estimation network is that the pilot matrix is integrated into the weight matrices, which leads to a joint optimization of both the channel estimator at the base station and the pilot signal transmitted from the users. We also develop a nearest-neighbor search method to further improve the data detection performance. Finally, via numerical results, the proposed machine learning-based methods are shown to be efficient and outperform existing ones. They are also shown to be robust against inherent computational issues in the low-resolution MIMO framework.
