UCLA

Ever increasing demands in mobile data rates have resulted in exploration of millimeter-wave (mmW) frequencies for the next generation (5G) wireless networks. Communications at mmW frequencies is presented with two keys challenges. Firstly, high propagation loss requires base stations (BSs) and user equipment (UEs) to use a large number of antennas and narrow beams to close the link with sufficient received signal power. Consequently, communications using narrow beams create a new challenge in channel estimation and link establishment based on fine angular probing. Current mmW system use analog phased arrays that can probe only one angle at the time which results in high latency during link establishment and channel tracking. It is desirable to design low latency beam training by exploring both physical layer designs and array architectures that could replace current 5G approaches and pave the way to the communications for frequency bands in higher mmW band and sub-THz region where larger antenna arrays and communications bandwidth can be exploited. To this end, we propose a novel signal processing techniques exploiting unique properties of mmW channel, and show both theoretically, in simulation and experiments its advantages over conventional approaches. Secondly, we explore different array architecture design and analyze their trade-offs between spectral efficiency and power consumption and area. For comprehensive comparison, we have developed a methodology for optimal design of system parameters for different array architecture candidates based on the spectral efficiency target, and use these parameters to estimate the array area and power consumption based on the circuits reported in the literature. We show that the hybrid analog and digital architectures have severe scalability concerns in radio frequency signal distribution with increased array size and spatial multiplexing levels, while the fully-digital array architectures have the best performance and power/area trade-offs.

The developed approaches are based on a cross-disciplinary research that combines innovation in model based signal processing, machine learning, and radio hardware. This work is the first to apply compressive sensing (CS), a signal processing tool that exploits sparsity of mmW channel model, to accelerate beam training of mmW cellular system. The algorithm is designed to address practical issues including the requirement of cell discovery and synchronization that involves estimation of angular channel together with carrier frequency offset and timing offsets. We have analyzed the algorithm performance in the 5G compliant simulation and showed that an order of magnitude saving is achieved in initial access latency for the desired channel estimation accuracy. Moreover, we are the first to develop and implement a neural network assisted compressive beam alignment to deal with hardware impairments in mmW radios. We have used 60GHz mmW testbed to perform experiments and show that neural networks approach enhances alignment rate compared to CS. To further accelerate beam training, we proposed a novel frequency selective probing beams using the true-time-delay (TTD) analog array architecture. Our approach utilizes different subcarriers to scan different directions, and achieves a single-shot beam alignment, the fastest approach reported to date. Our comprehensive analysis of different array architectures and exploration of emerging architectures enabled us to develop an order of magnitude faster and energy efficient approaches for initial access and channel estimation in mmW systems.