Compressed-sensing based channel state information acquisition in mmWave hybrid beamforming communication systems

Fifth-generation (5G) cellular communications promises 10x higher data rate, 10x reduced latency and high reliability compared against the fourth-generation (4G). The higher data rate is primarily enabled by the use of higher frequency millimeter wave (mmWave) bands. MmWave bands experience high signal power attenuation over distance, which can be overcome by employing large antenna arrays and advanced signal processing techniques to focus radiated power in a beam. Massive number of antennas and accompanying radio frequency (RF) circuitry, however, excessively increase system operating power.

Hybrid analog and digital beamforming (HB) architectures, which can significantly scale down the number of RF transceivers, and low resolution analog-to-digital converters (ADCs) are attractive in reducing power consumption for wireless communication systems with large antenna arrays. However, reducing power consumption comes at the expense of reducing communication performance. The HB architectures, due to fewer dimensions of the digital beamforming stage and hardware constraints in the analog beamforming stage, suffer fewer degrees of freedom compared with the all digital architecture. Low resolution quantizers inevitably produce higher quantization noise than high resolution quantizers do. Conventional channel state information (CSI) acquisition algorithms employed in all-digital beamforming architectures generally yield degraded performance in such power saving architectures. Therefore I consider compressed sensing techniques to acquire millimeter wave (mmWave) CSI in HB architectures. Compressed sensing is able to exploit the sparsity in angular mmWav channel responses.

In the point-to-point mmWave communications, I develop a deterministic HB codebook design framework for compressed sensing (CS) based channel estimation. The framework is versatile to be applicable to various HB architectures including phase shifters, switches and RF lens. The design approach is to configure analog and digital beamformers in the most favorable forms to CS techniques under the hybrid beamforming constraints. When one tries to reduce measurement overhead of CS-based channel estimation, extra randomness is usually considered: random RF precoder permutation. I propose a computationally efficient algorithm to find a deterministic order of RF precoders that can reduce the overhead down to a half without significant performance loss.

Low-resolution ADC is another means for further power consumption reduction along with HB architectures. However, the combination of a HB architecture and low-resolution ADCs makes channel estimation in such systems more challenging. Adopting the extremely low resolution, i.e., one-bit ADC, in the HB communication systems, I develop a CS-based channel estimation algorithm that is suitable for one-bit quantization.

In developing 5G NR, a new challenge has been arisen: beam management. Since beamforming became an essential component in 5G, beam search and detection are performed even in the initial access. I investigate CS-based downlink beam detection for mmWave HB systems taking the 3GPP standard into account. With the exhaustive search being a benchmark, the CS approach is evaluated using the random and the discrete Fourier transform (DFT) RF codebooks in terms of the beam detection probability.

Through the research contributions I present in this dissertation, it is shown that compressed sensing is the key to exploit sparsity in angular mmWave channel responses. Compressed sensing is beneficial in not only improving accuracy and reducing latency of CSI acquisition, but also the overall communication performance of the hybrid analog/digital beamforming system.