Hybrid Beamforming for Massive MIMO Backhaul (Working Title)

TL;DR

This paper proposes hybrid beamforming techniques for massive MIMO backhaul in small cell networks, focusing on reducing RF-chain complexity while maintaining high performance through correlation-aware analog processing and eigenvector-based digital precoding.

Contribution

It introduces a novel hybrid beamforming design using correlation-based analog combiners and eigenvector precoding, reducing phase shifter count and outperforming DFT schemes.

Findings

The row combiner clusters correlated antennas for improved performance.

The column spreader repeats signals based on transmit correlation, reducing phase shifters.

Eigenvector-based precoding achieves high sum rate without full correlation matrix knowledge.

Abstract

The uplink where both the transmitter and receiver can use a large antenna array is considered. This is proposed as a method of antenna offloading and connecting small cell access points (SCAP) in a Two-Tier cellular network. Due to having a limited number of RF-chains, hybrid beamformers are designed where phase-only processing is done at the RF-band, followed by digital processing at the baseband. The proposed receiver is a row combiner that clusters sufficiently correlated antenna elements, and its' performance is compared against random projection via a Discrete Fourier Transform (DFT) matrix. The analogue to the row combiner is a column spreader, which is dependent on the transmit correlation, and repeats the transmitted signal over antenna elements that are correlated. A key benefit of this approach is to reduce the number of phase shifters used, while outperforming the DFT scheme. When only the transmitter has correlation and is RF-chain limited, the baseband precoding vectors are shown to be the eigenvectors of the effective transmit correlation matrix. Depending on the channel correlation, this matrix can be approximated to have a tridiagonal Toeplitz structure with the proposed column spreader (CS). The resulting eigenvalues have a closed form solution which allows us to characterize the sum rate of the system. Most interestingly, the associated eigenvectors do not require knowledge of the effective transmit correlation matrix to be calculated using an Eigenvalue Decomposition (EVD) method.