Quasi-Newton FDE in One-Bit Pseudo-Randomly Quantized Massive MIMO-OFDM Systems

TL;DR

This work tackles reliable high-order modulation in one-bit massive MIMO-OFDM by introducing a PRQ scheme and a low-complexity projected quasi-Newton detector PQND. By formulating the frequency-domain equalization as a constrained log-likelihood optimization and applying Newton-based updates with diagonal and decoupling approximations, PQND performs per-subcarrier processing without matrix inversion. PRQ mitigates quantization distortion and, together with PQND, enables $256$-QAM and $1024$-QAM in wideband channels while preserving a practical complexity comparable to first-order detectors. The combination demonstrates enhanced BER performance over the 1BOX benchmark and broadens the viability of high-order modulation in one-bit massive MIMO-OFDM, particularly in channels with moderate delay spreads.

Abstract

This letter offers a new frequency domain equalization (FDE) scheme that can work with a pseudo-random quantization (PRQ) scheme utilizing non-zero threshold quantization in one-bit uplink multi-user massive multiple-input multiple-output (MIMO) systems to mitigate quantization distortion and support high-order modulation schemes. The equalizer is based on Newton's method (NM) and applicable for orthogonal frequency division multiplexing (OFDM) transmission under frequency-selective fading by exploiting the properties of massive MIMO. We develop a low-complexity FDE scheme to obtain a quasi-Newton method. The proposed detector outperforms the benchmark detector with comparable complexity.

Figures (3)

• Figure 1: A block diagram that summarizes the system model. S/P is for serial-to-parallel, and P/S is for parallel-to-serial.
• Figure 2: (a)-(b) Convergence analysis of the methods in terms of the negative of the log-likelihood and BER analysis in a $128 \times 10$ system with $V=32$, employing ZTQ and $16$-QAM constellation in the SDS channel. $\alpha=1,0.8,0.7,0.009$ for NM, its one and two-stage approximated versions, 1BOX, respectively. (c) Comparison of the BER performances of MRC, ZF, 1BOX 1BOX, and PQND methods with ZTQ in the LDS channel with respect to SNR.
• Figure 3: (a) The performance comparison of PQND and 1BOX 1BOX with respect to SNR in the SDS channel using ZTQ and PRQ where $K=2$. With 1BOX, $\alpha=0.02,0.01,0.007$ for $N=64,128,256$, respectively. (b) The BER performance against $K$ at $\rho=30$ dB with ZTQ and PRQ in the SDS and LDS channels. (c) The BER performance against $\log_2(N)$ when $K=1$ at $\rho=30$ dB with ZTQ and PRQ in the SDS and LDS channels.