A Deep-Unfolding-Optimized Coordinate-Descent Data-Detector ASIC for mmWave Massive MIMO

TL;DR

The paper tackles data detection for mmWave massive MU-MIMO-OFDM under realistic channel conditions by introducing a Gram-domain block coordinate descent detector enhanced with a deep-unfolding PME denoiser. It combines algorithmic innovations (GBCD with SINR-based UE sorting and PME-based denoising) with a reconfigurable VLSI PE array to achieve high throughput and low area/power, validated by a fabricated 22 nm FD-SOI ASIC. Results show competitive or superior error-rate performance compared with LMMSE, especially in correlated channels, while delivering up to 7.1 Gbps throughput at 367 mW and occupying 0.97 mm^2. The approach enables efficient parallel processing across subcarriers in OFDM, making it attractive for practical mmWave base stations needing many detector cores. The work demonstrates substantial gains in area efficiency and real-time throughput, supporting QPSK to 256-QAM across a 128-antenna BS with 16 UEs, and substantiates the viability of deep-unfolding techniques in hardware-constrained wireless receivers.

Abstract

We present a 22 nm FD-SOI (fully depleted silicon-on-insulator) application-specific integrated circuit (ASIC) implementation of a novel soft-output Gram-domain block coordinate descent (GBCD) data detector for massive multi-user (MU) multiple-input multiple-output (MIMO) systems. The ASIC simultaneously addresses the high throughput requirements for millimeter wave (mmWave) communication, stringent area and power budget per subcarrier in an orthogonal frequency-division multiplexing (OFDM) system, and error-rate performance challenges posed by realistic mmWave channels. The proposed GBCD algorithm utilizes a posterior mean estimate (PME) denoiser and is optimized using deep unfolding, which results in superior error-rate performance even in scenarios with highly correlated channels or where the number of user equipment (UE) data streams is comparable to the number of basestation (BS) antennas. The fabricated GBCD ASIC supports up to 16 UEs transmitting QPSK to 256-QAM symbols to a 128-antenna BS, and achieves a peak throughput of 7.1 Gbps at 367 mW. The core area is only 0.97 mm$^2$ thanks to a reconfigurable array of processing elements that enables extensive resource sharing. Measurement results demonstrate that the proposed GBCD data-detector ASIC achieves best-in-class throughput and area efficiency.

Figures (15)

• Figure 1: Exact PME $\mathcal{P}_{\mathcal{O}'}(x;\omega, \beta)$ and the proposed piecewise linear approximation $\tilde{\mathcal{P}}_{\mathcal{O}'}(x;\rho, \beta)$ for $8$-PAM with zero-mean Gaussian noise.
• Figure 2: General deep unfolded architecture of the GBCD algorithm with the piecewise linear approximate PME denoiser.
• Figure 3: Complexity, measured in terms of the number of real-valued multiplications of the considered algorithms, versus the number of transmissions $T$ per coherence block.
• Figure 4: Coded BLER simulation results for a $128 \times 16$ system with 256-QAM data in (a) non-LoS and (b) LoS channels, demonstrates that the proposed GBCD-PME with three iterations achieves better than or near LMMSE equalization performance in the practical BLER range, while RCG suffers a notable performance loss compared to LMMSE equalization even with six iterations. The coded BLER of a $16 \times 16$ system with QPSK data in (c) non-LoS and (d) LoS channels, indicates that in high load scenarios the proposed GBCD-PME can significantly outperform LMMSE equalization. The curves with "(fp)" in their labels correspond to the fixed-point performance of the implemented ASIC.
• Figure 5: Coded BLER simulation results with each proposed technique applied incrementally to the original BOX-based coordinate descent detector, OCD.