Recursive-Repetitive Truncated QAOA-MVSIC for mMIMO Detection: Matrix Product State Simulation and Experimental Evaluation

Quantum approximate optimization algorithm (QAOA) promises near-optimal decoding for n × n massive multiple-input multiple-output (mMIMO) systems. QAOA-MVSIC, recently formulated for Noisy Intermediate-Scale Quantum (NISQ) settings, combines truncated QAOA of small depth p with majority voting (MV) and successive-interference cancellation (SIC) for robustness to noise. We introduce RRT-Q, a recursive, repetitive extension of truncated QAOA-MVSIC, and demonstrate strong noise tolerance, markedly outperforming QAOA and QAOA-MVSIC. We report, to our knowledge, the first experimental bit-error-rate (BER) results for QAOA-based decoding: on an IBM Eagle processor, performance matches semidefinite relaxation with Gaussian randomization (SDR-GR) at signal-to-noise ratio (SNR) ≤ 10 dB for n = 120 (BPSK) and n = 32 (QPSK). Matrix-product-state tensor-network simulations up to n ≤ 256 reach SDR-GR with p = 1. We derive RRT-Q complexity bounds; empirically, we observe that for n ≤ 256 and SNR ≤ 10 dB with Ns QAOA samples, average cost per instance is n α +Ns n β (classical) and Nsn η (quantum) with α ∈ [2.8, 3.4], β ∈ [1.8, 2.2] and η ∈ [1.6, 2.2] while RRT-Q-ML version with maximum-likelihood detection at the final step adds n ζ term with ζ ∈ [2.7, 4.6]. The tunable, low-complexity design targets near-optimal NISQ performance and is expected to scale to the fault-tolerant regime (n ≫ 256).