Selecting a MIMO receiver for next-generation wireless systems requires more than comparing uncoded bit-error rates. Detector outputs must remain informative after channel decoding, while a reduction in search effort must translate into lower implementation latency. These requirements make the joint evaluation of detection accuracy, soft-output quality, coded performance, and measured runtime essential. We study these dependencies through a reproducible MATLAB benchmark of zero forcing (ZF), exhaustive maximum-likelihood detection (MLD), exact sphere decoding (SD), and a fully specified multilayer perceptron (MLP). Uncoded transmission and rate-3/4 IEEE 802.11 QC-LDPC coding are evaluated over a 2 × 2 independent Rayleigh channel with perfect receiver CSI. The experiment uses matched realizations, SNR-dependent sample caps, and 95% Wilson intervals. MLD and SD return identical decisions at all 46 uncoded modulation–SNR points, yet SD is slower for the small search spaces considered. Among the tested configurations, its pruning first becomes advantageous at 6 × 6 QPSK, where SD is 3.40× faster than exhaustive MLD. At coded BER 10−3, MLD gains 6.52 dB for BPSK and 6.27 dB for QPSK relative to uncoded detection; the MLP lies 0.32 and 1.16 dB from MLD, respectively. These results establish a reproducible methodology for evaluating coded MIMO receivers that is relevant to 5G NR and IMT-2030 research without claiming compliance with either framework.

MIMO detection, coded MIMO receivers, LDPC codes, sphere decoding, learned receivers, runtime, next-generation wireless systems