A QoS‐Aware and Reliability‐Enhanced Architecture for Optimizing Massive MIMO Systems

ABSTRACT Massive multiple‐input multiple‐output (MIMO) is now a key enabling technology for 5G and 6G communication, with its potential to deliver high data rates, low latency, and increased user fairness. In contrast, feedback‐dependent modulation and precoding techniques are not scalable, adaptable, and reliable in high‐mobility environments with high feedback overhead, signal degradation due to feedback among users, and a lack of quality of service (QoS)–aware precoding. To overcome these difficulties, we propose QARE‐MIMO, a massive MIMO framework based on reinforcement learning for adaptive, energy‐efficient, and fair communication. QARE‐MIMO consists of three main modules: (i) the Reciprocity‐Aware Multimetric Adaptive Modulation (RA‐MMAM) module, which exploits reciprocity to reduce the feedback overhead for modulation and minimizes the multimetric feedback overhead; (ii) the Hierarchical Constrained Graph‐Aware Reinforcement Learning module and Weighted Minimum Mean Square Error (HC‐GARL + WMMSE) module, which perform constrained precoding with dynamically optimized power allocation and beamforming under time‐varying channels; and (iii) the Cyclostationary Analog Denoiser (CAD) module, which suppresses analog‐domain interference prior to analog‐to‐digital conversion (ADC) to improve the fidelity and effective SNR. The simulations were all performed in the 3GPP TR 38.901 Urban Macro channel model with up to 100‐km/h mobility, with a total of 5000 realizations. The results also compare QARE‐MIMO with baselines (L‐GSM, CF‐DUCJP, and DnCNN‐GRU) and show that QARE‐MIMO outperforms the other baseline methods in terms of spectral efficiency, feedback reduction, latency, throughput fairness, and QoS satisfaction. The framework consistently improves major metrics by 2%–3% and an overall performance score of 95.6%, thus demonstrating that it is scalable and robust for next‐generation massive MIMO systems.