AI-Driven Secure Channel Estimation and Energy-Efficient Power Allocation for Cell-Free Massive MIMO

Cell-free massive multiple-input multiple-output (MIMO) improves user fairness by allowing many distributed access points (APs) to cooperatively serve users without fixed cell boundaries. Its practical reliability, however, is limited by pilot contamination, pilot spoofing, channel state-information (CSI) uncertainty, inter-user interference, and inefficient power consumption. This paper proposes an AI-driven secure channel estimation and energy-efficient power allocation (AI-SCE-EEPA) framework for uplink cell-free massive MIMO. The AI component is a lightweight supervised anomaly classifier trained on synthetically generated pilot observations representing clean and spoofed contaminated CSI conditions. Its output is used as a soft reliability gate before minimum mean-square error (MMSE) channel refinement. The refined secure CSI is then used by an adaptive regularized MMSE receiver and an energy-efficient power allocation rule. A reproducible Monte Carlo simulation is conducted under distributed APs, Rayleigh fading, large-scale path loss, log-normal shadowing, QPSK transmission, pilot contamination, and active pilot spoofing/eavesdropping stress. The proposed framework is compared with MR, conventional LMMSE, and adaptive R-MMSE baselines using bit error rate (BER), normalized mean-square error (NMSE), spectral efficiency (SE), energy efficiency (EE), secrecy rate, and power-allocation behavior. At 30 dB, AI-SCE-EEPA improves the reported NMSE, BER, SE, EE, and secrecy-rate indicators over fixed-power LMMSE baselines. The results indicate that combining trainable CSI anomaly detection with interpretable model-based estimation and energy-aware power allocation is a practical direction for secure and green cell-free massive MIMO.