An Autoencoder-Aided Multi-Objective Optimization Framework for Energy-Latency-Reliability Trade-Offs in 6G Networks

Abstract

The simultaneous optimization of energy efficiency, end-to-end latency, and communication reliability constitutes a defining challenge for sixth-generation (6G) wireless networks, where conventional designs that prioritize a single performance metric are no longer sufficient. This paper proposes an autoencoder-aided multi-objective optimization framework that jointly addresses these three objectives within a unified mathematical formulation. The proposed model couples a learning-based Maximum Likelihood Detection (MLD) module with a weighted multi-objective formulation over transmit power, computational complexity, and Block Error Rate (BLER), allowing the operating point to be tuned according to service-level requirements. The framework is evaluated through MATLAB simulations of an uplink massive Multiple-Input Multiple-Output (MIMO) system under Rayleigh and composite fading channels for Signal-to-Noise Ratios (SNRs) from 0 dB to 20 dB, with channel-error rates of 20% and 30%. Compared with Zero-Forcing Equalization (ZFE), Minimum Mean-Square-Error (MMSE) detection, and convolutional and recurrent neural network baselines, in single-run simulations, the proposed method reduces energy consumption by approximately 50–70%, improves the Bit Error Rate (BER) by up to two orders of magnitude at moderate SNRs, and achieves an end-to-end latency of 5.5 ms under the simulated low-latency configuration. These values represent representative single-run simulation results under the specified experimental configuration and should not be interpreted as statistically validated population estimates. Confidence intervals and significance testing are left for future work. Limitations relating to channel-shift generalization and statistical reproducibility are explicitly discussed.