Design of hybrid power channel MQAM demodulator based on deep learning

Abstract In the context of smart grids and the Internet of Things (IoT) for power systems, power line communication (PLC) and power wireless private network (VPC) channels exhibit complex hybrid characteristics such as strong nonlinearity, multipath fading, and impulse noise, leading to a sharp deterioration in the performance of traditional MQAM demodulators. Traditional methods, relying on accurate channel estimation and rigid mathematical models, are ill-suited to the dynamic interference environment of power systems. This paper proposes a deep learning-based MQAM demodulation algorithm. It utilizes a symmetric fully-connected deep neural network (DNN) to learn the nonlinear characteristics and joint statistical properties of signals in hybrid power channels, constructing complex decision boundaries to achieve intelligent compensation for signal impairments in an end-to-end manner. It utilizes deep neural networks to learn the nonlinear characteristics and joint statistical properties of signals in hybrid power channels, constructing complex decision boundaries to achieve intelligent compensation for signal impairments in an end-to-end manner. Simulation results show that, compared with traditional demodulation algorithms based on signal space and decision boundaries, the proposed algorithm exhibits superior performance under time-varying and non-ideal power channel conditions, and is particularly suitable for high-order modulation and complex power communication scenarios.The system achieves a 1.5 to 3 times reduction in Symbol Error Rate (SER) for 16QAM, 64QAM, and 128QAM configurations,providing an effective solution for improving the reliability and adaptability of power system communication.