Meta-Learning Empowered Receiver Design for MIMO-OFDM System under HPA Nonlinearity

In this correspondence, a meta-learning empowered detector is proposed for multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM) systems under nonlinear distortions of high power amplifier (HPA). By utilizing the residual connection that directly transmits the data flow into the last layer, a deep neural network (DNN)-aided detector is proposed to blindly compensate the nonlinear distortions and denoise. By introducing learnable parameters into nonlinear models, a model-driven compensator is proposed to further improve nonlinearity compensation without the aid of specific HPA characteristic parameters. To enhance the adaptivity of our proposed compensator, a meta-learning empowered scheme with offline training and fast online adaptation is developed by fully leveraging the supervised information from pilots. Simulation results demonstrate that our proposed detector can achieve satisfactory performance in MIMO-OFDM systems under HPA nonlinearity, while showing the superior online adaptivity with few pilots and gradient updates.