Over-the-Air MIMO Autoencoder Based Simultaneous Multiple Access for Error-Tolerant and High-Efficiency Communication

In this paper, we propose an over-the-air MIMO autoencoder-based simultaneous multiple access (OA-MIMO-AEbased SMA) scheme for applications that require high-efficiency transmission but can tolerate certain errors. Unlike existing nonorthogonal multiple access schemes, the proposed SMA scheme conducts simultaneous transmission to exploit the superposition nature of the wireless multiple access channel and transform it into an equivalent wireless linear processing layer (WLPL) of the MIMO autoencoder. The receiver aggregates the signals from different transmitters by computing their weighted sum via OFDM over-the-air computation (AirComp), which can fully exploit the correlations among the transmitted data to significantly reduce the communication cost. An alternating algorithm for OFDM over-the-air computation is proposed for accurate data fusion. Benefiting from the MIMO autoencoder, the receiver then recovers each transmitter's data from their fusion version in parallel, and the proposed SMA scheme can be optimized using the end-to-end method. Simulation results demonstrate that the proposed OA-MIMO-AE-based SMA scheme achieves accurate transmission with a lower communication cost than existing schemes such as OFDMA and NOMA.