Deep learning multi-user communication

A one to many downlink system (e.g. wireless base-station 110 and user devices 120/130). The transmitter and receiver devices are each provided with a deep learning function 112/122/132 (e.g. a deep neural network, DNN) encoding and decoding user data. The deep learning functions are trained to optimise error free recovery of input data streams 101 at each user device 121 (the overall system essentially functions as an auto-encoder). The encoded data is subject to channel impairments 103/104 before being input to the decoder. The training may involve varying the channel impairment parameters (e.g. signal to noise ratio). Once trained the en/decoders are used to process live user data. An uplink version with a deep learning encoder per user device and a common base-station deep learning decoder is also claimed.