End-to-end Learning for OFDM: From Neural Receivers to Pilotless Communication

End-to-end learning is a promising technology for future communication systems, but its benefits have not yet been quantified over realistic wireless channel models. This work aims to fill this gap by exploring the gains of end-to-end learning over a frequency- and time-selective fading channel using orthogonal frequency division multiplexing (OFDM). We present two approaches to enable communication without orthogonal pilots and with no bit error rate (BER) loss and hence higher throughput compared to a orthogonal pilot-based baseline. Both approaches leverage a neural network (NN)-based receiver, operating over a large number of subcarriers and OFDM symbols. The first approach consists of learning superimposed pilots (SIPs) that are linearly combined with a conventional quadrature amplitude modulation (QAM) and jointly optimized with the neural receiver. The second approach consists of learning an optimized constellation geometry together with the neural receiver. The learned geometry works for a wide range of signal-to-noise ratios (SNRs), Doppler and delay spreads, has zero mean and does hence not contain any form of superimposed pilots. Both schemes do not negatively affect the peak-to-average power ratio (PAPR) compared to conventional QAM. Thus, we believe that a jointly learned transmitter and receiver are a very interesting component for beyond-5G communication systems which could remove the need for demodulation reference signals (DMRSs) and achieve unprecedented throughput and reliability.