GRU-Based Equalization for CP-Free OFDM Over Frequency Selective Channels

We consider orthogonal frequency division multi-plexing (OFDM) over frequency-selective channels without cyclic prefix (CP) insertion. We design a specific recurrent neural network utilizing gated recurrent units to mitigate the effects of inter-symbol interference and inter-carrier interference for improved symbol detection. Numerical examples demonstrate that the proposed deep neural network outperforms conventional equalizers for CP-free OFDM in terms of bit error rate. Moreover, through block error rate simulations, we show that the proposed solution yields higher performance compared to OFDM systems with sufficiently long CP when the multipath spread is high, hence the new approach offers significant potential for enhancing the spectral efficiencies for certain wireless communication scenarios.