A Deep Learning Based Equalization Scheme for Bandwidthcompressed Non-orthogonal Multicarrier Communication

A Deep Learning Based Equalization Scheme for Bandwidthcompressed Non-orthogonal Multicarrier Communication

Spectrally efficient frequency division multiplexing (SEFDM) is a bandwidth-compressed non-orthogonal multicarrier communication scheme, which provides improved spectral efficiency compared to orthogonal frequency division multiplexing (OFDM) system. The loss of orthogonality yields the self-introduced intercarrier interference (ICI) complicating the equalizer design. In this work, a deep learning (DL) -based SEFDM equalization scheme is proposed to characterize the ICI and to detect the transmitted information bits. The DL-based equalization scheme is trained offline using randomly-generated data and then deployed online. The performance of the equalization scheme is tested by extensive numerical simulations. The results show that the proposed equalization scheme outperforms the linear equalization based equalization scheme, such as zero forcing (ZF), minimum mean squared error (MMSE) and truncated singular value decomposition (TSVD), under additive white Gaussian noise (AWGN) channel in terms of the bit-error rate (BER). Especially for BPSK, the uncoded BER performance approaches the traditional OFDM even for the compression ratio of 0.7, which saves the bandwidth by 30%.