Enhanced Signal Detection for Low-Order Modulated OFDM-IM via Deep Learning

Orthogonal frequency division multiplexing with index modulation (OFDM-IM) enhances bit error rate (BER) performance by partially activating sub carriers and utilizing the activation pattern as an additional source of information. How-ever, its performance declines with higher-order modulations. To address this, lower-order modulation assisted OFDM-IM (LOFDM-IM) increases the information embedded in the indices while reducing the constellation size. This approach necessitates reusing the activation pattern with distinguishable constellation rotations, which increases detection complexity. To mitigate this, the authors propose a deep learning-based detection method that outperforms conventional maximum likelihood detection by reducing complexity while still achieving optimal BER.