Mamba-Based Network for OTFS-IM Detection

Recently, the application of deep learning (DL) technologies to communication systems has garnered significant attention. The Mamba architecture can adaptively adjust the model parameters based on the dynamic changes of the sequence, effectively capturing complex dependencies in long sequences. Furthermore, Mamba provides similar modeling capabilities to Transformers while ensuring near-linear scalability with respect to sequence length. In this paper, we propose a Mambabased detector for orthogonal time frequency space with index modulation (OTFS-IM), termed as MambaIM, that can extract features and establish the global input-output relations from the received signal. Simulation results demonstrate that the proposed MambaIM detector outperforms current DL-based detectors in terms of bit error rate (BER) performance.