LSTM-Based Deep Reinforcement Learning for ISI Mitigation in Maritime LoRaWAN

For reliable, long-range, low-power Maritime Internet of Things (MIoT) communication (e.g., vessel tracking, ocean monitoring, and offshore automation), LoRaWAN offers attractive coverage and energy efficiency. However, sea-surface reflections, wave motion, and platform mobility create time-varying multipath with large delay spread, which induces inter-symbol interference (ISI) and degrades packet delivery ratio (PDR) and energy performance. This paper proposes a Long Short-Term Memory (LSTM)-assisted Deep Reinforcement Learning (DRL) framework LSTM–DDPG Adaptive Modulation and Coding (LD-AMC) that proactively mitigates ISI by predicting short-term channel evolution and adapting the LoRaWAN physical-layer parameters. An LSTM predictor learns temporal correlations in observed link metrics (RSSI, SNR, PER, and RMS delay spread) and provides one-step-ahead forecasts, which are appended to the agent state. A Deep Deterministic Policy Gradient (DDPG) controller then selects the spreading factor (SF), coding rate (CR), bandwidth (BW), and transmit power (P t ) within LoRaWAN constraints to maximize a reward that favors reliable delivery and throughput while penalizing energy cost and ISI severity. MATLAB/Simulink simulations under coastal and offshore two-ray maritime channels show that LD-AMC reduces ISI-induced symbol errors by up to 58%, improving PDR by up to 47% and reducing energy per delivered packet by up to 32% compared with standard and enhanced ADR baselines.