In recent years, Orthogonal Time Frequency Space Modulation (OTFS) has gained popularity in integrated sensing and communications (ISAC) system due to its robustness against Doppler offset and delay changes. Traditional pilot-based methods for accurate channel parameter estimation are complex and struggle with rapidly changing channel conditions. In this letter, we propose a deep encode-decode network (DED-Net). It uses DL to automatically learn and eliminate channel interference from OTFS signals. The framework employs a deep encoding and decoding network, similar to a filter, learning complex signal features to effectively remove interference. Our experiments demonstrate DED-Net’s ability to eliminate interference in OTFS modulation signals, offering an alternative to pilot-based methods and showcasing DL’s potential for ISAC systems.