CNN/LSTM Based AMR with Custom Dataset

Automatic Modulation Recognition (AMR) is a pivotal algorithm to recognize several types of signal modulations prior to demodulation in modern wireless communication systems and is essential for adaptive modulation and cognitive radio networks. Traditional AMR approaches rely heavily on manual feature extraction, which is often complex and lack of adaptability. The recent proliferation of Machine Learning (ML) and Deep Learning (DL) practices that has opened new avenues for automating and improving AMR performance. This manuscript conveys a ample analysis of the ML and DL practices used in AMR, highlighting their strengths, limitations, and potential future developments. Availability dataset of various modulation schemes is challenge, here, dataset is simulated using python. CNN and LSTM based AMR are implemented, tested on custom dataset. Comparison in between CNN based AMR and LSTM based AMR is presented. For small dataset, CNN based AMR outperform in comparison with LSTM based AMR. Challenges related to model complexity, computational requirements, and real-time adaptability were also examined, thereby providing a roadmap for future research.