Encoder-Decoder Based Complex-Valued Transformer for Signal Classification and Demodulation

In electronic communications, the capability to accurately demodulate signals is a cornerstone for effective data exchange. However, in adversarial or non-cooperative settings where system parameters are unknown, traditional demodulation techniques often fail to decode the complex waveform characteristics influenced by modulation and communication specifics. This paper introduces an novel solution—a complex-valued transformer with an encoder-decoder architecture specifically designed for the classification and demodulation of single-carrier signals. By leveraging the nuances of modulation, synchronization, and bit-level information embedded within I/Q data, our proposed model learns to discriminate and decode a plethora of communication signals through a unified transformer framework. This architecture encapsulates the complexities inherent in signal processing into an end-to-end trainable model, and demonstrates robust performance in both signal classification and blind demodulation scenarios. Our experimental findings confirm that this novel architecture not only excels in classifying single-carrier signals but also performs commendably in challenging conditions where demodulation occurs without prior knowledge of signal parameters which highlights it's potential as a versatile tool for communication systems operating under non-cooperative environments, offering significant advancements over traditional methods by providing concurrent classification and demodulation capabilities within a single forward pass.