IEEE RFID 2025

Join DeepSig Director of Machine Learning Raj Bhattacharjea, on the first day of IEEE’s RFID event for his tutorial on AI Software Radio. Raj will discuss RFML Beyond Signal Classification: ML Signal Processing, Generative AI for Channel Modeling, and Learned Comms for 6G. The abstract for his tutorial:

Machine-learning-based algorithms are slowly being adopted into RF signal processing chains. The main driver is that conventional signal processing algorithms of the past were derived by simplifying assumptions about things like linearity, noise, and interference. Still, these algorithms are only approximately optimal in the presence of more realistic effects (e.g., amplifier non-linearities, non-Gaussian interference statistics, etc.). Therefore, most conventional signal processing algorithms fall short of theoretical performance bounds while being practically very useful. In RF machine learning, algorithms in a signal processing chain are re-imagined as nonlinear input-output mappings with free parameters that fit the data. Individual learnable operations can be jointly optimized using end-to-end metrics, leading to a co-design of the entire signal processing chain for optimality without restrictions or assumptions on the kinds of channel models and impairments that can be used. The first examples of RF machine learning appeared in the literature about 8 years ago and were for modulation recognition and signal detection and classification; however, the field has advanced to other things, such as end-to-end communications waveform design via machine learning and generative AI applications for the RF domain are beginning to be explored. This tutorial presents an overview of topics in these areas: machine learning and generative AI for signal classification, signal processing and fully learned wireless communications waveforms.