There have been a lot of hot technologies to generate a buzz over the years, but this one feels a bit different. The rapid adoption of ChatGPT and other generative AI tech has led to a major investment and opportunity for Microsoft (followed by a buggy demo of Google''s AI ChatGPT competitor, Bard), and the halo effect of anything related to Artificial Intelligence has spread from tech stocks to crypto and across almost any function and industry you can imagine. The growth of ChatGPT's user base in particular has been incredible and has created a widespread excitement far beyond most of other recently hyped tech. It's not often that an early stage technology like this has captured such mainstream attention. Despite the mad dash to get involved in everything AI and feel of a potential bubble around the space, this technology has the potential to be a true game changer in a number of areas including education, programming, content creation and, yes, real estate.

Unlike areas such as computer vision and speech recognition where convolutional and recurrent neural networks-based approaches have proven effective to the nature of the respective areas of application, deep learning (DL) still lacks a general approach suitable for the unique nature and challenges of RF systems such as radar, signals intelligence, electronic warfare, and communications. Existing approaches face problems in robustness, consistency, efficiency, repeatability and scalability. One of the main challenges in RF sensing such as radar target identification is the difficulty and cost of obtaining data. Hundreds to thousands of samples per class are typically used when training for classifying signals into 2 to 12 classes with reported accuracy ranging from 87% to 99%, where accuracy generally decreases with more classes added. In this paper, we present a new DL approach based on multistage training and demonstrate it on RF sensing signal classification. We consistently achieve over 99% accuracy for up to 17 diverse classes using only 11 samples per class for training, yielding up to 35% improvement in accuracy over standard DL approaches.