Stack Type Detection Using Few-Shot Learning

Abstract

Wireless digital communication has become so saturated that it is harder for radar receivers to distinguish noise from desired signals, essential for tracking applications like air traffic control towers, defense systems, and communication towers. This is where signal detection is a vital capability of radar systems. Signal detection is the ability to detect signals from noise, and often these signals will be interleaved with noise and other signals. Noise can be alleviated by using filtering techniques, windowing, and transforms, which then can be used by a deinterleaving algorithm to isolate signals. Standard deinterleaving methods isolate signals using deterministic methods; however, more state-of-the-art methods may approach these problems using machine learning or artificial intelligence. Often these methods require copious amounts of data, which can vary from a few hundred to thousands. This might not always be possible in certain situations where privacy limits the amount of available data. This is where Few-Shot Learning (FSL) is utilized for training models on small datasets. This paper proposes a system that can generate interleaved signals and deinterleave them with the help of an FSL model. Various FSL models will be used to compare and determine the optimal configuration of the proposed system.