Integrating AI into Radar System Design: Next-Generation Cognitive Radars

Over the past fifteen years, cognitive radar has emerged as a new approach for incorporating learning and adaptivity on both transmit and receive to enable robust performance in dynamic environments. Although the term “cognitive radar” was coined by Dr. Simon Haykin in 2006, its foundations date back several decades to research in cybernetics, knowledge-aided signal processing, and adaptive radar design. At the core of cognitive radar systems lies the “perception–action cycle”—a feedback mechanism within the transceiver architecture that enables the radar system to learn information about a target and its environment and adapt its transmissions so as to optimize one or more missions (e.g., detection, tracking, classification) according to the desired goal. There are two main approaches for implementing this learning–decision–action mechanism in a radar: (1) a Bayesian approach, which relies on prior distributions and knowledge-aided models of the environment as derived from past measurements, and (2) a machine learning approach, which synthesizes data-driven action with knowledge of sensor and signal models. This chapter will provide an in-depth overview of the concept, architecture, and methods for modeling cognitive processes in a radar system. Both Bayesian and machine learning approaches will be discussed in detail with working examples given from tracking and spectrum sensing problems. Challenges to advancing the current state-of-the-art will be discussed as insights are provided into future directions of research.