Real-time Spectral Prediction and Metacognition for Spectrum Sharing Radar
Open Access
- Author:
- Kovarskiy, Jacob
- Graduate Program:
- Electrical Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- February 17, 2021
- Committee Members:
- Ram Mohan Narayanan, Dissertation Advisor/Co-Advisor
Ram Mohan Narayanan, Committee Chair/Co-Chair
Timothy Joseph Kane, Committee Member
Julio Urbina, Committee Member
Karl Martin Reichard, Outside Member
Anthony F. Martone, Special Member
Kultegin Aydin, Program Head/Chair - Keywords:
- cognitive radar
spectrum sharing
spectrum prediction
software defined radio
radio frequency interference
metacognition
reinforcement learning
multi-armed bandits - Abstract:
- The growing demand for radio frequency (RF) spectrum access poses new challenges for next-generation radar systems. Recent Federal Communications Commission (FCC) policies permit wireless communication networks to share the spectrum with incumbent radar systems. To operate in a crowded electromagnetic environment, radars must coexist with other RF emitters while maintaining system performance. The concept of cognitive RF provides robust and innovative solutions to efficiently share and build awareness of the spectrum. Cognition is actualized by the perception-action cycle (PAC) which iteratively senses RF interference (RFI), learns RFI behavior over time, and adapts the radar’s frequency band of operation. New developments in software defined radio (SDR) technology have enabled complex cognitive systems to be realized on hardware in real-time. This work 1) presents a cognitive spectrum sharing implementation based on spectral prediction, 2) compares this implementation against radars employing alternative cognitive strategies, and 3) introduces a metacognition architecture to optimize a radar's cognitive strategy with respect to the environment. The spectral prediction approach enables the radar to learn a stochastic model describing RF activity. Using this model, the radar adapts waveform parameters in anticipation of changes in the spectrum. Spectral prediction is demonstrated in conjunction with pulsed linear frequency modulated chirp waveforms as well as notched noise waveforms for coexistence. Additionally, this predictive implementation is compared to reactive and reinforcement learning-based spectrum sharing strategies. Experiments demonstrate that these different cognitive strategies are well suited to particular RFI scenarios. This indicates a need for radars to intelligently adapt cognitive strategies in changing environments. The bio-inspired concept of metacognition provides a framework for cognitive radar to achieve this via self-monitoring and regulation of the PAC. Here, we describe an algorithm selection process aided by metacognition theory. To demonstrate the efficacy of spectral prediction and metacognition for radar, real-time SDR implementations are evaluated. A comprehensive set of synthetic RFI, emulated long-term evolution (LTE) RFI, and real measured RFI scenarios are used to characterize performance. These experiments measure the impact of RFI on radar processing and assess the relative performance improvements due to spectrum sharing. In measuring performance, a metric to characterize target detection quality is proposed based on the Jensen-Shannon divergence. Overall, this work presents a state-of-the-art review for cognitive RF, describes the theoretical background for each approach, details a real-time implementation for both predictive and metacognitive frameworks, and evaluates the performance of these implementations in a variety of RFI scenarios.