Deep Learning for Low-Latency, Quantum-Ready RF Sensing

TL;DR

The paper tackles the challenge of real-time RF signal classification in dynamic, interference-rich environments and introduces quantum-ready processing pathways. It proposes a continuous wavelet transform (CWT) based input representation coupled with a lightweight recurrent neural network (CWT-RNN) to enable online, low-latency classification, and provides extensive latency-optimization results for both GPU and CPU hardware. The authors validate portability to quantum RF sensing by simulating a Rydberg-atom based QRF receiver with RydIQule, achieving high accuracy on SNR-based tasks and demonstrating real-time capabilities. Collectively, the work shows how latency-optimized AI/ML methods can bridge conventional RF sensing and quantum-enabled receivers, moving toward practical, real-time quantum-assisted RF sensing systems.

Abstract

Recent work has shown the promise of applying deep learning to enhance software processing of radio frequency (RF) signals. In parallel, hardware developments with quantum RF sensors based on Rydberg atoms are breaking longstanding barriers in frequency range, resolution, and sensitivity. In this paper, we describe our implementations of quantum-ready machine learning approaches for RF signal classification. Our primary objective is latency: while deep learning offers a more powerful computational paradigm, it also traditionally incurs latency overheads that hinder wider scale deployment. Our work spans three axes. (1) A novel continuous wavelet transform (CWT) based recurrent neural network (RNN) architecture that enables flexible online classification of RF signals on-the-fly with reduced sampling time. (2) Low-latency inference techniques for both GPU and CPU that span over 100x reductions in inference time, enabling real-time operation with sub-millisecond inference. (3) Quantum-readiness validated through application of our models to physics-based simulation of Rydberg atom QRF sensors. Altogether, our work bridges towards next-generation RF sensors that use quantum technology to surpass previous physical limits, paired with latency-optimized AI/ML software that is suitable for real-time deployment.

Figures (13)

• Figure 1: Left to right: plots of $S(t)$ (top), amplitude $A(t,f)$ spectrogram (middle), and phase $\phi(t,f)$ spectrogram (bottom) for Gaussian standard deviation $\sigma = .2, 6, 100$.
• Figure 2: Validation loss and classification accuracy as a function of training epoch for the 11-modulation classification task, using our CWT-RNN model. Final classification accuracy over the validation set approached $60\%$ overall.
• Figure 3: CWT-RNN validation loss and classification accuracy as a function of training epoch for the 9-SNR classification task. Final classification accuracy over the validation set approached $70\%$ overall.
• Figure 4: CWT-RNN validation loss and classification accuracy as a function of training epoch for the 5-SNR classification task. Final classification accuracy over the validation set reached $90\%$ overall.
• Figure 5: CWT-RNN classification accuracy as a function of timestep, averaged over the validation dataset, for all three classification tasks: Modulation, 9-SNR, and 5-SNR.