Attention-Refined Unrolling for Sparse Sequential micro-Doppler Reconstruction

TL;DR

This work tackles reconstructing micro-Doppler signatures from highly incomplete channel estimates in joint communication and sensing, addressing real-time constraints and high sparsity. It introduces STAR, a lightweight neural architecture that unrolls a single iteration of iterative hard-thresholding and refines the output with a temporal attention mechanism, enabling accurate MD reconstruction from up to 90% missing measurements. Evaluations on the DISC 60 GHz dataset show STAR outperforming state-of-the-art methods in MD quality and in activity recognition accuracy, while maintaining low computational complexity and reduced communication overhead. The proposed approach demonstrates the viability of interpretable, data-driven refinements to classical CS methods for real-time wireless sensing, with potential impact on low-overhead sensing in mmWave JCS systems.

Abstract

The reconstruction of micro-Doppler signatures of human movements is a key enabler for fine-grained activity recognition wireless sensing. In Joint Communication and Sensing (JCS) systems, unlike in dedicated radar sensing systems, a suitable trade-off between sensing accuracy and communication overhead has to be attained. It follows that the micro-Doppler has to be reconstructed from incomplete windows of channel estimates obtained from communication packets. Existing approaches exploit compressed sensing, but produce very poor reconstructions when only a few channel measurements are available, which is often the case with real communication patterns. In addition, the large number of iterations they need to converge hinders their use in real-time systems. In this work, we propose and validate STAR, a neural network that reconstructs micro-Doppler sequences of human movement even from highly incomplete channel measurements. STAR is based upon a new architectural design that combines a single unrolled iterative hard-thresholding layer with an attention mechanism, used at its output. This results in an interpretable and lightweight architecture that reaps the benefits of both model-based and data driven solutions. STAR is evaluated on a public JCS dataset of 60 GHz channel measurements of human activity traces. Experimental results show that it substantially outperforms state-of-the-art techniques in terms of the reconstructed micro-Doppler quality. Remarkably, STAR enables human activity recognition with satisfactory accuracy even with 90% of missing channel measurements, for which existing techniques fail.

Figures (10)

• Figure 1: Processing chain for human activity recognition: our model (STAR) is capable of recovering the md spectrum from very few CIR measurements.
• Figure 2: Block diagram of STAR.
• Figure 3: Framing cir samples into subsequent windows to be processed by STAR.
• Figure 4: Evolution of the rmse over the test set for increasing percentages of missing measurements, achieved by STAR, in comparison with iht at convergence, iht stopped after one iteration, DUST, DUST-V2, OMP, and LASSO algorithms.
• Figure 5: Evolution of the ssim over the test set for increasing percentages of missing measurements, achieved by STAR, in comparison with iht at convergence, iht stopped after one iteration, DUST, DUST-V2, OMP, and LASSO algorithms.