StarBOA enables real-time radar sensing with severe data loss
StarBOA: Real-Time Mamba State-Space Unrolling for Sparse Radar Micro-Doppler in ISAC Networks
Machine LearningInformation TheoryNetworking and Internet Architecture
Summary
Radar sensors often lose much of their data when working alongside communication systems, making it hard to track movement accurately. The authors designed StarBOA, a method that keeps track of long-term motion patterns without needing to revisit old data, making it fast and able to handle very sparse signals. StarBOA predicts missing radar data much better than previous methods, even when up to 90% of the signal is missing. It works quickly enough for real-time use without looking into future data.
What this means in practice
- •For radar system engineers: Deliver real-time radar signal reconstruction under high data loss for integrated sensing and communication devices.
- •For wearable device developers: Enable efficient processing of sparse radar signals to track human movements with low latency in compact hardware.
Authors
Mustafa Bora Çelik, Ceren Çelik, Orhan Gazi
Abstract
In Integrated Sensing and Communications (ISAC), radar sensing must operate under chirp subsampling with up to 90\% missing data. An attention-based baseline, limited to a 52~ms buffer, collapses toward maximum uniform entropy ($H=2.584$ bits) as sparsity increases, failing to capture long-range gait-cycle context. We propose StarBOA, which replaces attention with a causal Mamba state-space model that updates incrementally on a per-window basis without re-scanning past reconstructions. By maintaining a persistent state, StarBOA integrates over $100\times$ more temporal history at no additional per-step computational cost. StarBOA outperforms the baseline's published results across all sparsity levels, with SSIM gains increasing from $+0.0379$ at 50\% missing data to $+0.2472$ at 90\%. Each window is processed in 1.53~ms with zero lookahead, demonstrating efficient causal reconstruction under extreme chirp subsampling.