SR-TTA: Spatial-Redundancy Test-Time Adaptation for Interference-Robust Respiration Sensing

The gist: Spatial-redundancy test-time adaptation (SR-TTA) recovers respiration by maximizing consistency across random antenna subsets under an out-of-band spectral veto, closing the gap to a per-recording oracle label-free.

How it works

The core problem addressed is recovering a one-dimensional breathing waveform from high-dimensional CSI received at a cell-free massive MIMO base station, especially when strong in-band motion interference causes traditional hand-crafted fusion methods to collapse. The system models the sensing chain where the channel state information (CSI) is fused into a breathing waveform through a learned complex-weight beamformer.

The proposed solution involves a pipeline mapped onto the Open Radio Access Network (O-RAN) architecture. This pipeline includes:

1. Range-gating at the O-DU.

2. BeamNet acting as a nearreal-time (near-RT) RIC xApp to predict complex weights.

3. SR-TTA performing label-free test-time adaptation to refine these weights.

The Adaptation Mechanism (SR-TTA)

SR-TTA refines the initial learned beam weights w0 by treating the combiner as a free per-recording variable. The adaptation objective is defined by maximizing agreement across random antenna subsets under an out-of-band spectral veto. This is achieved by drawing S random keep-masks m(j) ∼ Bernoulli(pkeep)M and maximizing their pairwise agreement Jcons(w) = 2 S(S − 1) X j. This spatial redundancy criterion is crucial because breathing modulates many distributed antennas, whereas a localized interferer dominates only a few.

Failure Analysis and Discrimination

The paper rigorously demonstrates why frequency- and variance-based criteria fail to separate in-band interferers from breathing. Frequency criteria, such as maximizing band peakiness or spectral entropy, lock onto the in-band interferer because it lives inside the respiration band B. Subspace nulling fails because the breathing and interferer subspaces overlap, causing projection to remove both components. SR-TTA succeeds because it exploits spatial redundancy; a beam riding an interferer is fragile to antenna dropout while a breathing beam is robust.

Performance and Scaling

Experiments show significant performance gains under strong in-band interference. The respiration-rate error drops substantially under simulated interference, moving from 5.8 to a much lower value of 0.8 breaths per minute (bpm). Furthermore, the method repairs severe cross-domain shift on real recordings, reducing failures from 47% to 7% after one-time crosssubject calibration. The spatial redundancy scaling law confirms that service quality grows with more distributed APs. SR-TTA climbs from 0.78 to 0.96, closely tracking the oracle ceiling which rises from 0.92 to 1.00 as the number of antennas increases.

Conclusion

SR-TTA provides a label-free method for refining learned spatial combiners by maximizing consistency across random antenna subsets and incorporating an out-of-band spectral veto. This approach maintains high fidelity under strong in-band interference, achieving a respiration-rate error of 0.8 bpm and reducing real data failures to 7% after calibration ". This demonstrates that spatial redundancy is the key mechanism that frequency and subspace criteria cannot access. This single label-free front-end covers clean, interfered, and real operation with no detector or baseline switching.

--- Page 1 ---

SR-TTA: Spatial-Redundancy Test-Time Adaptation for Interference-Robust Respiration Sensing

I. INTRODUCTION

Future 6G systems aim to provide environmental awareness as a native network capability [1]. Passive respiration sensing can reuse cell-free massive multiple-input multiple-output (MIMO) channel state information (CSI) without additional spectrum or hardware [2], [3].

1. textbf Spatial-Redundancy Test-Time Adaptation (SR-TTA) Implementation in O-RAN Architecture: This improvement maps the sensing pipeline onto a specific Open Radio Access Network (O-RAN) structure, utilizing range-gating at the O-DU, adaptation as a nearrealtime (near RT) RIC xApp, and one time calibration as a non real time rApp to enable deployment.

2. textbf Robust In-Band Interference Rejection: The system can recover respiration where SSIR-WAC collapses by exploiting spatial redundancy, stating that Maximizing (11) pushes the beam onto the spatially-redundant (breathing) component, information that neither the output spectrum nor the dominant-variance subspace exposes.

3. textbf Adaptive Beamforming with Label-Free Refinement: The system refines learned complex weights by maximizing a surrogate objective Jcons(w) which uses spatial redundancy, distinguishing breathing from in-band interferers where frequency and variance-based criteria cannot separate an in-band interferer from breathing.

4. textbf Scalable Sensing Performance: The system's performance scales with distributed hardware, as the paper shows that SRTTA instead climbs from 0.78 to 0.96, closely tracking the oracle ceiling, demonstrating how a larger array offers more recoverable headroom for service quality under interference.

5. textbf Domain Shift Mitigation via Calibration: The system reduces real-data failures from 47% to 7% by performing a one-time crosssubject calibration plus SR-TTA on testbed recordings, effectively closing the gap between synthetic performance and real operational fidelity.

Future 6G networks aim to expose sensing as a native service by reusing communication infrastructure. We study respiration sensing on a cell-free massive multiple-input multiple-output (MIMO) base station, where a 64-antenna channel must be fused into a breathing waveform. The state-of-the-art hand-crafted fusion is near-optimal in benign conditions. It collapses, however, under strong in-band motion interference, whose frequency falls inside the respiration band. We show that a learned complex-weight beamformer recovers respiration by spatial nulling, and that the remaining gap to a per-recording oracle can be closed at deployment by label-free test-time adaptation. Crucially, we identify which label-free signal makes this work. Frequency- and variance-based criteria cannot separate an in-band interferer from breathing. Our spatial-redundancy test-time adaptation (SR-TTA), which maximizes consistency across random antenna subsets under an out-of-band spectral veto, preserves benign performance in our tests. The respiration-rate error drops from 5.8 to 0.8 breaths per minute (bpm) under simulated in-band interference, and the pipeline maps onto the Open Radio Access Network (O-RAN) architecture as O-RAN distributed-unit (O-DU) range-gating, an adaptation xApp, and a calibration rApp. On real testbed recordings, a one-time cross-subject calibration plus SR-TTA reduces failures from 47% to 7%, drawing level with the hand-crafted combiner using label-free test-time adaptation.