d1e6d28a4a
The first witness (Windows-generated hash 89704bfd…) failed on Linux CI
with a different hash (b36741bf…). Root cause: hashing `re`/`im` parts of
top-5 taps at 1e-6 precision is too tight against libm differences in
sin/cos/sqrt across glibc, MSVC, and Apple-clang. The previous
"top-5 sorted by magnitude" form also suffered from rank instability when
taps are near-tied — libm jitter could shuffle the ordering even when the
algorithm is unchanged.
New canonical form: full per-tap quantised-magnitude profile in natural
index order, no sort.
- 156 taps × 2 bytes (u16 le) per frame = 312 bytes/frame.
- Quantisation 1e-2 — robust to ~1e-3 float drift while still tripping
on real algorithmic changes (e.g., a 10× lambda shift moves magnitudes
by >1e-2).
- No top-K selection — eliminates the unstable magnitude-sort step.
Regenerated expected_cir_features.sha256 — new hash 120bd7b1…
If the next CI run still mismatches, the cause is structural (rustfft SIMD
code path selection or NeumannSolver internal ordering), not magnitudes,
and the witness needs further coarsening or to be made platform-tagged.
Co-Authored-By: claude-flow <ruv@ruv.net>
|
||
|---|---|---|
| .. | ||
| benches | ||
| src | ||
| tests | ||
| Cargo.toml | ||
| README.md | ||
README.md
wifi-densepose-signal
State-of-the-art WiFi CSI signal processing for human pose estimation.
Overview
wifi-densepose-signal implements six peer-reviewed signal processing algorithms that extract
human motion features from raw WiFi Channel State Information (CSI). Each algorithm is traced
back to its original publication and integrated with the
ruvector family of crates for high-performance
graph and attention operations.
Algorithms
| Algorithm | Module | Reference |
|---|---|---|
| Conjugate Multiplication | csi_ratio |
SpotFi, SIGCOMM 2015 |
| Hampel Filter | hampel |
WiGest, 2015 |
| Fresnel Zone Model | fresnel |
FarSense, MobiCom 2019 |
| CSI Spectrogram | spectrogram |
Common in WiFi sensing literature since 2018 |
| Subcarrier Selection | subcarrier_selection |
WiDance, MobiCom 2017 |
| Body Velocity Profile (BVP) | bvp |
Widar 3.0, MobiSys 2019 |
Features
- CSI preprocessing -- Noise removal, windowing, normalization via
CsiProcessor. - Phase sanitization -- Unwrapping, outlier removal, and smoothing via
PhaseSanitizer. - Feature extraction -- Amplitude, phase, correlation, Doppler, and PSD features.
- Motion detection -- Human presence detection with confidence scoring via
MotionDetector. - ruvector integration -- Graph min-cut (person matching), attention mechanisms (antenna and spatial attention), and sparse solvers (subcarrier interpolation).
Quick Start
use wifi_densepose_signal::{
CsiProcessor, CsiProcessorConfig,
PhaseSanitizer, PhaseSanitizerConfig,
MotionDetector,
};
// Configure and create a CSI processor
let config = CsiProcessorConfig::builder()
.sampling_rate(1000.0)
.window_size(256)
.overlap(0.5)
.noise_threshold(-30.0)
.build();
let processor = CsiProcessor::new(config);
Architecture
wifi-densepose-signal/src/
lib.rs -- Re-exports, SignalError, prelude
bvp.rs -- Body Velocity Profile (Widar 3.0)
csi_processor.rs -- Core preprocessing pipeline
csi_ratio.rs -- Conjugate multiplication (SpotFi)
features.rs -- Amplitude/phase/Doppler/PSD feature extraction
fresnel.rs -- Fresnel zone diffraction model
hampel.rs -- Hampel outlier filter
motion.rs -- Motion and human presence detection
phase_sanitizer.rs -- Phase unwrapping and sanitization
spectrogram.rs -- Time-frequency CSI spectrograms
subcarrier_selection.rs -- Variance-based subcarrier selection
Related Crates
| Crate | Role |
|---|---|
wifi-densepose-core |
Foundation types and traits |
ruvector-mincut |
Graph min-cut for person matching |
ruvector-attn-mincut |
Attention-weighted min-cut |
ruvector-attention |
Spatial attention for CSI |
ruvector-solver |
Sparse interpolation solver |
License
MIT OR Apache-2.0