6ed69a3d48
Major changes: - Organized Python v1 implementation into v1/ subdirectory - Created Rust workspace with 9 modular crates: - wifi-densepose-core: Core types, traits, errors - wifi-densepose-signal: CSI processing, phase sanitization, FFT - wifi-densepose-nn: Neural network inference (ONNX/Candle/tch) - wifi-densepose-api: Axum-based REST/WebSocket API - wifi-densepose-db: SQLx database layer - wifi-densepose-config: Configuration management - wifi-densepose-hardware: Hardware abstraction - wifi-densepose-wasm: WebAssembly bindings - wifi-densepose-cli: Command-line interface Documentation: - ADR-001: Workspace structure - ADR-002: Signal processing library selection - ADR-003: Neural network inference strategy - DDD domain model with bounded contexts Testing: - 69 tests passing across all crates - Signal processing: 45 tests - Neural networks: 21 tests - Core: 3 doc tests Performance targets: - 10x faster CSI processing (~0.5ms vs ~5ms) - 5x lower memory usage (~100MB vs ~500MB) - WASM support for browser deployment
1.3 KiB
1.3 KiB
ADR-002: Signal Processing Library Selection
Status
Accepted
Context
CSI signal processing requires FFT operations, complex number handling, and matrix operations. We need to select appropriate Rust libraries that provide Python/NumPy equivalent functionality.
Decision
We will use the following libraries:
| Library | Purpose | Python Equivalent |
|---|---|---|
ndarray |
N-dimensional arrays | NumPy |
rustfft |
FFT operations | numpy.fft |
num-complex |
Complex numbers | complex |
num-traits |
Numeric traits | - |
Key Implementations
- Phase Sanitization: Multiple unwrapping methods (Standard, Custom, Itoh, Quality-Guided)
- CSI Processing: Amplitude/phase extraction, temporal smoothing, Hamming windowing
- Feature Extraction: Doppler, PSD, amplitude, phase, correlation features
- Motion Detection: Variance-based with adaptive thresholds
Consequences
Positive
- Pure Rust implementation (no FFI overhead)
- WASM compatible (rustfft is pure Rust)
- NumPy-like API with ndarray
- High performance with SIMD optimizations
Negative
- ndarray-linalg requires BLAS backend for advanced operations
- Learning curve for ndarray patterns