francescopace/espectre 2.1.0

v2.1.0 - "Made for ESPHome" Compliance

on GitHub

All example configurations now meet "Made for ESPHome" requirements

WiFi Provisioning

• BLE provisioning: esp32_improv for easy setup via ESPHome/HA Companion app
• USB provisioning: improv_serial for web.esphome.io configuration (not yet supported on ESP32-C5)
• Captive Portal: Fallback AP "ESPectre Fallback" for WiFi configuration
• No hardcoded credentials: Removed YOUR_WIFI_SSID placeholders

Dashboard Adoption

• dashboard_import: One-click adoption from ESPHome Dashboard
• project metadata: Version tracking for firmware updates

Code Cleanup

• Renamed espectre_component.cpp/.h → espectre.cpp/.h
• Component ID standardized to espectre_csi
• Updated me CLI: erase_flash → erase-flash (esptool deprecation fix)

⚡ Performance Optimization

Unified variance algorithm and optimized Hampel filter across both platforms

This release focuses on code uniformity between MicroPython and C++ implementations, improving numerical stability and performance.

Algorithm Uniformity

• Two-pass variance: Both platforms now use the same numerically stable algorithm
  • Formula: Var(X) = Σ(x - μ)² / n (more stable than E[X²] - E[X]²)
  • Eliminates catastrophic cancellation risk with float32
  • Identical behavior between MicroPython and C++

Hampel Filter Optimization

• C++ (ESPHome): Eliminated dynamic memory allocation

  • Pre-allocated static buffers in hampel_turbulence_state_t
  • Insertion sort replaces qsort() for small arrays (N=3-11)
  • ~20-25μs saved per packet (no malloc/free overhead)

• MicroPython: Pre-allocated buffers and circular buffer

  • Eliminated list creation per call
  • Insertion sort for small arrays
  • ~120μs saved per packet

Validation

• New test script 16_test_optimization_equivalence.py using real CSI data
• Verified with 2000 real CSI packets (baseline + movement)
• Maximum variance difference: 9.41e-14 (effectively zero)

🧪 Test Suite & Code Coverage

• 140 test cases (+72 from 2.0.0) with real CSI data
• Full device testing: All tests run on both native and ESP32-C6 via IWiFiCSI dependency injection
• Codecov integration: Coverage badge, PR comments, 80% threshold
• 84% line coverage, 94% function coverage
• Refactoring: Shared utilities in utils.h, configurable CalibrationManager