13  FTM-Sense: Robust Sensor-free Occupancy Sensing Leveraging WiFi Fine Time Measurement

13.1 Overview

Fateme Nikseresht and Bradford Campbell. 2023. FTM-Sense: Robust Sensor-free Occupancy Sensing Leveraging WiFi Fine Time Measurement. In The 10th ACM International Conference on Systems for Energy-Efficient Buildings, Cities, and Transportation (BuildSys ’23), November 15–16, 2023, Istanbul, Turkey. ACM, New York, NY, USA, 9 pages. https://doi.org/10.1145/3600100.3623741

Application domain: Smart building (energy) management such as automatic HVAC control, lighting systems, security.

  • Sense: This is the primary innovation of the paper. FTM-Sense introduces a novel sensing approach that uses WiFi Fine Time Measurement (FTM) protocol to detect human presence through variations in round-trip time (RTT) measurements caused by multipath reflections from the human body. Unlike traditional occupancy sensors (PIR, cameras, IAQ sensors), this approach:

    • Detects static occupants (including those with minimal motion like breathing or typing)
    • Works through walls (NLOS scenarios)
    • Adapts to new environments without retraining
    • Uses existing WiFi infrastructure with low-cost ESP32S2 devices
    • Provides privacy-preserving sensing without cameras or personal device tracking

  • Plan: Though the occupancy estimates are not used to plan anything leading to the application domains described above, we could think of the specific plan/learn process needed to come up with the occupancy estimates as a proxy. In this sense, the paper employs machine learning classifiers (Random Forest, XGBoost, KNN, Logistic Regression) to process extracted statistical features (variance, standard deviation, root mean square) from FTM data bundles. The models are trained once on baseline data and are shown to generalize to new environments.

  • Act: While not the focus of this paper, the accurate occupancy detection enables downstream building automation actions such as automated HVAC control, lighting adjustments, and energy-efficient building operations.

13.2 Review of the paper

13.2.1 Summary

FTM-Sense presents a sensor-free occupancy detection system that leverages WiFi Fine Time Measurement (FTM), a protocol standardized in IEEE 802.11mc. The core innovation is demonstrating empirically, using larger samples than previously presented, how using variations in RTT measurements between two WiFi devices (caused by multipath reflections from human presence) can be used to detect occupancy with high accuracy and the overall performance of this approach (97.7%).

Key contributions:

  1. Static occupant detection: Unlike PIR sensors that fail to detect motionless occupants, FTM-Sense achieves 97.68% accuracy detecting occupants with only fine motions (typing, reading) and acceptable performance (70.09%) even with very fine motions (breathing only).

  2. Adaptability: The system adapts to new indoor environments within 80 seconds without requiring retraining. Tested across 7 different rooms (office and residential) with varying configurations, achieving >87.83% accuracy.

  3. NLOS capability: Can detect occupants through walls and doors (>92% accuracy through wood/drywall, >87% through aluminum sheet).

  4. Low-cost implementation: Uses standard ESP32S2 boards (~$5 each) with built-in WiFi, no specialized hardware required.

Technical approach:

  • Collects FTM measurements across 10 WiFi channels (2.4 GHz band) to avoid interference
  • Extracts statistical features (variance, std, rms) from sliding time windows of FTM data
  • Trains baseline models (Random Forest and XGBoost perform best) on 3 hours of data from single room
  • Generalizes to new environments without retraining

13.2.2 What do we know already?

This paper connects to several concepts covered in earlier lectures:

From Lecture 2 (Why Buildings):

  • The energy management motivation: buildings account for 40% of greenhouse gas emissions, and occupancy-based control can significantly reduce energy waste from heating/cooling/lighting unoccupied spaces
  • The Sense/Plan/Act framework for autonomous building systems - FTM-Sense provides the “Sense” component

From Lecture 7 (Thermal Comfort):

  • We know that occupant needs are the main reason why we cool/heat buildings. So clearly, understanding their behavior and presence is very useful.

General building automation concepts:

  • The need for real-time occupancy information to enable demand-based HVAC and lighting control
  • Trade-offs between accuracy, cost, privacy, and energy consumption in sensing systems
TipThings to learn more about

To fully understand this paper’s contributions and methods, students may need background on:

  1. WiFi protocols and the physical (PHY) layer:

    • IEEE 802.11 standards and amendments (specifically 802.11mc)
    • How FTM protocol works: burst exchanges, timestamp recording at PHY layer
    • Round-trip time (RTT) measurement and time-of-flight ranging
    • Channel State Information (CSI) vs. FTM for occupancy sensing

  2. RF signal propagation:

    • Multipath propagation in indoor environments
    • How human body presence affects RF signals
    • Line-of-sight (LOS) vs. non-line-of-sight (NLOS) scenarios
    • Signal attenuation through different materials (wood, drywall, metal)

  3. Machine learning for time series classification:

    • Feature extraction from time series data
    • Sliding window techniques
    • Supervised classification tasks:
      • Random Forest and XGBoost ensemble methods
      • K-Nearest Neighbors (KNN)
    • Cross-validation and generalization to new environments

  4. Understanding related sensing technologies:

    • Bluetooth Low Energy (BLE) RSSI-based methods
    • WiFi RSS (Received Signal Strength) methods
    • Channel State Information (CSI) approaches

  5. Occupancy detection metrics and evaluation:

    • Accuracy, precision, recall, F1-score
    • Static vs. dynamic occupancy scenarios
    • Adaptation latency as a performance metric

13.3 Methods

The paper employs several technical methods that should be expanded upon in class discussion:

13.3.1 1. Time Series Feature Extraction

  • Variance: Captures spread in RTT measurements - higher when occupant present due to multipath variations
  • Standard deviation: Measures RTT variability
  • Root mean square (RMS): Quantifies magnitude of RTT fluctuations

13.3.2 2. Machine Learning Classification

  • Random Forest: Ensemble of decision trees, best performer (97.72% accuracy)
  • XGBoost: Gradient boosted trees (96.58% accuracy)
  • K-Nearest Neighbors: Instance-based learning (83.48% accuracy)
  • Logistic Regression: Linear classifier (74.86% accuracy)

13.3.3 3. Hardware Implementation

  • ESP32S2 development boards and their use for WiFi diagnostics