⚡ electrical engineering

A Body-of-Revolution Human Model for RF Sensing with Measurement-Driven Calibration for Indoor Environments

This paper proposes a computationally efficient RF sensing prediction framework for indoor environments that combines a Body-of-Revolution human model with a measurement-driven background-field approach to generate realistic datasets for Device-Free Localization with approximately 85% accuracy, significantly reducing the cost and time associated with large-scale data collection and full-wave simulations.

Original authors: Haoqing Wen, Michele D'Amico, Matteo Oldoni, Federica Fieramosca, Vittorio Rampa, Stefano Savazzi, Qi Wu, Gian Guido Gentili

Published 2026-07-07

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Haoqing Wen, Michele D'Amico, Matteo Oldoni, Federica Fieramosca, Vittorio Rampa, Stefano Savazzi, Qi Wu, Gian Guido Gentili

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to figure out where a person is standing in a room just by listening to how their voice echoes off the walls. In the world of radio waves, this is called Device-Free Localization (DFL). Instead of a voice, we use radio signals. When a person walks through a room, their body acts like a giant, invisible mirror that bounces, blocks, and twists the radio waves. By measuring these changes, we can guess where the person is.

However, teaching a computer to do this usually requires a massive amount of real-world data. You'd have to hire people to walk around in every possible room, with every possible furniture setup, for days on end. It's expensive, slow, and boring.

This paper proposes a clever shortcut: a digital twin that thinks like a human but calculates like a super-fast robot.

Here is how they built it, broken down into three simple parts:

1. The "Spinning Top" Human (The BoR Model)

To simulate a human in a computer, you usually have two bad choices:

The Stick Figure: A simple cylinder. It's fast to calculate, but it's too smooth. Real humans have heads, shoulders, and legs that scatter radio waves differently. A stick figure misses all that detail.
The Super-Realistic Statue: A 3D model with every bump and curve. It's accurate, but it takes a computer days to calculate the radio waves bouncing off it. It's too slow to be useful for training AI.

The Solution: The authors created a "Body of Revolution" (BoR) model.

The Analogy: Imagine taking a human silhouette and spinning it around a central pole like a potter spinning clay on a wheel. The result looks like a human, but it's perfectly symmetrical.
Why it works: Because it's symmetrical, the computer doesn't have to calculate every single angle. It can use a "2.5-D" math trick (think of it as solving a 2D puzzle and spinning the answer to get a 3D result).
The Result: This model is almost as accurate as the super-realistic statue but runs 154 times faster. It captures the "bumps" of a human body (like the head and knees) without the heavy computing cost.

2. The "Ghost" in the Room (Measurement-Driven Calibration)

Even with a perfect human model, you can't predict radio signals in a real room just by drawing walls on a computer. Real rooms are messy. Radio waves bounce off the floor, the ceiling, the fridge, and the curtains. It's like trying to predict how sound travels in a cave with a thousand echoes; it's nearly impossible to map every single bounce.

The Solution: Instead of trying to map the whole room, they let the room tell them what's happening.

The Analogy: Imagine you are in a dark room and you want to know where the furniture is. Instead of measuring every inch of the room, you clap your hands once (a measurement) and listen to the echo.
How they did it: They took a few real measurements of the room without a person in it. Then, they used a math trick called Phase Retrieval. This is like solving a puzzle where you have the loudness of the sound but lost the timing (phase). By having a person stand in just one spot and measuring the change, the computer can "fill in the blanks" to figure out the hidden timing of the echoes for the whole room.
The Result: They can now predict how the radio waves will behave in that specific messy room without needing to know exactly where every chair or wall is.

3. The Grand Experiment

They put these two ideas together:

The Spinning Top Human (to simulate the person).
The Ghost Room (to simulate the messy environment).

They tested this in a real room with 20 antennas and a real human walking around. They compared their computer predictions to what actually happened in the real world.

The Outcome:

Their model predicted the radio signal strength with 85% accuracy.
It successfully reproduced the "ups and downs" of the signal that happen when a person moves (like how a shadow grows and shrinks).
It did this using a fraction of the computing power required by traditional, super-detailed simulations.

The Bottom Line

This paper doesn't claim to cure diseases or replace security guards yet. It simply offers a fast, physics-based tool to generate realistic training data.

Think of it as a video game engine for radio waves. Before, if you wanted to train an AI to find people using radio, you had to film thousands of hours of real people walking in real rooms. Now, you can use this "Spinning Top Human" and "Ghost Room" engine to generate that data instantly, saving time and money while keeping the physics realistic enough to be useful.

Technical Summary: A Body-of-Revolution Human Model for RF Sensing with Measurement-Driven Calibration

Problem Statement

Device-Free Localization (DFL) and Radio-Frequency (RF) sensing systems rely heavily on large-scale, diverse, and well-labeled datasets to train machine learning models for detecting human presence, movement, and activity. However, acquiring such datasets through extensive measurement campaigns across various environments and configurations is prohibitively costly and time-consuming. While numerical simulations offer an alternative for generating synthetic data, existing methods face a significant trade-off between accuracy and computational efficiency:

Full-wave methods (e.g., 3-D FEM, FDTD) provide high fidelity but require excessive computational resources, making them unsuitable for large-scale database generation.
Analytical or semi-analytical models (e.g., spherical or cylindrical approximations) are computationally efficient but fail to capture the complex scattering characteristics of the human body due to geometric oversimplification.
Indoor environment modeling is further complicated by multipath propagation and reflections from walls and objects, which are difficult to characterize accurately without detailed geometric knowledge or extensive measurements.

Consequently, there is a need for a physically consistent, computationally efficient modeling approach that can generate realistic RF sensing data while accounting for complex indoor propagation effects.

Methodology

The paper proposes a hybrid prediction framework that combines a specialized human body model with a measurement-driven environment reconstruction technique.

1. Body-of-Revolution (BoR) Human Model and 2.5-D FEM

The core of the human scattering model is a Body-of-Revolution (BoR) approximation. This model preserves major geometric features of the human body (head, neck, torso, hips, legs) while maintaining rotational symmetry.

Numerical Solver: A fast 2.5-Dimensional Finite Element Method (2.5-D FEM) is developed to compute the scattering fields of the BoR model under the excitation of a vertical polarized Hertzian dipole.
Decomposition: The method decouples the field contributions of different azimuthal harmonics, solving the problem in the $\rho-z$ plane rather than a full 3-D volume.
Field Reconstruction: To compute fields at distances beyond the immediate FEM domain, the method employs equivalent sources on a cylindrical surface surrounding the body, allowing for efficient post-processing of far-field and near-field data without re-meshing the entire environment.
Material Properties: The human body is modeled as a homogeneous muscle tissue with complex relative permittivity derived from the Cole-Cole model at 2.43 GHz ( $\epsilon_r = 52.7 - j12.76$ ).

2. Measurement-Driven Background-Field Modeling

To address the complexity of indoor propagation without explicit ray-tracing or full environmental mapping, the authors introduce a phase-retrieval approach:

Concept: The total field is modeled as the superposition of a background field ( $E_{bg}$ ) and a human-scattered field ( $E_{sca}$ ). While the amplitude of the background field is known from measurements (with no human present), its phase is unknown.
Phase Retrieval: Using a small "training set" of measurements where a human is present at known locations, the unknown background-field phases are reconstructed. The optimization minimizes the difference between predicted and measured Received Signal Strength Indicator (RSSI) amplitudes.
Calibration: A global calibration constant is jointly optimized to account for uncertainties in transmitter/receiver calibration and to convert electric field units to power.
Assumptions: The method assumes that mutual coupling between the human body and environmental scatterers (higher-order interactions) is negligible and that the excitation for scattering calculations can be approximated by the free-space field.

Key Contributions

The paper outlines three primary contributions:

Efficient 2.5-D FEM for Dipole-Excited BoR Models: A novel formulation for computing body scattering using a 2.5-D FEM solver specifically adapted for dipole excitation. The numerical accuracy is validated against commercial full-wave simulations (Feko™).
Assessment of the BoR Human Model: A comparative analysis demonstrating that the BoR model retains the principal scattering characteristics of a full 3-D human model (including spatial variations along the body height) while incurring almost the same computational cost as a simple cylindrical model.
Measurement-Driven Background-Field Reconstruction: A novel method to reconstruct unknown background-field phases from a minimal set of measurements. This implicitly encodes complex indoor multipath effects and environmental reflections, enabling accurate RSSI prediction without explicit environmental modeling.

Results and Validation

The proposed framework was validated through numerical comparisons and a real-world indoor DFL experiment.

Numerical Validation

Accuracy: The 2.5-D FEM results showed excellent agreement with Feko™ simulations. The average deviation was below 0.9% for near-field ( $\rho=0.3$ m) and 1.2% for far-field ( $\rho=2$ m) scenarios.
Efficiency: Compared to a full-wave 3-D FEM simulation in Feko™, the proposed 2.5-D FEM achieved a 154× speedup (37.4s vs. 5744.7s) and an 84% reduction in memory usage (3.1 GB vs. 19.3 GB).
Model Fidelity: Comparisons between cylindrical, BoR, and 3-D models showed that the cylindrical model failed to reproduce height-dependent scattering variations. In contrast, the BoR model closely matched the 3-D model's scattering trends across different body orientations, with negligible additional computational cost compared to the cylinder.

Experimental Validation

Setup: Experiments were conducted in an indoor room with 20 IoT radio nodes (IEEE 802.15.4, 2.43 GHz) and a human subject at five predefined locations.
Phase Retrieval Performance: Using measurements from a single human position (training set) to reconstruct phases, the model predicted RSSI for four other positions (testing set) with an NRMSE of -16.2 dB (approx. 15.5% linear error). This was a significant improvement over a "no-phase" baseline model, which yielded an NRMSE of -11.6 dB (approx. 26.3% error), representing a 35% reduction in relative prediction error.
Spatial Variation: The model successfully reproduced spatial RSSI variations, including local extrema caused by constructive and destructive interference, across different transmitter-receiver links.
Overall Accuracy: The proposed approach achieved approximately 85% prediction accuracy in reproducing the spatial RSSI variations observed in practice.

Significance and Claims

The paper claims that the proposed approach offers a practical compromise between electromagnetic fidelity and computational efficiency. By combining the BoR human model with a measurement-driven environment reconstruction, the method enables:

Rapid Generation of Realistic Data: It can generate large-scale, physics-consistent RF sensing datasets at a fraction of the computational cost required for full-wave simulations.
Reduced Reliance on Measurements: It significantly reduces the need for extensive measurement campaigns by using a small sample of data to calibrate complex indoor propagation effects.
Foundation for ML: The approach provides a viable pathway for training, evaluating, and generating databases for future machine-learning-based DFL and RF sensing systems, addressing the critical bottleneck of data scarcity.

The authors conclude that while the model assumes weak higher-order interactions and uses an average body orientation, it effectively captures the dominant scattering mechanisms and propagation characteristics necessary for practical indoor RF sensing prediction. Future work is proposed to extend the method to directional antennas and multi-target environments.