Trade-Offs in FMCW Radar-Based Respiration and Heart Rate Variability

This study evaluates a low-cost FMCW MIMO radar for non-contact vital sign monitoring, revealing that while it achieves high accuracy for average respiratory and heart rates at an optimal distance of 70 cm with sufficient chirp counts, it faces significant limitations in capturing high-resolution beat-to-beat and breath-to-breath variability.

The Gist

Imagine you are trying to listen to two different instruments playing in a room: a loud, slow drum (your breathing) and a tiny, fast violin (your heartbeat). Now, imagine you can't be in the room; you have to listen through a wall using a special kind of "sound radar."

This paper is about testing how well a low-cost, off-the-shelf radar can hear these two "instruments" without touching the person. The researchers wanted to know: How far away can we stand? How many "listening pings" do we need? And can we hear the tiny details, or just the general rhythm?

Here is the breakdown of their findings using simple analogies:

1. The Setup: The "Flashlight" Radar

The team used a small radar chip (like the kind found in some smart cars or automatic doors) that sends out invisible radio waves.

• The Analogy: Think of the radar as a flashlight that sweeps back and forth. When the light hits a person, it bounces back. Because your chest moves when you breathe and your heart beats, the "echo" changes slightly. The radar measures these tiny changes to figure out your vital signs.
• The Goal: They wanted to see if this cheap radar could replace the sticky sensors (ECG pads) usually stuck to a patient's chest.

2. The "Goldilocks" Distance (Too Close, Too Far, Just Right)

The researchers tested the radar at different distances, from very close (30 cm) to far away (150 cm). They found a classic "Goldilocks" scenario:

• Too Close (< 60 cm): It's like standing right next to a speaker; the sound is too loud and distorted. The radar gets confused by "multipath" (echoes bouncing off walls and furniture) and the "near-field" effect (the radar is too close to focus properly).
• Too Far (> 100 cm): It's like trying to hear a whisper from across a stadium. The signal gets too weak (low signal-to-noise ratio), and the radar struggles to pick up the tiny heartbeat.
• Just Right (~70 cm): This is the sweet spot. At about 2.5 feet away, the radar works best.
  • Breathing: It was very accurate (error of less than 1 breath per minute).
  • Heartbeat: It was good, but not perfect (error of about 3 beats per minute).

3. The "Chirp" Count: Taking More Photos

The radar doesn't just send one signal; it sends a burst of signals called "chirps."

• The Analogy: Imagine trying to take a photo of a hummingbird. If you take one blurry snapshot, you might miss it. If you take 100 photos and stitch them together, you get a clear picture.
• The Finding:
  • For Breathing: You don't need many "photos" (chirps). Even a few are enough to get a good average.
  • For Heartbeat: You need at least 96 "photos" (chirps) to get a stable reading. If you take fewer, the radar often fails to hear the heart at all.

4. The Big Trade-Off: The "Average" vs. The "Instant"

This is the most important part of the paper. The researchers found a major limitation, which they call a Trade-Off.

• The Average (The "Big Picture"): The radar is excellent at telling you, "On average, this person is breathing 16 times a minute and their heart is beating 70 times a minute." It's like looking at a time-lapse video of a flower blooming; you see the general movement perfectly.
• The Instant (The "Fine Details"): The radar struggles to tell you, "Exactly when the heart beat, and how much the time between beats varied."
  • The Problem: Heart Rate Variability (HRV) and Breath Rate Variability (BRV) are crucial for doctors to detect stress or heart problems. These require measuring the exact split-second difference between beats.
  • The Result: The radar was quite "blurry" on these details. The error rate for these instant fluctuations was high (15–30%). It's like trying to read the fine print on a menu from 10 feet away; you can see the menu exists, but you can't read the ingredients.

5. Why Breathing is Easier than Heartbeats

The paper explains this with a simple size comparison:

• Breathing: Your chest moves up and down by about 10 to 50 millimeters (like a big wave). This is easy for the radar to see.
• Heartbeat: Your chest only moves 1 to 9 millimeters (like a tiny ripple). This is much harder to detect, especially if the breathing "wave" is drowning out the tiny "ripple."

The Bottom Line

This study tells us that low-cost radar is a great tool for general monitoring. If you want to know if a patient is breathing or if their heart rate is generally high or low, this technology works well, provided you stand about 2.5 feet away and let it gather enough data.

However, it is not yet ready for high-precision medical diagnosis that requires tracking the exact millisecond-to-millisecond changes in your heartbeat. It's like having a good pair of binoculars: you can clearly see the bird, but you can't yet read the tag on its leg.

Future work will need to combine this radar with other data or smarter algorithms to sharpen that "vision" and catch those tiny, life-saving details.

Technical Summary

Here is a detailed technical summary of the paper "Trade-Offs in FMCW Radar-Based Respiration and Heart Rate Variability" by Mura et al.

1. Problem Statement

Non-contact vital sign monitoring (NCVS) is critical for clinical scenarios where wearable sensors are impractical (e.g., contagious diseases, severe burns, psychiatric conditions). While Frequency-Modulated Continuous-Wave (FMCW) radar systems have proven effective for estimating average Heart Rate (HR) and Respiratory Rate (RR), a significant gap exists in their ability to accurately capture instantaneous physiological variability (Heart Rate Variability - HRV, and Breath Rate Variability - BRV).

Current literature often focuses on average rates, leaving the trade-offs between estimation accuracy, sensing distance, signal processing parameters (chirp count), and the resolution of beat-to-beat fluctuations largely unexplored. This study aims to quantify these trade-offs to determine the operational limits of low-cost FMCW MIMO radars for high-resolution autonomic monitoring.

2. Methodology

System Architecture

• Hardware: The study utilizes a low-cost Infineon BGT60TR13C mmWave radar operating in the 60 GHz band (58–63 GHz).
• Configuration: A MIMO (Multiple-Input Multiple-Output) setup with 1 Transmit (TX) and 3 Receive (RX) antennas arranged in an L-configuration for 3D Direction of Arrival (DoA) estimation.
• Signal Parameters:
  • Bandwidth: 5 GHz.
  • Chirp repetition time: 124 µs.
  • Variable chirp counts (32 to 256) and distance variations (30 cm to 150 cm).
• Ground Truth: Simultaneous measurements were taken using:
  • A conductive silicone rubber strip for Respiration Rate (RR).
  • An optical pulse sensor (APDS-9008) for Heart Rate (HR).
  • Both synchronized via an Arduino Nano (50 Hz sampling).

Signal Processing Pipeline

The authors developed a rigorous processing chain to extract vital signs from raw radar data:

1. 2D Spectral Analysis: Range-Azimuth maps generated via 2D FFT to localize the thoracic region.
2. Optimal Bin Selection: An algorithm selects the best range-azimuth bin by maximizing a weighted combination of Phase Variability (indicating motion) and Power Concentration (signal strength), effectively filtering out static clutter.
3. Phase Extraction: The slow-time phase is extracted, averaged across chirps to improve SNR, and unwrapped to create a continuous displacement trajectory.
4. Rate Estimation:
   • RR: Band-pass filtering (0.1–0.5 Hz) followed by Hilbert transform for envelope extraction.
   • HR: Notch filtering to remove respiratory harmonics, followed by band-pass filtering (0.9–3.0 Hz) and FFT to identify spectral peaks.
5. Variability Metrics: Inter-Beat Intervals (IBI) and Breath-to-Breath Intervals (BBI) were extracted to compute standard clinical metrics:
   • HRV: Mean IBI, SDNN, RMSSD, pNN50.
   • BRV: Mean BBI, SDBB, RMSSDBBI.

Experimental Design

• Variables: Subject distance (30 cm to 150 cm) and Chirp Count (32, 64, 96, 128, 256).
• Subject: One adult male seated on an office chair, instructed to remain still.
• Duration: 2-minute recordings per configuration.

3. Key Contributions

1. Quantification of Trade-offs: The study explicitly maps the relationship between sensing distance, chirp count, and estimation accuracy for both average rates and variability metrics.
2. Variability Analysis: Unlike many prior works, this paper rigorously evaluates HRV and BRV metrics, revealing that while average rates are robust, instantaneous variability remains a significant challenge for low-cost radar.
3. Operational Guidelines: The authors provide concrete benchmarks for radar configuration (e.g., minimum chirp counts and optimal distances) required to achieve clinically acceptable error margins.

4. Key Results

Distance Dependency (U-Shaped Error Profile)

• Optimal Distance: 70 cm yielded the best performance for both RR and HR.
  • RR MAE: 0.8 bpm (~4.5% error).
  • HR MAE: 3.2 bpm (~4% error).
• Short Range (<60 cm): Performance degraded significantly (RR MAE 2.1 bpm, HR MAE 7.8 bpm at 30 cm) due to multipath reflections, near-field coupling, and uneven beam illumination.
• Long Range (>100 cm): Errors increased gradually due to reduced Signal-to-Noise Ratio (SNR). At 150 cm, HR MAE rose to 6.4 bpm.
• Robustness: RR estimation was more robust to distance changes than HR, attributed to the larger chest displacement of respiration (10–50 mm) compared to cardiac motion (1–9 mm).

Chirp Count Dependency

• Respiration (RR): Error decreased monotonically with more chirps. Performance stabilized (converged) at ≥96 chirps.
• Heart Rate (HR): Exhibited a threshold effect.
  • **<96 chirps:** Measurements frequently failed or had high errors (>15 bpm).
  • ≥96 chirps: Required for stable detection.
  • 256 chirps: HR MAE stabilized near 2.8 bpm.
• Variability Metrics: Increasing chirp count improved average rate accuracy but offered limited gains for variability metrics. Even at 256 chirps, variability errors remained high (15–30%).

Variability Metric Accuracy

• Respiratory Variability (BRV): Errors ranged from 2% (Mean Interval) to ~20% (RMSSD).
• Cardiac Variability (HRV): Errors were higher, ranging from 18% (RMSSD) to 30% (SDNN) at optimal settings.
• Conclusion: The system is capable of robust average rate estimation but lacks the resolution for high-fidelity beat-to-beat or breath-to-breath analysis required for advanced autonomic nervous system assessment.

5. Significance and Conclusion

This study establishes that FMCW MIMO radar is a viable solution for robust, non-contact monitoring of average vital signs (RR and HR) within a specific operational window (50–120 cm, ≥96 chirps). However, it highlights a fundamental limitation: the trade-off between average rate accuracy and instantaneous variability resolution.

• Clinical Implication: While suitable for general monitoring and triage, current low-cost radar configurations are not yet sufficient for diagnosing conditions relying heavily on HRV/BRV (e.g., stress assessment, sleep analysis, or early detection of cardiac arrhythmias) without further algorithmic refinement.
• Future Directions: The authors suggest that achieving clinical-grade variability resolution will require advanced techniques such as multi-domain data fusion, temporal super-resolution, or improved motion compensation algorithms to isolate subtle physiological fluctuations from noise.