For Clinicians & Therapists

What Is Resonant Frequency Breathing?

Resonant frequency breathing (RFB) is a specific form of slow-paced breathing in which the patient breathes at their individual resonant frequency — the rate (typically 4.5–6.5 breaths per minute) that maximizes heart rate variability by exploiting the resonance properties of the cardiovascular system (Vaschillo et al., 2006; Lehrer & Gevirtz, 2014).

At resonance, respiratory sinus arrhythmia (RSA) is maximized: heart rate rises during inhalation and falls during exhalation with the greatest amplitude, producing HR oscillations 4–10 times larger than resting baseline. This is believed to strengthen baroreflex sensitivity and improve autonomic regulation through repeated practice (Lehrer & Gevirtz, 2014; Shaffer & Meehan, 2020).

RFB is distinct from generic slow breathing apps. Being off by even one breath per minute substantially reduces the physiological effect (Steffen et al., 2017). Precise Breath identifies each patient's individual rate through HRV analysis rather than using a population default.

Evidence Base

Published research has examined RFB and HRV biofeedback across a range of clinical presentations.

These findings describe published research on resonant frequency breathing. Precise Breath is a wellness and breathing training tool — not a medical device. Individual results may vary.

How Precise Breath Implements the Protocol

The established clinical protocol for identifying a patient's resonant frequency tests multiple breathing rates in a single session using heart rate variability analysis (Lehrer, 2000; Shaffer & Meehan, 2020). Precise Breath adapts this protocol for ongoing multi-session refinement:

• Sensor: Any Bluetooth LE chest strap with R-R interval support. The Polar H10 (recommended) provides research-validated beat-to-beat precision — the reference standard for HRV measurement (Task Force, 1996; Gilgen-Ammann et al., 2019). Confirmed compatible: Polar H10, Garmin HRM Dual. Other standard BLE chest straps using HR Service 0x180D should also work.
• Block structure: Each breathing block uses a 30-second transition ramp, 15-second settling period, and 120-second measurement window — meeting the Task Force (1996) minimum of 2 minutes for LF power estimation.
• Scoring: FFT spectral amplitude at the breathing frequency (85% weight) combined with Hilbert-transform phase coherence (15% weight). Signal quality gating (artifact rate + coefficient of variation) filters unreliable blocks.
• Calibrate mode: Tests 5 preset rates (4.5–6.5 BPM) in randomized order in a single ~14-minute session to establish a baseline estimate.
• Explore mode: Adaptive multi-session algorithm that explores nearby rates session by session, converging on the patient's personal frequency with increasing confidence.
• IBI adaptation: When sufficient data is available, the app accounts for session-to-session shifts in resonant frequency correlated with resting inter-beat interval — consistent with findings in Lalanza et al. (2021).
• Privacy: All data is stored locally on the patient's device. No data is transmitted, no accounts are required.

HRV Analysis & Scoring Details

For clinicians and researchers who want to understand the measurement methodology.

Signal Pipeline

Each breathing block produces a sequence of R-R intervals from the chest strap sensor. These are processed in a five-stage pipeline:

1. Artifact removal: A two-pass adaptive pipeline: (1) range filter flags intervals outside 300–1500 ms; (2) adaptive successive-difference filter uses a running median of inter-beat differences with a both-sides criterion, flagging only transient spikes where both neighbors deviate. Flagged intervals are replaced by linear interpolation (not dropped), preserving timing. Blocks with fewer than 30 usable intervals are excluded.
2. HR tachogram: Clean RR intervals are converted to instantaneous heart rate (bpm) and placed on a continuous time axis.
3. Uniform resampling: The tachogram is interpolated via natural cubic spline to a regular 4 Hz grid — well above the highest breathing frequency of interest — making it suitable for FFT-based spectral analysis.
4. Detrending & windowing: The DC component is removed and a Hanning window is applied to reduce spectral leakage before frequency-domain analysis.
5. Spectral & phase analysis: FFT is applied to the windowed signal. A separate Hilbert transform pass extracts instantaneous cardiac phase relative to the breathing reference.

Composite Score

Each 120-second measurement window produces a composite score (0–1):

• Spectral amplitude (85%): Normalized power at the target breathing frequency (±0.015 Hz band) relative to total spectral power. Captures the magnitude of cardiovascular resonance at the target rate — the primary signal of RSA maximization. Self-normalizing (ratio, not absolute power) so it is robust to between-session differences in resting HR and HRV magnitude.
• Phase coherence (15%): Mean resultant length (MRL) of instantaneous cardiac phase relative to the breathing reference, computed via Hilbert transform. MRL = 1.0 indicates perfect phase-locking; MRL = 0 indicates a random phase relationship. Weighted 15% because the 120-second window (~10 breathing cycles) is below the ≥20-cycle ideal for stable phase estimation — this metric is advisory rather than primary. Note: phase coherence requires a symmetric (1:1) inhale:exhale ratio. The app supports extended-exhale ratios (1:1.5, 1:2) as a user option, but phase measurement is disabled for those sessions — asymmetric breathing creates a systematic cardiac phase offset that would make MRL uninterpretable. Extended-exhale sessions are scored on spectral amplitude only.

Signal Quality Gating

A signal quality metric (0–1) is computed from artifact rate and HR coefficient of variation:

Blocks with SignalQuality < 0.50 are excluded from the resonant frequency estimator. This threshold prevents motion artifact, electrode contact issues, or ectopic beats from distorting the frequency estimate. Excluded blocks are flagged in the session summary so the patient can identify sensor placement problems early.

Resonant Frequency Estimation

After each Explore session, the app re-estimates the patient's resonant frequency using all valid blocks from all historical sessions:

• Weighted Lorentzian fit: A quality-weighted Lorentzian (damped-oscillator) curve is fit to the (rate, score) pairs across all valid historical blocks (minimum 5 blocks, signal quality ≥ 0.50). The Lorentzian is the physically correct spectral shape for a resonant system and is fit via grid search over center frequency and width, with amplitude and baseline solved analytically at each grid point. Quality-squared weighting gives high-quality blocks more influence while retaining the full historical record. A weighted quadratic is used as fallback if the Lorentzian grid search fails to find a valid peak.
• IBI covariate model: When sufficient data is available, an extended model makes the Lorentzian center frequency IBI-dependent: x₀(IBI) = x₀ + β · IBI_c, accounting for the correlation between resting inter-beat interval and optimal breathing rate — consistent with findings in Lalanza et al. (2021). This enables mid-session adaptation when the patient's resting HR differs from their historical baseline, so the rate explored in Block 2 onward is adjusted to the patient's state that day.
• Fallback: If neither the Lorentzian nor quadratic fit is well constrained, a quality-weighted bin method is used as a fallback estimator.
• Confidence: Each estimate carries a confidence score (0–1) based on data quantity, goodness of fit, and peak coverage. This is displayed in the session summary so patients can see how settled the estimate is over time.

Patient Setup Guide

The following four steps get a patient from first download to active practice.