Nora DabbaghBiomedical engineering

04Analysis · MATLAB

ECG RR-interval variability

Two simulated rhythms with nearly the same average heart rate, separated by how much the time between beats varies.

Coursework · Simulated signals, not clinical data
Two 10-second simulated ECG traces. Top: normal rhythm with R peaks every second. Bottom: irregular rhythm with uneven spacing. Each detected R peak is circled; near 5 s on the top trace and 6.1 s on the bottom trace the circle sits below the peak, on its rising edge.

MATLAB output, seed rng(1). Circles mark detected R peaks; two sit on the rising edge, discussed under Outcome and limits.

Course
BME 240, Modeling Physiological Data and Systems, University of Illinois Chicago
Term
Spring 2026
Team
2 students, submitted as a pair
My part
All of the work: signal model, detector, analysis and paper

01Problem

Irregular heart rhythms show up as uneven spacing between beats. I tested whether one simple timing measure, the spread of RR intervals, can separate a regular rhythm from an irregular one.

02Requirements and constraints

  • Controlled signals: 10 s at 500 Hz, with the same waveform shape, so beat timing is the only variable.
  • The same detector on both signals.
  • Reproducible numbers: a fixed random seed.

03Design decisions and why

01Synthetic beats from Gaussians.
Each beat sums a small P wave, a tall narrow R peak and a wider T wave, a shape used in published synthetic-ECG models. Normal beats fall every second from 1 to 9 s. Irregular beats fall at uneven times over the same window. Added noise (σ = 0.03) keeps the signal from being perfectly clean.
02A detector with a refractory window.
A sample counts as an R peak if it is above 0.6, higher than both neighbors, and at least 0.4 s after the last peak. The threshold ignores P and T waves. The 0.4 s window ignores noise around a peak and caps detection at 150 beats per minute.
03Coefficient of variation as the main metric.
CV = σRR ÷ μRR scales the spread by the average spacing, so it compares rhythms at different rates.

Try it

The R-peak detector, on both signals

9 peaks detected in the irregular signal, which has 9 beats.

Peaks found
9 / 9 beats
Mean RR
1.025 s
SD
0.317 s
CV
0.310

Re-run in your browser with its own random noise, so the numbers differ slightly from the paper’s rng(1) results in the table below. Lower the threshold under the T wave and shorten the window to see why both settings exist.

04What I did

I wrote the MATLAB signal model, the R-peak detector and the RR, SD and CV analysis, plotted the results, and wrote an IEEE-format paper with the full code in an appendix. The October 2026 revision re-runs the same code with a fixed seed and marks the detected peaks on the plots.

Download the paper (PDF, 4 pages)

05Outcome and limits

RR-interval metrics, seed rng(1)
SignalMean RR (s)SD (s)CV
Normal1.0000.0070.007
Irregular1.0250.3170.309

The mean RR intervals were nearly equal, but the irregular rhythm’s coefficient of variation was more than 40 times the normal one. Marking the detections exposed a detector flaw: twice, noise on the rising edge of an R wave formed a local maximum above the threshold, so the detector fired about 8–14 ms early and its 0.4 s window skipped the true peak. That error accounts for most of the normal rhythm’s 0.007 s spread. The signals are synthetic, a fixed threshold fails when amplitude changes, and irregular timing is not a diagnosis.

RR interval by beat number. The normal rhythm stays flat near 1.0 s; the irregular rhythm swings between about 0.69 and 1.51 s.
RR interval by beat: the normal rhythm stays near 1.0 s; the irregular one swings from 0.69 to 1.51 s.

06What I’d do next

  • Fix the detector by taking the largest sample within a short window after each threshold crossing.
  • Run the pipeline on real recordings, such as the MIT-BIH Arrhythmia Database on PhysioNet.
  • Replace the fixed threshold with an adaptive detector, for example Pan–Tompkins.

Sources

Show all 4 sources
  1. Task Force of the ESC and NASPE. Heart rate variability: standards of measurement, physiological interpretation, and clinical use. Circulation 93(5), 1043–1065 (1996).
  2. Sayadi O, Shamsollahi MB. Synthetic ECG generation and Bayesian filtering using a Gaussian wave-based dynamical model. Physiol Meas 31(10), 1335–1350 (2010).
  3. Moody GB, Mark RG. The impact of the MIT-BIH Arrhythmia Database. IEEE Eng Med Biol Mag 20(3), 45–50 (2001). physionet.org
  4. Pan J, Tompkins WJ. A real-time QRS detection algorithm. IEEE Trans Biomed Eng BME-32(3), 230–236 (1985).