← Back to portfolio

Timing pulses to a beat
a hands-on explainer

Can the timing of a small electrical pulse, rather than its strength, make it easier to see a phosphene, a flash of light produced by stimulating the eye? Play with the three ideas behind our study below.

Based on: Re, D., Ibrahim, R., & Hanein, Y. Pulses timed to temporal-interference envelope peaks increase phosphene detection with wearable dry-printed transorbital electrodes. Manuscript in preparation. Preliminary results: IEEE EMBC 2025.
Everything on this page is a teaching simulation with made-up parameters. It shows the ideas, not our data or results.

00The setup in 3D

Three pairs of electrodes, all dry-printed. A pulse pair above and below the right eye delivers brief biphasic pulses. A front pair beside the eye carries a 1000 Hz current, and a neck pair on the right side of the back of the neck carries 1005 Hz. Where the two carriers overlap they beat at 5 Hz, and the pulses are timed to that beat. Drag to rotate, and switch conditions at the top.

Open the 3D view full screen → Head: the MIDA model (scalp, eyes, brain, and the real optic nerves, chiasm and tracts). Electrode positions come from our SimNIBS model on ernie, transferred by aligning the eyes; current paths are illustrative.

01Psychometric results

Do gel-free, dry-printed electrodes cost anything? Each curve is the average psychometric function, the probability of reporting a phosphene as the current rises, next to how uncomfortable the same trials felt.

Dry-printed vs gel electrodes Experiment 1 · 6 people · 13 sessions · 1040 trials

Detection
dry (mean of 6)gel (mean of 5)± 1 SE
Discomfort
share of trials at each rating (0 = none, 5 = barely bearable)
1.10dry / gel threshold, paired within the 5 people tested with both (90% CI 0.91–1.34)
2 / 5highest discomfort rating in 1040 trials, either electrode
0 / 70false alarms on catch trials, dry (gel: 4 / 60)

Logistic fits per session (detection against current), averaged within each person and then across people. Discomfort rose with current more steeply for dry electrodes (p = 0.016), but above threshold dry was rated uncomfortable more often in only 2 of the 5 people tested with both. Data from our manuscript in preparation.

02Timing the pulse

Our question was not whether the envelope stimulates on its own. It was whether the same pulse becomes more effective when it lands on an envelope peak. Switch between the four conditions in the 3D view (00):

The clever comparison is the last one. "On peaks" and "unmodulated control" deliver the same kHz amplitude at the moment of the pulse. If peaks still win, it's the beat that matters, not just extra kHz current. In the 3D view's signal strip the band has the same height on peaks and with the unmodulated carrier; only the beat is missing.

One more subtlety from the head models: "peak" is defined by the two sources being in phase. Inside tissue that instant is an envelope maximum only where the two fields point the same way (Efront · Eneck > 0) and a minimum where they point in opposite directions. With the montage as wired, the "peak" pulses met the local maximum on 99.7% of the retina and all along the nerve to the apex; reversing the neck pair would have flipped that.

03Closing the loop

A wearable stimulator has to keep working for minutes or hours, but the threshold for seeing a phosphene isn't fixed: it drifts upward as the visual system adapts, and drops back after a rest. A fixed intensity that starts near threshold soon falls below it.

Below, a simulated user answers "seen" or "not seen" on every trial. Pick a controller and see how well it keeps the stimulus near the target, a 70% chance of being seen.

hidden 70% point (what we want to hit) intensity the controller chose seen not seen rest
–trials in the useful range (55–85% seen)
–overall seen rate (target 70%)
0trials so far (rest every 20)

What the controllers do. Fixed never changes. Staircase steps down after two "seen" answers and up after one "not seen". Bayesian tracker keeps a probability distribution over where the threshold is, widens it a little every trial because the threshold drifts (like a Kalman filter), sharpens it with each answer, and picks the intensity that should be seen 70% of the time.

In the study we built this kind of simulation from the effects we measured, to choose a controller and to size the follow-up experiment. The design it points to is a two-level loop: a slow tracker for the drifting threshold, and fast timing of each pulse to the beat.