hkk.fyi/PM-TI
B.Sc. final project · University of Tehran · ICEE 2026

PM-TI

Sharper deep-brain stimulation.
No surgery. No new hardware.

My B.Sc. project, told as a chain: a problem, its answer, the new problem that answer brings — until the last answer, which is ours. Scroll to follow it, or use the arrow keys.

StudentHassan Keshtkar
SupervisorDr. Hoda Ameri
SchoolECE · University of Tehran
PaperICEE 2026 · IEEE ↗
scroll · or press →
Problem · 01

Millions live with a brain that misfires.

When the drugs stop working, the target sits deep inside the brain.

In Parkinson's, essential tremor and epilepsy, a small group of neurons deep in the brain fires in the wrong rhythm. Medicine helps many patients — but not all of them, and often not forever. Then the faulty circuit itself has to be reached.

Gowers' 1886 sketch of Parkinson's disease 01 / 1886 · Gowers
Parkinson's8.5 M+ people
Spiral drawing test of an essential-tremor patient 02 / spiral test
Essential tremor~1 in 100
EEG 3 Hz spike-and-wave discharge 03 / EEG · spike-wave
Epilepsy50 M people
Solution · 01

Deep Brain Stimulation works.

Electrodes at the target. The tremor stops.

A surgeon places a thin wire right inside the faulty circuit. A pacemaker-like device in the chest sends it small, fast pulses that override the bad rhythm.

Solution · 01

A few volts, right where it matters.

Parkinson's · tremor · dystonia · epilepsy — in clinics since the 90s.

The contacts sit inside the target, so only a few millimetres of tissue around them are stimulated. That precision is why it works so well.

Problem · 02

But it needs brain surgery.

skull openinginfection · bleedinghardware for life

A hole is drilled in the skull, the wire passes through healthy brain, and the hardware stays for life. So the next question is: can we reach the same deep spot without opening the head?

Lateral skull X-ray with two DBS leads
X-ray · two DBS leads · CC BY-SA
Solution · 02

Temporal Interference.

Stimulate deep — from the scalp.

Two pairs of electrodes are simply placed on the head — no surgery. The trick is in the two frequencies they use.

Solution · 02

Two kHz fields. Neurons ignore each one.

2000 Hz from one pair, 2010 Hz from the other.

A field that flips thousands of times per second is too fast for a neuron to follow, so each field on its own does almost nothing.

Solution · 02

Where they meet, a 10 Hz beat is born.

Neurons follow the beat — only where both fields overlap.

Two tones 10 Hz apart add up to one tone that swells and fades 10 times a second — like two guitar strings slightly out of tune. Neurons do follow that slow beat, and it is strongest where both fields are strong.

Solution · 02 · the formula

Two frequencies, one beat.

TI · 1
s1(t)=E1cos⁡(ω1t),s2(t)=E2cos⁡(ω2t),ω2=ω1+Δω
TI · 2 · envelope |EAM|TI =2min⁡(E1,E2)

The weaker field sets the envelope. Nothing else.

E₁ and E₂ are how strong each pair's field is at one point in the head. The beat there is twice the weaker one — so it is big only where both fields are strong.

E₁ · 2000 Hz
E₂ · 2010 Hz
E₁ + E₂envelope · 10 Hz
200 ms
Problem · 03

The hotspot is wide.

≈ 30 mm across a 70 mm phantom. The neighbours get stimulated too.

Both fields are fairly strong over a large region around the centre, so the beat is strong there too. The usual fix is more electrodes and more channels. We wanted to sharpen it by changing only the waveform.

more electrodes more channels bigger hardware → a smarter waveform
k = 1 · conventional TI · |EAM,y|
01 · envelope
Solution · 03 · our work · ICEE 2026

PM-TI: one carrier, opposite phase modulation.

PM-TI · 1
s1(t)=E1cos⁡(ωct+φ(t)),s2(t)=E2cos⁡(ωct−φ(t))
φ(t)=arccos⁡(cos⁡(ωmt)k),k≥1,ωm≪ωc

Both electrode pairs now share one frequency. Their phases are pushed in opposite directions by φ(t), and the number k sets how hard. With k = 1 this is exactly classic TI. With k > 1 the beat stays the same where the two fields are equal — at the target — and shrinks wherever they are not.

envelope · E₁ ≠ E₂
E₂ / E₁0.60

The beat at a point where E₂ is weaker than E₁. At E₂/E₁ = 1 all three are equal; the lower it goes, the more k > 1 cuts the beat.

Unequal fields get penalised. k turns the penalty up.
No new electrodes — a new waveform.

Solution · 03 · one number to tune

Turn k up. The focus tightens.

Same electrodes. Same current. Same peak.

k is just a number in the signal generator. Raising it from 1 to 3 shrinks the hotspot width (FWHM: where the beat is above half its peak) from about 30 mm to 9 mm, while the beat at the target stays as strong.

1.00
k · focusing parameter
29.5mm
FWHM · along y = 0
1 3
|EAM,y| · live · auto, or drag the slider
Results · simulation · 70 mm phantom

Same peak. A third of the width.

A finite-element model of the 70 mm salt-water phantom: same four electrodes, same current, only k changes. The width halves at k = 2 and drops by two thirds at k = 3.

k = 1 · conventional TI
k = 2 · PM-TI
k = 3 · PM-TI
0.0mm
FWHM · k = 1
 
0.0mm
FWHM · k = 2
−53 %
0.0mm
FWHM · k = 3
−69 %
envelope along y = 0k=1k=2k=3
peak change < 0.5 %
Live · sweep along x · compare the envelope

Off-centre, the beat dies faster with k.

A test point slides along the middle line. Each panel is the signal there, scaled to its own size so only the beat matters. At the centre all three beat fully; move away and k = 3 goes flat first.

k = 1
V/m
|EAM|—
k = 2
V/m
|EAM|—
k = 3
V/m
|EAM|—
x · mm · along y = 0
x = 0.0 mm
drag on any map or the profile to scrub
Results · measurement

We built it. We measured it.

A 70 mm glass of salt water stands in for the head. An Arduino with two signal-generator chips makes the waveforms, two current sources drive the four electrodes, and an oscilloscope records the voltage at 37 points on a 10 mm grid — once for each k.

Bench test: saline phantom, probe, oscilloscope showing the 10 Hz beat
21 Aug 2026 · test day · the beat on the scope
Probe in the saline phantom
probe · 10 mm grid
phantom70 mm saline
scan37 points · 10 mm
carrier · beat2005 Hz · 10 Hz
runsk = 1 · 2 · 3
Results · simulation vs. measurement

The phantom agrees.

Top row: the computer model. Bottom row: the real phantom, measured point by point. Same shape, same trend — the higher k, the tighter the spot.

k = 1
k = 2
k = 3
simulation
measurement
0.00
correlation · simulation vs measurement
0.0mm
measured FWHM · k = 2 · from 33.9 mm
−49 %
0.0mm
measured FWHM · k = 3 · from 33.9 mm
−58 %
recorded at the centre · k = 3 · band-passed
240 ms
The whole story · one screen

Three problems. Three answers.

Problem · 01The brain misfires

Deep circuits fire in the wrong rhythm, and medicine is not always enough.

↓
Solution · 01DBS

An electrode inside the target overrides the bad rhythm.

Problem · 02It needs surgery

The electrode goes through the skull and brain, and stays for life.

↓
Solution · 02TI

Two fast fields from the scalp; a slow beat where they meet.

Problem · 03The beat is wide

About 30 mm across — the neighbours are stimulated too.

↓
Solution · 03 · oursPM-TI

Same electrodes, a new waveform: k = 3 cuts the width by 69 % in simulation, and the bench agrees.

PM-TI · phase-modulated temporal interference

Sharper. Non-invasive.
Software-only.

Same electrodes. Same current. A third of the footprint.

−69%
focus width
0
surgery
0
new electrodes
StudentHassan Keshtkar
SupervisorDr. Hoda Ameri
School of ECEUniversity of Tehran
PaperICEE 2026 · IEEE ↗
scan · this page, live hkk.fyi/pmti