ProjectA month eachAbout $35 (fNIRS) or $50 (in-ear)Tier 2

Project C: Other windows on the brain

A DIY fNIRS prefrontal band, which is a pulse oximeter pointed at the cortex, and in-ear EEG on a custom earpiece, the frontier for sleep and consumer wearables. Two modalities almost nobody builds as a student.

AssumesProject A: The ADS1299 board Project C: Make your own electrodes SpineAnalog / mixed-signal hardware

On this pageA month each

fNIRS
In-ear EEG

Scalp EEG is one window. Two others are within reach of the Phase 2 bench and are far less crowded. Functional near-infrared spectroscopy shines near-infrared light through the scalp and skull and measures how much comes back to estimate blood oxygenation in the cortex beneath: the same physics as a pulse oximeter, pointed at the brain. In-ear EEG puts electrodes on a custom earpiece, out of sight and stable for sleep and everyday wear. Both have real research literatures and both are being commercialized now.

Predict before you look

fNIRS measures a haemodynamic response. How long after neurons become active does the fNIRS signal peak?

Four to six seconds. Active neurons demand oxygen; blood flow increases with a delay of a few seconds, overshoots, and the oxygenated haemoglobin peaks around five seconds after the activity. fNIRS is therefore slow, like fMRI, and cannot see the millisecond events EEG sees. What it can see is sustained activation of a cortical region through hair-free skin, without electrodes, in a moving person, which EEG does badly.

Infrared LED eye safety

Near-infrared LEDs are invisible and can be bright. Never look into one. Drive them at currents within the datasheet’s eye-safe limits (photobiological safety standard IEC 62471 is the reference), keep them in contact with skin or pointed at the detector, and add a hardware current limit so firmware cannot exceed it. Battery power. For in-ear electrodes: nothing that could damage the ear canal; use soft custom moulds made as an audiologist would, and never insert anything rigid.

fNIRS

What it measures

Time resolution

What is hard about it

fNIRS

Blood oxygenation in the outer centimetre of cortex, from how much near-infrared light comes back

Seconds. It is a haemodynamic signal, like fMRI

Hair, ambient light, motion, and separating scalp blood flow from cortical

In-ear EEG

The same electrical activity as scalp EEG, from inside the ear canal

Milliseconds, as with any EEG

An earpiece that fits, stays put, and keeps contact without gel

fNIRS is a pulse oximeter pointed at the cortex, which is both a fair description and a good way to think about the hardware. In-ear EEG is where consumer sleep and hearable products are heading, and the contact problem is the entire field.

Figure 1. Two windows, both buildable in a month, and what each actually measures. Neither replaces EEG; the point is that almost no student builds either, which is exactly why they are worth building.

The physics. Oxygenated and deoxygenated haemoglobin absorb near-infrared light differently, and the difference flips around 800 nm. Two wavelengths, one either side (760 and 850 nm are standard), let you separate the two. A source LED and a detector photodiode three centimetres apart on the forehead: the light that reaches the detector has travelled a banana-shaped path through skin, skull, and the top millimetre or two of cortex. Changes in detected intensity over seconds, converted through the modified Beer-Lambert law, give changes in oxygenated and deoxygenated haemoglobin concentration.

The hardware. Two LEDs (760, 850 nm) and a photodiode with a transimpedance amplifier, time-multiplexed so the detector sees one wavelength at a time plus a dark period, sampled at 10 to 20 Hz. The MAX30101-class pulse oximeter chips do exactly this on one chip and can be pressed into fNIRS service for a first prototype; a discrete design with a bigger photodiode and a more sensitive front end does better. Mount on a headband with the optodes in a 3D-printed holder that presses gently and blocks room light.

The experiment. Mental arithmetic versus rest, twenty-second blocks, ten repetitions. Prefrontal cortex activates during the arithmetic. Average the blocks; you should see oxygenated haemoglobin rise and deoxygenated fall, peaking about five seconds in. The heartbeat will be visible in the raw signal, which is your confirmation that you are measuring blood. Motion is the enemy; the head must be still or the optode-skin coupling changes.

In-ear EEG

Why. The ear canal is close to the temporal lobe, stable, hidden, and comfortable for hours. In-ear EEG picks up alpha, sleep stages, auditory evoked responses, and some SSVEP, with amplitudes a few times smaller than scalp electrodes and a very different reference geometry. Several sleep-wearable companies and hearing aid manufacturers are building it.

The hardware. An ear impression (an audiologist can make one, or a hobby kit does an acceptable job), a soft earpiece cast or printed from it, and two or three electrodes on its surface: silver-coated dry contacts, or conductive silicone pads, or Ag/AgCl dots with a trace of gel. Wires out to your ADS1299 board. Reference on the other ear’s earpiece or a mastoid; bias on the neck.

The experiment. The eyes-closed alpha test, again. It should appear, smaller than at O1. Then an auditory steady-state response: a tone amplitude-modulated at 40 Hz produces a 40 Hz response the in-ear electrodes see well. Then wear it to sleep alongside your forehead band and compare the stagers.

fNIRS: breadboard the two-LED, one-photodiode front end with time multiplexing. Confirm the heartbeat in the raw signal on a fingertip, then on the forehead.
Build the headband holder. Run the arithmetic-versus-rest blocks. Average and plot oxygenated and deoxygenated haemoglobin.
In-ear: make an ear impression and cast an earpiece with two electrodes. Measure impedance.
Alpha test in-ear versus O1. Then the 40 Hz auditory steady-state response.
Record a night with both the in-ear and the forehead band; compare hypnograms.

What goes wrong

fNIRS signal saturates or is flat. LED current too high (saturating the photodiode) or too low, or room light leaking in. Block ambient light and adjust current until the dark period reads near zero and the lit period is mid-range. Huge motion artifacts in fNIRS. Optode coupling changing. Better holder, more pressure, less movement. In-ear electrodes read nothing. Impedance too high; dry contact in the canal is hard. A trace of gel or a saline-wetted pad, and give it minutes to settle.

Stop here if you like. A haemodynamic response you measured with light, or alpha from inside your ear. Either is a portfolio piece few students have.

ExhibitAn fNIRS haemodynamic response averaged over blocks, or an in-ear EEG alpha and auditory steady-state response, from hardware I built, with the optode or electrode design documented.

Recall

What does fNIRS measure and what is its time resolution?

Changes in oxygenated and deoxygenated haemoglobin in superficial cortex, from the differential absorption of two near-infrared wavelengths; the haemodynamic response peaks four to six seconds after neural activity, so it is slow like fMRI.

Recall

Why is in-ear EEG attractive despite smaller amplitudes?

The ear canal is stable, hidden, and comfortable for hours, making it suited to sleep and everyday wearables; it captures alpha, sleep stages, and auditory responses well.