Your brain runs on electricity.

When you think, brain cells fire. That activity is electrical, and a faint trace of it reaches the surface of your head.

A brain-computer interface puts sensors on the skin, picks up that trace, and lets a computer react to it. Nothing goes inside you. It does not read your mind. It measures electrical patterns that may correspond to particular intentions or actions.

sensors sit on

the skin

signal size

millionths of a volt

what it reads

patterns, not thoughts

The equipment works. Nobody wants to wear it.

Recording brain signals properly means a cap covered in sensors, held on with sticky gel, usually fitted by someone trained to do it.

The data it produces is good. But no one is wearing that to school, to work, or on a train. That single fact is why most of this technology never leaves the lab.

sensors

32 to 256

setup time

minutes to an hour

wearable daily

no

The array shown is a rough illustration, not a real sensor layout.

So put it in your ear.

People already wear electronics in their ears all day. Nobody looks twice at a pair of earbuds.

The idea is to shrink the whole system into a small band that sits around one ear. Comfortable, invisible, and wearable for hours. The catch is real though. Far fewer sensors, packed into a small space, sitting further away from most of the brain.

sensors

roughly 4 to 12

placement

around one ear

the question

how much signal survives

Placement shown is illustrative. Sensors near the ear are not the same as sensors inside it.

Your jaw is louder than your brain.

There is a problem with sensing near the ear. Your jaw and face muscles are right there, and the electricity they produce is far stronger than the brain signal underneath.

So what actually arrives at the sensor is a mixture. Brain activity, muscle activity, eye movement, your heartbeat, and the hum of the mains supply in the walls, all stacked on top of each other.

brain signal

small and slow

muscle signal

large and fast

also in there

eyes, heart, movement, mains hum

These traces are drawn for illustration. They are not recorded data.

Everyone throws the muscle away. Should they?

The standard approach treats muscle activity as junk and filters it out to leave a cleaner brain signal.

Cerebral is testing whether that is the wrong move for a device worn on the ear. If some of that muscle activity is telling you what the person is trying to do, then deleting it is not cleaning up the signal. It is throwing away half the answer.

common practice

filter the muscle out

the hypothesis

muscle may carry intent

study 01

separate both, then use both

Read the study design →

Teaching software to read the mixture.

A model is trained on many examples. It learns to match a short slice of messy signal to whatever the person was doing at the time.

Fewer sensors means less for it to learn from. Two things need measuring. How quickly performance drops as sensors are removed, and whether treating brain and muscle as two separate inputs slows that drop down.

input

a short slice of signal

compared

brain only, muscle only, both

measured

how fast accuracy falls

This is a diagram of how a layered model is arranged. It is not a trained network.

Nothing has been built yet.

The goal is something the size of an earbud. A few sensors on the tip that touch the skin inside the ear, a chip to clean up the signal, and enough computing power to be useful on its own.

The shape here is a sketch. It exists to work out what has to fit where, and how far apart the sensors can sit. No hardware has been made, and no number anywhere on this site came from a real device.

status

planned, not built

sensor count

not decided

stage

four on the roadmap

No working prototype exists. This shape is illustrative.

See the build section →