5 min read
Updated September 9, 2025
An EEG channel is a difference between two points. The reference is one of them, so it is half of every measurement you make. Choosing it is a scientific decision, not a wiring detail.
Research use only. NeuroNexus products and the procedures described here are for non-clinical research use, and must not be used in human or veterinary medical procedures.
Reference and ground are different things#
The reference is the second input to the differential amplifier. It defines what "zero" means on every channel.
The ground is the return path that keeps the animal and the amplifier at a common potential. It does not appear in the measurement.
Conflating them is the most common wiring error on a grid array, and it produces noise that looks environmental.
Where to put the reference#
Whatever you choose, the requirement is the same: electrically quiet, mechanically stable, low impedance, and outside the field you intend to measure.
Bone screw, outside the grid. The usual choice for rodent grid work. Stainless steel or Ag/AgCl, placed away from the recording area — over cerebellum or frontal bone, depending on where the grid sits.
Wire in a small craniotomy outside the recording area. More stable potential than a screw, because it contacts CSF rather than bone. More surgery, and one more thing to fail chronically.
A site on the array itself. Some grids carry a reference site. Convenient, and it works when that site sits over inactive tissue. It fails quietly when it does not, because the reference then carries signal and subtracts it from every channel.
Configurations borrowed from human EEG — linked mastoids, a midline site, ear references — have animal-model equivalents, but the equivalence is approximate. State what you actually used rather than the human label for it.
The three schemes#
The reference you record against and the reference you analyse against do not have to be the same. Record against a quiet physical reference, then re-reference in software. That order preserves options; the reverse does not.
Monopolar (common reference). Every channel is measured against one reference. Preserves large-scale, spatially broad activity. Any noise or signal on the reference appears on every channel, inverted.
Use it when the phenomenon is spatially broad — sleep staging, global oscillations, slow waves.
Bipolar. Each channel is the difference between two adjacent grid sites. Rejects what is common to the pair, including reference noise and much of the EMG. It also removes anything spatially broad, which may be the signal.
Use it when you want local activity and strong artifact rejection, and can accept losing the global component. Note that a bipolar montage has no absolute polarity — the sign depends on which site you subtracted.
Common average reference (CAR). Subtract the mean across channels from each channel. Approximates a neutral reference without needing a physically neutral site.
Use it when the array covers enough tissue that the average is close to inactive. It fails on small arrays and on arrays where activity is spatially uniform — the average then contains the signal, and CAR subtracts it. It also propagates a single bad channel to every channel, so exclude bad channels before computing the average.
Consequences worth stating in the methods#
Every scheme changes what a result means:
- A phase relationship measured in a bipolar montage is not the same quantity as one measured monopolar.
- Coherence between two channels is inflated by a shared reference and deflated by CAR.
- Amplitude is not comparable across schemes at all.
State the scheme, the physical reference location, and any re-referencing applied. A result that does not name its reference cannot be reproduced.
Recording alongside stimulation#
Match impedances across sites where you can. Differential drift during stimulation comes from mismatch as much as from the stimulus itself.
Record differentially where the montage allows it — common-mode rejection removes most of the stimulus artifact before any processing.
Subtract a template for the residual. The artifact is deterministic and its shape is measurable; the signal is not.
Shield. A Faraday cage removes the environmental component so what remains is attributable to the stimulus.
Troubleshooting#
Every channel is noisy, in phase — the reference. Check its impedance and its connection before anything else.
Baseline drifts slowly — reference contact, or a temperature change at the reference site.
One channel is flat or wild — that channel. Exclude it before computing a common average.
Activity looks identical across the whole array — a reference that is carrying signal. Move it, or re-reference in software and see whether structure appears.
Support#
Technical questions: support@neuronexus.com · +1.734.913.8858