Common Average Reference (CAR)

Common average reference (CAR) is a re-referencing step that subtracts the mean of all channels — or of a chosen subset, such as one shank — from each channel at every sample, removing noise that is common across them. Movement artifact, mains interference and drift of the reference electrode appear on every channel alike, and the average is a better estimate of that shared component than any single reference site. Where a few channels have large artifacts, the median holds up better than the mean.

CAR is a substitute reference built from the probe, and it works as well as the probe's sites are alike. It needs many channels sampling the same kind of tissue — it is a high-density technique — and it removes any signal that is common too, so a field potential shared across a shank is subtracted along with the noise, and CAR is applied to the spike band rather than to the LFP. Per-shank CAR keeps the subtraction local on a multi-shank design. It is applied after filtering and does not replace a good reference electrode ↗; it corrects for what one could not cancel. On a NeuroNexus probe the reference and ground come out on separate wires for the same reason: a reference placed well needs less correcting than one placed by default.