Some recording situations can create signals that are highly correlated, but that are not the result of capacitive, inductive, or conductive coupling in the circuits of the recording setup. For example, it is possible that there is no electrical coupling in your circuit, but that all electrodes are experiencing a common signal. This common signal could be due to a very strong noise source that every electrode is experiencing equally. If you are experiencing strong noise in your recordings, especially if the noise is periodic as it may be in an environment experiencing strong line (power) noise or strong sampling noise from another machine (such as an MRI, additional biosignal sampling machine, or task tool), then it is likely not useful to take a crosstalk measurement as it is not valid.

There are other situations that have 'good recording' that still may show high correlation, especially depending on your method of evaluating crosstalk. One possibility is that you are making use of a very high density neural probe, such as one utilizing CMOS technology, dense tetrodes, or similar. Another is that you are doing experiments in brain regions that are known for high coherence, such as the brain an ictal state. In these cases, slower wave neural activity may show up as nearly identical on all channels. For this reason, crosstalk is usually best evaluated at higher frequencies due to the likelihood that those are independent on each channel. Even at high frequencies, though, electrodes may be accurately reporting common biological signal, such as muscle activity near the neural site.

Finally, some methods of referencing or signal processing can produce artifacts on multiple channels that may make these channels appear correlated more than they are. Consider, for example, a simple software reference/montage where each channel is referenced to a single neural channel. If this single neural channel is not a lower impedance reference, it may have strong neural signal and high frequency content of its own, and that simple subtraction will result in a situation where each other channel experiences the inverse of that channel, giving them all a layer of common signal that may include high frequency content such as spikes/action potentials. While this can be easy to spot (due to the inversed action potentials), it can also occur in a more subtle form if that channel is noisy or has high amplitude content that is more broad spectrum. Similar effects can be observed when choosing hardware reference channels for your setup. Choosing a reference channel that has a high impedance reference or a reference that is experiencing noise that the neural channels themselves don't experience can result in common signals on all channels.

All of the above effects are typically more likely than true crosstalk, so it is best to investigate these options first.

Crosstalk due to conductive effects is the most common cause of crosstalk because it can occur to effectively any cable through damage. If conductors somehow have their insulation compromised, due to crushing, cutting, rubbing, melting, etc, then the wires can directly contact eachother, shorting the conductors together and resulting in completely common signal on both conductors since all wires can be though of as equipotential paths. This type of effect can also occur if water or some other conductive debris bridges electrical connections.

Some electrodes have very high density, making them more susceptible to certain forms of crosstalk. Consider, for example, a set of microwires that are bunched together on a pcb. In many cases the wires themselves will touch eachother, but are insulated by a coating on the wire. If that insulation were damaged by an errant soldering iron, excessive rubbing during handling/surgery, or due to a manufacturing defect, then these wires would be in direct contact with eachother.

Similarly, imagine a situation where one is using a high density laminar probe that has layered metal traces. A very high voltage stimulation pulse across one lane of this device, may result in capacitive voltages/currents on the other traces.

As mentioned above, it is typically best to measure for crosstalk in higher frequency registers as signal in this band is less likely to be correlated.

A common tool available to electrophysiologists is the ability to take impedances of their electrodes. Crosstalking channels will generally show similar impedances. For channels that are experiencing conductive crosstalk/shorting, impedances will typically become lower, as the impedance of the channel will be evaluated on both contacts in parallel.

Spacial elements are important to consider when evaluating crosstalk. If you have a linear connector that has channels 1-2-3-4-5, and you are seeing similar signal on channels 2 and 5, it is unlikely that it can be attributed to the connector as those channels are too far apart to likely have an capacitive influence or conductive contact. Spatial organization of the crosstalking channels can be very important when finding sources of crosstalk.

Removing elements from the system is another effective way to locate sources of crosstalk. For example, if you are experiencing crosstalk in your full recording setup, taking a recording from saline/air without connection to the electrodes could eliminate the electrodes as a possibility (or prove them as the source). This can be repeated for each element in the system until it is isolated.

Finally, crosstalk can often be seen by electrical analysis/investigation as well. In the extreme example of a cable that has conductive shorting, one can use a multimeter to find continuity between channels that should be completely separate. In the case of capacitive crosstalk, one could pass a signal through one channel and use an oscilloscope to see if the signal is being capacitively coupled to other channels.

A common issue in electrophysiology is neural signal that is not necessarily correlated in broadband, but is correlated in high frequency events, such as spiking activity. This can appear as highly synchronous spiking activity on many or all channels. Typically, this is due to a common high frequency but infrequent noise source, such as a feeder, capacitive touchscreen, or some other quick effect, like the subject experiencing an impact. For this reason, a common method of artifact rejection in neural recordings is the removal of spike events that are in synchrony on too many channels at once. In the cases that these can be ruled out, the next step is to investigate the spiking activity itself. Do the shape, magnitude, and duration of the spikes match an action potential? Do spike shapes match between channels or are they unique per channel? Are the timings perfectly correlated or just close?

Blackrock offers a crosstalk app within the Central software that is capable of measuring correlation of signal between channels. The method of that software is shared below:

The crosstalk function attempts to measure correlation on LFP and Spike filter ranges. To be more specific, the algorithm temporarily overrides user settings to collect 1 second of 30 kHz, unfiltered data. This data is then filtered two ways: 500-1000 Hz Band Pass and a 1750-2250 Hz Bandpass with a 2nd Order Butterworth. After filtering the data, correlation coefficient matrices are calculated for both ranges and the diagonals are set to zero. In Matrix 1, Channels are cross talking if >0.5 and in Matrix 2 >0.35.

The software will report the crosstalk value primarily by color, generally green and yellow are acceptable colors, because some amount of commonality is expected in neural signals, but orange or red colors may indicate crosstalk issues (or highly correlated noise, as mentioned at the beginning of this article).

While there are many situations that can appear as crosstalk, real crosstalk in electrophysiology is relatively uncommon. Generally, crosstalk can be approached methodically to find the source and correct it.