Question `commen_reference` for multi-shank probes:

I had a brief conversation with the CambridgeNeurotech about commen_reference today. It seems that since I am using a multi-shank probe, it would be great to take channels from one shank as a reference for channels from the other shank, and the other way round (instead of calculating the reference based on all channels). Has anyone tried this method before? In addition, I understand that current version of commen_reference can take dictionary as an input, which helps process data with multi-shank probes. In this case, how does the function work when those reference methods: "global" | "single" | "local" are called? I was wondering if the easiest way to take this concern into account when taking the reference would be giving local_radius and not splitting channels into dictionary, so that for every channels there would be an unique combination of channels set for calculating common reference (depending on the radius). Does it make sense?

Start by inspecting the existing commen_reference implementation and how its global, single, and local methods handle dictionary inputs. Clarify the intended multi-shank behavior and whether local_radius should replace channel grouping; the issue needs a decided scope and explicit tests before completion can be assessed.