how to separate similar clusters more precisely

[happyqiu]

Hi ferchaure,

Thanks for your wonderful automatic sorting tool. But I am not sure how I can separate similar clusters, maybe just the difference in the amplitude.
units.pdf
As I attached, the offline sorter figures out two separate units for channel 6 and channel 21, however, we only got one cluster for each channel, which may combine two units. We tried to change the parameters like par.min_clus, par.template_sdnum, par.max_spk, but didn't make much difference. Do you have any suggestions how we may be able to separate these?

Thanks!

[ferchaure]

Sorry, I'm not used to seeing waveforms in that GUI, but it looks like the second unit in channel 6 is just detected noise with a bit of spikes from the yellow neuron. For channel 21 I'm not sure if the green cluster is a neuron o just filtered noise aligned to its peak. You could get "similar" results if you use a smaller par.stdmin on waveclus. Usually, the spikes have good amplitude or some particular shape that differentiate them from filtered and aligned noise

[ferchaure]

Just to show you, remember that just normal noise will generate fake "spikes" like the ones here.