Hacker News Reader

Comment by growlix | Hacker News Reader

This seems to be a pretty significant step in miniaturization. It seems noteworthy that the animals perform more trials of the task with this implant as compared to an older, larger implant. However, the article is misleading: this device does not record EEG signals, which are recorded from the surface of the scalp and comprise the aggregate activity of millions of neurons across multiple cubic millimeters of cortex. It records LFPs (local field potentials), which are the aggregate activity of thousands to tens of thousands of neurons in a hundreds of µm to single mm volume around the electrode, and single neuron activity. EEGs are also recorded non-invasively, which could lead one to think that this implant isn't invasive.

View on HN

Gah, i totally missed that they were getting LFP and multi-unit only. I (foolishly) assumed they could get putative single unit, too, but I guess that's yet to come.

The grantparent says as much ("... and single neuron activity"). I guess, this means careful separation in case of sparse signals, though I'm nowhere near qualified to judge.

My layman take: 9 Khz bandwidth in principle means resolution of 9000 evenly spaced peaks of different magnitude. Perhaps those correlate to different distances between each neuron and the sensor, especially if 16 channels allow for spatial resolution, and if the differences might actually be too small to resolve naively, I wouldn't doubt if I were told statistical tricks existed to filter single signals.

Overall, major features are interesting enough, I guess.

The paper very carefully does not say single units were recorded: https://www.nature.com/articles/s41598-017-08078-8 unless i'm totally missing it.

Spike sorting was done with this project, which i have used and found to be quite good: https://github.com/cortex-lab/KiloSort

19 kHz sampling rate gives you a maximum temporal precision of ~50 ms, which is pretty good.

50 us, you mean.

They claim a nicely low input referred noise (~4 uV), so 19 kHz should be great for spike detection, though as you note upthread it's a little low for spike sorting: it's nice to have at least 30 kHz.

I wonder how much weight savings they got over combining a wireless chip and a bare-die Intan [1].

1. http://intantech.com/products_RHD2000.html

yes, 50us.

I wonder if it's more of a form factor thing than a weight thing.

I wonder how it holds up in multi-animal cages, too.

Reply on news.ycombinator.com