After yesterday’s win, you would think things would be smooth sailing, but alas that isn’t in the cards! I mean yesterday’s good news still applies thankfully, but wow did I wake up to a flurry of emails and now a mountain of work that all needs to be addressed all at the same time. Wait, is there an echo in here? I feel like we’ve done this before…
Since lists seem popular these days (LOLsob), let’s just go through my to-do list. Obviously the response to last paper is coming up like a bad horror movie… seven days. Most of that is done, but hospital-PI gave me some revisions to my fourteen page response letter (14 pages of just text, for reference “last paper” is roughly 12 pages? and contains several figures). Then I need to get this new data processed, which new data? All of it. Oh I’m also teaching another course (sometime soon I think…?). Then I have my latest paper that I’m working on, we’re meeting Monday to talk about that. Oh and I need to do some stuff for a different project I help out with at school.
If that isn’t enough to make a person cry, I don’t know what is!
So the paper, “last paper” that is, has a bunch of edits I need to make to address hospital-PI’s comments. Most of them are mindless so it’s just a time suck and I need to make sure to do it correctly and be very certain I modify the actual paper the same way my response letter was modified (that is the difficult part FYI). It makes it slightly less mindless and makes a bunch of extra work for me, but whatever at least I’ll be done with the paper… hopefully.
Then I got a few texts from hospital-PI this morning. We’re friends — as much as coworkers can be, I guess I like to think we’re friends anyway! haha — so texts over the weekend aren’t uncommon. Sometimes he just checks in, other times he wants to make sure I know something needs to get done ASAP. I have a super flexible work schedule, so I don’t mind this arrangement (it’s good for my physical and mental health frankly). He mentioned that he has a presentation in a week (why is everything in a week?!) and wants to show off some of the stuff we just got from “big idea,” I’m assuming this would involve a funding pitch or possible funding. We don’t typically do this, but big idea is… well big, so we’re all excited.
The problem is that this is a whole new world of data that I’m working with. I wish I could be less vague, but just trust me on this. So I’m working through trying to figure out the best way to process the data. I wish it were mindless or there were best practices, but there really aren’t. Instead I’m left to make up my own “best practices” and hope for the best. No pun intended, I’m not allowed to pun when I’m in this type of mood, I’m being…. PUNished (okay one pun never hurt anyone). But seriously, I’m working with this data and it’s doing some weird things that data we process wouldn’t normally do and that’s making me nervous, really, really, really, nervous.
So he wants me to have all the data processed and fancy figures made using this data… data that we don’t know we can even really use at the moment mind you. Basically that’s all I’ve done today is try to determine the best way of dealing with the dataset we collected. I think I have something resembling data that make sense, somewhat? But there’s a lot of incredibly high frequency (> 500 Hz) signal in there that is confusing me. Normally in EEG we ignore stuff > 100 Hz at the max (most of the time we work at >50 Hz), but there’s this large increase in power spectral density that makes no sense to me.
I’m assuming it’s artifactual or some sort of spectral leakage effect, but it makes me nervous. I’ll have to break down what all those words mean some other day (sorry). Actually some of my old posts talk about it (probably poorly since I was just learning about it)… spectral leakage (here) oh and power spectral density (here). Man those are some old posts… memories! I should write new versions of those actually, I’ve learned a lot since then, but that’s beside the point. Moving on!
I also have two new meetings on Monday, one regarding my latest paper that I’m working on, one that I haven’t really named since I’m having name block apparently so I just keep referring to it as the latest paper, which will get confusing once I start on a new “latest paper,” so name it when I make a post on that paper specifically I guess. That paper has been on hold for a bit since I have the last paper response due soon, plus all this exciting new data to work with. So I should probably get on with it and DO something with this paper before our meeting Monday…. morning it looks like. Oh good so I won’t even have time to work on it the day we meet, fun.
I also have to make time this weekend to meet with some of the school labmates to discuss the next class I’m teaching. It’s on independent component analysis or ICA for short. I maybe teaching that next, maybe. School-PI surprised me by saying I would be teaching a second MATLAB course for those who missed the first so I need to make sure I heard that correctly. He also wants me to teach a figure making class since I’ve dedicated the past year to making cool figures and now that makes me somehow a teacher. Don’t get good at anything people, be mediocre! Learn from my mistakes, wait… scratch that, don’t learn anything!
If that wasn’t enough (how long can one post be?!) I also have a solid modeling project that I need to get done by Wednesday. So if it doesn’t get done this weekend, it won’t get done at all. I would’ve worked on it now instead of writing this post, but here I am. Ironically it’s not because I’m putting off the work (for once) since I’m already super anxious about getting all this stuff done, I would be doing it now, but I can’t. We share a certain number of solidworks licenses (solid modeling software, super expensive) and all the licenses are currently being used so someone either forgot to release the license when they finished or they are using it right now. Making the work I need to do impossible, joy!
Since I was so annoyed by that little bump in the road I figured I would take a breather and write this post. Now that we’ve arrived at the end of my list (as it stands now anyway, I’m sure more stuff will magically appear as the weekend progresses), I should probably get back to doing some work with the big idea datasets. Speaking of which, I am 70% confident I’ve got the pre-processing pipeline down, but the power spectral density of the data makes me nervous that something is wrong.
Fine, quick aside…
Power spectral density (PSD) of biological data tends to follow a 1/f curve where f is frequency so as frequency increases the PSD at that frequency decreases. That looks something like this…
Stolen from one of my very old posts! that big bump at 60 hz? That’s the stupid line noise I always complain about! If you imagined a smooth curve (just use your imagination folks) it would be basically 1/f. I’m not even sure what the 40 and 80 Hz bumps are, I would guess some sort of sensor movement or other noise since 80 is a harmonic of 40. I forget what this data even was from haha. As I was saying (so much for quick aside) the data should look like 1/f, roughly if you squint hard enough. That data follows the 1/f expectation enough for me to be like, cool this is okay data that needs a good cleaning. I’ll do an updated PSD post one of these days, but that’s a good enough rundown for now.
My big idea data on the other hand does not follow the same shape. It does, to a certain degree, but not enough that I feel comfortable about it. At the moment I THINK I may have over cleaned my data (removed actual data and not just the noise), since the dataset was INCREDIBLY NOISY!!! So much noise, imagine a plot like the above, but the line noise (60 hz) is like 4x larger. The raw data look just like 60 hz waves frankly, you can’t see any other signal because it’s so overpowered. But there is underlying data there, so we have something, but who the hell knows what it means? I guess I’ll figure it out once I start trying to compare conditions, but for now I’m a little worried that I can’t get a good pipeline in place before the new due date.
So yeah hospital-PI wants me to work miracles because I do work miracles on a semi-regular basis, but this is a bit much and I don’t think he realizes is because it’s outside his expertise. We’ll have a long talk about it Monday for sure, but for now I’m trying to at least have something to show, even if it isn’t the perfect or best solution I could get with the data. I wish I could blame hospital-PI, but I’m mostly doing this for me, I want (need) to see if we have something usable.
Now seriously, this post is getting far too long and I should get back to work, so off I go.
And here I thought I was having a busy day because I have a bunch of housecleaning and other random things to do. Sounds like you kinda got piled on. Good luck digging your way out.
So the oddity in the data – you make spectral leakage and artifact sound like the good options. If it wasn’t one of those, is there something specific and bad it could be (like a consequence of the excess de-noising), or does it just worry you because there’s no real explanation for it?
LikeLiked by 1 person
June 4, 2022 at 6:01 pm
Haha thanks! Turns out I’m mostly caught up (ish) so not too bad I guess. The paper edits went faster than I anticipated, which was helpful.
Spectral leakage or some other artifact from the processing would be the better option because that’s something I can control (I can go back to the original data and refilter it or adjust the way I’m filtering it).
If that’s not the problem then it makes me a little nervous because then I can’t correct/fix the issue. Then again, maybe it’s not an issue, this is all really new so I’m just comparing to other known quantities that may not be the same as the stuff I’m looking at. I did go back and refilter all my data again… and again… and again, today though and I’m a little happier with the outcome this time around.
Really I just need to just get to the part where I separate out the data and see if there are differences between the test conditions. If there is then I won’t stress about it too much, if there isn’t, well then there’s a non-zero chance that it’s because of the way I processed the data.
LikeLiked by 1 person
June 5, 2022 at 5:53 pm
Breathe, you’re the SME. Enthusiasm and showing you’re identifying issues and resolving them all is enough. I know you’re not a doctor yet, but imagine sitting in a room with someone in your position pitching another Big Idea. Show what you want to see.
From a technical standpoint, I want to point out this isn’t my field (this is when you stop listening. 😉 ) but good lines of inquiry or research.
40 MHz is 8m amateur radio. It propagates in the atmosphere incredibly well. 82-87 is channel 6 on television, if that’s active in your area.
Both of those are common clock rates for ESP and PIC32 microprocessors, as well as other, similar systems. I mention this because when a clock cycle clicks over, a lot of states change and circuits do their dance. If they happen to be poorly shielded, or those are in your sensing equipment, they can cause a problem.
This is all just spitballing, but if any of these ring a bell, it’s a good place to start.
If you think it’s a problem that must be resolved (I tend to subscribe to the school of 100% duty cycle, no fatigue limits, “what’s wrong with you” overengineering, since I don’t have to pay for it. 😉 ) I encourage you to look up TEMPEST standards, as NATO shielding standards for emissions control likely works both ways, and people love international standards.
I also encourage you to get a cheap SDR radio and spend some time studying your workspace. I expect you well understand how to triangulate sources and the like, so there are plenty of ways to determine the general source of interference. For tighter checks, you can use what I call a delta detector, where you run two parallel antennas and subtract one from the other. Signals at a great distance are eliminated, since they’re so close in power, but nearby sources are vastly different, due to the cubic dropoff of signal strength. I banged together a wifi detector once for proof of concept, but I build such imprecise equipment, I don’t know if it works as intended or just gives me false positives.
Anyway, if you have the spare cash, RTL-SDR works a treat.
It’s always this side stuff that sneaks up on you, isn’t it?
June 4, 2022 at 7:26 pm
Thanks for the advice! A radio scanner may not be a bad idea. It may just point out the obvious though, which is the hospital is noisy. I’m not sure how much we can correct for it, but that’s the hard part I guess. Looking over the shielding standards is helpful, I’ve only glanced, but I think you’re onto something with just trying to shield the equipment better. I mean people are hard to shield (since we put the equipment on them), but we may actually be able to at least partially get rid of some of the noise from sources inside the room.
June 5, 2022 at 5:48 pm
I wouldn’t put too many resources into anything before testing. Low frequencies tend to need longer antennas, so it’s possible your sensors and cables won’t pick up those low frequencies, while cables of at least a foot could pick them up.
Sorry to be all over the place with information, it’s like my brain has a little turnstile and after X number of functional ideas in a day, it reaches capacity and we’re done.
RF engineering is specialized, you’d probably need a specialist to get any major benefit, but if you’re in the middle of building, then filtering caps, RC tanks, and mass/cage shielding is a nice first order solution. Optoisolation is also fun, but I don’t even own an oscilloscope, so this is all theoretical.
LikeLiked by 1 person
June 5, 2022 at 6:15 pm
Oh you know getting a specialist involved may not be such a bad idea to be honest! At the very least it would give me an idea where the noise is coming from and even if there’s nothing that can be done about it, knowing the source would be helpful for documentation purposes.
LikeLiked by 1 person
June 6, 2022 at 6:13 pm