Marco Reps - 2021-02-24
Get your high quality PCBs from https://www.pcbway.com/ Win Test Gear at Keysight University Live http://bit.ly/MR_KULive Miniware Hot Plate https://www.banggood.com/custlink/K3KyZe7los Support me on https://www.patreon.com/marcoreps Cool new Brymen DMM https://www.welectron.com/Brymen-BM789-Multimeter_1 5% coupon: marcoreps5 HPM7177 Project https://ohwr.org/project/opt-adc-10k-32b-1cha/wikis/home CERN YouTube Channel https://www.youtube.com/user/CERNTV The Worlds best resistors http://www.vishaypg.com/foil-resistors/ Desktop Enclosure https://www.reichelt.de/de/de/systemgehaeuse-geh-sg-1-19-p50439.html My Multiinstrumentalist datalogging software https://github.com/marcoreps/multiinstrumentalist/ xDevs INL article https://xdevs.com/article/inlperf/ 00:00 Intro 02:25 Circuit Overview 05:18 Assembly 10:19 Soldering 13:51 THT parts 15:47 Mechanical stuff 25:10 Software 31:02 Result
What a funny, informative and interesting video! Your level of skill and diverse talent is just so inspiring! Really love your channel, please keep up the awesome work
I didn't understand 80% of what happened in this video but it was one of the most satisfying things i've ever watched.
idk what ppm is but i wont google it
@1marcelfilms
I think it’s “Peppa Pig’s Mother”, but I won’t swear to it.
Ditto :P
It measures electricity. Volts, specifically. Does it very precisely. You could use it for any number of very sensitive scientific experiments (such as trying to discover new, unimaginably small particles hitting detectors at CERN).
I think the rest is so many design considerations that it goes from engineering to art.
I understand so little of what you are talking about but can totally appreciate taking on a project of this size! Very well done!
Glad you can profit from the open sourced equipment designed for fundamental research. Thats one reason why research is important, in my opinion. You went full out with all the components. No cheap session at all. Very nicely done Mr. Reps.
You got an LOL out of me with the line: “It can’t be that hard, can it?”
Beautiful piece of equipment. Throughly enjoyed the build.
This is absolutely hands down the most impressive video you have ever made! Wow! Mr carson's lab quality process and signal path level circuitry. Amazing. I'm glued to my phone haha
I have not seen DIY vapor phase reflow soldering before! The potential lethal danger adds that special level of excitement!
For the front panel, look into Alodine or Iridite chromate conversion coating. If you know what alloy the front panels are made of, you may be able to find an anodizing house that can treat them for you. They both can be specified as electrically conductive, and would be suitable for your ground contact, while still protecting the aluminum. If youre feeling extra spendy, you can have done with a mask so the main portion of the panels are anodized, while a contact ring on the back side is a conductive conversion coating. Or get them nickel plated. Polished first, if youre feeling a bit garish, brushed or sanded if youre not.
That feeling when you take 30 minutes of your day to bask in the raw magnificence of a beautiful machine coming together. Wonderful work Marco and nothing but love to CERN for open sourcing this. Magnificent work by everyone. Thank you for sharing.
I didnt realise that CERN released their designs but making them available for self production is truly wonderful. Your manufacture and walkthrough is similarly wonderful. Thank you. I suppose next thing is to bin the FPGA and replace it with a re-programed Tesla AI chip, to get another 3 digits of resolution? 1pV you would need to dig a big underground well screened room to be able to use it.
Hey, great work! I would like to suggest an alternative to polynomial interpolation, and that is simply linear piece-wise interpolation. Numpy is using least squares linear regression, which can easily overfit these polynomials of higher degree, introducing the error wiggles your data seems to be showing. Also, with piecewise interpolation, the interpolation step is only a lookup table, and more precision consumes only more memory, while keeping the computation amount the same.
The dry delivery had me smiling, but the ring had me laughing out loud! This was phenomenal work for someone working at home, and at a level of quality I'll never see.
Since you get that much Samples with some Noise, you might look in to "ADC Oversampling" instead (or beside) using a simple averaging.
I'm no expert, but that way you can boost the ADC Bit count and smooth it out at the same time
Impressive video! Have you considered using Allan Variance/Deviation to analyze your long strings of readings. This technique was developed to analyze stability of oscillators for atomic clocks, but also works for long strings of other numeric data. Besides showing how many points to average for minimum noise, it can also show long term correlations due things like line frequency and day-night cycles.
I love LEMO connectors. Incredibly expensive, but widely used in medical instruments for very good reason. The quality is unmatched, and you can get ridiculously small connectors with amazing voltage and current specs.
yep. I've worked at companies that use LEMO connectors for as much as possible, and they are amazing.
Yes they feel great
When you mentioned thin film resistors in the beginning i started wondering: In precision applications like this, are those better for minimising Johnson-Nyqvist noise in measurements over carbon or metal film resistors?
I remember the talk of a doctor in physics at my university, he says: in CERN they have some of the finnest electronics working on the experiments, I never understand what exactly he refers with finnest up to now with this video, thanks for sharing your effort and for the pleasant build.
This is all exceptionaly impressive. Lots of new techniques, cool equipment and strategies. Bonus points for torx fasteners. Thank you for showing! You should really start a company around your skillset if you haven't already. You would probably do very well, and over time that position would allow you to make even stronger content and establish better relationships with various larger companies. Low and medium volume production is quite different from single projects like this, but I think you have the right kind of personality to be able to optimize it to be profitable while maintaining high quality. This channel would give you the necessary marketing. Kind of what Doug DeMuro did with his "cars and bids" site/business. Regardless, thanks again and good luck!
Love your assembly videos and the humor!
Interesting reflow method. I have been using Chipquik low temp Bi57Sn42Ag1 paste and reflowing in a toaster oven with the peak reaching only 180C.
Much less stress on the components and very easy rework.
What do you use for a vacuum chamber on the two part silicone?
If you’re looking to improve that linearity, polynomial regression is probably a poor choice for the calibration curve due to that “wobbliness” you talked about with your tenth degree function. You can avoid that, along with reducing computation cost and error by using cubic splines. In layman’s terms they’re glueing together third degree polynomials to form a smooth continuous curve without any ugly wobbles.
Hope that helps! 🙂
The point of the 10th degree polynomial is that he's trying to compensate for as much of the repeatable non-linearity as possible, so that he can determine an accurate figure for the non-linearity of subsequent measurements.
@AV3NG3R00 That’s correct, he’s using a polynomial regression to compensate for known error between the ADC and calibrator. I’m suggesting that instead of using a polynomial regression, he should try using cubic splines as they are cheaper to compute and don’t have the inherent instability of polynomials. It will still accomplish the same goal (mapping a continuous mathematical function to discrete error data) but it will do it more efficiently and without unwanted wobbles in the function.
@Zachary I agree - 10 degree polynomials are almost sure to overfit. i like your method better with little chance of overfitting.
Ohm law eat your heart out. fm.
Many ADC INL characteristic plots have discrete jumps at 1/16, 1/32, etc divisions of full scale (with linear deviation between).
"we" found that using evenly spaced 128 point linear interpolation reduced the error from DNL/INL with very little computational overhead
Use orthogonal polynomials, e.g. Legendre or other for the interpolating polynomial. The number of zero crossings of the function gives you the number of terms you need and will exhibit great numerical stability.
The detail & that nice PCB design is just OUT TO SPACE !! Nice work !!
It's incredibile the amount of cool stuff condensated in Just one video. I'm Always exited to see what have been going on behind the scenes for the passed months. Question: i can still hope for a video on the osmu?
I think this is my 6th time coming back to this video and it still looks like wizardry, brilliant work Marco
WOW, that DIY vapor soldering technique was brilliant, no need for an IR oven...
I can’t believe your incredible knowledge or how you achieve what you do, but hats off for a brilliant journey into the depths of physics and electronics so far removed from the average enthusiasts mental capacity.
Well done. Brilliant!
you can use the fpga like a DSP and program your signal processing function in hardware. That will get you plenty of speed to do the processing as long as you have enough cells.
Such perfection, components, voice and manufactoring made my life
Great video as always! I would love to see more videos about FPGAs.
Amazing! Great job!
And thanks for the hint with Grafana. As i recently build a sensor data acquisition systems that logs over network into a RAM buffered SQLite database, I will give it a try.
Hopefully it can take advantage of selecting data from the tables instead of bulk loading every ASCII line tokenized via string-to-float into memory. I find those CSV files annoyingly slow to handle ;)
Awesome project. You can try another method of cheating, and use scipy UnivariateSpline with lets say k=3, it should produce more interesting and smoother and more numerically stable and rich fit of data. Using polynomial for fitting, is not recommended with high degrees of polynomials, they get really bad and unstable.
Marco, you might want to try piecewise polynomial functions. High order polynomials have well known ringing effects making them not converge well -- in reality a higher order polynomial on well behaved functions will eventually perform worse than a low degree one. The fix is to use piecewise low order polynomials, of degree 2 or 3. The data itself could be smoothed either using Bayesian methods or a simple nearest neighbor average; from this curve you can then optimize the parameters of the piecewise approximator using a method like gradient descent or something fancier if you sense it's not converging to a global optimum. The third or second order functions also make it much less computationally demanding as well. Feel free to ask for suggestions :) This is fantastic! --GNR
I just found this video, it's really awesome congrats! I'm designing a lot of analog stuff I'm interested in low noise design. Keep up the good work.
I'm actually glad you only post a video every couple of months or so, considering how much psychic damage I take from each one. It takes nearly a month to completely process all the new things I learn each time.
Yay convective tombstones!! Your Rube Goldberg reflow technique is on point. Love the content, please keep going.
I personally find that solder paste in a tube is worth the slight loss every time you use it compared to the tubs of the stuff. I've noticed that the continuous exposure to air dries out the flux in tubs of paste, and even when you rejuvenate it with a bit more liquid flux, it's still never "factory new." I always get near flawless reflows with minimal bridging when using paste out of a syringe. In the end, I always use up the entire tube of paste while the tubs have to be thrown out prematurely.
Great project Marco. A bit off topic, but your CNC work is spectacular. All that micron chasing earlier seems to have paid off.
Absolutely superb! Really looking forward to that 10V source that may or may not be the subject of the next video...
In the software section I could not help but think about machine learning practices with the polynomial fit that was implemented. I would recommend reading about overfitting and underfitting. But also the fixes for that: testing and validation data sets and cross-validation. This would give some insight in whether the numpy fit was actually any good in explaining the variance in the data, without being too sensitive to this data.
You had me at mechanical stress from incorrect soldering. I love it you talk electro-mechanical stress related hysterisis. I have a box of Vishay bulk metal foil resistors in T03 packages that I use in audio amps and loud speakers. Tempco 25 ppm 0.5 percent tolerance.
Amazing what you can do at "home." I learned a lot. Before I would expect to need a whole factory to make this stuff. Thanks.
How much was the overall cost?
Holy shit it’s like Christmas!
I find you in every deep dark corner of the YouTube comments sections…metrology and 8.5 digit videos.
So I know you’ve been around. Glad to see you finally came back to your own channel!
Now get back to the lasers… Pronto! You can’t just buy a bunch of nice fiber lasers like that and then let them sit around 😂
It’s your fault I didn’t get a newer truck this year. But I do have a few nice fiber lasers (bought for pennies on the dollar on eBay. But still expensive). And soon I will have my 8.5 digits of glory recalibrated and displayed in all of her beauty. A year ago I was completely happy with my trio of 34401A meters. What have you done to us!?
You and Mr. Carlson are the only ones I’ve ever considered and I’m glad to donate/join patreon
It would be interesting to see a follow up a CERN video of the usage for this instrument.
Where is such precision required in CERN - is it for positioning purposes or
for measuring sample values from an even more precise sensor ?
Thank you for sharing your knowledge.
I've watched this so many times, just because that chassis is gorgeous. I have a terrible feeling I'm going to find a horribly expensive project just so I can own that.
An astoundingly ambitious project and build. Very impressive.
On a personal preference note, I don't usually like high priced builds without a specific purpose. This was somehow an exception and very enjoyable - perhaps because promoting open standards is dear to my heart. So thank you for this, please carry on.
I'm excited to see the results from the DIY vapor phase reflow setup 🚀
I kept thinking of sending someone to get a bucket of condensed steam.
Wow I love your enthusiasm! Having the balls to invest 2k BOM into a voltmeter is some heavy engagement. And your humor is just unique!
Whew! You're back. What a project. I was getting worried. Should have known you fell for some new gear. Great video as always. I still can't understand 5% of what you are saying. Have to admit the vapor soldering thing blew me away. All the technologies i don't even know exist. But watching your videos, it is one place where I am dumb and happy. All is right with the world when Marco is wrangling pixies.
xDevs_com - 2021-02-24
Need moar ppms! Nice work and thumbs up for CERN folks for pushing DIY metrology forwards!
daweim0 - 2021-02-25
or is it fewer ppms?
Sheila Walker - 2021-03-19
How to build a muon solenoid from common household appliances...
Александр Бегунов - 2022-06-09
Cancel all ppms at the legislative level and to penalize for ppms.