Need a Professional to take a quick look at design

[orchid rose]

you probably updated your libs since you started the project

you can update them all as one action

[exotic kernel]

ok

thank you

[finite pebble]

No

Ah Christopher beat me to it

Yeah you can’t assume anything about the internal connections of an IC

[orchid rose]

remember all that business about return paths? this is the return path

or one of them at least

[exotic kernel]

thats exactly what i was thinking. Its a return for the rf.

[orchid rose]

that seems reasonable enough. so then it's probably wirebonded to the rf section of the IC and if it's not connected those return currents will have to find their way out of a different ground pin which the bits of the IC in between will definitely not appreciate

[exotic kernel]

god i hope this thing works

[broken musk]

$250 is cheap for pcba tbr

[dry grove]

it is but for people who spend 2/3s of their income on housing it can be months of saving

so i get the hesitation

[finite pebble]

yeah out-of-pocket HW dev is rough

[exotic kernel]

yea its a lot of money for me personally. Gotta do what I gotta do to learn though.

is it ok to put silkscreen txt over traces?

[dry grove]

for most traces yes

[exotic kernel]

ok ill try to avoid them as much as possible and stay away from pads

[broken musk]

this isnt really the kinda project id use to learn but..

[exotic kernel]

Its not too bad. Might as well get a bunch of stuff into 1 project and learn it all at one time

I learned so much its ridiculous

[exotic kernel]

this is the same cap. I tried updating footprint, I tried updated PCB from schematic... I dont know why this footprint is like that. I dont know how to get it to update to whats in the symbol

[orchid rose]

what's wrong with it?

[exotic kernel]

its not the same footprint. i figured it out i had to check of update footprints from symbol in the option when updating pcb from schematic

god i mentioned that I asked chatgpt on reddit and now im getting eatin alive in the comments

[orchid rose]

you can also right click on the footprint and select update footprint from library

[exotic kernel]

yea but im really trying to update from symbol, either way im gonna have to double check my entire pcb now

Becuase this was my first pcb i hapahazardly added footprints without making sure they were absolutely the right size according to component datasheets

[astral knoll]

Impressive first board non the less. Yeah triple check footprints are correct.

[exotic kernel]

thank you. I literally almost sent it out to get manufactured but the little voice in my head was saying, "no you better get it reviewed professionally first" and I had it reviewed and he pointed out the strange footprint sizes.

[astral knoll]

The reason I haven't sent for my board to be made. The anxiety I'll screw up.

[exotic kernel]

the anxiety saves us

[astral knoll]

Sometimes. I hate my social anxiety

[exotic kernel]

The amygdala is like a sponge that gets filled up and needs to be squeezed out by facing fears. We evolved in constant danger. Facing fears in a safe way keeps the brain healthy

[finite pebble]

We all still feel that way

[astral knoll]

Some things never change. Noted 📝

[finite pebble]

An analogy I really like from cycling:

riding doesn’t get easier, you just go faster

[exotic kernel]

Put together my first emg on a breadboard today

Jesus it was a pain. There were no reference circuits available for my specific case. Took me 3 hours lol i was getting so pissed but i was so happy when I got it to work. Electrode/skin connection integrity is no joke.

[exotic kernel]

Have any of you guys ever worked on eeg, emg, ecg? Or anything related? Im trying to piece together the gold standard circuit for bio signal acquisition. So far I got the instrumentation amp and some output dc decoupling, i dont know whats next.

[stable leaf]

Have you explored app notes from sensor ic companies. Usual suspects like Analog Devices, etc.

[exotic kernel]

Yea but its hard to find what exactly i need

Like they have emg circuits but i have other experts telling me to add certain things that arent in the notes

Really im sifting through multiple sources of info and finding common reccomendations

And filtering based on the credibility of the sources

You guys have proven to be pretty smart so i would consider this chat of high value

[dry grove]

you really need to understand both the biological and electrical principles to be successful. I don't know jack about the biomedical engineering side

like if you're getting advice they should be able to say why, or based on which assumptions

[exotic kernel]

Yea sometimes people give me advice that even I know is wrong lol.

[exotic kernel]

It is a nightmare trying to find the right adc

[stable leaf]

Have you gotten to the point where you have a clearer set of requirements, key issues?

Fyi... I stopped following this thread a while back because there was too much thrashing while going up the learning curve and too many audio posts.

[exotic kernel]

Kind of. I understand all the op amp specs now, the Adc is a different story. I understand the noise, input range, bits, sampling rate and all that i just dont understand what minimum bits would get the job done. I know i need minimum 2.5ksps, but bits? I dont know if 10 or 12 would work i have no idea.

I would imagine 12 bits would get the job done with a 5vpp signal but i dont have a great understanding of thr benefits in emg of higher bit rates. I know its more precise but is it truely necessary

My average design session is 50% studying stuff and 50% me standing in my hallway alone drinking a capree sun repeating over and over “my name jeff”

[stable leaf]

Do you have a signal to noise ratio for the signal from your sensor? Adding more ADC bits has no benefit if those bits are just measuring the noise.

[dry grove]

flicker noise performance is going to determine your possible sensing performance

[orchid rose]

do you have an example of one of the signals you're trying to measure? that would likely be very helpful, you could put it into spice or something and see what your frontend actually does to it

[exotic kernel]

My end signal will be about 25mv rms dense noise and the emg will peak to around 2.5v to - 2.5v so deffinantly enough snr for what i need. I got a pretty good amplification and filtering cascade

Honestly i got the signal amplified so good a garabge adc would probably get the job done

I wanted a 5 to -5v analog input range but apparently they are commonly 2.5 to -2.5 range. Ill have to cut the signal level in half with a small value r resistor divider

[stable leaf]

If I understand you correctly that's a signal-to-noise ratio of 100 or 40dB. A 12-bit ADC has a theoretical 72 dB SNR. That would seem to be sufficient. Flicker noise, 50/60 Hz EMI, etc. probably need careful consideration in your application which I know very little about.

A resistor divider to scale the input signal before input to the ADC is probably not a good idea unless you follow it with an opamp buffer to yield a low source impedance. I'd have to look at your schematics to comment further. There are many other people here that are way more experienced than me for sensor analog front ends. But I might recognize some mistakes I have learned about the hard way.

RE: sample rate... what are your signal bandwidth requirements?

[dry grove]

Adding a divider is pretty risky for other reasons too

[exotic kernel]

Im aware of the sample hold 20pf capacitor in the input of the adc and its relationship to external resistance and capacitance. I think i can get away with a 10ohm R divider. My last op amp pga outputs 50ma. Otherwise im hitting the adc with twice the peak to peak. Maybe ill just gain less but the problem is stray peaks and god forbid an electrode comes loose, everything will rail basically lol

Ill find a decent adc and if it has any problems ill just point at it and say “bad” until it works.

[dry grove]

dividers will load your source though

Increase resistance, you get more noise. Decrease resistance, the source is loaded

So you need to know the behaviour of the probes and how the signal moves through the body

[exotic kernel]

It will load the last amp in the chain. It wont affect the electrodes input or anything with the initial amp. It will load my battery more. It might add noise but idk. I dont really know a better way to have a hard stop at +- 2.5v without adding another ldo and amp

[dry grove]

Your signal is smaller than 2.5 v you dont need a divider, just an amplifier

[stable leaf]

Is there a schematic of the input signal path? Can you post it?

[exotic kernel]

I left is electrodes the right end load is the adc

I will be using only op378 instead of opa333

Meaning i will still have the highpass filter but it will be opa378

I went through half of ti adc training course yesterday

Looks like im leaning towards something like this

https://www.ti.com/lit/ds/symlink/ads8681.pdf?ts=1770325497090&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FADS8681

[compact herald]

How about denoises & stablize current on-load ? Have you fab/test the board yet ?

[exotic kernel]

The adc input stabilization is what im learning about today, no theres no board yet just breadboard

[stable leaf]

Can you say what you are using for electrodes? Datasheet?

Are planning to notch filter 50/60 Hz EMI?

Think about a LPF for anti-aliasing.

[exotic kernel]

Im gonna try the 3M red dots.

https://ww1.microchip.com/downloads/aemDocuments/documents/APID/ProductDocuments/DataSheets/21034F.pdf

finally found my adc

now im lookin at adc driving op amps

and thanks for the low pass recommendation i completely lost track of that while on this unreasoanbly difficult search for the right adc

they dont make a lot of good throughhole adc's for breadboarding.

i will do filtering in digital

I was told by 2 seperate engineers on reddit that the 60hz filtering and full scale rectification are best done in digital these days

the high and low pass stuff i will do in analogue with the amps

the opa378's as 2nd order filters

Kinda liking how this passband is looking. what do you guys think

[stable leaf]

Yeah, that makes sense to me.

[stable leaf]

I'll have to look later.

[exotic kernel]

https://www.ti.com/lit/ds/symlink/opa228.pdf?ts=1770407035768&ref_url=https%253A%252F%252Fwww.ti.com%252Fproduct%252FOPA228

That will be the sar driver amp

@dry grove what do you think about my progress so far in EE. I started 8 months ago with 0 experience. I dont have anyone I know to use as a comparison so I have no clue if im progressing at a reasonable pace or if I should be more advanced by now.

[orchid rose]

why this adc?

[exotic kernel]

its a long story lol

its the simpliest with what i need, has a throughole version, enough bits, enough speed,

cheap

[paper harbor]

Makes sense

[exotic kernel]

grok, gpt, or gemini all on expert mode could not find a good throughhole adc

i found that one myself through some serious digging

also had to find the adc driving amp myself

man ai has hit a serious ceiling. I havent seen any significant improvement in a while

except for video generation, they have gotten super realistic. but everyone hates them lol

[orchid rose]

have you considered that there is probably a reason that there are very few good adcs in dip packages

if 12 bit 100ksps is all you need why not use a microcontroller internal adc

[exotic kernel]

i was thinking of that but I kinda want to get the experience

[orchid rose]

i would argue the experience of working with microcontroller peripherals is much more useful than being able to talk to some spi device

[exotic kernel]

Yea you might be right

Now that i think of it i already did all the external adc calculations and studying

Anyways

I kinda get it at this point

Might as well get a mcu

Then i can just go straight to the rf

[stable leaf]

If you need to do simultaneous sampling across many channels then a micro isn't going to be a solution.

But that gets back to requirements.

[dry grove]

the bandwidth of the signal is low so you can sample from many channels, multiplied by how many ADCs are in the micro

yeah youll have some slight phase differences but you'll still be able to perfectly recreate all the signals

[stable leaf]

@exotic kernel Have you tried to create a Spice model for the electrodes? There must be some examples out in the wild.

[exotic kernel]

yea i did

i made the imedance and mains hum and all that

Its only 1 channel

[stable leaf]

Double check SNR... I may have goofed that and not squared terms. You may need more ADC resolution (bits).

[dry grove]

40 dB is right

[exotic kernel]

Anything remotely near 40db will work

[exotic kernel]

Today is all about mcu for me. Gonna be doing a lot of studying

[exotic kernel]

Yea looks like a mcu adc combo is not gonna work. First of all they have very low input voltage 3.6V-4V. Im already cutting my signal in half for 5V. Then on top of that, the ones with good adc’s dont have ble. Then on top of that, theres none with pdip package.

so ill get the standalone adc and input into my esp32 dev board and when i need to make the pcb ill just do a esp32 as my mcu/rf

Or a simpler lower power mcu i dont know yet

All i know is im getting somewhere, i dont know where, but somewhere

[stable leaf]

Umm... Why do you want through hole components? Are you making a PCB or thinking that you'll use a 100mil solderless breadboard with wires?

If you are making a PCB... Why not smt?

[exotic kernel]

I need to test things first and understand more about how to not mess it up. Although it might not appear to be sometimes but i am a beginner. Like a beginner beginner. I barely am putting this stuff together

I dont want to go striaght to pcb i need to make my mistakes on a breadboard. I need to know exactly what values of every passive element is and all that

I already made a pcb but never got it manufactured yet

This is how it will be

Now i have to pull that off. As a beginner.

Notice how the inputs to the adc are noded from different stages of the front end

It gives the choice of what stage you want in the digital domain for processing

I am a big fan of back up plans on top of back up plans

[finite pebble]

I’d recommend making your own little modules of different sub circuits and plugging those into a breadboard. That hybrid approach will open up a whole new world of components to you.

Right now you’re limiting yourself to either old tech or super niche expensive parts that come in TH for some esoteric reason.

[exotic kernel]

What do you mean like pcb modules?

[finite pebble]

If you’re a beginner beginner, you might want to consider a less complex project to start with and then come back to these.

[finite pebble]

Yes

[exotic kernel]

Well i can put this stuff together

I have half of it built on a breadboard already

Its not too difficult

Im just making sure I get it perfect

Dont wanna waste money of manufacturing

[orchid rose]

you can get a qfn breakout board with some decoupling caps on it made for $5

genuinely

[finite pebble]

Unfortunately you’ll hit a point where breadboarding isn’t feasible and you’ll need to bite the bullet and get it manufactured.

[finite pebble]

This

[exotic kernel]

I know. Im just getting the hang of the amps and all that.

5$?

[finite pebble]

JLCPCB

[orchid rose]

the world has largely left breadboardable parts behind for a reason

[exotic kernel]

Well remember this thread is from the whole eeg pcb i designed already lol

[orchid rose]

i know, i'm honestly a little confused how we went from like 20 to 12 bits lol

[exotic kernel]

Im just refining first

[finite pebble]

But why not build it and then rework any issues with a soldering iron/hot air?

[exotic kernel]

Im terrified of rework on pcbs

I really dont want to have to do that

[finite pebble]

Well you’ll need to. That’s an essential skill.

[exotic kernel]

Yea

Im tryna avoid it as much as possible

[finite pebble]

Yeah but I think the avoidance to build the board and to rework is hampering you. You’re creating unnecessary roadblocks for yourself that’ll slow down your learning IMO

[exotic kernel]

I figure if i get all the stuff right on the breadboard and transfer it to smd it has a high probability of working

[finite pebble]

Anyways, if you do go that route seriously consider the pcb modules.

I’d throw out the idea of using purely TH parts. That’s a non starter

[exotic kernel]

Interesting

[exotic kernel]

Well i learned a lot from breadbording the amps first

Im using the breadboard to understand the parts more

[paper harbor]

But you can still do that

[exotic kernel]

I dont mind doing pcb

[orchid rose]

oh yeah I forgot about those adafruit breakouts those are nice

[paper harbor]

You can just make modules like Daniel mentioned

[exotic kernel]

I plan to

[orchid rose]

and you can solder decoupling caps across the legs if the package lends itself to that

[exotic kernel]

Im just breadboarding to understand the parts first

[finite pebble]

Ideally with onboard decoupling caps like was mentioned by someone else

[finite pebble]

That module is for a breadboard

[exotic kernel]

What module

[paper harbor]

.

[finite pebble]

The blue photo

[exotic kernel]

Yes

Im sayong

Saying

Im breadbording all these amps first to understand them

Before i put on a pcb

Im getting familiar with the parts

The adc the amps

The active filters

[finite pebble]

Yeah that’s fine, but I wouldn’t limit yourself to the crappy TH parts

[exotic kernel]

This is just for me to learn first lol

[finite pebble]

(Either crappy, obscenely expensive, or out of stock)

[exotic kernel]

Im not staying on a breadboard for long

The eeg will cost 200$

This little emg im making from scratch will be about 50

I figured it would make more sense to try and get this little one right first and see if i make a mistake to learn before the big one

I might not be aware of something yet about the big one that i could potentially come to realize after the little emg

Im seeing on my scope how different circumstances affect the parts

To refine my understanding

So i can apply that to my eeg design and possibly spot a flaw

Pcb is just the half of it, im going to make wearables

Im half way done with the stimulation circuit, im gonna be winding my own transformer with a ferrit core and tape and everything pretty soon here

[dry grove]

thats not how these work...

You get however many bits across the operating voltage of the ADC. A lot of precision ADCs will have a relatively small swing because their capacitance will skyrocket otherwise

You're not "cutting your signal in half" as such

[dry grove]

Wdym good mcu adcs dont have BLE?!

There's entire industries built on these chips!

[exotic kernel]

Are you saying that a 5v signal going into a 12 bit is the same as a 3.6v signal going into a 12 bit?

And are you saying that going from a -5 to 5. To a 0v to 5v adc isnt cutting the signal in half?

[dry grove]

The snr is going to be nearly identical

[paper harbor]

You can offset the signal

So -5 and +5 is essentially 10v, so you can inject 5v to get the range from 0 to 10v if im not mistaken

I'm not sure how you can do that. It's been a long time

[exotic kernel]

The adc input is 3.6v

[paper harbor]

Then use a voltage divider

[exotic kernel]

On an mcu

[paper harbor]

Which mcu

[exotic kernel]

I know

Almost all of them

[paper harbor]

You're going to put almost all MCUs on your board? Damn

[exotic kernel]

I have explained everything already lol

[paper harbor]

You seem to be changing your mind a lot because here you were talking about a separate ADC

Which is okay

But it's rather confusing for me

[exotic kernel]

I think your right

I was thinking about how the noise would be reduced the same as the signal when it goes through resistance

I was thinking that 3 days ago

When i put it through the esp

To test my inst amp

I squashed it in half

So i can literally just squash it into the adc and its the same

Wow

[dry grove]

If you need dual rail then a standalone ADC can help a lot, but usually a virtual ground is sufficient

[exotic kernel]

Well in that case ill just squash it to 3.6v lol

Yea im thinking about the bits and how they work and yea technically if the ref is set lower the distance between bits its set lower

So its the same quantization depth

Hmmmm

There a little bug in the back of my brain telling me somethings not right about it though

Ok wow i just did some research and yea you are right once again

The only thing that could affect it is resistor noise but thats very negligible at that point

[finite pebble]

yeah this is the "flicker" noise Christopher mentioned earlier. I actually just learned that from him recently; had never heard of it before.

[exotic kernel]

The flicker would get filter out in digital no problem anyways. The real conundrum is I dont know how much i can do in digital without causing issues. I have to experiment with the full wave rectification and envelope amps vs doing it digitally when I breadboard to see if i can pull it off digitally without it lagging like crazy

[exotic kernel]

The more I learn about EE the more I appreciate how nice the esp32 is

The newer ones are incredible

[dry grove]

You cannot filter out flicker noise if you need those frequencies

[exotic kernel]

Flicker is <10hz. Its pretty much useless for emg

For eeg its useful though

[stable leaf]

The less noise you let enter your front end the better,

Filtering noise after it has been amplified is less desirable.

Re: breadboards... Personally, I breadboard analog in LTspice and then it's smt all the way.

[dry grove]

has little to do with the concern here

flicker noise is often introduced during signal conditioning

[stable leaf]

Would a zero-offset amp vs. an instrumentation amp help in that regard?

[dry grove]

choppers help, yes

Instrumentation amp is just a configuration

[exotic kernel]

The flicker noise is not a problem

The real pain in the ass is mains hum

[dry grove]

comb filter @60 hz, use dsp if you cant get the precision you need w analog

[exotic kernel]

The mains would either be a analog notch filter or a mains filter program in digital and from everything Im seeing, its done digitally becuase there is also all the main harmonics which can be pretty decently strong at its first couple orders on harmonics

And the analog notch is meh

It kinda scoops out other critical stuff a little too much

[exotic kernel]

Isnt it crazy that an inductor with less loops creates a larger flux

[finite pebble]

I think you have it backwards

[dry grove]

Again, a comb filter

That is their exact purpose

[exotic kernel]

Haha I refused to believe it for 2 days. Its true. Im building a transformer and im learning about the core saturation and teslas and all that and the equations show if you have 2 turns instead of 4 or 6 or whtever in an inductor, the flux is larger (webbers)

[exotic kernel]

Theres another equation that has to do with the amps and turns and when you decrease the turns (loops) the current becomes disproportionally larger and cause a bigger flux

[exotic kernel]

Oh wow i didnt know it was called that, i should of known, it looks like a comb lol

[dry grove]

No, flux increases with the square of the turns

Only linearly with current

If you have a non-linear system then sure, it can happen, but not in usual operating modes

[exotic kernel]

Do some research on it real quick

Inductance increases with square of turns not total flux (webers)

[finite pebble]

Care to show the formula?

[exotic kernel]

1 sec

[dry grove]

AI 😬

[exotic kernel]

But not wrong

[dry grove]

It is

[exotic kernel]

No man its specific to transformers

Its not wrong

Its not just ai

I spent 2 days studying this

[finite pebble]

oh that's just wrong though

[exotic kernel]

No daniel

Its not

Theres engineering websites that say the same thing

[dry grove]

Sure, whatever, youre right

[exotic kernel]

On transformer design websites

[dry grove]

Go ahead and screw your designs up

[finite pebble]

Are we talking about a transformer here or an inductor?

[exotic kernel]

Why dont you check it out a little bit before being so certain

The inductors in the transformer

Thats what im talking about

How they behave in inductors

Transformers*

You have to have a certain minimum of turns to prevent the core from saturating because if theres too little the flux becomes too large and saturates the core

[finite pebble]

I think we were all getting turned around on magnetic flux (Φ) and flux linkage (Ψ)

I'll grant that my EM fundamentals are rusty, but that AI answer smells of being completely hallucinated.

@exotic kernel if you can link the engineering source that you mentioned says the same thing that would be rad.

[exotic kernel]

Ok leme see if i can find the pages

Heres a stack exchange post i was looking at earlier. I was looking at like 20 different things hold on ill find more

https://electronics.stackexchange.com/questions/624129/why-is-my-maths-saying-that-increasing-turn-count-will-decrease-magnetic-field-s

I found some other pages but they were describing more about the other physics of the transformer

But yea. I literally refused to accept it for like 2 days as I studied the transformer more and more, it didnt make sense to me at all but i researched it to death and it is actually true

Its a well known problem when people design there first transformer, they try to get strong ferrit cores so they need less turn to get a desired inductance and end up saturating their cores lol

I almost ended up doing that

[finite pebble]

So, @exotic kernel this was your statement:

Isnt it crazy that an inductor with less loops creates a larger flux

Let's look at Faraday's law: {\displaystyle {\mathcal {E}}=-N{\frac {\mathrm {d} \Phi _{B}}{\mathrm {d} t}}}

I think that's where the AI got tripped up. This ratio just means that if N is small (not many turns), magnetic flux needs to be big to get the same electromotive force (epsilon)

[exotic kernel]

Yes i sent that message without the context intially

Initially

I had just got done reading about it for a couple hours and was like wow

[finite pebble]

magnetic flux = B*A

[exotic kernel]

I was reading something on a transformer engineering website where they were describing how because the core increases the inductance and therefore resistance to change in current, so much per turn, that with the ac current it actually overpowers the turns increasing it normally like a stand alone inductor

[finite pebble]

anyways, all that's to say, I can find nowhere that would imply that magnetic flux is proportional to 1/N. Faraday's law and the formula for a magnetic field across a solenoid show the opposite

[exotic kernel]

Well you gotta remember its 2 coils

And a core

Thats really stronr

Strong

And a step up in voltage

And a resistance on the secondary side

And AC current

[finite pebble]

The issue I take with their wording is that they're conflating magnetic flux and magnetic field.

[exotic kernel]

Isnt it the same thing

[finite pebble]

no

[exotic kernel]

What is flux

I thought it was the field

[finite pebble]

a lot of these AI answers are like 60-80% correct

no flux and field are very different

[exotic kernel]

Oh dam

Is flux like movement?

Or the energy of the field

[finite pebble]

unfortunately I don't have time to go into it, but I'd really suggest a book over AI - I've been using AI recently as a helper when studying a textbook which has been useful for learning new stuff, but I've caught it be slightly wrong a few times. Having the textbook as a source of truth is super useful.

(gotta get back to dayjob)

will look up some textbooks for you later

[exotic kernel]

Ok thanks bro i really enjoy talking to you guys. Smartest guys on electrical engineering discord servers

[finite pebble]

I think you are right about flux density through a single coil if voltage is held constant, but I think it might be leading you astray a little bit. Good to familiarize yourself with maxwells equations in depth.

[exotic kernel]

Your right it was leading me astray

It was warping how i thought regular inductors worked

[finite pebble]

The total flux (N* flux) determines the voltage induced across the whole coil stack

flux on its own is the magnetic field lines passing through an individual winding alone.

[exotic kernel]

Oh jesus i know what ill be studying tommorow

More physics

Lol

[finite pebble]

The wikipedia page for "flux linkage" is good

[exotic kernel]

What do you guys think is gonna happen with AI and EE in the next 5-10 years? Do you think its gonna be a situation where only the best EE’s will survive and it will be a heavy combination of AI and human together?

[orchid rose]

right now it's pretty useless but it was pretty useless for software a few years ago and now it's actually pretty real

there are quite a few companies taking the idiotic hyperscaler approach to it right now with pretty brutal results. in a couple years we'll find out that someone was working on it in an intelligent way rn but we didn't know about it because they know they actually have a product and therefore don't need a flashy website

[exotic kernel]

What Im concerned about is that it will get so good it will be able to do all pcb design and part selection correctly and then once the robots start having intricate hand movements and embedded tools in their hands its over for us monkeys

[orchid rose]

what i'm concerned about is that an asteroid will hit my house and blow it up

[exotic kernel]

Daaaaaam

[finite pebble]

I’m not smart enough to even guess. I just want to be able to buy DDR5 for a beefy home server

[stable leaf]

I just want energy for my home and not have to subsidize data centers when paying for it.

The asteroid thing worries me too. Of all the houses in the world... Why my house?

[exotic kernel]

I just want to own a little house and help fix biological issues

Chatgpt has officially lost my trust

Grok and gpt have been telling me for a week that my core would saturate

I figured out it wouldnt because the pulse times were too small

And now look what its saying

Ridiculous

Yea these AI are farrrr from taking ee jobs

Thats forsure

[dry grove]

It depends on a lot of factors, and specially designed transformers can deal with pulses better than constant signals. Hence pulse transformers.

[exotic kernel]

This is a nightmare lol

[dry grove]

lumped element modelling of magnetics is pretty similar to electrics, but you're going to get way more parasitic effects

it helps to know the models

[exotic kernel]

The biggest issue is the pulse

Its 100us 10v being stepped up to 80V

The core saturates because of the low turns required to get that inductance

If i add more turns i get more inductance and it cant respond to the transient fast enough

Im trapped lol

[dry grove]

why are you using a transformer for this

Instead of driving a switch across 80v

[exotic kernel]

What do you mean?

How am i gonna get the 80v

[dry grove]

Boost converter?

Or potentially a flyback

[exotic kernel]

Thats not gonna work for a stimulator

[dry grove]

why not

[exotic kernel]

The peak to peak has to be directly opposite

And the current has to be 10ma

[dry grove]

So use a switch

[exotic kernel]

Ive seen some example of flybacks

I gotta look into it a little more

The 1:10 100uh transformer is the classic way to make it i just dont understand how they did it yet

[dry grove]

probably driving it with a buffer

Either way 100uH is going to present a big impedance

[dry grove]

indeed, the fet is acting as a buffer

[exotic kernel]

I am feeding a 555 into the gate of a fet

[dry grove]

Do you have high enough Vgs?

[exotic kernel]

Yea

The Ron is 100ma and there a full 10v going straight through it from the battery

Power is not the issue

Or switching

[dry grove]

Ron is 100 mA?

[exotic kernel]

Mohm

Sorry

mohm

[dry grove]

Pretty high but it might be ok

[exotic kernel]

Everything is perfect except the transformer

It will saturate in like 10us

[dry grove]

Well youre effectively sending DC across it without a load on the other end for one

[exotic kernel]

Theres a 1k load

And 2 ohms on primary side

1k on secondary

[dry grove]

80 mA is pretty small

[exotic kernel]

The resulting secondary output is perfect

Im actually gonna add more resistance to make it 10ma

Thats what i need 160v peak to peak 10 ma

Thats a stimulator

[dry grove]

That will make your saturation problems worse

[exotic kernel]

I wouldnt be suprised

This thing is wierd

Shared inductance is hurting my brain lol

1 inductor is confusing enough

2 of them coupled is annoying

With a core

And loads

I can avoid saturation with higher inductance values but it creates another problem

[dry grove]

have you tried a ferrite?

[exotic kernel]

The initial pulses are not fully stabilized as quickly

Yea

Im trapped lol

I cant have my cake and eat it too

[dry grove]

i dont see a flyback diode

[exotic kernel]

Theres no fet either

[dry grove]

Well you kinda need the flyback diode

[exotic kernel]

Explain please

Is that supposed to let the inductor release or something

[dry grove]

I'll leave it as an exercise to the reader

[exotic kernel]

Ok

I thought the flyback diode was the diode that protects the fet

In that case that diode is the flyback

[dry grove]

Yeah this isnt the same circuit configuration as your reference schematic

So

It wont work the same way

[exotic kernel]

Yea

Theres a bunch of em

Bunch of way people make them

The way i did it can work

I just need the right core

And probably a couple pulses to let it stabilize

I just want it to stabilize immediatly

I dont want it to take 2 seconds

And im not saying yea in a condescending way im saying it like i agree with you

Lol some guy on reddit just gave me a link straight to the right core

https://ferroxcube.home.pl/prod/assets/e53272.pdf

3f3

Wow

https://ferroxcube.home.pl/news/gate drive trafo.pdf

[exotic kernel]

Microcontrollers are so cool

Im over here making a square wave with a 555 like an idiot and I relaized wait a minute why dont i code different frequency waves out of different pins of my esp and use a logic mosfet and a rotary switch to select which output i want

This thing output square waves of various timings and on times so good.

I got it outputting 1 100us pulse once every 100ms

Perfect

[exotic kernel]

Hey do any of you have experience with rf receiver amplifier circuits?

[stable leaf]

For .. ???

[tawny wasp]

did you ever end up making the EEG PCB?

[tawny wasp]

If you want a proven circuit, here.

[exotic kernel]

im making my own data transmitter to send 1ksps data out of a serial register thats being fed from a adc thats being fed from my emg circuit

[exotic kernel]

No i learned from my emg circuit that eeg is nearly impossible unless in a isolated room, this is exactly why i decided to build the emg first before making a final decision on sending the board out to get manufactured

[exotic kernel]

Im going to begin wrapping my transformer today I got all the parts

this is my post explaining everything. Im going to try wrapping the circuit in tin foil right now and see how it blocks rf interferance. https://electronics.stackexchange.com/questions/767136/why-isnt-my-rf-receiver-amplifier-circuit-working?noredirect=1#comment2037463_767136

it was all fun and games until I started working with rf lmaooo now everything is a capacitor

[finite pebble]

How do you mean? I’m working on a similar project and they’re having good success with EEG.

[tawny wasp]

you better ground it very well, otherwise it'll do nothing

[finite pebble]

I agree with the advice given on stack exchange. Breadboarding RF is a non starter.

Along the lines of the XY problem (https://en.wikipedia.org/wiki/XY_problem) why do you want to build your own radio? What’s wrong with an off the shelf BLE module or wifi module?

[exotic kernel]

Yes it can work buts its kinda vague, like for example... the most youll get from it is good signals from the visual cortex when closing the eyes and maybe some useful signals when relaxing but its minor. My inteded purpose was to use it to see when I am meditating, and it will barely work unless external noise and movement is extremely low.

[exotic kernel]

LOL I just wrapped the entire probe in foil and was experimenting with connect it to ground and it didnt do anything at all. what ground are you refering to I tried to connect the foil to the scopes ground pin on the oscilliscope itslef and it did nothing. I tired connecting it to a random rail on the breadboard and it did nothing, (no battery was hooked up to the breadboard). These 100mhz radio signals from local radio stations do no decrease at all

[exotic kernel]

I am doing a alot of seperate fundamentals projects to grow a deep understanding of everything involved in brain computer interfaces. They need custom transceiver systems, cutsom amps, custom circuits, etc... everything they do is cutting edge so its more along the lines of invention instead of the general combining premade ic's together. I could easily jsut use an ic but I would learn pratically nothing relative to what these research labs are looking for in an employee

[exotic kernel]

I have been visually seeing the effects of breadboard parasitics on the varying frequencies of my oscillilator 4mhz-60mhz. Under 10mhz the transmitter does not emmit anything visible. around 16-24mhz works pretty good. 48mhz appears to still exist pretty strongly in a breadboard and I can recieve around 40mv peak to peak of 20-40mhz with harmonics which is enough for what i need, its just the recieving and amplifying that is not working and I dont know why. If 4vpp 48mhz sine wave can exist on my transmitters oscilliator in the breadboard, then I dont see why I couldt amplify that 20mv recieved signal to atleast a volt

there is just so much dam noise I cant stop, hence the tin foil but thats not working. Im learninng a ton and thats really the whole point. Im going to take a break from the receiver circuit and wrap this transformer for the stimulator circuit

[finite pebble]

The drive is great but I doubt they’re looking for you to design a radio from scratch if you want to work in EEG. There a reason premade ICs exist.

If you want to be an RF engineer, go for it! But it doesn’t sound like that’s what you want

The data rates you’re dealing with can easily be sent over off the shelf wireless links. There’s nothing there that necessitates a custom solution IMO.

[finite pebble]

There’s nothing magical about “ground”. There’s no single place you’ll connect it that will solve all your issues.

It comes down to EM field theory and how shielding works.

[finite pebble]

This is a great primer but it’s very short for the price so see if you can find a free copy somewhere.

https://www.abebooks.com/servlet/BookDetailsPL?bi=32221808711&dest=CAN&ref_=ps_ggl_21011545875&cm_mmc=ggl-_-CA_Shopp_ISBN-_-product_id=CA9781119183747NEW-_-keyword=&gad_source=1&gad_campaignid=21011545875&gbraid=0AAAAAD3Y6gvMM_YG_DbKwfIs5djnZ28g3&gclid=Cj0KCQjwve7NBhC-ARIsALZy9HXGnP84FYnJiOJGYO4qrI7XjUCuvs8LzJVMBps0QklRbAJdGLJCEEYaAnWMEALw_wcB

If you haven’t seen it already, watch this: https://m.youtube.com/watch?v=ySuUZEjARPY

[exotic kernel]

yes i have seen that, i need to watch it again I forgot

[finite pebble]

Fin foil is going to be leaky

What you want for prototyping is a PCB. And to make shields, solder thick copper tape to make DIY shield cans

Or, but your circuit in a steel cookie tin with edges sanded down for good electrical contact

[exotic kernel]

im going to have to deep dive into rf shielding

wrapping this transformer is a pain in the ass lol

250 turns

[finite pebble]

Trust me. Tin foil wrapped is going to be bad. It’ll not work at worst, or at best be inconsistent. Both bad outcomes.

[exotic kernel]

i tried it it didnt work at all lol

https://www.youtube.com/watch?v=QJdgHXyJh7M&t=973s skip to 15:45 and watch for like 30 seconds it will give you an idea of what i mean about the inventing stuff.

[finite pebble]

Excellent app note https://www.analog.com/en/resources/app-notes/an-159.html

Check out figure 7

[finite pebble]

Yeah it’s all very cool but, it doesn’t mean a beginner should learn all those things. If someone told you that, that’s bad advice.

Nothing wrong with being a generalist! But it’s important to understand they don’t have the same principal eng working on the analog frontend as they would developing a brand new wireless frontend

They’d hire a vast array of specialized engineers to develop these systems

[exotic kernel]

I absolutely agree with that

I am just doing this once

to learn a little about it

[finite pebble]

An that’s good 🙂

Curiosity is great. If the goal is learning, awesome.

[exotic kernel]

what better way to learn about parasitics than to annihilate my mind with breadboard oscillator tranceivers lol

[finite pebble]

Yeah if anything you’ll gain respect for the challenges in each of these specializations

[exotic kernel]

and it helps with indetifying issues with analog front ends

which is what i want to specialize in

Im learning how transistors work

linear regions,

capacitances between base and emmiters and how that affects it etc...

it goes on and on

If i just throw this into an ic and get it over with ill barely learn anything

but yea like you said im making sure not to generalize too much

[exotic kernel]

lmfaooo I get to turn 220 and the wire is protruding significantly past the bobbins edge and im like wait a second and i try to fit the core into the bobbins and it doesnt fit. Now im gonna go back in ltspice and see if it will work with half the inductance and then rewrap this thing. Unanticipated failures lololol

[tawny wasp]

they're a scam

[exotic kernel]

what...

[tawny wasp]

would recommend to read into them some more. Like their monkey computer game controller demo, where the exact experiment had been demonstrated and published like 20 years prior (and, completely unrelated I'm sure, one of the head developers of neuralink was doing an internship at that lab), of course without all the musk hype surrounding it

[exotic kernel]

well what other company has fully functioning proven brain computer interfaces currently working where paralyzed guys are controlling computer mouses better than a human can with their hands?

[tawny wasp]

you mean where a paralyzed person can interact better with the computer than without an implant? There's a whole field of research focussed on it, although most tend to be quite careful when it comes to human experiments.

[exotic kernel]

I think you are coming from a perspective of some kind of personal negative bias towards Elon. from an objective point of view neuralink is not a scam and is actually helping people. There are other bci companies on the cutting edge along with neuralink that are helping people which i would also be glad to work for, but you cant deny they are doing a good think for humanity.

[tawny wasp]

as a primer: https://spectrum.ieee.org/monkeys-control-computer-with-thought
The issue here isn't the technology, it's the biological problem of cutting open your brains meninges and implanting an incredibly fragile device in a way that functions a long time and does not cause more problems than it solves. Conductive material and biocompatibility really do not play well together, any implant material currently in use has long term issues with biofilm formation, and the brain is somewhere you really don't want that to happen. Considering that the first human patient had most of his electrodes detach within a year iirc and the general secrecy surrounding other test subjects I wouldn't put too much faith in this.

my negative bias towards Elon is not personal, but historic based on every single company he's owned. Of course if you put enough smart people in a room and throw money at them they'll come up with great things, I don't remember for what product it was exactly but one of their EEs had a great technical talk on how vibrational modes of their PCB were an issue in one of their pcoducts and how they solved it. At the same time you have the CEO running around claiming things like "point to point starship flights", " 10-100$\kg to LEO!!!", " Robotaxis coming next year, it's financially insane to buy anything other than a Tesla", "Electric trucks more efficient than rail", or starlink, a product that has no financial viability with even the most generous of calculations. For me, that's a scam.

[stable leaf]

These meandering conversations make it hard to follow any particular thread.

Much of it isn't relevant to the question(s) asked / topic.

Too much noise, not enough signal.

[stable leaf]

Posted a response on StackExchange re: RF where that discussion is more focused.

[exotic kernel]

Thank you tc

[exotic kernel]

Yes i have heard of different failures including retraction of the sensor fibers. The primary point im making is that although there are complications, it is fundamentally a necessary technology to develop and there will be mistakes made along the way. Now whether they are moving a little too quick at the expense of making more mistakes or not is something I am not qualified to determine. Regardless, I wouldnt consider it to be a scam. About Elon… yes he talks a lot and predicts things that never end up happening but he is pushing innovation in multiple fields significantly. Nobody is perfect.

[exotic kernel]

His name is Megatron

[finite pebble]

Now you just need an LCR meter or VNA to verify it

[exotic kernel]

Literally was looking at one on amazon an hour ago 🤣🤣🤣

[stable leaf]

You can measure inductance with a function generator and an oscilloscope. Various methods via a simple web search.

[exotic kernel]

Bro… so im testing my transformer and i notice on the scope that the peaks of the pulses are different for each pulse and i spend hours trying to test and read why only for it to be the sampling rate of the scope in the end… As I decrease the time interval the peaks became more even and more even until i zoomed all the way into 1 pulse at a time and it was perfectly even.

When I started EE, I never would have imagined how many weird things would be involved like the act of measuring and its effect. I was under the naive impression that the scope is the scope and thats it. Its not only a scope but part of the circuit and will show inaccurate results based on its capability… god I have a lot to learn

[stable leaf]

Aliasing and other sampling artifacts are your companions on this electronic journey.

Probing and measurement techniques are often overlooked as well.

[exotic kernel]

I was pulling the core apart as it was running and the uneveness was still there. Thats when i knew something was weird

[stable leaf]

Images of a cat with a ball of yarn come to mind...

[stable leaf]

@exotic kernel I was watching this video for a different reason but at 3:25 or so he shows an oscilloscope method for tedting a coil. https://youtu.be/YQNW4ZTsg2w

[exotic kernel]

just shocked the ish outa myself.

set up my stim circuit and had it dialed in and was testing it on my forarm and everything was going nice and smooth until i accidently brushed the gate square wave input jumper wire against the giant mosfet heatsink drain and jolted over 100v through my forarm.

im still shooken up lmaooo

luckily Im smart enough to know to set the electrodes close together on only one arm and I had the battery going through a regulator and had resistors set up but I wasnt smart enough to predict the random things that could happen. Note to self... use a smaller battery lol, Laziness is unacceptable with stim circuits. I said in forums that i would use a small battery but ignored my own plan...

[paper harbor]

Please try not to k*ll yourself 🙏

[exotic kernel]

Well ill definantly take it alot more seriously now

thats for sure

i thought controlling the crrent and slowly bringing it up by replacing resistors was enough and it was going perfect. Did not expect accidently touch conductive metals and what it could cause

[stable leaf]

Get a pair of gloves and wear them when playing with these things.

[exotic kernel]

JLCPCB vs PCBWAY?

[finite pebble]

Both are good. JLC is cheaper, PCBway has better quality control and more options

I default to JLC nowadays

But have use both a lot

[exotic kernel]

Ok ill try jlc

My first PCB took me 3 months. This is my 2nd one. Took me 3 days and only had 1 drc error.

[tawny wasp]

post a schematic and an editor view that at least has the ref designators, no clue what's supposed to be what here

[exotic kernel]

Ok give me 30 mins

Ill post the layers to

[exotic kernel]

layer 1 - Components. Layer 2. (green) Ground. Layer 3. Orange (power and ground). Layer 4 (blue) Ground

The schematic components go from left to right. On the PCB it is layed out right to left

[orchid rose]

why are you only using 1 opamp per dual opamp

[exotic kernel]

becuase im stupid

didnt even realize thatr until you just mentioned it

its becuase when breadboarding I have been experimenting with adding single amp filters, pga's, etc... and I got in the habit of only using one side of the amp

now im debating whether its even worth it to redo the pcb becuase the lm358's are cheap

i might just leave it like that this time and make sure i dont do it again

[exotic kernel]

Im just gonna redo it.

Im gonna be adding rf and stuff to this design later so i might as well get it right the first time and make sure it works instead of doing it over and adding more on top of it and then not knowing what went wrong

[tawny wasp]

will comment tmrw, long day and I'm going to bed

[exotic kernel]

ok

[finite pebble]

bit of a nit-pick but you're mixing conventions. Should either be (VDD/VSS) or (VCC/VEE), not (VCC/VSS)

[exotic kernel]

Ok I was wondering about that when I first did it. Thank you

[exotic kernel]

You pointing this out led me to redo the design and I found that i had misplaced a resistor in my rectifier amp. Without you mentioning that I would have ordered the board and it wouldnt have worked. Thank you

[exotic kernel]

Nit-picking is a great trait to have as an engineer

I feel like making it as perfect as possible is something that a potential future employer will respect.

[stable leaf]

Well ... There is a tradeoff. Perfection takes time, and honestly, it's elusive.

There is always a tradeoff. It just has to work ... well enough.

The experienced engineer knows how to make it good enough, and know when it is.

[exotic kernel]

Ive always been a half asser so I think i need a lil more attention to detail. I agree at a certain point its good enough

I think it also depends on what the goal is to. Like if its a project where it just has to work or something where human safety is at stake

[stable leaf]

I've worked on my share of safety systems. It's not a black and white thing either. But yeah, always requires more care. Depends on the risk.

[tawny wasp]

can be shrunken down by a factor of 2, use polygons instead of branched traces, especially if they lead to a sharp internal corner.
ldo datasheet recommends caps on input and output. Don't route on your gnd layer, especially if you're jumping over reference planes. That single track on B.Cu can be routed on layer 3. Or probably on F.Cu with better layout. No need to branch out this far just to drop a gnd or power via.
Layout in general seems a bit odd, I recommended putting more time into laying out the components in a way that makes routing easier.

[exotic kernel]

@tawny wasp

@stable leaf

@finite pebble

layer 1. Signal. Layer 2. Ground. Layer3. Power. Layer 4. Ground. All signals on layer 1. All ground planes full (no cuts)

[stable leaf]

I won't be able to look at this today.

[exotic kernel]

no problem

[tawny wasp]

looks much better imo. Any reason to use multiple pins for the same net?

[exotic kernel]

Thanks. The right jumper J2 are 2 pins per input incase a pin breaks. Thats the same principle for all the headers. Im gonna add a couple more headers just for ground.

[tawny wasp]

fair. Don't be afraid to use polygons for routing btw.

[exotic kernel]

Like this?

That top left pin

On the right of U1

[tawny wasp]

yeah for example. I just see a lot of tree branching with your traces, which isn't great. won't make much of a difference at this size but recommend just putting a polygon there instead

[exotic kernel]

Ahh ok i see exactly what you mean.

[stable leaf]

Is there a schematic for this?

[exotic kernel]

@stable leaf

[exotic kernel]

is there anything I should add or change to this step by step system for ee desgin

• Ask yourself if what you are making is the best option for what you are trying to do
• Research circuits that people have already made
• Make sure you dont exceed the voltage specs for all parts
• Make sure you dont exceed power (watts) spec for all parts
• Take note of the current on all nodes in simulation
• Falstad/Ltspice, then breadboard, then PCB
• Show your designs to proffesionals before getting manufactured

[finite pebble]

A few notes:

• "... circuits that people have already made" -> I'd replace this with "reference designs from proven sources". There are a LOT of crap designs out there.
• For most things I work on, breadboard doesn't provide much value. Sim + design reviews -> PCB is fine IMO, even for beginners. Boards are so cheap nowadays that I see little value in breadboarding. The parasitics that the breadboard introduce make it less accurate than spice a lot of the time.
• A commonly overlooked one to add to that list is, "Do I have good thermal management?" - parts often overheat if you're not careful, especially LDOs. Big trap for newbies. If a part says it can do 1A, that's only true for certain conditions (diff from Vin to Vout not too big for LDOs, near perfect heat sinking through the board).
• My gut feel (having not used it much) is that falstad isn't good enough for design work. I'd use a proper simulator like LTspice, Pspice, Qucs, ngspice.

[exotic kernel]

Nice

[finite pebble]

Another big one, before running a sim or doing a bench test: ripped from Dr. Eric Bogatin's rules of thumb:

Rule #9: Never perform a measurement or simulation without first anticipating the results you expect to see. Never believe a measurement or simulation blindly.

[exotic kernel]

yea ive been noticing a lot of surface level googles circuits are trash

[finite pebble]

Bogatin's 20 rules: https://www.signalintegrityjournal.com/blogs/4-eric-bogatin-signal-integrity-journal-technical-editor/post/1539-bogatins-20-rules-for-engineers

[exotic kernel]

oh wow yea i think i heard eric bogatin mention that in a youtube video before

hahaha

i sent that message before i saw your bogatin message lol

thank you Daniel, great advice as always.

[finite pebble]

I didn't really buy into this until a few years ago, but there are some great textbooks out there. There are some bad ones too but in terms of finding "sources of truth" among the bad advice, I defer to industry experts and well-regarded textbooks

This in particular applies for contentious topics like PCB design best practices, ESD mitigation, EMI mitigation, grounding techniques. Topics that are more "pure theory" like control systems, filtering, power supply design, analog circuit design, etc, are far less contentious IMO.

[exotic kernel]

right now im learning about gate driving techniques like the totem pole. Does it ever get to the point where you learn about almost all the techniques and you pretty much know whats going on in 90% of circuits just by looking at them and does it get to a point where you can pretty much work on almost any circuit in the world or is there some deeper level that you face after 5-10 years where its like omg theres still so much more to learn

[finite pebble]

It never gets easier, you just get better and work on harder problems

[exotic kernel]

Wow

Thats interesting

[stable leaf]

Scale/scope/complexity/detail ... Different layers of the same onion.

[exotic kernel]

So basically as you level up, so do the projects you work on

[stable leaf]

If your career follows a good path, yes.

Truth is, you have to work hard to acquire skills to advance your career, and to make sure you have the opportunity to apply them, and pursue your interests.

Way too many people get trapped into narrow lanes and never find their way out.

So it takes a commitment to life long learning and a focus on where you are heading, where you want to go with your career.

With that you can figure out what you need to learn to get there, acquire the skills, and find opportunities to use them.

It is like sand dunes in the desert. They keep shifting with the wind and you still need to get across.

[exotic kernel]

Im finally starting to understand the capacitances of a MOSFET. So basically theres really no way to get rid of the dg ds gs parasitics? Im building an hbridge and from what i know so far I dont see how it would be possible to get rid of the parasitic effects without significanly disturbing the function of the circuit

[stable leaf]

You can't get rid of inherent characteristics of a device. Your design to has to control/mitigate/accommodate the effects, and achieve the performance you require.

Fortunately for that application there is extensive information, design guides, application notes, reference designs, etc.

[exotic kernel]

Do you guys think its possible, if I study and use MOSFETs long enough, one day i, myself, could physically transform into a MOSFET

[stable leaf]

Anything is possible with electronics. Or maybe that's software. I forget. 🥴

On the flip side of parasitics...

Sometimes these intrinsic characteristics are exploited.

Think of a diode used as a temperature sensor, a pad on a PCB used as a capacitance touch sensor, the voltage variable capacitance of a tuning diode, the impedance of a transmission line, or the resonant frequency of an antenna, NTC and PTC resistors, etc.

[exotic kernel]

After trying to build my first rf receiver and transmitter i have so much respect for rf engineers

[stable leaf]

I used to feel that way but some of the mystery is gone for me. I forced myself to take the time to learn the basics.

You definitely have to be comfortable with impedances, frequency dependent effects of capacitance and inductance, impedance matching techniques, VNAs, and Smith charts.

It's hard to learn all of it on your own but thankfully, there are many good teachers on YouTube, etc.

[exotic kernel]

Long term goal = work on brain computer interfaces on a pcb design level (not chip design)

Should I do..
100% just pcb with 0% ic design
75/25
50/50
should i even be designing anything from subcomponents
Is it worth the experience

It feels too easy just doing pcb with premade ic’s. It makes me feel like im not learning and/or its not a skill that is rare. When i design things from scratch it seems a lot harder and therefor would be more sought after and have a longer career path.

[finite pebble]

IC design is pretty niche IMO, but that’s my view as an outsider.

If I wanted to do IC design I’d have to be ok with relocating.

While I won’t say it’s impossible to self teach IC design, the only people I know who work in that sector have either a BSc or MSc in engineering, so you may need to consider going to school if you want to work in semiconductors.

Also, VERY few companies do end to end design, from semiconductors to end devices. If you want to work on the brain interface people use, you’ll most likely not be designing semiconductors. Conversely, if you design, say, EEG frontends, that would be typically at a company like TI or ADI, who then sells them to other companies to implement into products.

[finite pebble]

I would say, slightly defensively, PCB design and product development are not easy.

[exotic kernel]

Ok yea I think its very likely I am ignorant to the level of complexity of real industry standard pcb design

[finite pebble]

Another thing: mastery in any field will be sought after. Some are more in demand than others for sure, but you can become rare by excellence anywhere.

Advanced PCBs take months to develop, and months to years to validate

[exotic kernel]

Yea i have a weird tendency to try and do everything myself. Id design down to the atom if I could. I have to actively stop myself

[exotic kernel]

Oh wow

Im really starting to think i should refrain from any further circuit designing after this h bridge.

[finite pebble]

Doing silicon internally only makes sense when you can’t buy what you need off the shelf. It rarely makes sense as a product developer

If you’re apple, sure. But most companies, even multi billion dollar companies, don’t roll their own silicon

[exotic kernel]

Yes and on top of that brain interface companies have separate roles for pure pcb deisgn

Even though they do their own chip fab and design

They separate them into 2 specialties

And i have 0% chance of having access to chip fabrication

[finite pebble]

Which brain interface companies are making silicon?

[exotic kernel]

Neuralink

They do chip design

They have the ic designers and the pcb designers as separate roles

[finite pebble]

Yeah those people would almost never do both

It’s very much “pick a path”

[exotic kernel]

I think pcb would be better anyways becuase during my learning i can make devices i can physically use

[finite pebble]

Are you going to enroll in uni for EE?

[exotic kernel]

No

I am poor

[finite pebble]

May want to look into grants

Not sure if neuralink hires without a degree

[exotic kernel]

I think it depends on how good i can get

I think if i can get really good and prove it with projects i will have a chance

But i dont really know

[finite pebble]

Might want to talk to someone at neuralink, or snoop on LinkedIn to see if anyone in their design team doesn’t have a degree

[exotic kernel]

Theyre probably all phds lol

[finite pebble]

Well, better to check

And if they all are phds, maybe you could find another way in to the tech sector first, like as a technologist, or pcb designer

Where I am, both technologist and pcb designer are diploma programs

If you choose PCB designer, many designers I’ve worked with have IPC CID and/or CID+ certifications

[exotic kernel]

I know its very likely i will start at a different company

This is a long term goal

10 year goal

Im just making sure i put this train on the right track

Those certs are interesting i will have to look into all that becuase i know nothing about emi or anything

[finite pebble]

Also, if you’re not aware, MIT has a ton of free course material online

[exotic kernel]

Yes ive been watching them for years, and standford

For various things

Recently i have been contacting bci researchers from different universities. Its hard to get in touch with them but i will keep trying

[dry grove]

They almost certainly dont do their own fab, they dont have the money

Unless its old equipment. Theyve only raised ~a billion which isnt near enough

[exotic kernel]

I know they do their own chip design. I heard a former employee say they had a fabrication team.

[finite pebble]

yes, actually forgot to mention that. They're surely fabless

[exotic kernel]

I also have heard someone mention they have to wait s while when they send stuff out to get fabbed

[finite pebble]

I'd hazard an estimate that < 1% of companies do their own fab

[exotic kernel]

I believe they dont have a fab

I dont know if the employee was refering to some other kind of fab

Ive seen a youtube video of the 4nm fab machines or whatever. That stuff looks crazy

Thats a huge reason im not really considering chip design cuz its impossible for me to actually get the designs fabed

Its like 10,000$

People desperately try building their own fabs in there garages. Im good on that, id rather do pcb all day

Ok so its not chips

Its the other parts of the device

[broken musk]

your missing quite a few 0s

if you want it fabbed on a node thats anything more recent than the 80s

just the software is gonna require a few more 0s

let alone the design and validation, fabbing, testing, and packaging

[tawny wasp]

that's not what he's saying. You can get silicon fabbed for a couple k if you wanna play around with that

[broken musk]

Yeah but your gonna be very limited

[stable leaf]

I spent 15+ years in the custom IC business. It is an elite business model. As others have said, few companies can afford to pursue it. In addition to the investment required projects take a long long time and you are just a cog in the wheel on huge teams.

FPGA design is worth learning but do something small. Like Lattice ICE. I'd suggest trying a function not in a micro that is data path related.

If you want to expand your skills beyond PCB design think about embedded software as an excellent compliment.

[stable leaf]

Well... Nothing wrong with having a curious mind. But you will earn an income from getting stuff done. That takes focus.

Spend your learning time on things that will help you on the job, and prepare you for the next position/job you want.

The hardest part is figuring out what you want to do in 3-5 years and what skills you need to develop to be able to do it.

Most don't have a plan.

[exotic kernel]

Do you think its worth mastering pcb design over the next 5-10 years considering how ai is getting better and better at it? I guess by the time ai can do pcbs we will have universal income anyway but idk

[stable leaf]

I don't have the crystal ball on AI influence in PCB design.

But my limited experience with AI in software development is that you will need expert humans to drive, set requirements, evaluate results.

I've taken algorithm research papers with heavy mathematics and complex algorithms and fed them into ChatGpt 5.4 with impressive results. But I definitely had to steer it to get results specific to my application.

My observation is that when there aren't a lot of low-level constraints and AI had a ton of freedom on implementation it was great.

AI isn't at that level yet for PCB development. There are so many minute low-level details in PCB design that I wonder if its role will be more constrained.

But as I said... I don't have the crystal ball. I lost it in a bet. 🥴

[exotic kernel]

Lol yea i agree that the future will be expert with AI assistants. I guess the only real answer is to be in the top 10% and youll probably still be able to find work no matter how good ai gets. Because inevitably we will combine with the machine anyways

I think ai being at the point where it needs no human supervision through the entire process is pretty far away, maybe 20-30 years and by then we will have ai in our brainchips and we will be the ai

[exotic kernel]

God im glad i have you guys to talk to man. Saving me a lot of time.

[stable leaf]

Giving back in appreciation for those that mentored us.

[exotic kernel]

And i will do the same one day

[finite pebble]

Yeah it’s very much “pay it forward” 🙂

[exotic kernel]

orange dots*

[wet badger]

Well usually diodes are anode to cathode so by the polarity u drew it its the cathode

[exotic kernel]

...

[jolly saffron]

Command Error

The input text has too few parameters.

Usages

.quoteprint text

Run .ve to disable these prompts.

[wet badger]

i would assume its the cathode

[exotic kernel]

well I cant really assume

pin 1 is usually the white dot

[wet badger]

cathode has the stripe

[exotic kernel]

thank you sabo but that is not what im asking

[dry grove]

What does the datasheet for your diode say

[exotic kernel]

It barely says anything. The other indicator is the bottom copper has a different shape but other than that there is no indication

I know which direct the diode is in but the footprint doesnt make sense

Other diodes have the white dot on cathode and these are on the annode

[dry grove]

can you please share the datasheet

[exotic kernel]

Yes 1 sec

https://www.digikey.com/en/products/detail/liteon/LTST-C190GKT/269230?gclsrc=aw.ds&gad_source=1&gad_campaignid=120565755&gbraid=0AAAAADrbLljSnOrKSwpYRphvOTMII6lk6&gclid=CjwKCAjw4ufOBhBkEiwAfuC7-eOTQdfMX3Gu8Lzhe-ST_7I2IRvaIJNtY1OJXN33sJfjUaUDvEDNsRoCLV8QAvD_BwE

[dry grove]

It says right on the first page

[exotic kernel]

So basically im assuming there is a global standard of where the dot is supposed to be on the diodes

I understand that

What im asking is how is the manufacturer supposed to know

Based on the gerbers

If there is no other indicator but the dot

And other diodes have the dot on the other side

[wet badger]

can always tell them

[orchid rose]

diodes always point towards the dot

it will likely also be marked on your fab layer if you are nice to your mfg

[exotic kernel]

Ok i was assuming this must be the case. Ok so whoever made the footprint did it wrong

Ill add extra notes just to be safe

[dry grove]

The image you sent is correct, its not wrong

[exotic kernel]

Yea the image is one set of diodes

There are other where the dot is on the anode

[dry grove]

Are you certain?

[exotic kernel]

100%

[dry grove]

Can you please show us?

[exotic kernel]

yea

1 sec

[wet badger]

are they indicator LEDs?

[exotic kernel]

yes

[dry grove]

Wait, are you asking if the silkscreen is how the mfrs determine the orientation of the diodes? Because that's not correct. Its in the p&p file

[exotic kernel]

hold on a sec

let me export the gerbers and see if the dot flips around on theses

[dry grove]

The dot on the silkscreen is for you

[exotic kernel]

i know i thought it shows up the same on gerbers

[dry grove]

Yes

[exotic kernel]

why would it be backwards lol

[dry grove]

Because either the footprint or symbol is wrong

[exotic kernel]

right

[dry grove]

I'm assuming the way youve drawn it matches the nets in the schematics

Because we cant see that

[exotic kernel]

i drew the diode the way its supposed to be facing and the way the schematic symbol had it ]

[dry grove]

Which footprint is this?

[exotic kernel]

dot same way in gerbers

[dry grove]

Yup

[exotic kernel]

footprint came straigfht from digikey

it is backwards

[dry grove]

OH

Yeah dont use those 😂

[exotic kernel]

omfg

lol

cant even trust digikey

[dry grove]

Theyre automatically generated and may or may not have anything to do with the footprint you need

[exotic kernel]

well im glad i caught it because i woulda had 6 non working leds lol

[finite pebble]

The standard is defined by IPC-7351B and is pretty well recognized, but this is a common footgun

[wet badger]

super true for transistor too with them

[exotic kernel]

omg

[dry grove]

Mayhaps. Use kicads diode fps

[finite pebble]

The way the "CM is supposed to know" is by reading your assembly drawing, which should make it clear (to add more context than the silk has).

[exotic kernel]

im just gonna leave gigantic notes for manufacturer

yep

[finite pebble]

I'll find an assembly drawing I provided for a board getting built

[wet badger]

always a good idea

[exotic kernel]

thatd be cool

check it out btw I added some experimental stuff to the board

[finite pebble]

not my cleanest, but was rushing to get it manufactured.

[exotic kernel]

nice

[finite pebble]

Some text is upside down and not centered, which I didn't bother fixing. But, the critical thing is that the PDF is text-searchable

[exotic kernel]

ok i understand

[finite pebble]

In the past I've added "K" and "A" on the pins for anode and cathode (convention from IPC)

[exotic kernel]

yea AI isnt gonna be able to do this stuff for a long time

[finite pebble]

but on this one, I had a silk drawing showing the diode direction, so I omitted it from the assembly drawing

IPC-2581 (the file format which is an alternative to Gerbers or ODB++) has support for pin names as "ANODE" or "CATHODE" to avoid the pin 1 confusion, but I'm not sure if my lib is set up right for that actually.

But that format theoretically supports it to avoid this problem.

[exotic kernel]

im going to blatently draw the entire diode lol

i will leave no chance for confusion

[finite pebble]

that's not a bad idea

IPC-7351B calls out pin 1 as the cathode. Interestingly, its successor, IPC-7352, makes no mention of using pin 1 as cathode, but they have an example of just naming the pins "C" and "A" and putting C on the left.

(I think older EDA tools didn't let you use letters for pin identification, but now you can)

Either way, the fool-proof approach is to draw the diode. CMs get this wrong all the time.

also, unfortunately can't screenshot these PDFs because they're DRM-locked

I use the "bar" in my assembly drawings instead of the dot for pin-1 because on the physical diode body, the bar marking represents cathode.

[exotic kernel]

im glad im learning all this before i sent a board out. My intuition told me not to get the eeg manufactured because there would be things i didnt know about that could cause an expensive mistake

[finite pebble]

You could argue that a SMT line operator doesn't know what a diode does, but they can look at an assembly drawing and see "drawing has bar, part has bar, part placement matches drawing ✅" which has served me well in the past.

[exotic kernel]

im am suprised by the lack of global coherance between all these seperate entities, but then again... we are dealing with humans.

[finite pebble]

The line operators are not EEs, so the diode drawing may or may not be better than the bar

[finite pebble]

manufacturing standards evolve like anything else. Hard to align companies.

IPC attempts do to this, and they've succeeded by a lot of metrics

things were much more "the wild west" a few decades ago

[exotic kernel]

hahaha

ill send out a board and leave a note saying everything is in non conventional current

[finite pebble]

a LOT of the callouts in my fab notes relate to specific IPC standards

there are hundreds of standards. I only own a few. A fortune-500 will own all of them

[exotic kernel]

you own standards?

[finite pebble]

they define, among other things:

• copper plating thickness
• soldermask thickness
• min annular rings
• via barrel plating thickness
• ENIG thickness
• substrate composition

[finite pebble]

yeah a handful

[exotic kernel]

how

you made them?

[finite pebble]

Own as in I purchased them.

[exotic kernel]

woahhh

i have no idea what these standards are

like i didnt know they could be purchased

and stuff

[finite pebble]

I have:

• IPC-6012F
• IPC-2221C
• IPC-7351B
• IPC-7352
• IEC 62368-1

all regulatory bodies/consortiums will sell you a standard

How else would people design to them? 🙂

Oh I also bought some IEEE standards, but forgot the names

IEEE 370-2020 I think

Some are free, most aren't

[exotic kernel]

are they like insider information or something

I dont get it

[finite pebble]

wdym?

[exotic kernel]

i dont understand why someone would need to buy them

[finite pebble]

IPC is a consortium (now called electronics.org) which attempts to solve this problem you referenced

[finite pebble]

this

It's an industry consortium which creates standards that they encourage the industry (designers and fabricators and assemblers) to all agree to, to speak a common language

IEEE is where you find definitions for prototols like Ethernet, WiFi, Bluetooth

IEC has a bunch of standards for consumer, industrial, medical product categories

everything has a standard in electronics

[exotic kernel]

right

but why do you have to buy them

[finite pebble]

because capitalism

most things in this industry aren't free

[exotic kernel]

why cant you just read them and then design accordingly

[finite pebble]

well, why would they give them away?

when they can charge you

[exotic kernel]

ok so its insider information'

[finite pebble]

well no

[exotic kernel]

i guess i worded that wrong

[finite pebble]

anyone can buy it

[exotic kernel]

i mean like secret info

that you have to buy to read

[finite pebble]

well no, it's not secret

[exotic kernel]

thats what i meant

[finite pebble]

it's "secret' in the same way a book from the library is secret

[exotic kernel]

right

[finite pebble]

you can't copy it and redistribute it

[exotic kernel]

right

ok

thats what i was assuming

ok wow

[finite pebble]

like, anything you purchase. a book, a video game, a movie

they all have DRM

so do standards. They're documents that the orgs sell you, for a (hopefully reasonable) price

[exotic kernel]

so when people figure out new stuff they sell the info and it is copyright if you redistribute

[finite pebble]

to cover the costs of writing it

uh ya!

[exotic kernel]

interesting

i thought all info was free

[finite pebble]

same thing with scientific papers. Most you have to pay for (at least the ones I deal in)

[finite pebble]

why would you think that? 😛

[exotic kernel]

because I believed that the engineering / science community was very open

[finite pebble]

in some respects, ya

[exotic kernel]

i guess not as much as I thought

[finite pebble]

it's the exception, not the rule

[exotic kernel]

yea

seems like a bank of info

[finite pebble]

Engineering is traditionally a for-profit endeavor, so everything related to it is built around making money at every step

[exotic kernel]

yea

[finite pebble]

IPC standards don't teach you how to design. It's a common language for you and your manufacturer to speak, so that you can hold them accountable.

and they can hold you accountable if you send them something unmanufacturable.

[exotic kernel]

hold them accountable?

[finite pebble]

to a quality standard

[exotic kernel]

im confused... so why would a standard have to be purchased to learn?

[finite pebble]

I'm saying you don't buy them to learn

you buy the IPC standards to say "hey, manufacturer, do this to X standard"

that's particularly for IPC

[exotic kernel]

what if you know the standard and the name and just tell them to do it to that standard without owning it

[finite pebble]

IEC 62368-1 are design specs to meet to design safe things (bunch of rules for high voltage equipment to not kill people), so you kind of do learn from it as you go through it

[finite pebble]

you can 100% do that and I do that for a bunch of stuff

but, it's a little hard to audit the results if you don't know what the specs are yourself 😉

[exotic kernel]

but arent you supposed to design your board to that standard in the first place?

[finite pebble]

For a bunch of stuff, I do just call out the standard #, even though I don't own them. For the critical ones I use a lot in my business, I bought them to know them inside and out

[exotic kernel]

how would telling them change anything

[finite pebble]

well, things like ENIG plating thickness are decoupled from design

[finite pebble]

telling them what? I don't understand the question.

[exotic kernel]

well you were saying that you tell them the standard

the manufacturer

[finite pebble]

mhmm

[exotic kernel]

whats the point

isnt the design the design

[finite pebble]

if I expect a certain ENIG plating thickness, I tell them the standard to plate it to

if I expect to have no scratches on my board, I tell them the IPC standard that calls out maximum allowable scratch depth and length.

If I expect that all holes to have a specific tolerance and not have them land outside the via annular ring, I call out the standard that defines it.

[exotic kernel]

is this becuase if you dont do that they are not legally obligated to do it

[finite pebble]

And, if I get a board back that doesn't meet my criteria, I have leverage with the vendor that said, "see here? my drawing says to make my board to this spec. You failed to meet the spec" and I can reject those boards

[exotic kernel]

isnt that in the manufacturer files already

[finite pebble]

It's less about legally binding (although that sometimes does become a factor). It's a common language.

[finite pebble]

none of the points I describe are called out by gerbers

[exotic kernel]

whatttt

[finite pebble]

well, yeah

how is the vendor supposed to know your ENIG plating thickness from gerbers?

[exotic kernel]

arent some of the files that you export from kicad including the baord specs/?

[finite pebble]

gerbers are vector art like a SVG file

same with NC drill data

IPC-2581 and ODB++ attempt to solve this problem of missing metadata, but the old school way is, provide a fabrication drawing PDF and assembly drawing PDF that says what you want built

just like with MCAD, we don't just send a STEP model. We send a toleranced drawing to accompany it

Making good documentation is a key differentiator between amateur and professional boards

[exotic kernel]

right

[finite pebble]

like I said, all these specific examples I gave are NOT in the gerbers you'd typically send out

[exotic kernel]

i remember seeing phils lab mention the standards numbers

[finite pebble]

Daniel [::<>], — 8:02 PM
if I expect a certain ENIG plating thickness, I tell them the standard to plate it to
if I expect to have no scratches on my board, I tell them the IPC standard that calls out maximum allowable scratch depth and length.
If I expect that all holes to have a specific tolerance and not have them land outside the via annular ring, I call out the standard that defines it.

[exotic kernel]

i thought all the hole size and placement and copper thicknesses and all that are included in some of the kicad export files

[finite pebble]

nope

not unless you use ODB++ or IPC-2581

[exotic kernel]

wooahhhh

[finite pebble]

or send in your kicad project file, which isn't a great idea IMO

[exotic kernel]

interesting

[finite pebble]

most vendors will not accept native EDA files

Gerbers are the most universally accepted.

ODB++ next most common in adoption, followed by IPC-2581.

[exotic kernel]

why wouldnt there just an extra sheet that included board specs from kicad

[finite pebble]

That landscape of export format could change in 5-10 years, but Gerbers are an old format and have outlived others

[exotic kernel]

you know where the layers properties is in kicad

[finite pebble]

well, you basically do that with a fab drawing. KiCad outjobs support setting that up

[exotic kernel]

why cant you just export that as a file

[finite pebble]

yeah. You export that as fab-notes

[exotic kernel]

right

so why isnt that enough

[finite pebble]

the vendor expects a PDF, so you send them a nice PDF with all that data

[finite pebble]

that is almost enough, sometimes. Pointing to an IPC standard is gold-standard in addition to those others

[exotic kernel]

i believe you. im just trying to understand why it would be necessary

im kinda looking at it like "my board layer setting say 1.6mm" now heres this extra spec that says 1.6mm

like i dont get it lol

like whyyyy

[finite pebble]

here's a pretty awesome blog post of good fab notes

https://resources.altium.com/p/decoding-pcb-fabrication-notes

If you were to call out every fabrication spec in IPC-6012A in your fab drawing, it would be wasteful. Far easier to say "meet this class spec defined in this standard"

seems pretty sensible no? 😉

[exotic kernel]

ok what im starting to understand is that the setting in kicad and the way that information is exported and conveyed is not persisely the same language in every single instance of the manufacturing.

[finite pebble]

Much of that information is not codified in KiCad

[exotic kernel]

there is a gap of translation

[finite pebble]

KiCad has no idea what flammability rating you need, or what IPC-6012A class of board you're making

[exotic kernel]

ok

i get it now

all of the design is not done in kicad

there are extra parameters

[finite pebble]

there is a lot of CAD agnostic information that an EDA tool doesn't embed or codify

[exotic kernel]

wow

[finite pebble]

TL;DR: engineering drawings are important, and are still pretty old school

places like PCBway or JLCPCB won't read a fab drawing though, typically. They want to make their process smooth and touchless, so they have a portal where you punch in this stuff in the UI buttons as you order your board

but, the good old fashioned way for real production volumes through traditional CMs and fabs is with a good PDF and emails

[exotic kernel]

there is a format in which this is supposed to be done that involves these ipc standards which are used as a form of compactified communication

[finite pebble]

the slick UIs are great, but at some point, there are edge cases the smooth UI doesn't cover and you need to talk with a real process engineer at the fab to smooth things out. THAT'S when these drawings come in.

[finite pebble]

unfortunately not really.

every company has their own format that they come up with. But you'll see commonalitites after seeing a few example pics online

That altium blog post I sent is a great base

but every employer you work for will have different process

[exotic kernel]

I had no idea there were things outside kicad that needed to be specified seperatly

[finite pebble]

lots 🙂

at least for a high quality board

[exotic kernel]

right

[finite pebble]

for hobby stuff, not really.

[exotic kernel]

for real engineering not just hobby

right

[finite pebble]

I do this professionally, so I do things the "right way"

[exotic kernel]

yea

exactly

there is a standard

that must be ipheld

upheld

[finite pebble]

it gets even more important for safety critical stuff

[exotic kernel]

yes

[finite pebble]

but even for the area I'm in, in high volume consumer, we care a lot about quality control

incoming inspection for PCBs will use the IPC standards for quality control inspections, and do all sorts of tests and inspections to make sure stuff is up to snuff

this is also used a lot for qualifying new fabricators. If they can't produce boards that meet the standards, they won't get approved as a vendor, and because there are IPC standards, the board shop knows what metrics the customer cares about

[exotic kernel]

ok so being a great pcb desinger isnt just about routing a good board, its about anticipationg and meeting the specific standards for the use case

[finite pebble]

yeah pretty much

Some companies don't expect their engineers or even their board designers to know all this stuff, and leave it to NPI/process engineers to provide feedback and know all this, but I think a holistic approach makes you a good engineer/designer. and, at some smaller companies, you may be the guy who wears all the hats

[exotic kernel]

I think if I implement some of these standards into my designs it will be a bonus on my resume

[finite pebble]

I think it's a good thing to know about for sure