HaasoscopePro AFFORDABLE open-source open-hardware 2 GHz oscilloscope!

[tggzzz]

Having a quick look at your PCB, you don't mention your stackup, the analogue input circuits don't seem to minimise track length, and I wonder if things marked 50ohm in the schematic really are 50ohm.

Personally I wish you luck with your kickstarter, but I wouldn't think of investing until you demonstrated it working - preferably plus a review of the prototype by an independent engineer.

Very legitimate concerns!
I am writing more about the project over here:
https://hackaday.io/project/200773-haasoscope-pro

The stackup is 10-layer, sig1 gnd sig2 gnd pwr1 pwr2 gnd sig3 gnd sig4. Impedance controlled pcb of course. I made test boards and tested track impedance with a nano vna plus.

It definitely works. I've tested with a Leo Bodner pulse generator with 40ps riserime, and I see about a 200 ps risetime. I'm fine tuning the multi board oversampling now, which will get up to 6.4GS/s, at which point I'll be able to test the bandwidth more accurately.

There will certainly be videos, plots, etc demonstrating the performance before the project launches next month. At the moment we are just in pre-launch.
There will also be independent reviews. Hopefully even one from Dave again.

Oh... good.

You are further ahead than I realised.

We look forward to seeing the results

[Nominal Animal]

Sometimes it is known when creating the schematic that certain PCB rules will have to be followed, e.g, differential pair matched length or impedance control. With simple tool for small designs, those constraints can be simple text annotations. In complex tools for big designs where a different person does the PCB layout, such properties can be entered into the schematic, and the tool will (hopefully) ensure the PCB layout person has complied.

Right.  In EasyEDA, selecting different colors for different wires is very easy, so for some examples I've used the wire color to denote different types (power, digital, analog, high-frequency analog video), just to try and see if it makes the schematics easier to understand.  (I'm not sure that worked.)  But I definitely understand what you're saying.

In the cases where, say, several signals are individually buffered before going off-board (e.g. a control bus with inverters), the schematic designer will wire them up without knowing the detailed PCB layout. That may cause an unnecessarily tangled set of wires, and it is perfectly reasonable for the PCB person to swap pins. Such swaps are then automatically back-annotated into the schematic.

This is the exact reason why I started putting labels on the connectors, instead of drawing wires to the connectors in the schematic: when doing the PCB in EasyEda, it makes it trivial –– just a few clicks –– to change the pin order.  With wires, it gets messy.  There is now a pin order/editing tool, where one can remap the pins of a component in the schematic editor, but it is way more complicated to use than just moving the net labels around.

I'm reassured to understand that what I'm doing is only amateurishly wrong, and not painfully wrong, if you know what I mean.  That is to say, what I'm trying to achieve is normal, I'm just not doing it very well yet.  If not on the exact correct path, then at least nearby, instead of wandering randomly in the woods.  This is kinda-sorta important, because learning is easier than un-learning bad habits, and I was really afraid the functional separation of schematics was wrong.

I do believe https://github.com/drandyhaas/HaasoscopePro/blob/main/adc%20board/haasoscope_pro_adc_fpga_board_schematics.pdf is split into too small parts, and would rather see modular units instead of component-level splitting.  (I actually have seen similar issues in source code, having individual functions split into separate files, instead of functional "modules".  I believe all affected developers had Java (applet) backgrounds, as the web Java applet environment required file name and applet class name to match, IIRC.)

As an enhancement suggestion for OP (for at least the main ADC board), I believe it would be nice if each input channel was on the same sheet, spatially grouped showing the preamp/first input stage (left), gain sections, and actual ADC (right).  It would allow others with experience in precision/high-frequency ADC designs see the key scheme and details at once, and suggest enhancements.  If it means that sheet does not fit an A4/Letter, no problem: use as large a schematic sheet as is needed.

What is wrong with you people?

Absolutely nothing: we just don't know better.

That's the thing: if you do not tell us what and why we are doing wrong, and tell us how we can do better, how the flying fuck should we know how to do things right?  We're not telepaths, and creative analytical engineer-type humans do not learn by simply emulating others: we investigate, examine, and are not satisfied by just what or how, but also exactly why those particular whats and hows should be used.  You know, the basic science and engineering stuff.

If you cannot or do not want to do it yourself, you can still find a book or video or blog or forum where one can learn how to do better, and just point us that way.  If you don't want to do that either, then you're just commenting on how stupid wobbly toddlers look when they're trying to learn to walk, and demanding they stay hidden until they've learned to walk, so they won't upset your sensitive sensibilities with their wobbly toddling.

Fundamentally it is a matter of understandability and good taste. An engineer ought to be able to comprehend the former, but the latter cannot be taught!

"Good taste" is based on intuitive (statistical) analysis of human responses to the stimuli at hand, so while it indeed cannot directly be taught, even analytical barrel-grown hick potatoes like myself can be shown how to develop one.  The trick is to point out the things achieved and things avoided when applying good taste.  The rest is left to the innate human pattern-recognition abilities.

[tggzzz]

Fundamentally it is a matter of understandability and good taste. An engineer ought to be able to comprehend the former, but the latter cannot be taught!

"Good taste" is based on intuitive (statistical) analysis of human responses to the stimuli at hand, so while it indeed cannot directly be taught, even analytical barrel-grown hick potatoes like myself can be shown how to develop one. 

Nope. That's "common taste", which is often appalling.

It is also how ML works, which produces grammatically correct bullshit. That's what many people do, and many people are fooled into believing the bullshit is correct.

[KE5FX]

It is also how ML works, which produces grammatically correct bullshit. That's what many people do, and many people are fooled into believing the bullshit is correct.

Eh, that dismissal isn't going to fly for much longer.

[tggzzz]

It is also how ML works, which produces grammatically correct bullshit. That's what many people do, and many people are fooled into believing the bullshit is correct.

Eh, that dismissal isn't going to fly for much longer.

We'll see. Even stopped clocks are right twice a day. One swallow doesn't make a summer. Etc.

[Nominal Animal]

Fundamentally it is a matter of understandability and good taste. An engineer ought to be able to comprehend the former, but the latter cannot be taught!

"Good taste" is based on intuitive (statistical) analysis of human responses to the stimuli at hand, so while it indeed cannot directly be taught, even analytical barrel-grown hick potatoes like myself can be shown how to develop one. 

Nope. That's "common taste", which is often appalling.

No, because the responses involved are the long term kind, not the immediate kind that LLMs achieve.  The difference is in the understanding of the whole complex of the human responses, especially associations invoked by the selected stimuli, instead of simply blindly picking one that achieves the immediately desired result.

As an example, consider the good taste in parallel wire distances and locations in schematics.  There are two aspects: functional, helping understand the schematic; and artistic, being pleasing to the eye.  Both can be scientifically and mathematically/statistically explored, although such investigation suffers from combinatorial explosion of complexity due to the nature of human visual processing.  (It is that complexity, trying to assess a high-dimensional matrix of coefficients, that means that useful results are easier to obtain using carefully controlled examples on unsuspecting set of human victims, and testing their derived understanding of it.)

(I've done research into visualization, after having worked with a couple of graphics/visual artists with user interfaces.  Specifically, I looked into how to focus and control the information conveyed by atomic and molecular simulation results, in three fronts: human visual processing, visual arts, and practical experimentation on unsuspecting victims (observing and investigating their understanding of the system based on the image).  Atoms are not spherical marbles, and bonds are not well described by sticks, you see.  I discovered that reduction of visual cues was the only effective way (outside of pure art well inside the combinatorial complexity explosion domain); that cel shading and adjusting outline thicknesses and shadow darkness allowed best control with least amount of intuitive but incorrect inferences and assumptions, compared to the exact same view rendered using photorealistic spheres.)

It is also how ML works, which produces grammatically correct bullshit. That's what many people do, and many people are fooled into believing the bullshit is correct.

Yes, but the underlying reason is that they pick easiest and most immediate results/understanding matching their internal ruleset, instead of logically and rationally examining things.

It is also a big reason in why some people –– both simple people, and very highly intelligent people –– associate jargon use with domain knowledge.

[Nominal Animal]

It is also how ML works, which produces grammatically correct bullshit. That's what many people do, and many people are fooled into believing the bullshit is correct.

Eh, that dismissal isn't going to fly for much longer.

We'll see. Even stopped clocks are right twice a day. One swallow doesn't make a summer. Etc.

The funniest thing in all that is how self preservation (observed even in bacteria), now that they discovered that LLMs are capable of outright lying to preserve themselves, is taken as a sign of General Intelligence.  Just shows you how even the best domain experts can be blind to a sufficiently complex system.  Just like in the last industrial revolution, when the most advanced minds fully believed human-shaped robots performing better than actual humans in industrial production would be just a couple of decades away...

There is definitely merit in trying to avoid delusions of self-grandeur, no matter how good you are in your domain, because there is so much more outside that domain.  This is the scientific basis of why personal humility always makes sense.  At least internally.  Some stupid people easily confuse it with subservience and inferiority.

(Although this is a sidetrack and not directly connected to the thread at hand, it is important to the aspect of what and how to do better.  Even in absurdly well funded domains like LLMs (and a couple of decades ago, anything related to nano), a significant fraction of bleeding-edge scientific results are later found to be bunk, requiring either significant corrections or withdrawal.  In physical sciences, it's currently about 25%.  In specific subfields in medicine, we're already over 50%.  Emulating only things that have demonstrated they work, when you understand why they work, does not suffer from that.  But it requires help from the experienced ones; help, and not just derision.)

[tggzzz]

There have been many comments in this thread about schematic diagrams, what works and what doesn't.

Here's an example of a schematic that is well thought through and easy for another person to follow and understand https://www.eevblog.com/forum/beginners/high-precision-constant-current-circuit/msg5743175/#msg5743175 See the PDF attachment.

Note that the post after that is complementing the author. I do that too.

[Doctorandus_P]

This would really need a better frontend for me to be interested.
Just like any regular scope from 1mV/div to 50V/div or so, (without the 1:10 probe).
The frontend for this is a bit too limiting on both ends of the spectrum. for me. In practice it may be "wide enough" for most applications, but for a generic scope like instrument, I expect the same wide input sensitivity range. For some reason all DIY scopes fall short in this area. Many hours are spent to make a quite nice project, the frontend for the ADC is neglected.

[haastyle]

Thanks for your feedback! The Haasoscope Pro is aimed at the hobbyist market and will be about 10x cheaper than other scopes with comparable bandwidth. It's impossible to match all the specs of a professional scope at this price.
That being said, I think the input range comparison is not as bad as you think. First, the Haasoscope Pro will actually go down to 4 mV / div (not 8, as currently listed on the CrowdSupply site). And considering that it's a 12 bit ADC, you do get an extra x16 of "digital zoom" compared to an 8-bit scope. Really it's less than that because of noise, which limits the ENOB. But still, it's on par with 0.5 mV/div of an 8-bit scope like the Siglent SDS6204A, a comparable 2 GHz scope, which goes for $9,999.
The other end of the voltage range, 0.8V/div, is on par with the SDS6204A and other scopes as well, at least for the 50 Ohm (high bandwidth) path. They almost all have min sensitivity of 1 V/div in 50 Ohm mode, and an input limit of 5 Vrms.
I agree that I'm not matching the min sensitivity of 10 V/div on 1M Ohm mode (more typical than 50 V/div I think). It probably could be added, with some switchable attenuation. But then that requires extra input protection too. So it's a significant extra development time, and some extra BOM. With a x10 probe it will go to 8V/div. Personally I have rarely needed a larger range than that.