This looks like a cool project in the same spirit!
One of the hardest 10 weeks of my life, but also one of the most rewarding. Our team was one of the few that actually got it working in the end. I had to custom-make a gigantic breadboard to hold the entire circuit.
Today I still work in hardware, but mostly with digital circuits. While my analog knowledge has decayed over the last decade, that project and it's success gives me great confidence any time I have to deal with the domain.
If you want to take a look, here's a pretty similar project: https://www.analogmuseum.org/english/examples/bouncing_ball_...
Also what components did you have access to, just op amps?
That idea was shamelessly inspired by the soft fluidic robot some years back.
- To get position, 2 integrators were applied to an adjustable voltage representing gravity.
- The FETs were used to set initial states of the integrators.
- A comparator used to detect the table (y=0), flip the velocity and apply a scaling factor for restitution
The math was actually quite simple given its just the standard velocity equations — the challenge was in handling state changes in the electronics.
I looked around a little more and this video is a very close replica of what we built: https://www.youtube.com/watch?v=qt6RVrmvh-o
Even better, it would only be affected by relatively rare phenomena, such as vibration, temperature change, orientation and rotation.
Do you think about the analog qualities of your traces when laying things out? If so then the course was well taken.
In my observations I've found that too many digital engineers assume a differential pair will save them without actually fixing the impedance and parasitic issues. Particularly as the timings of things become so much more precise analog is so important. People forget that a digital circuit is just an analog one under the covers.
The way you usually run an analog computer is to put it into fast repeat mode (which they call REPF), where it cycles between initial condition mode and run mode. Outputs go to a scope. Then you can twiddle the knobs and see the output respond immediately.
The other modes are used mostly during setup and debug.
Hours of fun. Ages 14 and up.
[1] https://www.alibaba.com/product-detail/YIXINTAI-DSO138-Digit...
Veritasium explains it really well in general here (and demos the device) https://www.youtube.com/watch?v=GVsUOuSjvcg
I also learned heaps! (Including after a few weeks when the circuit stopped working properly because one of the relays started to work just a little slower than another one, heh.) If anyone's interested, https://blog.qiqitori.com/2024/08/implementing-tennis-for-tw...
In what concrete problems do you (or Veritasium) think analog computing could beat a GPU?
With synthesizers you think mostly in the frequency domain and speed is more important than accuracy. Integrators from AC become lowpass filters, adders become mixers*, multipliers become ring modulators, etc.
* In audio, a "mixer" is an adder. In RF a mixer is a multiplier.
TL;DR: No.
I looked into it hook one up to my rack and ultimately decided that even with the high price of euro modules it's still not really sensible/useful for what it includes. There's a lot of more useful modules you can try the same vibe with for euro though.
For analog logic, Maths from Make Noise is the obvious. ANA from Mystic Circuits is pretty neat too.
For digital logic (Combining both types of logic is very fun IMHO) O_c (or uO_c) running hemisphers has a mode to construct complex logic gate changes in the module, which can be fun.
Also, consider that a LPF is effectively an integrator and a HPF a differentiator and you already have 90% of it already, probably.
Solving (systems of) ODEs without the issues that can arise from numerical solving like numerical instability. Oh and it does that instantaneously.