Debugging VCO Pitch unstability continued 3

Aaaaand testing continues. NB: Most measurements are now with the VCO connected to power ground.

Reducing pw to 25% to see if it affects the noise

The noise clearly follows the DCO even when pw changes, meaning it is highly likely that a lot of the noise comes from the pulse wave generation. It is a bit delayed/rounded off though.

Turning up DCO at mixer VCO to see if noise increases

It doesn't look like it affects the noise notably

Turning up square sub oscillators 1 and 2 

It's a bit hard to tell, but it looks like the noise drops whenever the pulse wave drops, even if the output from the waveform mixer has the sub oscillator cancelling out every second drop.

With a longer sample time however, we can see that it looks like the noise rises to a higher level when the mixed signal rises too. Not quite sure how to interpret it but it does seem like the sub oscillator has some effect on the noise.

Removing the mixer pcb

Noise is highly visible even if no waveforms reach the output

Trying to measure noise on the pitch CV 

It's a bit hard to measure the noise. There are huge spikes during SPI traffic, but this is probably due to crosstalk within the Logic Pro. 

Measuring again without SPI probes (those are needed at the moment for stable 50MHz SPI)

Much better, the SPI noise is mostly gone. As a side note, the CVs become unstable as the DAC sees wrong values, for example the VCA CVs for the DCOs (probably):

Comparing DCO output and Pitch CV (with SPI probes on):

Markers 0, 1, 2, 3 are at the start of the DCO edges. It is possible to see that the noise in the pitch CV follows (albeit slightly delayed) the same pattern.

Removing the SPI probes

The stepepd noise becomes more visible. It is clearly there still (though - is it there or is it on the ground of the Logic Pro? Nobody knows!

Zooming in on the CV noise

The CV noise seems to jump up and down approximately 10mV. If run through a 39k resistor, it will have the same effect as 100k/39k = 2.56*10mV =26mV would have on a 100k input (e.g. a 1V/oct input). One octave is 1200 cents, so this should result in a noise of about 31cents pole to pole. This matches well with what we see.

Still, I am not sure exactly WHERE the noise is

Moving VCO GND back to the mainboard. 

Connecting the ground of the Logic Pro to the GND on the VCO, measuring pitch CV at input to mainboard to measure more accurately what the VCO sees. Presumably, since the pitch CV summer is ground referenced, the difference between the CV and GND at this point is what the VCO sees - whether it is the CV or the GND that carries the noise. The 10mV difference is still there.

Controlling that the probes measures a silent GND

The CV probe and probe GND is now connected where the GND is tapped for the CV board. Strangely, we still see the DCO. This may be crosstalk in the Logic Pro though.

Disconnecting the output probe

This is weird, we can still slightly see the DCO but definitely not as clearly

Disconnect probes from the filter (!)

I've had probes connected to the JP6 filter the whole time (whoops). Disconnecting them changed nothing though, or not for the better at least.

Disconnecting the SPI probes again

Smaller variation but still periodic noise.

Reconnecting SPI, tapping GND and CV from same probe pair 

Did the noise go away? It's hard to tell. It does look less noisy though. We're up against the limits of what the Logic Pro can measure but it looks like using the ground probe from the same pair is a good idea.

I'd say the ground measurements are inconclusive. There is definitely noise there but I can't figure out where it gets into the measurements.

Using CV generated by my modular

The CV generated externally looks clean. It has a lot of noise, but it doesn't seem to be periodical

Using CV from modular to control VCO

Now, this was surprising! This is the worst noise so far, +/-40 cents. Perhaps the VCO VC and VCO GND both contain parts of the same noise, but of inverse phase, so it partially cancels out? 

Revert to normal but disconnect linear FM

Back to the 'normal' noise again (20 cents this time, but remember that we moved the VCO ground)

Connecting both waveshapers and VCO to power ground 

 This is also a bit interesting. The max noise is not reduced (sttill around 20/20), but there is a second signal bleeding through, modulating the noise at 7-8 times the DCO (between 2330 and 2660Hz). Also, at the negative going poles there are half the peaks. What is going on here, is this some sort of self modulation? 

 

Conclusion

We're really not that much closer to a solution here. 

Star ground from power may not be helping as other parts of the circuit will reference the noisy ground and that may even result in a higher noise since none of it cancels out. 

The noise seems to mostly follow the pulse wave, though something else may also affect it

I am inconclusive as to where the noise propagates. The one thing I haven't checked is the power lines

Noise and crosstalk on probes seem to mess with the result (and especially disconnecting the SPI probes has a great effect) though I don't think this affects the frequency modulating noise, just the measurements of noise on a signal/gnd line. The frequency modulation is still the best way to see what the VCO actually experiences.

Next

The most important thing left to try is rebuilding the comparator parts of the waveshaper, remove the original waveshapers and see if the noise is still there. If it is, we can try to isolate the noise, and if we manage to do that, try to improve it.