Chorus I+II mod rate appears wrong

[jpcima]

Hi, the LFO rating at 15.175 Hz seems very suspicious.
We are making a Juno chorus and we have used your sound samples as a reference.

It has current source available and there is experimentation in our repository.

This is a rapid analysis of your 1+2 saw sample, and we deduce a completely different mod rate.
Prior to this analysis, I have deduced from a ear a similar rate, and the ~15 Hz value seemed to absurdly fast to be believable.

Link to analysis:
jpcima/rc-effect-playground#2 (comment)

[pendragon-andyh]

Thanks for highlighting this. I did some more measurements of this recently when I was working on JunoX. After upgrading my analyser software it looks like 9.24Hz.

For reference: The earlier Juno-6's Service Notes helpfully labels its frequencies as 0.4Hz, 0.67Hz and 8.06Hz (page 13).

[jpcima]

Thank you for rechecking. If I may ask, how have you deduced such a rate?
My previous experimental result gave 9.75, based on a sound sample of this repo. (cf link in OP)
So our measures seem in disagreement. Back then, I have used this program for analysing.
https://github.com/jpcima/rc-effect-playground/blob/master/tools/analyze.py

cc @SpotlightKid

[AlpesMachines]

in the service manual ... page 7.
T = 1 sec or 1,2 sec or 2 sec
http://manuals.fdiskc.com/flat/Roland%20Juno-60%20Service%20Notes%20(%20HI-RES%20).pdf

[pendragon-andyh]

@AlpesMachines - That page is right at the back of my PDF of the service notes. The "2 sec" (0.5Hz) and "1.2 sec" (0.83Hz) values approximately match up to my observations. The I+II value is way-off. In reality this is definitely somewhere between 9Hz and 10Hz (which is the reason that @jpcima raised this issue). Maybe the notes intended to say "0.1".

[pendragon-andyh]

@jpcima - Most of my analysis has been done using Sonic Visualizer - so they are very approximate. The nature of the "Chorus I+II" makes it quite difficult to measure visually.

I took a look at your Python script ... looks like your approach is very accurate. I revisited Sonic Visualizer again, and found that the "Peak Frequency Spectrum" pane can be re-scaled to give better visualization of this waveform. It clearly shows 34 cycles take approximately 3.49 seconds. This equates to about 9.74Hz. So if you don't mind then I will update the page to show 9.75Hz and give you credit.

Interestingly, I can see a slight variation of the delays between the left and right channels. It feels like about 0.0001s difference. This kind-of makes sense because the original analog electronics (pointed-to by @AlpesMachines) uses an LFO signal that is very approximately 1.6v peak-to-peak - and the nature of all-things-analog means this could quite easily by +0.85 to -0.75 for one of the channels. Maybe that is something that you could measure accurately if you have some spare time.

Also I can see that the delayed signal appears to have been low-pass-filtered slightly, and there seems to be some element of non-linearity in there also (maybe that can be emulated with some TanH saturation).

The chorus effect on my JunoX synth project is feeling a bit weedy - so I will continue to analyse this further.

[SpotlightKid]

Also I can see that the delayed signal appears to have been low-pass-filtered slightly, and there seems to be some element of non-linearity in there also

This is the effect of using bucket-brigade-devices to implement the delay (2 MN3009 chips on the Juno-60).

[jpcima]

So if you don't mind then I will update the page to show 9.75Hz and give you credit.

I don't mind sure, it's no need to ask me :)

Interestingly, I can see a slight variation of the delays between the left and right channels. It feels like about 0.0001s difference.

You would appear to be correct! I just ran the analysis again to give this a check, and did a manual fit of the parameter on gnuplot for each channel.

Left: o=0.032; f=9.75; d1=3.22e-3; d2=3.56e-3;

Right: o=0.032; f=9.75; d1=3.28e-3; d2=3.65e-3;

Which makes left channel 3.22 to 3.56 ms variation, right 3.28 to 3.65 ms.

Also I can see that the delayed signal appears to have been low-pass-filtered slightly, and there seems to be some element of non-linearity in there also (maybe that can be emulated with some TanH saturation).

As @SpotlightKid pointed out, this may be an effect of the delay based on BBD. It uses LPF for antialiasing steps in and out.
In our repo I have implemented the algorithm of a DAFX publication that has supposedly an accurate emulation of the filtered BBD used in Juno60. If that's of any use in your synth, feel free to reuse this.

[jpcima]

For I and II, delay range may be approximated to
Left: 1.54 to 5.15 ms
Right: 1.51 to 5.40 ms

This one is a more rough approximation than the previous, for a variety of reasons, so don't sue me for it. :)
Regardless, it follows the shape moderately well, except that the triangle is observed to be slightly asymmetric in up and down half-periods.

a plot below (disregard glitch of the peak analyzer..)