etheory
Well-known member
I've been looking into compressor side-chain designs for some of my own compressor designs and have become interested in how you can make a faster and smoother side-chain.
I've noticed that in most designs the main source of distortion in the VCA is the superposition of the CV-rectifier ripple on top of the audio signal at the output, and the main issue is the discontinuity at the rectification "change of direction" that you want to "get rid of".
Here-in is the infamous real-world trade-off part
The more you filter the CV after rectification, the less your multiplied distortion (a vca is, after all, just a multiplier, hence your output distortion is only as good as your operands), but the slower your attack time, due to the increased filtering increasing the filter step response time.
So you have two options. Filter less (but that increases distortion), or derive "more push" from your signal to charge your cap faster, or, indeed, maybe both?
Here is an interesting approach I came up with which might never leave the simulator (due to issues I will discuss in a bit), and it is very likely not an original concept, but it is interesting and combines the idea of fast attack with a smooth signal. Well, kinda....
Take the area of a sine wave from 0 to π. The area under this curve is 1. This is the amount of "push" we have to charge a cap over π units of time.
How can we improve this?
Well, if we phase shift the signal by 90 degrees (π/2 time delay), or shift the signal -45 degrees (-π/4 time delay) to 45 degrees (π/4 time delay), and rectify both the phase shifted signals, then we have more area under our curve per unit time. Hence our integral area increases, and the amount of energy we have to charge the cap also increases.
However, our CV frequency has now also doubled, and we have twice as many discontinuities to remove (twice the zero crossings per unit time). So there is a trade-off here I'm still figuring out.
But, a schematic speaks a thousand equations, so, this is the idea (I've referenced back to the original sources for the sub-circuits in the schematic).
The way it works is this. Take an input (shown in green). Derive a 100Hz to 10kHz 1% accurate -45 degree and 45 degree split phase-shifted signal representation for each signal, and then rectify each independently (shown in red and blue).
From here, charge a cap with the summation of the two signals.
This would be an interesting thing to try on full-frequency audio signals, as no doubt the use of phase shifters will make full frequency content much messier than a nice sine wave. However the reference at http://webpages.charter.net/wa1sov/technical/allpass/allpass.html suggests the results should be OK.
Downsides:
A LOT of opamps, and probably a lot of noise (but most of that should be filtered out).
Less correlation to the original signal, since we are "generating" data instead of working with it directly, so no doubt this will introduce response lag (cause you can't look forwards in time, even in the analog domain).
Thoughts?
I've noticed that in most designs the main source of distortion in the VCA is the superposition of the CV-rectifier ripple on top of the audio signal at the output, and the main issue is the discontinuity at the rectification "change of direction" that you want to "get rid of".
Here-in is the infamous real-world trade-off part
The more you filter the CV after rectification, the less your multiplied distortion (a vca is, after all, just a multiplier, hence your output distortion is only as good as your operands), but the slower your attack time, due to the increased filtering increasing the filter step response time.
So you have two options. Filter less (but that increases distortion), or derive "more push" from your signal to charge your cap faster, or, indeed, maybe both?
Here is an interesting approach I came up with which might never leave the simulator (due to issues I will discuss in a bit), and it is very likely not an original concept, but it is interesting and combines the idea of fast attack with a smooth signal. Well, kinda....
Take the area of a sine wave from 0 to π. The area under this curve is 1. This is the amount of "push" we have to charge a cap over π units of time.
How can we improve this?
Well, if we phase shift the signal by 90 degrees (π/2 time delay), or shift the signal -45 degrees (-π/4 time delay) to 45 degrees (π/4 time delay), and rectify both the phase shifted signals, then we have more area under our curve per unit time. Hence our integral area increases, and the amount of energy we have to charge the cap also increases.
However, our CV frequency has now also doubled, and we have twice as many discontinuities to remove (twice the zero crossings per unit time). So there is a trade-off here I'm still figuring out.
But, a schematic speaks a thousand equations, so, this is the idea (I've referenced back to the original sources for the sub-circuits in the schematic).
The way it works is this. Take an input (shown in green). Derive a 100Hz to 10kHz 1% accurate -45 degree and 45 degree split phase-shifted signal representation for each signal, and then rectify each independently (shown in red and blue).
From here, charge a cap with the summation of the two signals.
This would be an interesting thing to try on full-frequency audio signals, as no doubt the use of phase shifters will make full frequency content much messier than a nice sine wave. However the reference at http://webpages.charter.net/wa1sov/technical/allpass/allpass.html suggests the results should be OK.
Downsides:
A LOT of opamps, and probably a lot of noise (but most of that should be filtered out).
Less correlation to the original signal, since we are "generating" data instead of working with it directly, so no doubt this will introduce response lag (cause you can't look forwards in time, even in the analog domain).
Thoughts?