The project started on an italian magazine, that's why the first video is italian, but loading some more videos in english
No secrets of course, trying to describe it here.
Active line interfacing we said. That chips go from bal to unbal loosing 6dB, unbal to bal getting 6 dB back.
No input gain, we choose to use threshold here (as opposite to our Vastaso which uses input gain).
L R get mixed to mono, then through a common sidechain (this comp is always Linked LR, you need two PCBs if you want things like Mid/Side).
Audio is allpassed (to shape time response to freq ranges), then goes through a pair of boost/cut filters (low and high sidechain controls, to shape freq response to freq ranges).
Signal is double wave rectified with timing options (peak/rms, which actually are two different time constants for the rectifier's output).
Then the DC signal goes through a shaping network, which combines some "logifying" passes with some shaping passes. It is done by some diodes and resistors in the right places.
In the meantime, in the middle, you get timing and hard/soft control.
Timing is done by the most classic RC nets, including two double and one triple release constants (by stacking RCs).
Curve smoothing is done by setting the working level of a diode by in and out (to the diode) gain. It is somewhere between "logifying" and "log smoothing" passes, so it is not exactly what you get by, for example, That papers.
This DC signal gets scaled (ratio) and goes into a xx5001 (can't remember now, should be easy) PWM controller, working @500kHz and doing some shaping, too, while PWM range is limited bottom and top, as usual, to avoid 0% and 100% pwm.
Audio stereo signal is filtered, switched, filtered, switched, filtered. First pass because your audio could contain high freqs which could alias to audible range, second and third passes are after a double PWM switching. This is done to reduce the need for PWM range for the same gain reduction. Like using two cascaded VCAs to limit their needed GR range.
Switching is through a quadruple can't-remember-now cmos analog switch. Nothing fancy, after some tests I preferred the non ultra fast ones. You can see its shape @500KHz in the second video. Double pass compensates for its non super fast decay.
Compressed audio gets output gain (here it is gain, not GR compensation) and is mixed to dry for simple and precise parallel compression.
Input to sidechain can be switched from comp input mixed to mono to compressed audio output (feedback mode).
Output gain can be wired in several ways. I prefer two separate gain knobs for L and R but gain and balance is an optional wiring. Pots don't give good stereo alignment, you know.
Ext sidechain input is exactly like L and R ins. It goes to the sidechain starting from the output of LR mix, so it passes through all of the sidechain. So, all controls are active, but the feedback mode which could have different effect because it is fw from ext sidechain input and fb from audio output. Uhm, perhaps it just switches to internal, feedback...
Here you can have fun and extreme control on the sidechain if you start using something else on the aux audio before going to ext sidechain input, because it is an audio input and not some kind of control voltage.
The meter is a double input, 40 steps bar meter.
You can wire DC (for gain reduction), which travels right to left, linear (scale to log is done by graphics, to get a good zooming on low GR), and audio (for level), which goes left to right and is rectified, transcended to log and scaled, slowed down with switchable option. You can set the combinations of readouts through the 3x4 rotary switch (peak, VU, GR). The main choice to get is what to show while in GR mode. I like GR + input level. In the video it is wired for GR+disabled.
What else?
Ah, all filters are Sallen-Key second order lowpass tuned close to 25 kHz