A Quadrature Power-Summed Compressor Sidechain For "The Uno Compressor"

[mediatechnology]

I just changed the above circuit to make the cap multiplier op amp (connected to the bottom of Ct) to a conventional inverting amp with a 10K Rfb.

Edit 4/4/24. If one were using a switch connected in a rheostat configuration:
For 600 dB/s Rin = ∞; 480 dB/s = 40KΩ; 360 dB/s = 15KΩ; 240 dB/s = 4K42Ω; 120 dB/s 2K87Ω; 60 dB/s = 1k1Ω.
R varies over a 36:1 range since, at the faster settings, fractions of Ct are being added.

Most people would want to use a stepped switch but someone using a pot could either use a log or inverse log pot depending on how one wants to define "more" (CW rotation) as either faster or slower.

There's not really an issue with clicking as a 2.2µF is always in circuit and 15° switches are usually MBB by default.

[emrr]

Doug what does NLC bring to the table for you that you don't get with RMS? Less distortion on low frequencies?

Missed this until now. Advantage over RMS mode - less audible overshoot on things like vocals, acoustic guitars. The recovery 'hiccup' isn't audible in complex waveforms, or is at least less intrusive than the RMS overshoot, which can then require peak control. Distortion isn't a thing that has been audibly appreciable, though it may be what keeps me away from the old Peak mode, or use of VAR much.

VAR - the only time I use that mode is for mix bus work, most similar to what Paul is doing, and then to tailor the pumping factor rhythmically for creative effect. Knobs seem to both live in a fairly narrow 8-11 o'clock range, no matter what.

[Gold]

VAR - the only time I use that mode is for mix bus work, most similar to what Paul is doing, and then to tailor the pumping factor rhythmically for creative effect. Knobs seem to both live in a fairly narrow 8-11 o'clock range, no matter what.

I’ve noticed that most of the time I’ve been using short attack/short release, medium attack/medium release or long attack long release. Kinda like a raw VCA. I’ve experimented with other settings only possible with VAR and they never sound good to me, even if the settings are doing what I expect. That’s why I think either the VAR circuit needs work or could be eliminated.

[mediatechnology]

I’ve noticed that most of the time I’ve been using short attack/short release, medium attack/medium release or long attack long release.

That's the beauty of the variable timing capacitor approach - the attack/release relationships remain proportional to each other as the Ct is varied. It also remains RMS and allows a smaller film cap to act like a much bigger one.

The same result could be achieved by switching or adding timing capacitors but with a synthesized variable capacitance it can be controlled by a pot and continuously-varied. And for those wanting fine adjustment granularity with a stepped switch, typically 12 or 24 positions, switching resistors is a lot less expensive than multiple film caps.

I never built the variable A/R but IIRC we settled on using a super slow OP07 op amp as a comparator to get it to be stable. I'm not crazy about using it.

I'm thinking the controls will be Threshold, Ratio, "Speed" and Makeup along with switches to bypass QPS and switch it to TPS for "color" as well as a Process Bypass and a 5 dB/10 dB meter scale switch.
I want to keep it simple.

I may experiment with a Peak timing but I'm not sure it would be that useful.

[mediatechnology]

Updated Schematic.

Gary Hebert of THAT suggested that I use a detector that is not "one Vbe up" at the timing capacitor node to reduce temperature drift between channels arising from the two channels being on different transistor arrays.
The original dBx 303 detector and 2252 are one Vbe up designs.
The later 43XX series have a 0V output at the input reference current.
We did this in the more complex RMS detector here: https://proaudiodesignforum.com/forum/p ... 149#p10149

In this version I bias the X2 gain stages down by one Vbe (at their outputs) with Q1 on array #3 biased at the timing current of approximately 8 µA.



The capacitance multiplier has also been re-drawn and matched transistors are used for the soft knee diode Q2 and Vbe reference Q1.

[mediatechnology]

I needed to assemble some boards for a client and had to move this off the bench for a few days. I should be getting back to it soon.

One of the issues I found in the original Pico, and some of the earlier 4301 app notes, is that the "Over Easy" pre-bias diode was pulled up by a large value resistor to provide a reference voltage to set pre-bias current. In the Pico it was 169K.
The reference voltage was then converted to current by a 10K resistor connected to the Threshold Detector's inverting input.
I found that the 10K significantly loaded the reference due to its high Thevenin equivalent.
Due to the loading three interacting factors set the pre-bias - the 169K pull-up, the 10K and diode Vf.
A stiff reference with pre-bias current set by the 10K provides more control.

On the newest schematic I use the added detector Vbe ref to provide a stiff source.
The Vbe reference is about 580 mV so the 10 K is contributing 58µA.
The actual pre-bias is set by the sum of currents into the Threshold Detector's inverting node.

The app notes also show a Threshold Control with a range more suitable for tracking.
I decided that +/-15 dBu would be a good range as a starting point with 0 in the middle.
So with the Threshold centered the 820K in the wiper is not contributing current.

At a 0 dBu input (-6 dB at the sidechain in) the detector output is calibrated to 0V so at that level the 4k99 is also not contributing current.

To keep Q2 from hard-conducting and to set the pre-bias to provide a useful dynamic resistance for Over Easy, the 332K pulls out about -45µA to set the actual diode pre-bias current to around 13µA.
Rdiode at 13 µA is 2KΩ (0.026/Id) making it about 4% of the value of the following 49K9 at the onset of Threshold.
As the level rises above Threshold, Rd drops.

This simple change seems to provide more control and the amount of pre-bias can be varied by either the 332K or 10K.

The next area is to look at Ratio calibration.

Promised some sound files that I have in long form.
Once I edit them down to around 1-2 minutes per cut I'll post them.

[mediatechnology]

After clearing off the workbench to build some boards and spring yard cleanup I'm finally back to working on the Uno Compressor.

During my break I thought a lot about the ratio control and wrote a spreadsheet to calculate trial values for a stepped switch and a means to linearize adjustment with a pot by slugging it.

Measuring a soft knee compressor's ratio slope is somewhat challenging because it varies as you move up and down the curve. If its measured far above the knee it's still got curvature and is an asymptote. So you have to pick your points on the curve and define the ratio based on them. Put another way slope is a constantly moving target.

For a hard knee feed-forward compressor operating above threshold ratio calculation is a simple thing. THAT in their app notes define ratio as dBin/dBout = 1/(1-G) where G is a control attenuation factor ranging from 0 to 1. When G=0 the ratio is 1:1, when G=1 there is infinite compression.

This seems simple enough until you put "G" on a linearly-scaled control where you find that 50% of the rotation range spans 1:1 to 2:1. While there is a lot to be said for having fine adjustment at low slopes things turn south in the final 50% where the slope goes from 2:1 to infinite. 4:1 is at 75% so the final 25% of rotation spans 4:1 to Inf:1. The rotational range becomes "compressed" both literally and figuratively.

Using a stepped switch we can get almost any taper we want. It would seem that all we need to do is pick the Ratio values we want and calculate target resistor values based on the Ratio formula given above. It turns out however that with a soft knee values need to be fudged a bit.

Slugging a pot is electrically easier - getting the desired taper requires quite a bit of thinking.

More...

[JR.]

Sounds like a handful....

JR

[mediatechnology]

The original Pico Stereo compressor used a threshold and ratio circuit where the pot element becomes part of a voltage divider.



At higher compression ratios small changes in control attenuation make for big changes in slope.
Without the optional slugging resistors in circuit the overall attenuation at higher ratios becomes highly dependent on the track resistance tolerance usually 20%. It's part of the voltage divider.
The optional slugging resistors complicate matters because they enter into the end-to-end track resistance which now varies with rotation.
Thirdly, the load on the soft knee diode changes with rotation.

In THAT app notes they often show the Ratio pot driven by an op amp output and, to reduce interaction and allow slugging, I also chose to buffer the diode and have the pot driven by an op amp output. I don't need a minimum component count design and would prefer accuracy at the expense of an op amp.

This is THAT DN115's approach:



The pot is actually loaded by R31||30K (the 30K is off page) which is actually 10K.
I decided to see how THAT's values work so I wrote a spreadsheet and plotted it with 1K43 and 10K "sluggers" in parallel with the 10K.



The attempt here was clearly to put 4:1 in the middle which gives good low ratio adjustment and good high ratio taper up to about 8:1.
I plotted the first ten steps of the pot but didn't plot the final "infinite" ratio.
It does a nice job of linearizing the Ratio curve and after spending the afternoon listening to it works really well.



The taper of the pot is quite interesting. The final "100%" value is shown which turns out to be about 15:1.

[emrr]

I don't recall any publicized application of the 'optional 1/2' resistors.