>
cut down on the wasted heat
There IS a non-issue here. My threshold level is one silicon junction, and thus temperature sensitive. I am assuming "shirtsleeve" environment: varying no more than +/-10 degrees C. Over that range, we get a +/-22mV variation in a ~600mV threshold, 578mV to 622mV. This is a +/-0.32dB variation in theoretical threshold. Few studios work to that precision. If they do, they probably would not want to rely on a sloppy 12AU7 scheme. Because it is not a brick-wall limiter, a few-dB change in peak input levels will cause more than +/-0.3dB change in peak output levels, making thermal drift just irrelevant.
If NYDave needs something to put in the company attic, where he can't work more than a minute before heat castration, this may not be ideal. A +50 deg C rise would cause a -1.6dB change in output, which might confound the monthly testing.
However, we have a tube heater plus a Class A 2N3055 in the box. The threshold transistors Q2 Q3 should be well away from heat sources. If this rig were lashed too tight, even setting levels by ear, there could be enough drift from a 5-minute warm-up over the next hour to slightly upset your limiting action.
>
just stick a mosfet in there..that old BJT technology is out to pasture!
In the words of a notorious experimenter: "I reject your reality and substitute my own." Yes, FETs
are wonderful things. If you insist, use one at Q1 and fiddle the gate bias to put about 6V at the cathode (whatever you call it). Won't raise the input impedance unless you find a very low-C MOSFET. A jellybean JFET will be low-C but may have low enough Gm to lose gain (yeah, it is only unity gain but it can't be much less than unity).
In the Q2 Q3 Q4 strip: FETs won't reduce parts-count, don't have a law-of-nature turn-on voltage, don't have the high Gm of a BJT.
And the real reason: Neve didn't have MOSFETs, and didn't use JFETs. I'd like to keep it all of similar vintage, even in the places that don't "sound". (Yeah, that could mean another 2N3055 to regulate the +24V, and using good British transformers on each side of the 12AU7....)
I've nearly melted a 3GHz Pentium flogging a simulator. I've proved that 500,000 transistors can't keep up with four transistors (to sim a millisecond of audio takes the Pentium over a second, with much lower accuracy than 3906 jellybeans would really give). I am reminded how complex non-linear systems can be. In the end, I think my design above works with a minor change (resistors between phase-splitter and Q2 Q3 to reduce bias-shift effects), but I reduced most impedances to speed-up the attack. It is now in the speed class of a Fairchild, it can catch any likely audio peak. I don't think it needs to be anywhere near that fast on systems that clip politely. I think a 1K Attack resistor, ~1mS, is plenty fast for music into digital recorders or well-behaved speaker-amps, but you can reduce the attack resistor to zero and get near 50uS. Without melting Q4. (Using 12AU7 at 100V, instead of a low-Mu tube at 240V, makes a 10:1 difference in voltage and 100:1 difference in power, which is why a jellybean can do what it took two 6V6 to do.)
Here is the chalkboard tonight.
The other sketch was right-to-left, as sidechains are often drawn in context; this is left-to-right because I was considering it alone.
The V1 V2 circles are sim sources; in real life an unbalanced main output must be fed through a pot to C1.
For stereo, that pot must be 2-gang, and everything is duplicate up to the "Other_Channel" point. The stuff around Q4 is common to both channels.
I thought it was clever to derive the +12V from the 12AU7's cathode current. That may work but may crap-out at high gain-reduction when we starve the 12AU7. More to the point, it may be very slow to start-up from power-on. Run a 1K resistor from the +24V supply to a 12V Zener to ground. The +12V side of the Zener accepts 12AU7 cathode current and feeds Q4.
The Ratio pot may or may not be useful. The Attack pot should go over 50K for slow attack, to 1K for audibly instant attack (and towards zero if you must NOT ever overshoot, as in disk cutting). The Release pot can run 1Meg for slow to 50K for fast. If all the times are too fast or too slow, change C4.
C1, C2, C3 are electrolytic: assuming input is ground referenced, the + side of the caps faces Q1.
All resistor values may be replaced with handy 10% values (4K7 for "5K"). The R5 R6 R11 R12 set should be matched better than 10% for 1dB symmetry on + and - peaks. For stereo, the two R7s should be matched better than 10%. All the Q2 Q3 parts should be from the same box.
Wild idea to ponder: if you go to this much trouble, I suspect that
6BC8 will be better in some ways. You can't steal a 6BC8 from the nearest guitar amp, the way you can a 12AU7, but it is hardly expensive (and always good USA/EUR NOS).
About $5. It needs similar plate voltage at a little more current, it needs a little more heater power with different pin connections, and it needs 6V which is not as serendipitous as two 12AU7 with a +24VDC supply nearby. But distortion and noise could be less, gain a bit more, limiting action a little firmer for the same sidechain response.
![Electronics Soldering Iron Kit, [Upgraded] Soldering Iron 110V 90W LCD Digital Portable Soldering Kit 180-480℃(356-896℉), Welding Tool with ON/OFF Switch, Auto-sleep, Thermostatic Design](https://m.media-amazon.com/images/I/41gRDnlyfJS._SL500_.jpg)
