DIY 12AU7 compressor

> I could build a Gates/Collins/DISA thing and sub in a 12AU7... and as long as I'm not looking for heavy limiting it would work just fine?

You can build a "Collins/Gates-type" compressor with a lot less parts and power than they needed in the old days, if you accept that it won't be overload-proof. Or at least it sure looks that way on computer simulation.

This is not going to put Collins, Gates, Fairchild, RCA et al out of the compressor business, nor does it compete with the fine recreations and reinventions done by Gyraf, Manley, and others. It certainly will not take the over-levels that a few of the old boxes could (the GE BA6 would swallow 35dB-over without fuzz). It isn't going to protect you from an idiot-helper plugging a tape deck into a mike input (one of the things the old live-radio limiters had to handle with some grace).

This unit is pretty good to 12dB gain reduction, and won't stink bad for peaks 20dB over, but anything more is badly bent in an odd way. It expects you to have your peaks set pretty good already; then it will trim the peaks so the average and soft parts come up 6 or 12 dB.

And this is hardly the toughest DIY project on the block. A heck of a lot easier than recreating a Collins, actually simpler than the SSL. For a stereo mastering compressor: 4 line transformers, 2 12AU7, a 5532 chip and a TL074 chip, a 100VDC supply plus op-amp and heater supplies.

I have not built this, only simulated it, with simulated tube models known to be imperfect. My instinct based on some years messing with tubes tells me it will "work", and roughly as the model says. However I do not know how it will sound. It could be better than an old Gates because of simpler audio path. It could be worse because a 12AU7 is not an ideal gain-control amplifier.

So if you try it, start with a breadboard, don't commit a ton of money in a case, shielding, sexy meters and other frills. A design goal was "low-cost, no fancy parts". Ordinary 600Ω:600Ω, +18dBm transformers, not necessarily center-tapped. An all-tube line output stage to professional specs is very costly, a good low-noise chip like 5532 does the job fine. Tube plate voltage is low so the grid voltage needed is within reach of common chip op-amps. This low supply voltage also allows using a $17 power transformer.

From the left: R27 and V4 are a simulated signal generator; you would feed signal in at "in+" and "in-". It is not necessary for "in-" to be grounded; that was just to simplify the simulation. The input can be fully floating ("balanced" only better).

TX_in, the input transformer, is a 1:1 line transformer. A good 600Ω:600Ω, +18dBm transformer is the logical choice. If it has a center-tap, you can omit the R4 R33 3.3K resistors and put control voltage on the center tap, but it will work fine with the resistors splitting an un-tapped winding. For a lowest-cost breadboard, another workable transformer is the "Automotive Ground Loop Isolator" used in car-sound installation, such as RadioShack #: 270-054 $14.99, which gives two 2K:2K line transformers plus a shield can. You can find "modem transformers" very cheap: these may work for voice or initial tests but are likely to distort bass, and some will not like any DC current. Using the car-sound transformers, the sound should be good and the total breadboard parts cost is around $100.

The tube is 12AU7, universally available either as NOS or as new production, at affordable price ($10-$40) and likely to be available far into the future. There are many types similar to 12AU7; some of these will work the same, some won't give as much gain range as 12AU7.

TX_out, the plate transformer, is another 1:1 line transformer. The audio level here is lower than in TX_in, but the DC current may be very high, so this too should be a +18dBM type. Use the same type transformer as TX_in. R25, 100K, on TX_out secondary is only there to keep my simulator happy and should not be needed with real parts.

U3 is the output booster and buffer. It needs low noise at 1K input source impedance, plus the usual audio line-driver attributes of good drive with low distortion. 5532 should work fine, has been a favorite of many designers, and is cheap enough for reckless breadboarding. One 5532 dual can handle both channels of stereo.

U6 U7 and associated parts are a full-wave rectifier, with gain and bias. Signals levels here are high, so it needs to run on at least +/-15V. "Sound" is unimportant because it is not in the audio path. Most Bi-FET opamps are plenty fast enough to boost the peaks. One TL074 or TL084 quad can handle both channels of stereo, will work fine, and is cheap. Pot R39 allows a small range of threshold trim to allow for different 12AU7 tubes. Bias voltage V10 can be derived from any handy source, preferably regulated.

D7 D8 rectify the audio peaks into the time constant network: C1, Attack, and Release. For stereo, these three parts are common to both channels; all other parts are duplicated for each channel. Attack and Release are simulated as fixed resistors but will normally be user-knobs. Attack should be a 100K audio-taper pot with a 1K fixed series resistor, giving attack times from 100mSec to 1mSec. Release should be a 1Meg audio pot with 50K fixed to give release times 1Sec to 50mSec. For initial breadboarding, you can make Release = 500K and Attack = 5K, for a generally-useful time characteristic.

R3 1Ω is not needed; however this is a good place for a meter. Put in a 22Ω resistor with a 100Ω pot across it. Put a 1mA meter from pot wiper to ground. A DC millivoltmeter will work for breadboarding.

Pot R32 should be set in the center initially. Once you have the compressor basically working, feed sharp transients through it at 3dB to 6dB peak compression. Listen through a tone control that blocks highs and boosts bass. Adjust R32 for minimum "thump" (DC bias shift) as the compressor goes in and out of limiting.

This compressor was designed as a Mastering Limiter, the last stage before the mastering recorder. Therefore the output level is limited to a fixed standard level, matching the mastering recorder's input. You adjust the amount of compression by how much signal you push into the compressor. It has been scaled for "semi-Pro" operating level, -10dBv at 0VU or 2VRMS at digital full-scale. I feel that the old +8dBm level is absurd in this modern day, at least for small studios. The output of this limiter is designed not to exceed 2.8 volts peak (2VRMS) under large limiting. Pot R39 allows about +/-3dB of trim for tube variations.

The overall gain, when not compressing, is fixed at about 10:1 (20dB). The idea of a "unity gain" limiter is only valid when you hope to never use the limiter, and have plenty of headroom for the peaks you hope won't happen. When you plan to compress a lot of peaks, and don't have gobs of headroom like the old tube studio systems, a compressor/limiter needs to have gain. I assume that you have a level control on the device driving the compressor (usually your mastering mixer) where you can adjust "drive" and thus the amount of compression.

Simulator results for various input levels. Audio levels are Peak, not RMS, and in milliVolts unless otherwise noted. dB_in is referenced to the 3dB compression point, 200mV peak or 140mV RMS. dB_out is referenced to 2.828 volts peak or 2 volts RMS, the standard "-10dB" semi-Pro operating level. THD distortion numbers are only a rough guide. CV is the control voltage at the 12AU7 grids, mA is the current through both halves of the 12AU7. Both of these are likely to be only approximate at deep compression due to tube model inaccuracy.

12AU7, compressor5
   in   dBin    out   dBout    Gv     GR    dBgr     THD    CV     mA
   10   -16     100   -29      10     1.0     0      0.02   0.0    6
  100    -6     940    -9.6     9.4   0.9     1      0.1    0.7    5
* 200     0    1440    -5.9     7.2   0.7     3      0.2    2.2    3.3
  300     3    1710    -4.2     5.8   0.58    5      0.2    3.1    2.4
   1V    14    2400    -1.4     2.4   0.24   12      0.2    5.2    1.0
   3V    24    2840     0       0.9   0.09   21      2      6.6    0.5
   6V    30    2960    +0.4     0.5   0.05   26     15      8      0.2

Input levels may be up to 3V peak (2V RMS) with fairly low distortion (for a tube compressor). 6 volt peak level will be visibly distorted, but occasional transients to that level may not be audibly mangled. Input threshold for small (3dB) compression is 140mV RMS. I assume you will be asking 6 to 12 dB compression on peaks: input levels to 0.7V RMS and output levels within a few dB of maximum output level. It is not "brick wall": smack it with 6V for 26dB compression, and the output level may clip your recorder's digital input.

Input level across the top. "0dB" is an arbitrary 5dB of limiting.

Output level up the side. "0dB" is design maximum output level, 2VRMS.

The straight line is a non-compressing amplifier of the same gain, just for reference.

"Level" is the output for varying input.

THD is for a sine test constant level input at different input levels. With speech and music, while limiting, this is the distortion on peaks. The sound between peaks will be lower in level and distortion, generally well under 0.1%, and more like 0.02% for soft passages.

This is a very gentle compressor/limiter, as would be expected from using the grid curve with very little offset for our control law. Taking the 1dB GR point as the start of compression, and 1% THD as a maximum, it crams 24dB of input into 6dB of output. That averages out to 4:1 compression, but from -6 to 0 input is 2:1 compression, while from +8 to +18 input is 5:1. Or taking 3dB to 12dB of gain reduction, the ratio is 12/5= 2.4:1.

If you only take 12dB maximum gain reduction, distortion on peaks is about 0.3% and because of the soft curve the output level is 2dB lower than the ultimate limit. You give-up 2dB of S/N on your recorder, which is negligible. And if the singer suddenly screams 8 dB louder than you expect, it goes to 20dB gain reduction and 0dB output with 1% THD, no gross sounds.

A DIY compressor won't get built if it needs a costly power supply. If you live with 120VAC wall-power, you can power this with just one dual-primary 120-120:12-12 volt 30VA transformer like Parallax VPP24-1250, available from DigiKey as 237-1078-ND $16.22. Use one primary as the 120V wall-power input. Use the other "primary" as a secondary to give 160V DC. A stack of 2W resistors and 1W Zeners knock this down to regulated 100VDC (the output level depends on tube supply voltage, so it should be regulated). The 12V-12V windings can be rectified and filtered to +/-17VDC. By using an oversize filter cap on one side, with a dropping resistor, we can get 12V at 0.3A to power the 12AU7 heaters. Another stage of dropping resistor and filter cap gives about +/-16V to power the op-amps. Our 12VAC windings do not give enough DC voltage to feed a 15V regulator well, but the 5542 and TL074 will be fine with unregulated 16V.

If you must have an all-tube audio path, here's an output stage to replace the 5532. It will not feed a dozen loads like a 5532, nor will it give as low distortion. But it will drive several typical "-10dB" loads of 47KΩ or 25KΩ at around 0.2% THD, mostly sweet second harmonic.
PRR, May 2003