new compressor design..diode based studio comp

[Johan]

..not a stompbox, but since I used my gallery here to store it, I feel I must share it with this crowd too..
http://www.aronnelson.com/gallery/main.php?g2_view=core.DownloadItem&g2_itemId=40958&g2_serialNumber=1

..sound with accoustic guitar..(sm57 laying on table, noisy background.no other FX. same track twice first 20 sec uncompressed for reference..quick and dirty, take it for what it is..)
http://www.aronnelson.com/DIYFiles/up/compressortest.mp3

I'll be out for a few days, but will be happy to answer any Q's when I get back..
j

[majormono]

Is this the same thing I saw over at prodigy-pro a while ago? Looks familiar IIRC...

[Johan]

Is this the same thing I saw over at prodigy-pro a while ago? Looks familiar IIRC...

yes it is...

[Nasse]

Heja Sverige and thank you again, interesting circuit.

I have old papers somewhere about diode circuits and tried similar voltage controlled attenuator some years ago. My paper suggests putting a suitable cap across diodes, for some distortion cancellation (no differential amp in my circuit doin that, it was only single ended arrangement in my circuit) and puttin more diodes in series if more headroom before clippin is needed. I tried with four and imho it was quite nice voltage controlled attenuator and low distortion too. But sure guys at prodigy know more.

I think that diode arrangement is wonderful diy building block for fx.

[R.G.]

The diode-ring modulator is a very old circuit. It's neat, simple, and tempting.

I've tried to mod it to pedal usage several times, and found that it's hard to get it completely sounding good, although modest results are easy.

About the best implementation I've found is the tremolo in the Thomas Organ Vox amplifiers, or the "big head" models at least. These seem to work well in the environment of those heads at least.

The problem is that since it's using diode resistance as the basis for modulating, you're limited to quite small signal levels if you don't want the tremolo to also distort. This happens to be how it's used in the Thomas Vox stuff.

The incremental resistance of a diode varies from quite high in the region before it begins significant conduction to quite low after you pass the conduction knee. The conduction knee is that region where the diode is beginning to conduct but is not yet conducting hard. In that region the resistance changes from perhaps 1M to 100K down to maybe a few ohms. The catch is that the voltage across the diode from the signal changes the resistance as well. So you need to keep the voltage change small to keep it from distorting by self-modulating the diode. 10mV to 25mV is usually a good place to start.

In all diode rings, the input side feeds a pair of diodes which are connected through control nodes to an output pair of diodes. The signal path is from the input node, through the control nodes and out to the output node, passing two diodes on the positive side and two diodes on the negative side. The control nodes have to be fed from a complementary current source or a complementary voltage source through high resistances to make them look like current sources. If this doesn't happen, the signal is attenuated by the impedance of the control sources bleeding signal off. Thomas Vox used 1M resistors. This design looks to use 4.7M, and probably a bigger voltage drive.

A secondary problem is that of control signal cancellation. If you feed this from a complementary control source, to a first order the control signals cancel from top to bottom and theoretically the input and output nodes should have zero control signal feed through.

In theory. In practice, you're fighting it every step of the way, because the signal you're controlling has to be so small. Practical levels of signal are in the tens of millivolts, practical levels of control signals are in the volts or more, so there's a 20 db difference in levels the wrong way to start with. You'd like to get control signal feed through down by over 60db compared to the signal, so there's maybe 80 db (that's a ratio of a million to one) of control signal to get rid of. You have to get rid of both the common mode (i.e. mismatch between the top and bottom control feeds) and the differential mode (i.e. mismatch between the diodes in the ring) differences.

Thomas fed the top and bottom of their rings from the outputs of a transistor phase inverter, which was OK-ish. That's a match to within maybe 1%, or most of your 20db for the control signal. Doing this with opamps or with some kind of complementary current mirror setup (Hmmm... how did I think of that one?) gets most of the rest of the initial 20db out. This still leaves you with suppressing another 60db. In this incarnation, it appears to be done with a full-blown differential input instrumentation amplifier (those three opamps), but this design gives up the suppression from matching the control signal feeds by grounding one side of the ring through a resistor. It would be much better to feed it a control signal in a complementary fashion from the top and bottom of the ring.

Thomas did a crude version of cancellation by feeding a buffered and attenuated version of the input to the output of the ring. This produced a further cancellation of the signal itself to get to a lower signal level on signal valleys.

Thomas used a cap across the diode ring to keep noise down (the high impedance feeds are quite noisy).

If I were building one today, I'd use a complementary control feed, many more diodes (bigger signal in the ring before distortion) maybe with a matched monolithic eight-diode array or two if I could find one, and I'd use the instrumentation amplifier to cancel control voltage offsets as well.

Diode rings are a neat idea, but it takes some work to get them to today's levels of performance.

[puretube]

Some did the diode-modulating with "tubes"... 

symmetrically controlled asymmetric signalpath, though...

[George Giblet]

The problem with those is when the diodes aren't matched you get a lot of feedthrough.

Old ways of getting a better matches was to use CA30xx transistors arrays.

These days you might be able to use dual diodes with SMT packages, one for the top and one for the bottom.  I'm speculating the diode pairs may match better but I haven't actually done any measurements.  IIRC RG had a build you own transconductance amp project perhaps the those transistors are matched better?


(IMHO, any tube based ckt is doomed to fail because of the matching problem - that goes for diode bridges and differential amplifiers.)

[R.G.]

The problem with those is when the diodes aren't matched you get a lot of feedthrough.

Yep. All circuits like this have that problem. In fact, the tremolo in the Thomas Vox amps is mildly astounding for how well it works. I've held off on reducing it to a stompbox for years because I didn't want to get into trying to make sure I could make it non-feedthrough.

Old ways of getting a better matches was to use CA30xx transistors arrays. These days you might be able to use dual diodes with SMT packages, one for the top and one for the bottom.  I'm speculating the diode pairs may match better but I haven't actually done any measurements.  

There are - or were - diode arrays intended for use with diode-ring modulators (which is different from diode ring-modulators!) which were monolithic arrays specifically for matching. Monolithic transistor arrays would let you do this.

IIRC RG had a build you own transconductance amp project perhaps the those transistors are matched better?

Nope. If you can get monolithic arrays, it gets better, but it's still very difficult for discretes to do matching. There are some matched pairs available from ALD, THAT, and a few other places.

(IMHO, any tube based ckt is doomed to fail because of the matching problem - that goes for diode bridges and differential amplifiers.)

Matching is an issue on all of them. The diode modulators for RF were first. I'm not surprised to see the 1961 patent for a tube diode-ring modulator; a similar circuit was used to impress audio on a carrier before that.

If you follow the path, diodes are OK-ish but lacking. But we can do transistor-connected diodes, which are more "ideal" than real diodes. But then if you have matched transistors, you could ditch the diode connection and use them as a current-modulated diffamp. In fact, since you'd need four of them, you could do a dual diffamp, and get second order cancellation... in fact... hmmm... and in about ten minutes, you re-invent the 1496 modulator and eventually the Gilbert Cell multipler.  I've caught myself doing this about eight times.

[George Giblet]

> diode arrays intended for use with diode-ring modulators (

Yes, I believe the lingo is "quad matched", at least for the HP devices.

> Nope. If you can get monolithic arrays, it gets better

I thought you had a DIY OTA project. It had a little board with a couple of SMD monolithic transistors.  Maybe it was someone else.

> I've caught myself doing this about eight times.

Yeah, designing is a vicious circle.   In fact the people at THAT saw all this long ago, they have the volumes to get the matching issues solved and down on chip - then we all try to find another way to do it!


[Forgot to mention, I had a look at some datasheets for SMT diode pairs and they quote 20mV match on the match part suffixes.   Still not that great for audio but better than nothing.]

[puretube]

George: Son`m`Ota ?

[alanlan]

The Neve 33609 uses a diode bridge.  People seem to like it!  Obviously there is more to it than just the bridge, but it goes to show that it is certainly a viable design for a studio use.

[puretube]

The Neve 33609 uses a diode bridge.  People seem to like it!  Obviously there is more to it than just the bridge, but it goes to show that it is certainly a viable design for a studio use.

http://www.celestial.com.au/~rosswood/data/33609c/33609_ca.jpg

[Nasse]

http://i837.photobucket.com/albums/zz293/Nasse2/diddex-1.jpg

I had a drawing I made few years ago, did a vero few times and tried it as a tremolo part and voltage controlled level

[Johan]

the Neve 33609 was indeed the inspiration for messing with this in the first place...that one just uses soo many parts thou..wanted to simplify things..
I know diode compressors aren't perfect in any way...it's more about a sound...so is pretty much all other compression technices too...if you want perfect, go OTA...or go downtown and track down a perfectly good used Boring-er for less money than it costs to build one..

Nasse, that distortion canselling cap wont work in a compressor. together with the resistor it creates a timeconstant, in your case attack time around 22seconds...no good for fast compression ( time=RxF  47000x0.00047=22,09)

the main problem with diode compression, as RG allready stated is that the signal needs to be so small...noise becomes a reall issiu...this one attenuates the input about 30db, so a 0db (0.775volt) becomes no more than ~25mV..enough not to get clipped. then you need to bring it back up without bringing up the residues of the controllvoltage(thumping) and niose..thats why it needs a fully ballanced differential amplifier..
j

[alanlan]

Yes, CMR acronym of choice

[George Giblet]

George: Son`m`Ota ?

Thanks Puretube - it wasn't a dream after all!

[puretube]

Johan: how about "higher-threshold-clipping" (e.g.: LED) diodes,
or a few in series?

[Johan]

Johan: how about "higher-threshold-clipping" (e.g.: LED) diodes,
or a few in series?

do you mean so the signal doesnt have to be so weak?...Hmmm..I don't know..
I don't know if LEDs behave the same way as regular diodes in this application, if they do, it would work. Several series diodes might work, but just thinking about it, it would seem that the risk of residual noise from missmatched diodes would increase significantly?...
..as an old repairman, I like designes ( and try to design my own gizmos ) that aren't too fussy about the exact component and component matching. with just two diodes on each transformer-leg, "from the same roll" seems to be good enough..with several in series, I'm guessing a more carefull matching would be necessary?...
j

[puretube]

Yes: high-threshold diodes (LED or several in series), but without attenuating the signal before they hit the "quartet"...
IMHO, it would give a much higher S/N ratio 

(I haven`t played around with these kind of "clean" circuits, though - it`s just an idea!)