Poor Man's Tube Compressor

[Joechris]

A real poormans solution would be a Fairchild 663 that is a discrete opto that runs from tube heater supply.
http://audio.kubarth.com/fairchild/fairchild_663.jpg
Could be made with a Vactrol and a low voltage ic today, and use the poor man EQ makeup amp as...makeup.
Not a real tube solution thou....but only the opto in the signal chain

 

[gemini86]

That would make it a poorman's whatever compressor, rather than a tube compressor...

....and I hate the way that schematic is drawn. Just sayin.

 

[ruairioflaherty]

Yes, I have come across the PRR vari-mu but it has semiconductors in the signal path and I could never work out why he used a 12AU7 as a vari-mu tube. However, It makes me wonder if a long tailed pair could replace his output transformer which would save a lot of money.

PRRs design is often critiqued without remembering the context in which he sketched it.  It was intended as a design exercise and an introduction for beginners to the vari-mu type of compressor.  Of course there are elements than can be improved and he mentioned many in the initial notes - the goal was not ultimate performance at any cost it was useful compressor and learning experience for pennies.

Looking forward to any new approach, I have a yearning for a tube compressor.

 

[mylesgm]

I reckon the PRR is actually quite a useful tube comp and use it regularly though it is ripe for modifying.  In fact an updated version modified with some of the additions on this site would be a great project.  The addition of a balanced line amp stage for variable input/output (valve or solidstate), a sidechain filter, more flexible power supply etc should be relatively easy, most of the information is already drawn and tested its just a case of consolidating and doing a pcb. Cost would still be pretty low I expect.  Sadly I have no PCB design experience and so its beyond me but I'm a little surprised someone else hasn't done one yet.  I'm sure there'd be lots of takers...

 

[ruffrecords]

Prr's vari mu has semiconductors in the side chain not the signal path.

The circuit I saw had a 5532 as the line driver. Is there another version I don't know about?

Cheers

Ian

 

[ruffrecords]

A real poormans solution would be a Fairchild 663 that is a discrete opto that runs from tube heater supply.
http://audio.kubarth.com/fairchild/fairchild_663.jpg
Could be made with a Vactrol and a low voltage ic today, and use the poor man EQ makeup amp as...makeup.
Not a real tube solution thou....but only the opto in the signal chain

That is an interesting circuit. The audio path is entirely passive - just a resistor and the LDR. All the active stuff is in the side chain.

Cheers

Ian

 

[ruffrecords]

PRRs design is often critiqued without remembering the context in which he sketched it.  It was intended as a design exercise and an introduction for beginners to the vari-mu type of compressor.  Of course there are elements than can be improved and he mentioned many in the initial notes - the goal was not ultimate performance at any cost it was useful compressor and learning experience for pennies.

As always, context is everything.

Cheers

Ian

 

[Rob Flinn]

Prr's vari mu has semiconductors in the side chain not the signal path.

The circuit I saw had a 5532 as the line driver. Is there another version I don't know about?

Cheers

Ian

I think you're right it's been some time since I studied the circuit diagram.

 

[gemini86]

It would be super easy to pop Ian's my follower into the prr circuit in place of the 5532. But i suspect it's going to be hard to beat that circuit as far as cost goes.

 

[Hank Dussen]

http://headfonz.rutgers.edu/comp5/Comp5.html
At the bottom there's a suggestion for a tube-output stage for the PRR VariMu.

 

[PRR]

> that schemo I posted is a dud... the pots do nothing, and those who have built it say it just brings the level down a ton, with no actual compression

It sure is a lossy loser if loaded with typical modern 10K inputs..... it is a 470K pass-through with a 5Meg-60K shunt. Basic circuit principles. Dunno why you can't get it to work in sim where loads are typically infinite. Yes, the sidechain is ugly. And some sim tube models don't model the grid diode.

> PRR's vari mu has semiconductors in the side chain not the signal path.

I recall a blameless IC to bring up the level and drive modern low-Z loads. If you insist that sound never touch silicon, replacement should be obvious. If you also ask that it drive any box in the studio, it will be hot and costly.

> a high slope vari mu pentode which might well make it more suitable

Secret tip: you generally want LOW slope. What you generally want is high current for the Gm you get. An un-thrifty amplifier. You want large input swings yet still have plenty of output current left in deep GR.

> circuit I posted did not specify a tube type

However that circuit is heavy-NFB. Hi-NFB suggests a high-slope valve. However there is so much NFB, and any likely pentode has so much gain, that it may not be at all critical (A Good Thing).

> why he used a 12AU7 as a vari-mu

It generally does NOT matter a lot what tube you use. All tubes reduce gain as current is reduced. Generally along Child's Law, except with exponent of 1.2 to 1.4 (Child showed 3/2 or 1.5 for ideal geometry). The "Super Control" (6BA6 and kin) tubes deviate below -20dB, but at this point their plate current is so very-very small that we can't get decent output in audio loads (RF/IF loads are different; ponder).

12AU7 is hot-rodded from 6SN7 and a whole family of General Purpose Triodes. The things they did to spiff-up the high-current gain actually reduce the low-current gain. The deviation from Child's is significant in the first 20db GR; moreso than most other tubes people insist on sticking in.

When people try this or that other tube, not just for sexy/fad/copycat reason, they are probably not driving the sidechain well and mis-matching the tube to suit the sidechain.

> the cross coupled 6BA6's somehow cancel out the thump

A more-readable image with part values:
GE Ham News Volume 13, Number 6 page 6 (700KB PDF file)

It is Current-Steering. A topology better used with fixed input level and output turnable down to zero (was a VERY popular volume control for TV chips).

In a pentode, cathode current is just about K, G1, and G2. The division of plate and G2 current is (simplified) mostly about G3 voltage. Put audio in at K-G1 and the electron stream flows up past G2. Dinking G3 passes the stream to plate or steers it to G2. The sum of P and G2 current is nearly constant.

So as a simple plate-loaded amp, it would GR and thump.

Add a second matched pentode to the same loads with P and G2 swopped. As P(V1) current drops, G2(V2) current rises. Total load current is near constant. Thump tends to cancel.

> wonder if a long tailed pair could replace his output transformer

You said it yourself:

> The dc shift at the plate is very large - almost the entire supply range.

Devise a long-tail with rail-to-rail common-mode range. Don't reduce the voltage across the GR tube: the max input level (and therefore the S/N ratio) is roughly proportional to that voltage. Possible approaches would include a stiff tail (a pentode) and a second even higher rail for the longtail.

There is a cheap trick possible. If you set the bias and coupling well, the longtail will clip the initial transients that the sidechain is too slow to react for. Short neat clipping is inaudible, certainly preferable to gross overload in filmsound or disk-cutting or broadcast license citations, and the sidechain can be much cheaper.

> which would save a lot of money.

$17?

Ian, run all the variations. Cheap hi-THD thumpers and over-complicated mish-moshes. I don't think there are any tricks un-tried (there certainly are a LOT of tricks tried). You may find that a twin-triode VCA can do studio-quality work as well as any.

 

[ruffrecords]

Hi PRR,

Thanks for your comprehensive reply. That's the really nice thing about groups like this. You don't have to wonder what the author was thinking; he can tell you direct. There's just one point I would like to pick up on.  You said

It generally does NOT matter a lot what tube you use. All tubes reduce gain as current is reduced.

I admit I am a bit of a stickler for being clear about the meaning of technical terms and in this context I am not clear what you mean by gain. If you look at the 12AU7 published curves, mu only varies by about 25% from 2mA to 10mA  plate current; hardly enough to provide the required gain reduction. However, if you look at the value of plate resistance over the same range it varies threefold which with a 22K plate resistor will change the stage gain by about 6dB which seems to be borne out by the results you published. It is probably more accurate to define stage gain a gm*RL (where RL is the plate load resistor). Since gm = mu/rp and mu decreases as rp increases this means gm drops quite rapidly at low plate currents and so does the stage gain. The data sheets do not show mu, rp and gm below 2ma but your own results seem to indicate it continues much the same down to a small fraction of a mA plate current.

So, to answer my own question I guess by gain you mean stage gain.

I am not sure I would call this a vari-mu since mu does not vary very much. However, I suspect the 'real' vari-mu circuits are called such only because they use a tube with a variable mu and they do so as much because the more rapidly lowering mu means a greater reduction in gm than in a non vari-mu tube so the gain range is greater without resorting to very low plate currents.

The other interesting thing I note from your results is the high level of distortion that occurs as gain is reduced; presumably because ra is increasing. A vari-mu tube ought to improve on this simply because ra has to increase less for a given gain reduction because mu is falling more rapidly. This would imply that a pair of triode wired 6BA6 vari-mu pentodes might achieve better distortion figures in your circuit. I cannot find data for a triode wired 6BA6 so I guess my first experiment when some arrive will be to characterise them as triodes.

Again, many thanks for your detailed input.

Cheers

Ian

 

[Rob Flinn]

Ian

I like a few other people round here have successfully used a pair of 6BA6 triode wired instead of a 6386.  They sound really nice & to my old cloth ears there is no real difference in sound.

 

[ruffrecords]

Ian

I like a few other people round here have successfully used a pair of 6BA6 triode wired instead of a 6386.  They sound really nice & to my old cloth ears there is no real difference in sound.

That's encouraging. Comparing the data sheets of the two types they both exhibit similar characteristic curves.

Cheers

Ian

 

[PRR]

> by gain you mean stage gain

Not sure where I was confusing. Monetary gain? Solar gain? In most audio electronics, gain is gain, boxes have fixed gain, unless they have a gain-knob, or a little demon inside turning-up/down that gain-knob. Inside the box, there's fixed-gain stages (the postamp) and sometimes variable-gain stages (the tube VCA).

> I am not sure I would call this a vari-mu

I certainly did not call it that in the write-up!

Aside from being untrue, and futile(*), that phrase is used as a product name by a powerful studio-gear company.

It is a dandy figure of speech. But misleading.

((*) Mu does NOT vary enough to be much use; we must "fight" the quasi-constant nature of Mu by violating the simplification which makes it useful.)

> high level of distortion that occurs as gain is reduced; presumably because ra is increasing.

No, the plate resistance in that design is always MUCH higher than load resistance. That's how to make Mu drop out of the gain factors. And the ra/rl ratio does NOT directly affect distortion. Distortion rises when signal current becomes a "large fraction" of bias current. Swing 0.001mA signal in a tube flowing 10mA, THD is low. At any load resistance. Swing 1mA signal in a tube flowing 1mA, THD is gross.

Look at my sim numbers. 2.8V out (limit level) is 0.28mA in the 1K tube load. We must have 0.14mA in each side of the twin triode to cover 0.28mA peak. More for stray loss in the 22Ks, the tube plates, and the fact that each tube's curve is bent and push-pull is not perfect cancellation. Indeed THD is low at 1mA, not awful down to 0.5mA, soars at 0.2mA.

> cannot find data for a triode wired 6BA6

I think there are enough stray hints to get an idea.

But if you must dink with pentodes (losing the better matching of twin-triode bottles), then use them AS pentodes. The plate resistance is much higher. We don't have to slag the load way-low to swamp Rp variation. The higher load (and similar or greater currents) means larger output level, less post-amplification.

A drawback of pentode is partition noise. Can be 4db to 15db higher than a triode of similar vintage/construction and current. Cascoding is another path but getting away from the "Poor Man" concept.

Something to ponder. A given tube with reasonable operating conditions has a hiss level and an input overload level. Some forms of NFB will raise input overload but also FIX gain; we can't do that (with the common topologies). Pick some numbers. Say 10uV hiss and 1V input overload. Input dynamic range is 1V/10uV or 100db. Pretty good.

Now run it as a limiter. Say brick-wall 30db off the top. The hiss is still the same, 10uV. Input overload is 1V, so we must begin to limit 30db lower, 0.031V input. OUTput dynamic range will only be 0.031V/10uV or 70db. Good enough for almost any music, but a bit ugly on the spec-sheet in this world of 96db CDs. Taking a less brutal limit is a little better: 20:5 gives 85db.

Input hiss is set mostly by cathode current and grid spacing (Gm), plus pentode partition hiss and dirty oxides.

Input overload is basically plate (or screen) voltage divided by Mu. In general, we want large B+ and _low_ Mu. We would also like Mu to fall-off, but even the renowned 6386's Mu falls off 20:4 or 5:1 at most, and in the F660 probably only 3:1. (He did not run them wide-open, and if shut-down too far even the quad can't make line-level output.) Since we want more than 3:1 GR, and would like 10:1 or more (when a ""vari-Mu"" overloads it gets real ugly, so we need some margin), "variation of Mu" does NOT get us where we need to go.

We want variation of Gm, but over the top 20db of current the Gm/Ik curve for most tubes is nearly square-root (6386 is a bit faster from 2mA to 10mA; it IS a fairly special grid).

The main free variable (forgetting Poor Man) is tons of current. High idle current should be lower hiss voltage. Large shift from high current still leaves a non-teeny current to be driving the load (generally a postamp). This thought leads to 6L6 or 6080, both able to pass 100 times more current than we need, therefore may be turned-down 20dB and still pass signal. The problem here is that 100mA idle with 10% unbalance is a 10mA thump, so the "push-pull" plate transformer must tolerate big hits of unbalanced current, will be a large hunk of iron (now it does get costly, plus power iron and filtering).

 

[alexc]

Awesome ...

 

[ruffrecords]

I have just taken delivery of some 6BA6 and E99F tubes, both of which are remote cut off pentodes. Time to do some experiments.

I have also looked at the characteristics of some of the popular and commonly available tubes. The good old ECC88/6DJ8 looks promising. Its gm varies nicely over a wide range for a sensible set of plate currents and the frame grid construction should make matching easier.

Cheers

Ian

 

[PRR]

> good old ECC88/6DJ8 looks promising.

Be aware that genuine 6DJ8 are beloved by audiophiles; therefore rare and costly and drowning in "fakes" (re-labeled lesser tubes which often "work" in DJ sockets). You could probably source a couple for yourself. As a general public project, it seems dubious.

6DJ8 has VERY high Gm. Great for lowest-noise on weak-signal TV reception. IMHO, this is not useful in a compressor/limiter.

6BQ7 is a similar semi-remote twin triode, not sexy, readily available genuine NOS. 6BZ7 may be a hair better for our use, though some stores figure they are interchangeable (near-enough for TV/FM restoration). $4 a bottle.

 

[ruffrecords]

> good old ECC88/6DJ8 looks promising.

Be aware that genuine 6DJ8 are beloved by audiophiles; therefore rare and costly and drowning in "fakes" (re-labeled lesser tubes which often "work" in DJ sockets). You could probably source a couple for yourself.

What is your view on current production versions?

Cheers

ian

 

[MeToo2]

I posted elsewhere about how triode wired 6K4P-EV remote cut off pentodes, seemed to deliver similar tube curves to a 6386 (after a tip by lewilson).

http://i958.photobucket.com/albums/ae69/MeToo2_Prodigy/6k4p_triode_mode_data_plate.jpg
http://i958.photobucket.com/albums/ae69/MeToo2_Prodigy/6k4p_triode_mode_data_transfer.jpg
http://i958.photobucket.com/albums/ae69/MeToo2_Prodigy/compare6k4p-to-6386-plate.jpg
http://i958.photobucket.com/albums/ae69/MeToo2_Prodigy/compare6k4p-to-6386-transfer.jpg

Though they still (for the moment) cost pennies each in large quantities, so you can really match them into pairs quite well.
[I bought a box of 50]

You need a lot of negative grid bias to really turn them off.... up to -70V.

Charging up a side chain timing cap of say 1uF to that voltage in a short attack time of say 0.2mS - 1mS takes some power/current in the side chain. (I = C dV/dT)

So thinking out loud, a long tailed pair of 6k4p tubes (no iron on input, or else a Sowter 8504) + a fully balanced Broski Aikido low noise make up gain 6n1p/6n2p/ 6n6p (again no iron on output or an Edcor) together with a solid state side chain might be interesting (if you can figure out a way of avoiding compression thumps by balancing the long tailed pair well enough). Russian tubes & PIO caps are cheap if you select unpopular models...... then you can use a switched mode LED lighting controller as the heater supply to keep the cost of power transformers down. hmmmmm.