PMsix61 Limiter DIY

[alexc]

Thanks guys - the encouragement helps when one thinks it's all gone to the dogs.

I've been rebuttoning up the signal amp and now have it back where it was, running higer currents and so on.

Right now, I have signal amp net gain of 16dB and have incurred noise floor worsening of around 10dB, spl 20-20K.
It's not too bad but I'm hoping that's the worst of it.

Right now, with a 'perfect' input signal, the noise floor is completely independent of the input t-pad attenuator setting.
That will change when I feed it with a real world signal that already has it's own levels of mains harmonics noise.

I'll be spending more time to see if I can find improvements but I'm pretty sure that's where it is at.
Most of it is coming from the UTC, but of course, one expects a rising noise floor as one adds gain.

----------------

Things that makes quite a difference are :

- the output transformer secondary termination
- input transformer secondary balancing network termination

The other thing  I noticed was removing pairs of GR tubes worsened the noise floor even as gain was reduced.
ie. 1 pair  of tubes and noise floor was worst even though gain was least
    Adding pairs reduced noise floor even as gain was increased

All related to the impedance of the combined triode-connected plates decreasing as pairs are added I suppose.
I guesstimate the plates are probably something like 4K or so, so 4 in parallel is 1K or so. (in the 'idle' case)
The 600R terminating means 8K plate-to-plate or 4K each phase (class A). So connecting 1K to 4K which is better than connecting 4K to 4K.

Or something like that  ;D

--

The output transformer loading at 600R is my default setting. Decreasing the loading of the traffo by increasing the terminating resistance increases the gain and therefore also worsens the noise floor.

There is something like 3.5dB increase in net gain going from 600R to 10K and higher. That does also incur about the same in noise floor worsening. At this point, I'm leaving the 600R as is; later on I wiill revisit when I get further down the testing track.

--

The input traffo balancing network which is required as I have no winding CT on the hi-Z side (and no knowledge of inductance and resistance as yet), plays a big role in loading down the input side of the signal amp.

In general, I would expect that a termination value similar to the spec'd 'typical use' impedance of 15K is the ballpark way to go.
Too low a value and the input traffo secondary appears too heavy a load and the input signal is attenuated. Not to mention that the CV is also kind of attenuated by the heavy load (light resistance).

Too high and the noise at the grids, which is reasonably high (around 1.5mVpp) from EM into the UTC worsens the noise floor more than the increased input level provides. Normally one would expect higher grid-ground resistances. With 4 pairs of grids, that is probably reduced I suppose.

I did compare the results of lower and higher, taking into account the gains achieved and noise measured, and for now, I conclude that lower is better subject to maintaining the CV. Again, I will be examining this more as I get further on.

--

So, given I am deviating a long, long way from the original, for many, many reasons, the thing is to decide what gain staging to run throughout the signal amp. Not to mention the setting of the 'idle' bias and current.

Things to consider are in deciding what gain to run relate to :

- amount of GR I want to achieve  vs  the limit of acceptable noise floor
  ie. tradeoff of output level for amount of GR limited by noise

- how hot of an input applied  vs  'overdriving' the grids in the 'no GR' case
  ie. tradeoff of 'no GR' distortion for 'deep GR' level achievable

All this stuff happening at once so a balance is needed across all of these. My knowledge is pretty slim in reality, but I'm gradually getting an intuitive feel of all the competing things - which is a bit part of the motivation for experimenting 

Anyway, onwards and upwards. I do think I have a worst case performance now which is acceptable.

Things are significantly better than they were a couple of weeks ago

 

[alexc]

Decided to try changing the t-pad attenuator on the input traffo primary for a dual-rotary attenuator on the input traffo secondary.
I think that may well improve my signal-noise ratio.

Also I have some UTC A33 20dB  shields on the way so that should help a bit too I think. Lucky to pick up a few for a reasonable price.

Plenty of things to get on with while I wait for these parts to materialise. I do have some instability in the control amp when connecting to the signal amp so I need to work thru that.

----

On the subjective side, I've been playing some guitars thru the signal amp at the current idle settings.

With the full circuit wired in but no CV connected, I'm seeing a noise floor of around 66dBu for a net gain of 16dB.

I'm finding I'm using the 'level' t-pad at some 15dB of atten and the input is a good strong signal from one of my channel strips.
Probably around +4dBu to +6dBu at the XLR in.

No audible  noise at all - I find around -60dBu is where I begin to hear hum at my reasonably-loud-but-not-crazy listening levels.
Probably something like 8W per side tube amping I would estimate.

Certainly sounded good to my ears - no overt distortion, pretty clean really.
Coloration was mostly from the channel strip (neve+pultec+1176) and the EL84 pp hifi I use to monitor.

I'm pretty familiar with the sound and would say not much distortion at all.
From memory, the distortion was more 2nd harmonic at around 55dB below fundamental with 3rd harmonic some 10dB less than that. Rest of harmonics fall off very quickly after that.

That's running 90mA thru the edcor 8K:600 with a 3K terminating resistance. I found that is about right for extended LF (-3dB at 35Hz, limited by the UTC input traffo). That cost me 1 dB in noise, taking me down to -66dBu.

Response is flat to some 35K with a very slight increase to 44KHz which is my limit.

Anyway - sound is surely there  - clean and clear up and down the spectrum. Surely gave me a good feeling - enough to keep me going while I work thru my issues  8) ???

 

[alexc]

I made the change from 600 ohm t-attenuator pre input transformer  to a 50K+50K rotary switch 23pos 1dB per step attenuator post input transformer.

I expected the signal to noise ratio to improve dramatically since most of the noise I was experiencing was occurring after the input transformer. Adding the atten after it means I would benefit noise wise when the atten is .. attenuating .. as is normal in operation.

What I'm seeing is that noise floor is much improved with the change, by perhaps 10dB or so but remains mostly constant  with  attenuation (< 2 dB variation over 22dB atten range) - which is as it was before, but a wad less noisy!

I need to think about that. I'm certain I had the wiring correct before - it's a pretty straight forward change.

At my signal amp setting of 45mA current per phase at 130V B+,
I am seeing an overall gain of 22.5dB for a worsening of the noise floor of 8.8dB (spl style 20-20KHz) which is pretty good

Irregardospectively , it's the result that I wanted. A truck load of gain with a still acceptable noisefloor.

I can live with that. Perhaps I can improve that some from here even.

Now I have a good baseline performing signal amp I can go back to the control amp.

And boy does it sound good.

I am playing a noiseless pickup tele into the Motu mic/di input (which is about the worst thing I have for a guitar DI)
thru the PMsix61 signal amp and back to the Motu, monitored out to my old el84 hifi amp and some fair but cheap JBL 8" hifi speakers.

AWESOME!

The minimal signal chain utc -> bunch of pentodes running pretty hot -> edcor X series 10W 8K:600 is wonderful.

Thickens what I know would be a grey and anemic motu DI into a classy, refined but muscular and harmonically rich velvet beast.
Not too much distortion but a definite thickness.

Very pleased with the basic channel. If I can retain that under deep GR I'll have succeeded.

 

[alexc]

Control amp is holding together nicely - I did some moving around and shielding of the control amp input traffo.
Doesn't really contribute much at all to the noise floor of the signal amp. Which is just fine by me.

So the signal amp and control amp are integrated - my signal amp I have set to 20dB overall gain input XLR to output XLR.

I did this using the trimmers for bridge reference voltage (which appears at the CV and so the signal amp grids) and the signal amp cathode balance reference voltage. I haven't done any real 'positioning on the 6K4P transconductance curve' yet. I'll look at all that later.

Now with 20dB overall gain, and no GR I get a noise floor which is 7.8dB (spl style of measurement 20-20K BW) worse than 'RTA loopback'. As expected, under GR I can see the noise reduce as the signal amp gain is reduced. GREAT!!

I have a decently quiet unit 
I've set the 'dc threshold' pot to begin compressing pretty early - I'll look at this setup later on too.

My control amp was easily generating in excess of -85Vdc.
I'm easily getting my desired 20dB of GR with a -10dBu input signal and 'Level' rotary switch at max (minimum attenuation)

Frequency response under GR is holding up very well -  flat with a bit of roll off at the bottom, but with no worsening under heavy GR.

Happy to say the noise is remaining low under GR.  8) 8)
Distortion under GR is higher, as expected but still pretty good 

This is the first time I have a basically working unit.  ;D ;D ;D

Now I'm going to have a listen for a while - this is a real milestone. 

 

[alexc]

Whoa! That's a lot of gain reduction! My tele sounds like a pedal steel  8)

Instant attack and very even release.

Distortion is pretty good - I am playing with an input around -4dBu or so and with maybe 20dB of GR.

Talk about an audio tramopaline! This thing is like a hifi sponge.

I'm watching the dc CV on the cro as I play - I have about 2s  release going.
I am getting around -30Vdc for 20dB GR or so and getting pretty soft > -45Vdc

All nice and quiet, with no twitchiness or other weirdities. I do have to check my time constants though, a couple of release settings are inducing oscillation in the control amp under GR. And some attack settings increase the distortion under GR.

All in all, not bad at all. And that really is a WHOLE LOT OF COMPRESSION MAN!!

 

[alexc]

Well the dc threshold pot really makes a big difference to the sound.

Next thing to do is check out what's what with my time constants and get the faster settings working properly.
Happy to say that's the only thing on my 'sh**t list' for now.

After that I need to do more thinking and investigating with respect to the signal amp biasing.

- where do I want it to sit at idle
- how much max gain do I want
- how hot are my typical input levels going to be
- do I want to run higher GR B+ voltage to swallow higher input levels

and so on.

Before buttoning up I need to

- make sure I am happy with the control amp feedback factor
- add the GR metering circuit and lights
- add the VU buffer circuit and lights
- add the post output traffo attenuator


Just need to be extra careful now - usually at this point of happy when something dramatic bad happens to set you back 

 

[alexc]

Oh yes - almost forgot.

NO THUMPING!

What so ever 

No tube matching to speak of either, so far, at all.
I did balance the quiescent (idle) currents at each phase plates as best I could, tho'.

I don't know if it's just my good luck or wonderful skils (as if) or  if it is indeed Analag's Thump-Be-Gone cathode clamping circuit

but,  it's in there and sure enough I don't have any thumps or audible CV break thru. 

Good enough for me 

 

[alexc]

What's even more insane is I already want to do another  : : 

With the Sowters, a full tube PSU  and DaveP level wiring!

Good thing I'm out of the game for new units  now 

Unless, of course,  I *sell* off some s**t 

 

[alexc]

Just some more tweaks - nothing major.

Turns out rotating the control amp input traffo utc does make a sizeable improvement to the mains harmonics noise under deep GR. Reduced 50Hz, 100Hz, 150Hz by more than 10dB with a 90degree turn, opposite to the direction of the input transformer utc rotation

Funnily enough, adding the utc shields makes no positive difference in either case. Yet another example in my opinion, of the unpredictable nature of noise - a complex mixture of induced EM, ground loops and psu impedance coupling between active stages.

Anyway - thats dropped the contribution of the control amp to noise where the CV signal is significant > -25Vdc or so.

I've also been looking at the 'ripple' on the dc CV signal with different time constant networks and it's effect on signal.
ie. CV breakthru into audio but not from 'compression' action, just plain old rectification ripple.

This is mostly impacted by the choice of the bridge 'hardwired 220K resistor and 2uF capacitor', as distinct from the switched TC networks.

As expected, the smaller the 'smoothing cap' at the bridge, the more CV breakthru one will see - I see it in the form of an integration of the sine wave input signal ie. a triangle ripple at the CV not unlike psu ripple.

Exactly like that, actually.

So, just investigating my TC settings now. And figuring out how to set it.

At present I have :

  bridge // fixed (220K // 2uF)    then  all that in parallel with   
  switched cap network //  switched resistive network

I can certainly see the effect of the switched attack resistor setting on the CV ripple - it varies mainly the slope of the triangle and the amplitude to a smaller degree. With a smaller 'fixed' bridge cap, say 220nF instead of the stock 2uF it becomes quite large. Just like in a psu bridge.

When the amplitude of the CV ripple is made to become significant in relation to the CV dc value, CV breakthru occurs and you get some superposition of the triangle CV onto the sine signal at the grids of the signal amp.

This causes some sharp transition points on the sine (high order harmonic distortion) as well as the beginnings of 'ticks', which is CV ripple breaking thru into audio. Not the same thing as classic 'thumping' which arises from the rapid CV transitions in response to the audio. This can happen with well smoothed CV if the CV is moving about fast enough to be mistaken for a musical note.

No surprises there - I am however surprised that it performs as well as it seems to be right now.
However, I haven't measured the time constants so it may be slower than I expect.

Distortion under heavy (>20dB) GR is more concern right now than CV thump at the fastest attack setting.
It seems pretty forgiving of release, so I am just checking that to make sure it is in fact having an effect.

One thing for sure, if I change the 'bridge cap' from 2uF to 0.22uF I get some strong 'tick'.

Anyway - lots of TC investigations here for a while, I think, while I button up some other things.

Still pretty happy - at least I have the noise beasts conquered.
That's the first deal breaker firmly behind me.

 

[alexc]

Update - I went thru all the combinations of orientations and shields for the control and signal input traffos.

I did find one combo which was the best for both no GR and deep GR scenarios - traffos 'facing each other' with shield on signal. So I keep that one. I'm certain I could have arrived at that thru theory and design. NOT.

Now I have the mains harmonics components from -100dBu to -85dBu in the 20dB of gain + no GR case and also the  20dB gain + 20dB of GR case.

So that's my new and bestest baseline. It's up there with the quietest I've done in a limiter.

 

[alexc]

Now that I have the basic limiter going, I can see that my next challenge is to reduce distortion under deep limiting (20dB GR).

I can see the distortion rises quite a bit under that operation. And it is exacerbated by the faster attack settings.

Clearly now I need to match the GR tubes.

While I'm thinking about how to do that, meaning what test rig to build, I'm preparing some audio tracks of signal generator at a range of output amplitudes.

Then copying them and inserting silence periods - this is to measure attack.

Then copy again and insert stepped fade outs - this is to try to measure release. I need to play about with this to figure out a workable way to go.

Then, I record those tracks playing out to the limiter for each of the attack and release settings.

Initially I am looking only at deep GR, so setting ac and dc threshold in order to achieve a constant 20dB GR in each case.

So - I am spending a while with Cubase and the signal gen

-------------------------

Here's a typical pic of an attack in this unmatched tube condition.

Showing a period of some 0.04s from turn on to steady state.

There is some large turn on spike, followed by one phase conducting and one not. Then it settles into a more even conducting state.

I think that spike is the program material, being initially subjected to the 20dB gain in the signal amp before the CV ramps up and clamps down on the program material.

It's solid because of the distortion that happens is highest during this time.

As the program is rapidly reduced by the onset of GR, there is probably overshoot in the clampdown.
An equilibrium is then reached between the signal energy and the CV generated and the second phase of the attack completes, arriving at the steady state.

  OR it could just be some level shifting shmutz impressed on the signal by the rapid ramping of the CV.
ie. CV breaksthru

Not sure yet - the higher the input level and therefore the CV, the shorter the duration of the spike phase and the traditional attack phase. I've seen it go from the 50ms I show here to as rapid as 1ms in the harder slamming cases.

I'm sure my understanding will grow as I look more at this.

 

[alexc]

Another of the waveform under steady GR. Sine wave 1KHz output still looks quite 'shapely' but the RTA showed quite a bit of distortion in the even and odd harmonics of the fundamental.

 

[alexc]

Here's an RTA pic which is a typical good case. Lots of distortion, though not obvious in the pic above

Significantly higher than the no GR case with 20dB gain.

Looking with the cro, it is one side of the waveform with a sharpened point and a distinct point where it transitions from smooth to triangular.

At a particular setting of GR (CV setting) I can visually trim the distortion with the cathode balance pot and see the reduction in the harmonic peaks on the RTA. But of course, it's different for each CV value.

I have 20 tubes, so I guess I need to do as Dave  showed earlier - make up a test rig, characterise the tubes at a set of bias points and plot the result. Then pick the closest performing set.

What a chore! All I've ever done before is eliminate the 'outriders' with fairly crude techniques.

I also need to investigate that big spike on the start if GR.

One thing that does seem to be good, is that I'm not seeing any significant cross talk from the high level output of the control amp (under deep GR) into the grids of the signal amp. Signal amp grids look very clean and clear even when the signal amp output is at it's most distorted.

 

[alexc]

More on 'thump' - I can induce it pretty easy with program material by attenuating the incoming signal whilst still having a strong CV signal. eg. 4Vrms at the grids with 20V CV or so.

Obviously the program material masks the thump, so when the program is louder at the output, (not so attenuated at the input) the thump is not as audible. But beyond that, there is clearly thump when the ratio of grid signal to CV is low.

On the upside, it does sound pretty good with the program material when adjusted for least thump. 

So many, many variables  ???

 

[alexc]

On the subject of characterising tubes ..

I'm lucky in that I'm also working on a tube preamp/eq at the moment. I'm writing about that one in another thread.
http://www.groupdiy.com/index.php?topic=51925.0

I've finished building up some of the basic modules for the preamp - JLM psu for phantom/utility, the heater rectifier and pi-filter, the Orange86 (redd47) preamp with onboard hv rectifier, pi-filter and regulator.

I have the chassis in place with the psu traffo and some edcor xsm as well.
I'm adding a 24-0-24vac traffo to generate my -ve ref rails for this exercise.

So, this is the perfect environment for making up a tube characterisation rig.  ;D ;D

I'm making up a little pcb with a 9 pin and 7 pin tube socket plus connectors for heater, hv and grid voltage rotary switch.

It will take a regulated hv from the Orange86, which I will tailor with some additional, easily modified regulator reference zeners  so that I get the desired regulated plate voltage.

I have a 23 position rotary switch which I will 'deck out' as a linear grid-voltage switch - 1V steps to to -37V thru a rotary attenuator. The JLM psu board gives me variable  reference voltage if I need further control.

Cool!  That way I can easily set the rig up for characterising 9pin or 7pin pentode or triode. I can set the regulated hv and I can select the range and resolution of the applied grid voltage. I can do my 6K4P for this project as well as some other 6BD6, 6BC8 and 6N5P which I use for other projects.

I can even replicate my vari-gm environments more closely with the edcor traffos if neccessary.

So - I plan to get that going in the next day or two and get busy characterising them toobs 

 

[alexc]

The other thing I've decided to do is to raise my signal amp B+

I'm seeing that higher signal voltages at the grid are good - in this unit where I have CV in the 20-45V dc range pretty frequently when I have a decent output level from the signal amp, I have the feeling that higher signal amp grid voltages are best for minimising thump issues.

The CV is pretty heavy and fast in most cases, so the ability to grid higher levels makes it easier to run without thump.
Also, I have run into grid current limiting happening in some cases with hot signals already.

Given that I am planning to drive this unit from a fairly hot tube pre/eq, as well as the observation that at lower signal amp grid signal voltages I seem to get my worst thump, it makes sense to raise the B+ and allow me to apply higher amplitude signals without distortion.

Another thing is that I am running my grid-cathode difference voltage quite low at around 4.5V
ie. high up on the 'transconductance curve'.


If I want to position the operation 'further down the transconductance curve' in order to 'take advantage of' or 'avoid the problem of' possible 'humps and bumps', I need to increase that grid-cathode difference voltage quite a bit  ie. push further down the curve.

The real unit runs things quite significantly down that curve, at some 18V for example.

That means I need to 'swallow' higher signal levels so that I still get decent output levels, as the gains involved will be lower.
ie. higher B+, say 200V and a grid-cathode difference of maybe 15Vdc or so

*phew*  ;D

Also, I seem to have settled on 20dB overall in the signal chain, for reasons of noise floor optimisation as well as that's already a truck load of gain/GR. 

I am running around 30mA per phase now, which is well down on my previous maximum, so I can raise the B+ without meltdown at idle.

So - rejig the second psu regulator  section and check that out  Should make a significant difference to the input signal at the grid before grid-current limiting or excessive thump audibility happens.

---

As things are now, I can get decent thump free operation by setting the dc threshold quite high to rein in the control amp and resultant CV levels in proportion to grid applied signal. As well as, of course, slowing the attack/release down.

But that means the available GR is all 'bunched up'
ie. twitchy  rather than  more 'spread out' which I think is more desirable, in that it provides a broader pallette of GRs.

I may also have a look at the frequency response of the control amp and possibly reduce the low freq sensitivity to reduce the tendancy to respond so much to bass energy. Plenty of places in there to look at tailoring the response.

 

[alexc]

I did the labelling of each of my 20 6K4P tubes and have my test rig working in the chassis of a preamp build I'm doing.

It is virtually the same circuit as the  intended usage in my limiter - regulated B+ now at 200V, dc unregulated heater.

I am grounding the grid thru a grid leak resistance and have the cathode thru a resistance to a constant-voltage current sink.
ie. the 'Analag cathode clamp circuit' from the original PM670 and as used in my limiter.

I am varying the regulated -ve rail supplied to the cathode clamp thru a rotary attenuator.
I have up to -37Vdc or so available at present. That should do for now.

So - now to get going on 20 tubes  ;D  Switch and measure x24 then do that 20 times. Graph the result, pick the closest set of 8.

Then to look at my distortion components of the limiter. Hopefully to be seeing an improvement.

As I saw from playing with the balance pot at just the idle current setting, the amplitudes of the distortion components are significantly affected by current balance thru the transformer. I could vary them by balancing/unbalancing.

Mostly the amplitudes of the even and odd harmonic sets would change - at balance I could find the minimum of both.

So - prolly a couple of days of doing this exercise.

 

[DaveP]

Its useful to average the data from all the plots and put that in the chart.  You can then use that plot as a datum target and take two tubes, one above the datum line and one below, these can then be a pair.  You repeat this with another pair and so on.  These pairs are placed on the same side of the Push-Pull where they balance out.  You should be able to get 4 pairs out of 20 tubes.
Good luck
DaveP

 

[alexc]

Thats a plan ! Conjugate pairing on the spread.

Reminds me of my qa eng days  

Thanks Dave

 

[alexc]

I changed my mind a bit regarding characterising my set of tubes ..

I am firstly checking them in the limiter unit - I want to see what is happening in the actual setup.

So, I've populated a single pair of 6K4P in the unit. I'm leaving one in place and varying the other thru the set, measuring at a set of grid voltages thru the CV range basically -1V to -50V. I've picked the steps such that the current readings are not too large in increment.

When I've varied all the tubes in one side I'll repeat it for the other.

This is a 'real life' measuring as it retains the current circuit fully, including the inherent 'dynamic' dc imbalances in the rest of the signal chain.

I'm allowing about 15mins for each tube to warm up as I can see there is a period of warm-up, then stabilisation, then equilibriating then to some reasonably steady state.

I'll then analyse it all make some choices to conjugate pair the set.

------

I'm getting my first results in and yes there are some differences - up to 0.5mA at the high current case and 0.25mA at the low current case.

Fascinating stuff. This is at the 130V B+.  So far it is fitting pretty close with the curves that were posted years ago for the 6K4P - from memory it was Larrchild, but I could be wrong.

Once I've done this for a bunch, I'll adjust up to B+ 200V and do all of them. I'm interested to see the basic curve changes - I expect some flattening and left-shifting to the larger -ve grid voltages.

I think I am currently working in and around the 'knee' of the curve and the compression is pretty abrupt and 'short in the range'. Moving to the flatter area should mean a softer knee type of operation with a more spread out gain reduction. We''l see 

Here's a curve for a couple of tubes in the 'A' position at 130V.