PMsix61 Limiter DIY

[alexc]

In this pic, it is zoomed out - you can see the whole 'settling' time of the signal - modulated programme material

That's about 250ms  of 'settling' time to steady state or 250x 1ms which is the wave period at 1KHz.
It's at the fastest of the current time constants.

I'm going to to do some tracks thru a plug-in comp and get an idea of what they do.
Not at all ruling out a fault in my transient response - this is my first time really looking.

I could well have something wrong, apart from possibly inherent CV breakthru -  like some stray capacitance charging/discharging in a voltage reference or mismatched impedance between the CV/time constants /grid circuit at signal amp.

But that's the part of the process - understanding in detail the characteristics of interest in your nice new build.


*** the more I look at it, the more I think I am getting an initial slam, overshooting to less than threshold, then releasing up to the equilibrium amplitude  and finally, CV to steady state value. I have to test of course, but that's my current thinking.

*** probably what happens when the dc threshold is set too low - but it could also mean a mismatch between the control amp and the signal amp, like a damping mismatch

 

[alexc]

And here's the waveform again zoomed out, after steady state.

The GR is still happening, and pretty significantly - it's just steady as the CV is constant.

No visible distortion there and the waveform is well centred about the centre line

 

[alexc]

Lifted the idle voltage again - grid-cath diff of -9.0V.
That's now 41.6mA per phase, 10.4mA per tube.
Plate voltage dropped some 8V and regulatoed B+ by 2.5V or so with the extra load.

Signal amp gain is up a little but not much, but again, the CV required for a given GR is reduced.

And once again, I think it sounds better, at least on the 'tele' test - much more spongy without as much gain loss.
Less distortion again under deep GR.

I'll sit here a while. When I get my boundary, I'll redo the phase currents profile as the 'final' one.

For now, I'm at least 19% more excited than I was with the idle at -12.5V 

---

Can't resist - now at my boundary -8.0V grid-cath diff.
That's B+ 194.5V, plate 185.6V, phase current of 45mA and per tube 11.25mA and 2.1W on the plate.

There's no doubt to my ear - the distortion under deep GR is much better at the higher idle currents.
The 'sponge' is back routinely - I was missing that at the lower current settings.
I should say 'clean' sponge - pretty deep GR routinely without getting too distorted.

With a slower attack and faster release, it does the 'pedal steel' thing. Which I love. Having always wished I could do that highest form of the electric guitar.  8)

Just doing some more measurements and that's done.

-- actually dropped it back a couple to grid-cath -9.5V. Best compromise for me so far.

 

[alexc]

Now, back to the time constants.

I have to rejig the 'release' switched-cap network - I just don't seem to get much release variation.

The 'attack' switched resistor - definately noticeable.

Need to review, remod and rebaseline.

-> OK - it was a bad solder.

Now I can see the release as I expect. I have a range open (fast) to 6.8uF (slow) in six steps.
I think I'll extend that out to 10uF at the end. My slowest is still too quick.
I want around 3s in the slow.

And the effect of the 'attack' switch as a 'release' multiplier is evident.
When attack is fast, the range of release is faster.
When attack is slow, the range of release is really slow.

But yes, that is certainly on the right track now!

Back to cubase to check how plug-ins look like for the same test tracks.

One other thing I'll do before the end is running it without the 'cathode clamp'- I want to look and hear more into it. Just to figure what's doing and not down there.

 

[alexc]

That same piece of audio thru a plugin slamming hard also

First, the zoom in - same 1KHz, 1ms period.

It looks inverted, wrt to my attack envelope, but shares some characteristics.
CV breakthru impressing a dc offset, discharging away to a steady state, centred around 0 line.

But unlike min, the first few periods are unsmashed or unskewed. Could be that it's not smashing as much.
So I'll try some more.

 

[alexc]

And the zoomed out view

Again, similarities and differences

An initial slam which goes well below the eventual steady state amplitude, with a quick release to the steady state, followed by a period of CV to stationary  ie. the move to centred zero line.

Makes me think this shape is pretty characteristic of a hard slamming CV.

 

[alexc]

Here's a 'maxi-smash' plugin view

Not so much 'overshoot' visible because the threshhold value is much closer to the equilibrium value. ? Maybe !

 

[alexc]

Going back to the PMsix61 transient, here's one with +4dB 1KHz input, with Threshhold and Level set for moderate GR and around +4dB out.

So this one clearly does maintain the first cycles without destruction and quickly settles to SS.
There is some CV breakthru but is small in comparison to the amplitude of the initial cycles, so all looks OK.

So it does seem that the first recordings I put up were (are) a case of small input (-10dBu) with large signal amp gain and maximal smashing, such that the early cycles  are reduced to oblivion on the initial slam.

So the attack can be divided up into an 'early attack' (almost an overshoot) and a 'mid attack' (reaches gain equilibrium) and a 'later attack' where the CV breakthru discharges away.

How much GR is dialled in with the Thresholds, both ac and dc influences the durations and amplitudes of these stages.


*** the 660 manual has some useful definitions of attack and so on. Just going thru all that now.

 

[alexc]

Same input, more GR and increased output.

Same thing - early cycles look good, some overshoot on the early GR with amplitude releasing to the SS level and some CV breakthru which fades out as the CV moves to stationary.

 

[alexc]

One of the things that really sticks out is :

- increase in distortion with faster attack and/or faster release settings

Not sure why this is - the CV remains the same dc level more or less and is steady - it's not tradional CV movement breakthru.

Looking at the 'ripple' on the CV for a high CV setting (eg -50V) and 1KHz signal I see a CV ripple of  551m at fastest settings, and 35mVpp at slowest.

At lower freq of signal, say 70Hz, it gets huge, like 330mVpp to 2.75Vpp.

The periodicity of the ripple is the same as the signal input.

Changing the time constants changes the 'slope' of the resultant sawtooth ripple waveshape within that period.

Looking at the worst, the 70Hz case, the distortion is something like 8dB difference in the 2nd harmonic and maybe 5-8dB for many of  the higher orders.

In the 1000Hz case, around 3dB in the second harmonic and 12dB in many of the higher than 5th orders.

So - thats a lot of distortion changes for variation in time constant. 

Of course you don't see all that CV-to-grid ripple at the output - some is CMRR'd out but obviously a fair bit of it is getting thru. There's balance issues to consider as well as all kinds of other things I'm sure which determine the CMRR.

But that would be my number one bad :

- significantly higher levels of harmonic distortion at  faster time constants for a given GR, particularly at the low freqs due to CV ripple (as distinct to CV break thru). 

- It worsens as CV is greater in amplitude.


I guess that's why firstly you got to balance as best you can the tubes, the rotaries AND have transformers suited to this app

 

[alexc]

Now this CV ripple is not related to what I understand as the release time constant.

If I toggle the signal on/off, I see the dc CV level reduce from it's high value (eg -50V) to the idle level at a speed according the 'release' switch setting  (and the attack switch too !)  ie. fast is around 0.5s and slow is 5s or so.

That's what I expect. (more or less)

But the ripple - that's something else.

And it does seem to most depend on the 'attack' setting. So I think I probably have the 'min equivalent resistance' in the rectifier load // time constant network  too low.

** this makes sense because the period of the ripple is fixed to the period of the signal. But the slope of the integration is varied by the attack resistance - too low and it gets higher slope, so it will have a higher max excursion possible within that same period.  Dig ?  

ie. more charging current due to less charging resistance means higher voltage achieved before discharge time comes around.

So   What to make of it? 

Measure the effect of the resistance on the audio attack time and work out how to increase that resistance without sacrificing the speed. ie. lowest ripple for still fast attack.

Or - accept it is a limitation of this build.

Lowering the CV ripple feed-thru critically important for getting lower distortion in the bottom end at strong GR levels.

----

The real deal has setting '1' which is fixed load 220K//2uF in parallel with 68K.
Setting '2' is fixed load // 470K

The others  parallel  a large cap alone or a combo of large cap + series 100K

And this all probably is affected by the grid network it connects to -> the impedance of the traffo+other stuff+grids circuit as well, which in my case is very different from the real deal.

I have the stock fixed load and it looks like my minimum 'variable parallel' resistance should be around 120K or so instead of the current 47K. That massively reduces the distortion. Hopefully I still have some fast attack.

So - back to the abacus and slate to continue trying to measure attack before I rejig the timing to minimise that ripple whilst retaining a reasonable attack and release time. IFF it's possible

 

[alexc]

From the original specs :

Firstly the time constants

 

[alexc]

and attack time definitions

 

[alexc]

So my cubase test session will need to go to the 96KHz sampling rate to get the best resolution possible of 10.42us sample period (from 25us at 44.1Khz) and my test files will need to go up in  frequency to try to catch attack envelopes of 200us.

I'm game

I need to think some more on those attack definitions. But in the meantime I have more to test.

The highest resolution event I've yet seen is that 1.5ms (1500us) inital mega-clamp.

Still plodding along in terms of the potential stomp factor of the real deal.

It's a riddle, wrapped in a mystery, shrouded in enigma, I tells ya!  ;D

 

[alexc]

OK so - attack time is the time taken for output to fall to 63% of it's final reduction (amplitude).

So in my case, I was seeing the initial stomp down at some 1.5ms (1500us) - a massive overshoot of the steady state output, which then 'released' up to a steady state value which was much, much larger.

So it seems reasonable that the '63% of final value' was achieved well within the first part of that 1500us stomp.
ie. some small, early  fraction of that stomp duration, in the 'hundreds of us' range.

In other words, could well be in the ballpark.

I just need the eyes, to see.

 

[alexc]

I have to say, about the 'poormans 660' limiter 'thing', at my now mkII level ...

It embodies everything  I do know about audio electronics
and also everything I don't know about tubes

So much stuff that seems simple, after years of playing about.
and so much, complicated stuff too - that you *should* know about,
but only glimpse at.

It's about the 'zen-ist' thing there is

 

[alexc]

I finished up the meters, light show and heater resistors.

The PMsix61 Limiter is now closed up for studio trials. I'll do some clips with it and post some links if I can.

I've done the basic testing at the bench, now I'll play with it my room for a bit and take notes. In a while, I'll open it up, make some tweaks then re-characterise it on the bench.

And that, my friends, is that.  14kg and 1200usd of glass and heavy metal 

---

Here's the final inside pic

 

[alexc]

And glow pic

 

[alexc]

And outside pic

 

[alexc]

And last of all, in the wild as lord of the flies