PMsix61 Limiter DIY

[alexc]

So I did the psu mod more a B+ of around 200Vdc - it's all relatively lightly loaded so that will drop some later I think.

Certainly has made a big diffference - peak current up to 19mA from 12mA or so.
Cuttoff now at around -62V up from -45V before.
Sig gen output is now at +6dBu up from 0dBu for a similar ease of adjust CV.

This is closer to where I want to be so I'll do the tube thing here.

Here's the curves for the same tubes at 200V B+

 

[alexc]

SO the measurements do seem reasonably stable and repeatable.

Tube 4A for example is a weakly outrider at both B+  so far
With a sample size of 3 tubes it is a  * g o n e r *  8)

----

Now assuming the signal amp gain remains in the ball park as before with the new higher B+ for the same current, I would be looking at around 30mA total per phase to achieve what I want.

So that would mean that I am looking at max idle of around 7.5mA per tube.
I can set my trimmers accordingly.

So I should focus my tube characterisations to a  current range of say 8mA down to cutoff which is grid of some -15V to -71V with a focus on the region most likely to be used, say -18V to -50V or so.

So that's what I will do. Start again and narrow the measurement range that way.

 

[alexc]

Another short cut - after doing a few full tests and getting an idea of the likely average current at a few key grid voltages, I am doing a brief warm up test on each of the remaining tubes and doing and quickly checking those.

There are some that are too low or too high by a widish margin, so they can get eliminated first.

That way I can spend more time on more detailed critical region measurements before I go nuts. 

-----

After the first pass, from the 20 tubes I have eliminated 3 as too high. From the remaining 17, I end up with 2 groups, one set of 8 clustered about a lower mean and another set around a higher mean. There is one in between.

So - I'm dividing now into two sets of 8 to do the detail work and also to verify previous results.
Of course, it would be great to have a decent 'burn in period'  - these were all nos and unused.

The first 8 in  the series were significantly burned in, as they were my 'build set'.
The rest, no so much. Until now.

-----

Early indications are quite repeatable and stable   Also, changing the 'control' tube (other position in pair) does not appear to alter the measurements much at all - by < 1/2 my basic margin of error.

Seems my basic error margin on repeatability of same measurements on different passes is around 0.05 mA at best and 0.1mA at worst.

-----

The deviation from best to worst within a grouping of 8 seems around 0.5mA - ie. 0.25mA above or 0.25mA below an average. Interesting to look at the error in current as a fraction of the increment of current change.
I'll do some of these things when I have a full set.

That's pretty good really - so if that relationship holds, I'm going to be quite well balanced.

Each of the detail tests is taking about 35mins including 10mins for warmup. Thats for 60 or so readings at a standard set of grid voltages from -11.0 to -62.0V which is the cuttoff. 

---

Interestingly, my 'first 8' or 'build set' has had 3 eliminated as the worst outliers
So hopefully I will ultimately see a good improvement in balance and hence distortion under deep GR

 

[alexc]

Starting to 'come together' now - literally.

One useful thing to do is to take the readings, graph it and before switching tubes, re-measure any slight anomalies in the curve. There are the occasional reading which doesn't settle as quickly and can cause a bit of ripple in the curve.

As long as the reading is done at the same measurement session, it will be accurate.
I found in needed to recheck a value or two here and there.

The overall session-session repeatability seems to hold at max of 0.1mA at the 'big end' of the current scale.
ie. 0.1mA in 9mA. At the small end of the scale, it is more like 0.01 in 1mA and quite a bit less.

The spread so far is around 0.8mA at the 'big end' of current (9mA or so) and 0.3mA at the 'mid end' of -30V grid. They all converge from -40V to cutoff at -60V.

The 'eliminates' were more like 3mA out of the spread at the top end.

Just finishing up my first cluster of 8. I'll put that one up then do my next cluster of 8.
The cluster 1 and 2 mean values vary by around 0.7mA. The spread of each is about the same.

So far I think I'm getting a reasonable tolerance - should get 2 fair groups of 8 from 20.
I would say 25 tubes would have made a significant difference, allowing removal of another outrider or two in each set.

Seems to me like 1 in 5 is a large outrider, another in 5 a smaller outrider - so 3 in 5 from a min sample of 25 will give 2 good sets of 8. OK?  ;D To be a total d**k, look for one set of 8 in 30.
-----

After a couple of sessions of this, and as a 'first time' observer, I can say I am truly amazed as to the repeatability and accuracy of these tubes.

Even down to measurements of <10uA the stability is pretty solid, once warmed up for 20mins or so.

I expected them to 'dance' around alot more - not so *at all*.

I think doing measurements like these could yield a pretty startling matched set of 8 from 30.

Mine are a little looser than that on this set of 8 of 16 (4 eliminated from 20). Don't think I can improve this given my sample size. I have 1 or 2 outriders within the set if 8 which drag it down - the other 6 are very close.

But by arranging hi-lo pairs I can average the differences down. Can do that either within a phase or across phases - I'll look at both.

However, the differences here are pretty small in reality. I don't know how tubes age and I don't know at what point of 'imbalance' the distortion components vary to any significant degree.

From my observations with the 'balance' pot at 'idle' (which is the high current mark where the errors are highest and have most 'puisaance'), the relative amplitudes of the odd-even harmonic distortion sets *did* flip in dominance pretty easily, but I don't think the overall THD raised or lowered all that much with balance.

But I'll find out more soon enough.

(Then of course, there is variations due to temperature, line voltage fluctuation and so on !)

Now to be taking a break please. I have another 8 to do   Tomorrow

 

[DaveP]

Alex's work on this project is a lesson to us all in how persistance and sheer hard work pays off. ;D
Too many projects get shelved at the first sign of difficulty....... not the way to go.
Much better for our overall wellbeing to see stuff through to the end, even if it gets tough on the way.
Well done Alex 8)
best
DaveP

 

[alexc]

Thanks kindly Dave.

And I have really appreciated your guidance on most every part of this project.

I recall some years ago reading where you first described the use of the Hammond control amp output traffo.

The detail of those postings really is what started me down this path.

And it works just great!

Here is the graph of the set of 8 I selected as my closest along with the average curve (black, bolder line)

 

[alexc]

Just completing the second set - it can sit in the cupboard in case I ever need to replace my best set.

I also added the GR meter - no probs. I found the resistance to add around the Triplett VU meter in the signal amp plate supply that gives me the movement I'm after.

----

I've selected my set to use and I have selected the positions in order to average out the differences in the 2 outriders.

To test the selection, I'll populate 1 tube in each phase with the most equal. Then I'll measure the currents in each cathode for the same grid volt range as before.

That should set a baseline for balance performance and to some extent 'best' harmonic distortion profile.

Then I'll populate the rest of the positions and repeat the test to get my final balance and distortion profiles.
If I think I can do better, I buy more toobs.

At that point, in theory I should have my signal channel complete. 

----

After completing the signal channel, I will do some more things to do :

- check distortion and GR at different CV to determine the final 'idle' CV
- check the time constants are working and as I want them
- add the vu metering

Final testing is back to Cubase and the pre-recorded test tracks to measure attack, release and schmutz.

And that's it.

-----

For reference, here is the noise floor spectrum of my test bench Motu 828MKII analog channel in 'loopback'
ie. output to input

RTA calibrated for dBu standard.

Basic noise floor is -78.3dBu 20-20KHz SPL stlye of measurement.

 

[alexc]

I rebalanced the cathodes to equal reference voltage with no tubes - the balance is not much use really.
If anything, I would try to balance at a nominal GR rather than idle.

------

Here's the first pair of my selection - this is the closest matching pair I have.

Perfect at top half, deviates a bit at lower half to around 0.15mA at it's worst.
That's actually quite a bit, ocurring as it does around 1mA  - some 15% error there.
Phase A is the lesser current.

This would likely happen at around 12dB of GR (to be confirmed), so it will be hit quite a bit.
(this is not going to be used as a tickle the top limiter - I want stopping power!)

----

So this is the finals, pair 1 and represents the best balance I have, and therefore, the best distortion at all grids.

 

[alexc]

And here's a pic of spectra :

LHS is  no GR
RHS is  20dB GR

Same scale so you can directly compare left and right.

Note the

- drop in amplitude of the 1KHz signal, as well as reduction in the overall noise under GR
- increase in harmonic distortion spectra under GR

It's actually not too bad under the deep GR - it was quite a bit worse before.

The max amplitude increases only a little, but there is added 5th and higher components at significant levels.

 

[alexc]

And a simple shot of the 1KHz sine at that 20dB GR.

Only very little distortion visible. Overall not too bad.

Now I'm going to play around a bit before adding my next pair of close matching.
Same sort of checks.

After that, I add the in-phase complementary pairs  ie. I pick a pair of tubes on opposite sides of the average line and as equal as possible. They go together in the same phase to cancel out. I do this for the remaining two pairs.

And hope for the best

 

[alexc]

Now here is the Phase A and B currents with pairs 1+2.

As one would expect, current is doubled, cutoff grid is extending.

The B+ has dropped a fair bit with loading - it doesn't matter much wrt to balancing and I will re-adjust at the end of the next stage.

---

Still looking good - max deviation is still around 0.15mA and is happening in the same region at 1mA or so.
Some imbalance at the top end, but as I will be idling probably at -15V or so, still looks good at the top half.
Phase A still the lesser current at the bottom end.

Next one is trickier - in-phase conjugate pair averagingTM 

 

[alexc]

Now I have my full graph, B+180V (still to adjust to final value) with all pairs.

That's one combination of the conjugate pairs - I could still reverse them.

There seems to be a constant-ish difference of 0.45mA - that's something that I could trim with the balance.
So I'll try that and do a quick check.

 

[alexc]

Did the rebalance on the cathode pot and quick check :

Grid (V)    A (mA)    B (mA)
11          26.43      27.19
30          5.26        5.29
40          1.60        1.58
50          0.32        0.31
63          0              0 


Now that's pretty darn good - I think that is basically as close to my repeatability error as I'm likely to get.

I would feel comfortable saying that at this time and with these calibrations, and until proven otherwise,
I have an accuracy of :

+/- 0.38mA    throughout the CV range of -11V to -20V 
+/- 0.05mA    throughout the CV range of -20V to -30V 
+/- 0.01mA    throughout the CV range of -30V to -63V 

which is hugely better than I ever expected before this exercise.

I wouldn't make such claims unless both the voltage and current readings were repeatable to a high degree and had a short settling time and long/stable/unchanging measurement period. It did - to a degree I found wholly unexpected.

That's a wrap  ;D

---

Here's the basic noise floor spectrum - it's a little worse than before, probably due to the changes in B+ which have raised some of the voltages. I'll do another optimisation effort before closing up.

 

[alexc]

Here's the basic No GR spectrum with signal amp at max gain of 16.3dB gain input to output

I need to re-optimise my cathode resistances for the new B+.
I'll probably return to my desired 20dB gain.

 

[alexc]

Here's spectra for 22.5dB GR over the 16.3dB gain signal amp  (  -10dBu + 16.3dB -22.5dB -> 16.2dBu)

As before, the 2nd and 3rd harmonics a tiny bit higher, but higher order components added at significant levels.

This is quite a bit of improvement - I recall the harmonic components being not much better than 18dB below fundamental. In the 'hairier' operations forming a more solid wall of distortion - couldn't really make out individual components.

Also, these are with 'Attack' at the fastest setting - they range down to about 1/2 as much distortion with the slower settings - so these are a worst case.

So A BIG WIN  8) 8) 8)  and worth the effort.

Mains noise at 50Hz is a bit higher, but a the rest are mostly lower. Again, I think I can lower this with my final noise optimise pass.

 

[alexc]

Next up :

- reset my final voltages and check all the op-points again
- set my final idle CV and cathode currents

then

- check time constants circuit
- record pre-prepared test tracks and examine time constants

Well that's been a fantastic result for a solid day's work 

 

[alexc]

I did the change for B+ 200V under load - no probs.
The change was simply to increase R9 in the PM660 PSU pcb from 182K to 740K.

-----

Signal amp is now set up for 18.9dB of gain input to output with 6dB noise added (20-20KHz spl)

That is with grid-cathode difference of -12.5V set as idle for 7.55mA per tube, 30.8mA per phase.

With 6dB attenuation dialed in on the "Level" rotary switch, I have 12.9dB gain for 3.4dB noise.

This is the final setup and is intended with nominal +4dBu inputs from a 600ohm source tube pre.

Typical gain staging would be +4dBu in, -6dB on level, +19dB gain -> +17dBu and 13dB GR average

That gives me +4dBu at the output on average to feed the Motu analog ins with another 6dB of Motu headroom.

See how that fares

-----

Just doing some sweeps - I am checking out signal amp for ringing and freq response.

Freq response is pretty flat to 35KHz or so with -3dB lower at 43Hz.
The UTC input traffo is spec +/- 2dB at 40Hz as it is an 8mA unbalanced dc design.

Even so, it looks like my low end rolloff is a little early - I'm seeing -2dB at 55Hz.

Since the change to the input attenuator (T-pad pre input traffo to rotary post input traffo) I don't see any change to the bottom end with output traffo loading, so that's out.

So - I'm thinking of anything to extend the lower a bit. It's fine where it is though.

Given that I'm driving this from a tube pre+eq I can simply boost the bottom if I feel it lacks.
So far, no probs at all. I'm just like to tweak.

---

Square wave have some ringing and what not - I may spend some time tuning that a bit but I'm not too concerned right now.

 

[alexc]

For reference, for the following grid-cathode differences (which I'll call CV from now on), I get GR as below :

  GR                CV
------------------------
0 dB            -12.5 V
3 dB            -21.1 V     
6 dB            -29.9 V
12 dB          -42.2 V
18 dB          -50.4 V       
20 dB          -52.8 V
24 dB          -56.0 V
30 dB          -62.5 V
36 dB          -65.4 V
40 dB          -69.0 V

----

The first listening test - with the tele thru the Motu DI input, bussed out to the PMsix61 then back to the Motu finally out to the el84 pp amp with 8" 2 way jbl speakers.

Pretty much all the low GR distortion I was getting before (at the much lower idle CV and B+) is gone.

I'm hitting it with around +4dB nominal.

With strong compression dialled up and the fastest time constants, I'm seeing the grid-cathode difference going from -12.5V at idle to around -40V or so.

No thumps at all with the guitar test

However, with CVs > 35V and the fastest settings, I can hear it begin to distort in the lower registers.

With mid-speed settings, I can get CVs > 45V without much distortion in the bottom.

It's pretty clear also in the spectra where the high speed settings (resistive switched network predominantly - 'attack' switch) have twice the distortion of the low speed ones.

I'll need to look at that more closely when I play with the time constants some more.

----

Now for the program material test - this is the hard one where I could get thumps before pretty easy when I had strong GR (high CVs) and low program level.

Here goes !

-----

Well, with program material, only the occasional thump with the most extreme settings : CV  exceeding -50V (around 20dB GR) and the fast settings, with music that is very soft-loud dynamic.

Basically thump isn't a problem  anymore   The move to the higher B+ as well as the balancing has reduced it very nearly to nil.

As before, under deep GR with CV exceeding -50V. the distortion is evident on the faster settings.
At mid speed, music sounds quite good, even with heavy GR.

No too bad at all.

 

[alexc]

Just doing some 'tuning' of the idle CV with programme material under the worst sounding of the GR.
ie. faster and heavier

I do think it gets better with a lower idle CV - I took it from -12.5V to -10.5V and I feel it sounds better under those worst conditions.

That reduces the CV required for the same GR by around 3V. Max gain only slightly higher at 19.7dB for a new phase current of 36.2mA or 9.05mA per tube. Thats still only 1.7W at idle so nowhere near the anode max of  3W or 15mA or so.

As usual, I'll try to find my boundary - I'll take it up a little bit more.

Originally it was my belief that the higher the idle the better. Best to have highest gain before GR so as to have room to limit.
The negative tradeoff was the reducing grid-cathode difference voltage resulting in thump and grid limiting on higher programme.
I had got to around -4.5V from memory.
Raising B+ pushed the idle grid-cathode down significantly down to -12.5V allowing safe raising of the idle current.

Now the limiting factor is deep GR distortion - coming from balance and low deep GR phase current.
So, working the balance and pushing the idle to -10.5V has helped.
Maybe a bit more will help some more.  Only one way to find out

Anyway I'll keep trying different listening scenarios but I'm pretty sure that sounds quite a deal better under strong, fast GRwith the hotter idle  ;D ;D

And still no thump pretty much at all.

----

Next thing is the cubase tracks. See what that throws up

 

[alexc]

Here's a zoomed in transient with GR - fairly typical of 'fast attack setting'

You can see the initial large waveform spike part which represents the (significantly) amplified input signal very, very rapidly clamped down on as the sidechain slams in on the order of 1.5ms for that spike.

So that's 2 periods or so of the 1ms which is 1KHz test tone where the signal is gone completely.
Then there's another 3 periods or so where the signal is heavily skewed.

---

Then there's a decaying negative dc offset impressed on the signal as it moves to it's steady state GR condition.

Control amp breaking thru to the signal as the amp interprets the CV as audio while the CV is in it's 'fast moving' phase.

As the CV reaches it's steady state value, there is no 'CV movement' on the grids being interpreted as signal, and the programme material returns to symmetry about the OV line and with a steady amplitude.

Looks like around 5ms from where the spike turns back into a sine wave of reasonable shape and some actual signal starts to override the 'thump'. I haven't yet heard that 'thump'  - too quick.

Inside the 5ms, you can see a sine wave skewed by 'momentum factors'  - its sort of bent due to the rapidity of the CV change.

The signal reaches its steady state amplitude after about another 30ms, although it takes a further 150ms-220ms or so to lose the negative offset (depending on the point at which you decide it's no longer offset)

---

I have to review the definitions of terms used to describe limiter transient behaviour as well as general transient analysis. It's been more than 25years or so since I last looked in detail at transient analysis  :

I certainly would expect some CV breakthru and quite a bit of overshoot and ringing and what not on that CV signal as well as the same thing at the plates.

It's not a characteristic I would expect on transients in the basic signal amp, however.

I can see from the square waves and pulses of 1KHz the that ringing and all that is of the order of 10x less or greater, in periodicity.

The question therefore becomes 'how soon is now?'.


**** if anyone can interpret this better than I, please feel free to contribute   ***