*
* Generic triode model: 12AX7
* Copyright 2003--2006 by Ayumi Nakabayashi, All rights reserved.
* Version 3.01, Generated on Wed Mar 22 17:21:36 2006
.SUBCKT 12AX7 A G K
BGG GG 0 V=V(G,K)+0.59836683
BEP EP 0 V=URAMP(V(A,K))+1e-10
BEG EG 0 V=URAMP(V(G,K))+1e-10
BM1 M1 0 V=(0.0017172334*(URAMP(V(EP)-1e-10)+1e-10))**-0.2685074
BM2 M2 0 V=(0.84817287*(URAMP(V(GG)+V(EP)/88.413802)+1e-10))**1.7685074
BP P 0 V=0.001130216*(URAMP(V(GG)+V(EP)/104.24031)+1e-10)**1.5
BIK IK 0 V=U(V(GG))*V(P)+(1-U(V(GG)))*0.00071211506*V(M1)*V(M2)
BIG IG 0 V=0.000565108*V(EG)**1.5*(V(EG)/(V(EP)+V(EG))*1.2+0.4)
BIAK A K I=URAMP(V(IK,IG)-URAMP(V(IK,IG)-(0.00058141055*V(EP)**1.5)))+1e-10*V(A,K)
BIGK G K I=V(IG)
* CAPS
CGA G A 1.7p
CGK G K 1.6p
CAK A K 0.5p
.ENDS
Thanks. I should be able to give that model a go Monday.
Next step is definitely to check out the specifics of this behavior in the real amp with different bias. Also I'll get waveforms both with and without the power grids connected. Without will allow us to see movement without the power rail sag confusing things. I'm coming up with ideas about test waveforms I can create in Cubase so we can see how things move at the start of signal, from no-signal steady state. Thinking a sine wave burst in one channel and a trigger pulse in the other channel. I can then move the trigger pulse in the DAW software to capture both start and once everything settles.
Only trouble is I'm back at work so things are going to move slower now. Amazing how productive you can be when at home on holidays.
Haven't tried that 12AX7 model yet jazbo8, but I have had the chance to test PI bias changes on the real amp instead.
Normal warm bias with the 470R resistor is just copy and pasted in from the previous waveforms posted. But I added resistors in series with the existing PI bias resistor and re-tested for both 740R (existing 470 + additional 270) and 1030R (existing 470 + additional 560). Once again triggered on the rising edge of the output to keep phase consistency. If it wasn't for the small differences including different pixelation even I'd wonder if they were just the same set of images pasted twice, they're that close.
So the simulation was right. Asymmetry doesn't change with PI bias and in fact the behaviour remains exactly the same!
Next up is dynamic response on the real amp to compare to what the Spice sim is telling me......
You do not have the required permissions to view the files attached to this post.
Well, luckily that was easier than I was expecting. I recorded a sample of a roughly 333Hz sinewave from my function generator into Cubase. Cut it off cleanly at the start with a bit of zero silence before the wave starts. Then I recorded a square wave from my function gen onto a separate channel, then went into the waveform editor and silenced all but the first positive pulse. Left channel mixer out with the sinewave into CRO channel 1, and right channel mixer out with the trigger pulse into CRO channel 2.
I can position the trigger pulse anywhere I want (once I remembered to turn off time quantising!) relative to the sinewave. The CRO triggers off the pulse and captures a screen worth of waveform centred around the trigger, for whatever time/div you set the horizontal axis for. Here's the result for a couple of tests.
Now I can feed the sinewave output into my amp, set the Volume knob right for a few different levels and capture the dynamic response including how stuff moves around under signal.
You do not have the required permissions to view the files attached to this post.
katopan wrote:
So the simulation was right. Asymmetry doesn't change with PI bias and in fact the behaviour remains exactly the same!
Right, I think this had been mentioned before, the hot/cold bias of the PI is not the main cause for the asymmetry. I have been under the impression that the tail resistor has more of an influence on the imbalance/asymmetry.
Can you suggest an experiment with the tail resistor including values? If so I'll do it - what would you like to see tested?
I did a lot more testing today but it won't be until tomorrow at least before I can format the waveform captures into Visio and convert to pdf for posting.
katopan wrote:Only it had mentioned from both Aiken and Merlin.
Can you suggest an experiment with the tail resistor including values? If so I'll do it - what would you like to see tested?
I did a lot more testing today but it won't be until tomorrow at least before I can format the waveform captures into Visio and convert to pdf for posting.
The larger the tail resistor - the better the balance but at the cost of lower voltage swing. Perhaps you can put a 50k pot in its place and see if varying it drastically effects the asymmetry as the theory suggests.
I'm more than happy to test it out. But my understanding of the tail resistor balance is more about balance between the two sides under normal 'linear' range of operation. This extreme overdrive is way past that.
Looking at the previously posted waveforms, it's hard to differenciate the cause of the rounded top on the 3rd stage plate & PI pin 7 grid. Possibilities are PI grid conduction clipping or 3rd stage cold bias and compression as the Vg curves bunch together before reaching cutoff. So the next test was to centre bias the 3rd stage and see what it does. This was done by soldering a 1K8 resistor in parallel with the 10K cathode resistor, and confirmed with a multimeter as reading the expected 1.53K.
As per my previous experiments and simulation, this increases the amount of duty cycle shift available before it resets back toward symmetrical after 3rd stage cutoff. The attachment is a set of waveforms around the PI showing the middle sweet spot with max duty cycle shift.
Things to take note of:
3rd stage grid and 3rd stage cathode indicate that the 3rd stage is operating in its linear range. It's no where near grid clipping or reaching cutoff. And yet the plate output still has a rounded top. The grid amplitude really isn't big enough to see a non-linear output from a centre biased stage with no cathode bypass cap. But this is still a triode and so won't be perfectly linear.
Comparing the PI grids, you can see that pin 7 input grid signal has shifted down relative to the pin 2 grid midpoint, the cathodes and the top of the tail resistor. At this point I was open to this shift either being from PI grid clipping clamping the input signal peaks, or the movement of 3rd stage plate signal from power rail sag.
The extreme positive peaks of input pull up the cathodes and tail resistor waveforms as expected and as shown in the simulations. On negative swings the cathodes and tail resistor are held flat by the pin 2 grid clamping.
The duty cycle shift in the PI still results in position shift at the power grids and the bottom part of the output has the corner cut off by the PI clipping becoming dominant and sneaking underneath power grid clipping on that side only, as described earlier. All that can be see in the speaker output.
You do not have the required permissions to view the files attached to this post.
Last edited by katopan on Wed Apr 30, 2014 4:09 am, edited 2 times in total.
Next I wanted to capture what was happening dynamically with this apparent movement of the PI pin 7 grid signal. Switched everything over to use the Cubase file I had set up. This allowed me to feed the amp with a many second long sinewave burst from my mixer left channel output, and position the trigger signal on the right channel output anywhere I wanted in time. Using that I can capture long samples which show the signal envelope over time. I can also 'zoom in' placing the trigger pulse anywhere I like in time with a much shorter time/div on the horizontal axis of my CRO.
So here's the same amp set up as before with the 3rd stage centre biased (1K5) but otherwise everything else is standard.
Just the PI grids, but they clearly show the movement of the PI input grid signal. Horizontal axis is set to 200ms/div so the envelope shown in a bit over 1.8 seconds. Notice how the pin 7 input grid envelope moves downward, but the pin 2 grid bumps up and then settles if anything a little above it's beginning no-signal position. The envelope start and end waveforms show the beginning and end of the envelope response zoomed in to 1ms/div. When the amp is first hit with the start of the sinewave the pin 7 input grid is centred around the same centre position as the pin 2 grid and everything is very symmetrical. After it shifts, the pin 7 input grid signal is centred lower than the pin 2 grid. The crossover points of the pin 7 input grid relative to the pin 2 grid are placed higher up on the sinewave, which makes the positive peaks narrow and the negative peaks wider. Notice the tops of the pin 7 input grid are rounded. The envelope end waveforms look the same as the static waveforms posted just before this with the 3rd stage centre biased, as you would expect.
If the asymmetry was anything to do with PI component choice, it should be there right from the start of the envelope. But it is dynamic over time and moves into a position to create the asymmetry.
You do not have the required permissions to view the files attached to this post.
Last edited by katopan on Wed Apr 30, 2014 4:05 am, edited 2 times in total.
Lastly I wanted to rule out the influence on the asymmetry of 3rd stage power node sag and 3rd stage non-linearity. So I disconnected the coupling cap off the 3rd stage plate so I could directly inject signal into the PI. I needed a fair bit of signal swing here that a mixer output wasn't going to provide. So I connected my mixer left channel output into a solid state power amp and took the speaker output as my source (no speaker attached). This amp is an old SS bass guitar amp and has well over 30Vpeak clean headroom. To give it some source impedance I connected a 22K resistor in series in between the SS power amp output +ve and the PI input coupling cap. As well as providing source impedance it also allowed me to see any difference in voltage across it to indicate flow of current.
So the top two waveforms are either side of the series 22K resistor, disconnected (or unloaded) and then connected to the PI input coupling cap. The first one totally unloaded shows a very small difference or error between the two CRO probes or inputs - they should be exactly the same. When feeding into the PI coupling cap, you can clearly see the waveforms overlay except for clipping at the top. If this doesn't indicate that the PI grid is drawing current I might as well give up electronics now.
Next row we have the PI grids again, both envelopes, and start and end zoomed in captures. Once again we see the pin 7 input grid signal shift downwards over time, and that the pin 2 grid moves to be centred slightly above it's no-signal starting point. The zoomed in captures show the pin 7 input grid to be centred around the pin 2 grid centre point at the start, and then it's moved downwards creating the asymmetry as previously described. The tops are rounded from clipping despite being driven from a completely independant source. This source has no sag, no clipping and is completely linear.
The bottom row is the speaker output envelope. Zoomed in captures show the output is relatively symmetrical with full headroom available at the start. After pin 7 input grid signal movement the output is what we've become familiar with from all the previous static waveforms I've posted - asymmetry generated in the PI, lower half wider with the trailing corner cut off from the PI clipping/headroom sneaking in under the power grid clipping threshold, top half is narrower and all power grid clipping dominant. You can see the top half peaks are a little bit lower at the end than at the start, which is due to B+ and power screen sag.
You do not have the required permissions to view the files attached to this post.
katopan wrote:As well as providing source impedance it also allowed me to see any difference in voltage across it to indicate flow of current.
When feeding into the PI coupling cap, you can clearly see the waveforms overlay except for clipping at the top. If this doesn't indicate that the PI grid is drawing current I might as well give up electronics now.
This source has no sag, no clipping and is completely linear.
I'm still a bit lost... if indeed there is grid current on the PI and the source is able to source the current - then why doesn't changing the bias resistor result in a significant change on the asymmetry (as your earlier CRO shot showed), i.e., if the grid is "less clipped" so less grid current is drawn, then shouldn't the asymmetry be reduced - but we don't see that.
Because of how hard it's being overdriven, it's beyond all that. Look at how big the PI signal swing is compared to what it has to be to clip the PI (not power grid clipping which is even earlier). It's huge, and just drags up the cathodes, saturates the inverting side of the PI and draws grid current same as a gain stage when the grid exceeds the cathode. Just in this case the cathode can be dragged up for a while so it needs a much bigger input than a standard cathode bypassed gain stage. The input signal swing is way beyond differentiating where the bias sits.
katopan wrote:Because of how hard it's being overdriven, it's beyond all that. Look at how big the PI signal swing is compared to what it has to be to clip the PI (not power grid clipping which is even earlier). It's huge, and just drags up the cathodes, saturates the inverting side of the PI and draws grid current same as a gain stage when the grid exceeds the cathode. Just in this case the cathode can be dragged up for a while so it needs a much bigger input than a standard cathode bypassed gain stage. The input signal swing is way beyond differentiating where the bias sits.
Got it. Ok, back to the original question, why does the PI in the TWE behave differently than other well-known amplifiers? Since its PI is a relatively standard configuration then it means that the behavior is all due to the level of the overdrive, i.e., its PI is driven with much higher voltage than in the other amps - could it be as simple as that? So if we just pile on more gain before the PI in the other amps, it would also result in the C2M bahavior?!
No, it's only the duty cycle shift created in the PI that would work for any LTP PI driven with enough signal. It's the PI headroom reduction from the 1K screen power chain resistor sag that allows it to sneak under the power grid clipping on the wider side. Any other amp that doesn't have that screen resistor sag can still have the asymmetry (I believe this is happening in a Marshall as well, but won't say that for sure until I test one, but I've seen the output of one on my CRO before and it wouldn't surprise me given it's basically the same PI & power stage) but will stay power grid clipping dominant on both sides because there's no reduction in PI headroom. The asymmetry at the power grids will still cause movement there, but with no reduction in PI headroom it stays above the power grid clipping threshold. So it won't have the change in tone and reduction in clipped rms output that the mixed mode distortion gives in the Express. My friend's JCM800 that I've had my CRO connected to (set for clean preamp and higher master to overdrive the power section) had some asymmetry, but not as much as the Express, and the height on each side of the waveform was consistent with power grid clipping and flat topped on both sides. It certainly didn't reach a peak and then back off on the wider side like the Express does. And even if you add the 1K screen supply resistor, it won't have the same dynamic unless it's got that slowed response to the PI supply node from the PI supply chain 18K2 resistance and cap.
