Ultrapath

[dcoyle]

Here is Broskie when he didn't dislike the circuit and some more thoughts about the noise.

AikidoAmp10 - Tom Clearwater - Tube DIY Asylum

Here is the URL for his circuit http://www.tubecad.com/2004/AikidoAmp10.gif

Dan

[dcoyle]

M S said:
" I think it is strange that you cannot hear any increase in hum."

I don't. The same power supply ripple that is making the B+ at the top of the transformer dance is presented to the cathode by the ultrapath cap. The hum on the cathode tracks the hum on the anode.

Dan

[jazbo8]

Is it the correct link? It seems to be someone else commenting on Broskie's Aikido vs. the ultrapath connection.

[Dave H]

The same power supply ripple that is making the B+ at the top of the transformer dance is presented to the cathode by the ultrapath cap. The hum on the cathode tracks the hum on the anode.

I think it works like this. The hum on the B+ at the top of the transformer is presented to the cathode by the ultrapath cap but it is then amplified by the tube to make a larger signal at the anode. i.e. with the ultrapath cap there is more ripple on the anode than there is at the top of the transformer. I'll try to do a sim.

EDIT: Green trace is B+, Blue trace is anode.

SE 6V6 Standard


SE 6V6 Ultrapath

[Mike Sulzer]

M S said:
" I think it is strange that you cannot hear any increase in hum."

I don't. The same power supply ripple that is making the B+ at the top of the transformer dance is presented to the cathode by the ultrapath cap. The hum on the cathode tracks the hum on the anode.

Dan

What about the grid-cathode voltage? That is the input to the tube. You know, vary the grid-cathode voltage, and you vary the cathode current (and plate-cathode voltage). When you connect the B+ to the cathode at the top of the cathode bias resistor you are varying the grid-cathode-voltage. How can this not act like any other input and get amplified?

[J M Fahey]

To put some numbers into it, just estimated values to understand the basic concept, a simulation will of course be far more precise but may show "a number" without explaining the concept:

Suppose it's half of a 12AX7, with 100k load , 1k5 cathode resistors , a real classic stage.

Why a 12AX7 which wasn't even mentioned, and why those values, instead of the example we were using?

Because, as you'll see later, it does not change the basic problem at all, and saves me a lot of Math, just by quoting widely known parameters.

In a 12AX7:

Plate impedance (the internal tube value) is often given as ~68k under those circumstances.

We use a 30uF cap as cathode bypass.
I don't mention brand or construction type because that's irrelevant, all that matters is *capacitance* .
Of course, it must also stand the applied voltage.

We will have some ripple at the +V line, unknown value as it stands, we should need to build the actual supply and rest of the circuit to be sure, but for analysis let's throw some values and see where that leads us.

Let's say we have 1mV ripple.

In a conventional circuit, cathode bypass capacitor from cathode to ground, stage gain around 50X, ripple at the anode will be that present at the +V line , attenuated by a voltage divider: load resistor/plate resistance: 100k/68k .
1mV * 68/168=0.4mV ripple at the output

In an Ultrapath circuit, 1mV ripple at +B will be present at the cathode and will be amplified, now we'll have : 1mV * 50X = 50mV
Am I missing something?

Let's adjust this very crude analysis a little and answer objections before they are rised:

1) The waveform problem:
the ripple waveform is not a sinewave but a 120Hz sawtooth

2) the attenuation problem:
the coupling cap is a highpass filter in series with the cathode resistor, and will attenuate the 120Hz signal (we must consider it signal by now, since it's being injected in the signal path and amplified) so it won't be just amplified 50X , it will also be attenuated something before that.

How much?

I'm not a human simulator nor have tube models loaded in my brain, but know what's going on and can roughly estimate values.

At the cathode we have two resistances in parallel : the 1k5 cathode resistor and the tube internal cathode resistance: 1/transconductance .

*Could* find it from the datasheet graphs but trust me, it's about 1k , so at that cathode we'll have some 600 ohms impedance to ground.

Capacitor impedance at 120Hz? : ~44 ohms

*IF* ripple were sinusoidal, we'd have it attenuated by: 600/644 = 0.93X so instead of 50mV ripple at the output we'll have "only" some 46 mV .

Compare 46mV undesirable ripple mixed with desired signal in the Ultrapath connection to ~0.4mV in the conventional, boring, old fashioned connection and I fail to see any advantages in the "new revolutionary" circuit.

Chapter 2:

now to estimate (or "Calculate Light" [tm] ) possible influence on stage distortion.

To have any influence (audible or measurable) , it must either provide some kind of negative feedback which reduces distortion, or provide some kind of nonlinear load where its nonlinearity runs the opposite way of the original one which is annoying, so they compensate.

a) feedback.
we already know that it sends to the cathode, to be amplified by the tube, a slightly attenuated sample of whatever is riding on the +V line .... do we have Audio/signal there?

How much?

Whatever signal is at the plate, will appear at the +V line attenuated by an RC lowpass filter: load resistor, in series with +V filter cap impedance.

Suppose we use a 22uF cap for local decoupling, and let's estimate signal levels at, say, 1kHz .

Why 1 kHz?

Well, it's as good as any other, a typical Audio test signal, and is *significative*, it's smack in the area where we all hear very well, transducers reproduce it well; anybody worried by my choice can repeat it using any other Audio Band one he chooses.

Well, a 22uF cap shows some 7 ohms impedance @ 1 kHz, so signal attenuation will be some 100007/7=~14000X
So possible feedback signal will be some 14000 times smaller than plate signal.

What does it mean?
Let's see a few choice cases and compare them:

* output signal is 100% present at the cathode (and opposite phase, of course) : 100% NFB , we end up having stage gain=1
It will be slightly less because open loop gain is not infinite like in, say, a perfect OpAmp but we are close enough for a conceptual analysis.

* output signal is attenuated by 10X .
We still have measurable and audible NFB because open loop gain of this triode stage is around 50X and we are reducing it to around 10X ; that 5X gain difference works to reduce distortion by 5X , a sizable and useful value, let's say that we go from 5% distortion when the stage is driven *loud* (to 10V or higher) and we reduce it to some 1%
COOL !!!!!!!

NFB was a truly Revolutionary concept which advanced Audio enormously.

* output signal is attenuated by 100X , so *possible* gain is "reduced" to 100X ... but .... we only have 50X available ..... we may still have a little gain and distortion reduction, barely measurable in a Lab, practically impossible to hear

* output signal attenuated by 14000X
Absolutely unmeasurable and impossible to hear gain reduction/distortion lowering/whatever.

So Ultrapath, as shown, can not "improve distortion", not even in the Lab , even less by hearing.

And it will increase ripple injection, big time.

"But some claim to hear an improvement, and write articles about that"

You can write *anything* , specially in the Internet ... I won't argue, just wrote above a little conceptual analysis, with rough nombers to see what order of magnitude we are talking.

A simulation will basically confirm it, with more precise numbers.

[dcoyle]

Yeah it is the link, inside that there is a second reference, which is the 2nd URL I added to my post. Aikido is Broskie's signature circuit and he's got the ultrapath connection there, without the cap cathode bypass he favors over ultrapath in the article cited expressing his negative views.

Dan

[dcoyle]

M S said

"How can this not act like any other input and get amplified?"

Ahead of the output tube there are two inverting (12ax7) stages, which also have the imprint of the p.s. noise, so the the noise is impressed on the signal at the grid of the output tube, as well as on the anode and cathode, all in the same phase.

Dan

[uneumann]

At the risk of yet another line of inquiry... how about a measurement? Can you put a scope on the B+ point where you tie the film cap? I'd say AC couple it at about 1V/div (or thereabout). When you're playing do you see any signal there? If there is anything in the range of a volt or more, then it will have an impact on the cathode voltage and likely on the amp tone. If there is no signal there, I'm at a loss for what could cause any difference.

My 2 cents of theory: A class A stage should show little or no signal at that B+ point - until you hit stage overdrive. It's possible that at that point, if the clipping is asymmetric, you'll see a change in current draw from the PS. Once that happens B+ may rise or fall - which may impact the cathode bias (and distortion) depending on the time constants involved.

Since nobody measured anything yet, I took to the bench with my THD Univalve. It's a Class A amp with bypassed cathode bias. Its driven by a 200ms burst at 220Hz, sort of like a short plucked note. The idea is just to get some impulse to the output stage and see how the B+ line behaves. Since the whole premise of this discussion is what happens when you hook a cap from B+ to the cathode, lets see what the B+ line is doing.

First picture is at high level, but not clipping. The B+ line is pretty stable as expected in a class A stage.
The upper trace is the B+ line right at the OT. Its over 400V, but it's AC coupled here to show variations and the scale is 5V/div.
The lower trace is the speaker output into an 8 ohm dummy load. It's also 5V/div, so there is about 5V peak clean output (~1.5W)



The second pic shows what happens as output stage clipping happens. The asymmetry of the waveform causes a decrease in average current from the B+ supply. This happens when the output tube is warm biased so saturation happens before cutoff. The opposite shift would occur for cold bias. I measure no signal cathode bias as +22V.



The third pic shows high overdrive. Output clipping is more sever and B+ variations are higher.



I'm not changing the cathode bypass cap in this amp - it's still connected to ground. BUT, if I tied it to B+ I suspect these variations would couple into the cathode and impact bias and therefore clipping and tone in a dynamic fashion. Since the B+ variation is positive, if coupled to the cathode it would increase or cool the bias and decrease the clipping asymmetry. It would make the amp cleaner sounding, which is what dcoyle claims he hears. Basically, this provides a NFB for DC - not AC. It's a NFB to center the tube bias. Since the FB is AC coupled to the cathode, much also depends on the PS caps and the cathode bypass cap. You have to scope the actual amp (or spice it) to see what's happening in that exact circuit.

There may be more to this circuit change than these B+ shifts. (I know I'm ignoring the noise issue here.) But, this is an example of a time-varying overdrive behavior which happens all over the place in tube amps. I suspect many people know about these shifts in voltage, but these behaviors are also often overlooked in discussions about tone and designs.

Lastly, having just written about book about this with Malcolm, it's sort of fresh in my mind.

[Mike Sulzer]

M S said

"How can this not act like any other input and get amplified?"

Ahead of the output tube there are two inverting (12ax7) stages, which also have the imprint of the p.s. noise, so the the noise is impressed on the signal at the grid of the output tube, as well as on the anode and cathode, all in the same phase.

Dan

And exactly the same amplitude? These are triodes, closer to a controlled voltage source than a controlled current source. How utterly unlikely it would work out the same! Have you actually shown that this is true, or are you just assuming it?

[J M Fahey]

I think it works like this. The hum on the B+ at the top of the transformer is presented to the cathode by the ultrapath cap but it is then amplified by the tube to make a larger signal at the anode. i.e. with the ultrapath cap there is more ripple on the anode than there is at the top of the transformer. I'll try to do a sim.

EDIT: Green trace is B+, Blue trace is anode.

SE 6V6 Standard
[ATTACH=CONFIG]37108[/ATTACH]

SE 6V6 Ultrapath
[ATTACH=CONFIG]37109[/ATTACH]

Why, of course, Ultrapath is a guaranteed way to increase ripple present in signal, with no distortion/noise/whatever advantage.
No wonder *Nobody* uses it.

[gui_tarzan]

Why not add some carefully tuned spark gap from +B to cathode, so we also add some tasty buzz to the signal?

Would that be .030, .045 or .060?

[Mike Sulzer]

I'm not changing the cathode bypass cap in this amp - it's still connected to ground. BUT, if I tied it to B+ I suspect these variations would couple into the cathode and impact bias and therefore clipping and tone in a dynamic fashion. Since the B+ variation is positive, if coupled to the cathode it would increase or cool the bias and decrease the clipping asymmetry. It would make the amp cleaner sounding, which is what dcoyle claims he hears. Basically, this provides a NFB for DC - not AC. It's a NFB to center the tube bias. Since the FB is AC coupled to the cathode, much also depends on the PS caps and the cathode bypass cap. You have to scope the actual amp (or spice it) to see what's happening in that exact circuit.

Why not just re-bias the circuit so that the clipping is symmetrical?

[uneumann]

Why not just re-bias the circuit so that the clipping is symmetrical?

sure. you certainly could. the point in this thread is that dcoyle hears something and I'm just try to explain what may be happening. I don't have his amp so I'm showing something that happens on a similar amp. I wish he would put a scope on his.

The larger point though is that time-varying shifts do happen and they can and do impact tone. IMO there should be more awareness of them and consideration of them in amp designs. Its worth considering them as a means of making an amp responsive. If I'm right about dcoyles case, he now gets a variable clipping *during* the notes he plays. It can make the amp very responsive to how he picks and mutes his strings. He may like it, or not, but its worth knowing about these effects if you make mods or design amps.

I could go on for pages... but I already did. 

[dcoyle]

Mike S said

"And exactly the same amplitude? These are triodes, closer to a controlled voltage source than a controlled current source. How utterly unlikely it would work out the same! Have you actually shown that this is true, or are you just assuming it?"


I agree Mike, it is unlikely to be the same.

I do not have an explanation to offer for why I am not observing the noise you predict.

If this amp is having a change in noise, it is at a level that is not apparent to me. A slight reduction in distortion is apparent to me.

Dan