Just trying to get some discussion going on this classic design.
In the Williamson, two features of the original design were a DC-coupled cathodyne (using the plate voltage from the preceding gainstage as the bias voltage for the grid), and 20dB of global NFB. (Citing Jones "Valve Amplifiers" 4th Ed. p485)
This arrangement of the cathodyne means you need the plate voltage of the preceding gainstage to be low enough to bias the following stage correctly. Bias it too hot and you'll run into grid current. "As a rule of thumb the grid voltage of a cathodyne will usually end up around 1/6th to 1/5th of the HT voltage, for typical operation. " (From Blencowe's online chapter about cathodynes)
Here's the problem... Triode gainstages run best when the quiescent plate voltage is around 66% of HT (Blencowe again, "Designing High-Fidelity Tube Preamps" 2016 p137)
So on the one hand you need the gainstage to ideally be around 66% of HT, but DC coupled to the following stage at around 20% Seems to me that this arrangement can only result in either the gainstage or the cathodyne operating well outside the most linear and lowest-distortion part of the curves
So could it be that the 20dB gNFB was a necessity to reduce the distortions caused by this heavily out-of-ideal operating condition necessitated by DC coupling?
I've built a Williamson using DC-coupled 12AX7 gainstage and PI, and I found I needed to use 390K on the plate resistor to get the voltage suitable for the following DC-coupled cathodyne.
And bonus question... has anyone ever had success with a Williamson design but with AC-coupling the gainstage and PI, to get both operating in the ideal part of the curve?
In the Williamson, two features of the original design were a DC-coupled cathodyne (using the plate voltage from the preceding gainstage as the bias voltage for the grid), and 20dB of global NFB. (Citing Jones "Valve Amplifiers" 4th Ed. p485)
This arrangement of the cathodyne means you need the plate voltage of the preceding gainstage to be low enough to bias the following stage correctly. Bias it too hot and you'll run into grid current. "As a rule of thumb the grid voltage of a cathodyne will usually end up around 1/6th to 1/5th of the HT voltage, for typical operation. " (From Blencowe's online chapter about cathodynes)
Here's the problem... Triode gainstages run best when the quiescent plate voltage is around 66% of HT (Blencowe again, "Designing High-Fidelity Tube Preamps" 2016 p137)
So on the one hand you need the gainstage to ideally be around 66% of HT, but DC coupled to the following stage at around 20% Seems to me that this arrangement can only result in either the gainstage or the cathodyne operating well outside the most linear and lowest-distortion part of the curves
So could it be that the 20dB gNFB was a necessity to reduce the distortions caused by this heavily out-of-ideal operating condition necessitated by DC coupling?
I've built a Williamson using DC-coupled 12AX7 gainstage and PI, and I found I needed to use 390K on the plate resistor to get the voltage suitable for the following DC-coupled cathodyne.
And bonus question... has anyone ever had success with a Williamson design but with AC-coupling the gainstage and PI, to get both operating in the ideal part of the curve?
Some of the reason behind the feedback was to decrease distortion, improve damping factor, and improve overall performance. The out-of-ideal operating range for the cathodyne wasn't a huge reason, as this was really limiting swing more than anything else. The cathodyne operates under such overall degenerative feedback that it is still highly linear when at sub-optimal operating points.
A good solution to the operating point of the cathodyne is to either self-bias it, or use a step network to drop some DC voltage. This wasn't done on the original Williamson more for the sake of stability, as there are multiple time constants wrapped up in the feedback loop as it is, and phase shift would become excessive otherwise.
A good solution to the operating point of the cathodyne is to either self-bias it, or use a step network to drop some DC voltage. This wasn't done on the original Williamson more for the sake of stability, as there are multiple time constants wrapped up in the feedback loop as it is, and phase shift would become excessive otherwise.
Ah, but that first stage does not have to run "best". There is another gain of 15 before the power tubes. The first stage never sees huge signals.
I dislike the idea because it can be hard to get the plate of a triode down around 1/4 of supply, as the cathodyne likes. However it does eliminate a troublesome pole.
Agreed, we don't need to maximise the gain from this stage - however with such a high value resistor and low current, you're right down in the dregs of the curves - the most non-linear area of the triode, thereby introducing an avoidable source of distortion. Plus it's gonna be prone to noise as well.
It just occurred to me that much of the distortion and noise that the NFB seeks to cure, is in fact an artifact of the DC coupling, which was specifically done to keep the NFB stable.
To this end I was speculating on how a Williamson-derived amplifier, but with AC coupling, having the initial gainstage and phase splitter working optimally, and with lower gNFB, say 6-10dB instead of 20, might work.
Willamson not so good
I have built Williamson in the past. Never liked it...
Mullard circuit is better. But:
a) Use Triode input
b) Direct coupled long tail pair as driver, and
c) Use triodes or triode strapped pentodes in output, and
d) Little or no global feedback
Start here:
http://www.preservationsound.com/wp-content/uploads/2011/09/Mullard_520_schem.jpg
Good luck
I have built Williamson in the past. Never liked it...
Mullard circuit is better. But:
a) Use Triode input
b) Direct coupled long tail pair as driver, and
c) Use triodes or triode strapped pentodes in output, and
d) Little or no global feedback
Start here:
http://www.preservationsound.com/wp-content/uploads/2011/09/Mullard_520_schem.jpg
Good luck
In almost any power amplifier most of the distortion comes from the output stage, so quite unaffected by the DC coupling. Almost all of the noise comes from the input stage, which will only be mildly affected by the requirements of DC coupling. You need NFB to improve things (including output impedance), so it is worthwhile ensuring that sufficient NFB can be used.aros71 said:
It appears that you missed some of the design details from the original Williamson articles as well as subsequent articles published by other authors to improve the design. Perhaps a visit to fellow member trobbins' site with its fine collection of Williamson-related articles will be beneficial.
As indicated, the distortion contribution of the assumed operation of the input triode in the 'dregs' is inaccurate - that stage's distortion is just so insignificant, compared to the driver stage, which has itself a distortion that is significantly below the output stage/output transformer.
One reason that input stage exhibits negligible distortion is that it incorporates the feedback node, making the stage gain only about x1.7 (5dB). That gain was very wideband in the original circuit, but the addition of the step network drops feedback at the end of the audio range (circa 17kHz), and so that stage gain increases at higher frequency, and so would distortion (but still pretty much negligible I'd suggest).
Another reason is likely the use of the 6SN7 - that valve beats the 12AU7 hands down in the driver stage for distortion.
One reason that input stage exhibits negligible distortion is that it incorporates the feedback node, making the stage gain only about x1.7 (5dB). That gain was very wideband in the original circuit, but the addition of the step network drops feedback at the end of the audio range (circa 17kHz), and so that stage gain increases at higher frequency, and so would distortion (but still pretty much negligible I'd suggest).
Another reason is likely the use of the 6SN7 - that valve beats the 12AU7 hands down in the driver stage for distortion.
OP here.
So what I'm getting from this is that the philosophy of the Williamson design regards the first stage running under less than ideal conditions as an acceptable cost for the benefits of DC coupling to the next stage, notably low-frequency stability, necessary due to the high amounts of NFB employed, and also due to the fact that distortion introduced in subsequent stages will greatly exceed that of the first stage anyway.
Sound about right?
So what I'm getting from this is that the philosophy of the Williamson design regards the first stage running under less than ideal conditions as an acceptable cost for the benefits of DC coupling to the next stage, notably low-frequency stability, necessary due to the high amounts of NFB employed, and also due to the fact that distortion introduced in subsequent stages will greatly exceed that of the first stage anyway.
Sound about right?
My comment is that you are inferring 'less than ideal' as an apropriate comment, when imho it isn't.
Here's my thoughts...
If we run the gainstage with a high resistor resulting in low anode voltage, the loadline will cross some of the most non-linear parts of the curves. Hence the historical wisdom about the so-called "Golden Ratio"... to maximise headroom and linearity.
The alternative to this wisdom might be to select valves which display greater linearity in the curves, particularly at lower mA, so they are more forgiving of the selection of operating point.
Perhaps this is why the 6SN7 is so favoured? Because its curves at low mA are particularly linear? As such it would suit operation in this region, where others might start distorting.
Opinion piece... I see a parallel with guitar amplifier designers here. In guitar amps, valves are selected for their overdrive characteristics... ie how they sound when operated well outside their linear area. It seems this philosophy is applicable to hifi to some extent also, when the design forces a bias point which puts the operation outside the centre, most linear area.
The "dregs" is a misconception. The curve is the same down to currents far lower than we use in audio; it looks "dregy" because the plot is linear, and the old techs did not measure low-low current. A "high" load resistor *linearizes* a triode.
The signal at the plate is maybe 2V peak, on a stage which would do 60V peak (if biased for that). THD is under 0.2%, compared to 2++% in the output bottles.
The added hiss of a stage that has, through studio to microphone, gain of 30 to 1,000, will be utterly nil.
20dB is not that high. I think I read somewhere that the Mullard designs use nearly 30dB.aros71 said:
That's shading into adding a gain stage just so that you can have enough gain to support 30dB NFB. It's an extra 32x
Yes, that is how high quality amplifiers are designed. Ensure you have enough gain to provide for enough feedback. Note that this is different from design, then add another gain stage, then add feedback.
- Status
- Not open for further replies.