cathode follower: why the bad rep, and when sounds good?

GroupDIY Audio Forum

Help Support GroupDIY Audio Forum:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

jhaible

Well-known member
Joined
Jun 24, 2004
Messages
530
Location
Germany
I must admit I'm a little bit confused about cathode followers.

Why do they have a reputation of sounding bad, and when do they sound good?

Apart from the obvious (bad design which makes them oscillate at RF etc.), I mean. And apart from loading them too much.

For instance, in phono preamps, you often see a CF between the last voltage gain stage and the volume potentiometer. Or you see them as output stage of a HiFi preamp, to improove cable driving capability.
Which all makes sense from an impedance point of view. And yet they are frowned upon by many "High End" audio builders. Is there a known technical reason for that?

And if so, does it also apply to the WCF, or to CF with current source loading? I've seen one circuit that uses a transitor constant current source, and a transistor emitter follower to feed the CF's output signal to the anode, to have the CF tube working with both, constant current and constant voltage. Constant current only for small load currents, though: Which makes me wonder if I need a CF then, at all ...

But really, at the heart of it, I don't understand the mechanism that would make a CF sound bad at all.

Can someone enlighten me?

(If it's been discussed before - and I suppose it has - just point me there.)

Thanks,

JH.
 
I've done some CF designs with supertriodes and sand current sinks that I am told sound fine, and this from someone who pays a lot of attention to the buzz and if anything was prejudiced against the CF.

Here is an opinion (warning---a very messy translation) which if I read correctly suggests that despite the abundance of higher harmonic energy, with light loading and moderate signal swings the CF is satisfactory:

http://digilander.libero.it/paeng/design_frame.htm

The White follower probably compounds the higher order harmonic generation and is more global in its feedback, but I haven't seen any results or analyses. If you know well your load Z you can do things to support the output current demands and unload the CF (or any other output) which always helps, and is potentially much faster than "Stasis" style topologies like the WF. But such alternatives won't handle arbitrary loads, though you could get fancy and have real-time Z characterization (I'll let JR write that code :razz: ).



I'm completing (I hope) a design right now that uses cathode-coupled pairs but no CFs per se, partially to address anti-CF prejudice but also as part of another trick. Boyk has assured my friend that there is no new tube topology under the sun, and he may be right, although I haven't seen this one before.
 
[quote author="bcarso"]
http://digilander.libero.it/paeng/design_frame.htm
[/quote]

Thanks for this link!

If I got that right, I could sumarize it like this:

(1)
The CF turns a well-natured, low-harmonics distortion into something with more harmonics at lower level, just like any other method of negative feedback.

(2)
As long as levels / loading are small, the resulting level of distortion products is so low that its nature of many harmonics doesn't matter

(3)
This changes with larger levels and with heavy loading.

What I still don't quite get, is the transition from (2) to (3). I was under the impression that with such heavy (and un-delayed) feedback as in a CF (or even an emitter follower), you'd have a very rapid transition from (2) to (3): when clipping occurs.

What is it that I still don't see here?

JH.
 
Yeah that guy has a lot of good stuff on his site---I hadn't seen most of it until a couple days ago. I had only looked at the one section about high performance triodes. Someone who is fluent in Italian and English and has some time could do a great service by assisting him in translations.

I'm guessing the short answer to your questions is to look (see some previous exchanges in threads recently about this) at the curve-of-growth of the harmonics as a function of harmonic number. For the typical series expansion, representing typical weakly nonlinear systems, second grows, as a percentage of total output signal, directly as the signal; third grows about as the square. And so forth. I've seen, in sim, systems for which this rule does not hold, but I haven't been able to investigate them to satisfaction, and some of the "results" may have been computational errors.

(Also when you are in hard-limiting typical of sand state, the series expansion is getting pushed probably beyond its domain of applicability)*

So as low as they might be at a given level, one could get some significant high-order well before scope-obvious clipping.



*As Prof. Virginia Trimble once said in a lecture, pointing to a region of a state diagram for some astrophysical system, "Over here the equations of state go soft."
 
CFs get bad raps when people don't know the difference between low small-signal impedance and low power impedance. Or just because they run out of things to knock.

> The White follower probably compounds

A well optimized WCF gives high cancellation of Gm variation with load current, much more so than any SE stage. To a first approximation, distortion becomes dang near zero. Even with peak load current higher than idle current.

WCFs are rarely found optimized.

If the load is higher than Rp, there is probably a better way to deploy two triodes.

WCF does not make sense until the load is near 1/Mu.

If you can use a transformer, you do not want a WCF.

Most often the top resistor is far too high. If it isn't in the neighborhood of 1/Gm, it's wrong.

It is worth noting that White was looking for very low-Z and asymmetric signals. In that case, a too-high top resistor is not fatal.

When you get near optimum, you can start hunting for the null(s). For lowest 2nd, the top resistor must be offset from 1/Gm by an amount which is load-dependent. For lowest 3rd, the input grid must be near but not exact halfway up the rail.

My old WCF sims are a mess of lost libraries. Simulated THD can be point-oh-oh, and this does not rise for loads down to zero. That's amazing for two triodes. However small changes in bias and resistors bring THD up to merely "point", and typical designs are lucky to do so well unless level and load are trivial.

runeight at HeadWize may have more notes, but he went off onto Futtermans.
 
On Pultec MB1 it sounds very good. It´s a pair of 12au7 triode sections in parallel cathode follower for driving a very high inductance triad 15kohms transformer primary... WOW! :shock: Talk about overkill!
 
Following this discussion so far, I did some sims using a rough model for the 6C45. As expected, with a current source and no loading a single triode is damn near perfectly distortion free. With resistive loading you see mostly second, as expected, a little third, and much lower higher-order (the latter may be at the limits of the simulator---the numerical limit looking at the generator directly is at about -290dBr).

Note all the "dB" below are dB relative to the output fundamental. Also the source is a voltage generator with zero Z, so there are other distortion effects not accounted for. The small modulation of input capacitance with swing will degrade things some with nonzero source Z.

A White follower helps the overall distortion driving the load, but the energy in higher-than-second goes up a lot. However, it doesn't grow according to the typical behavior as a function of level. There may be something fortuitous going on based on the behavior of third for example.

Examples: 1kHz, 500mV peak in: Single triode follower biased at 40mA, ~150V, loaded with 1kohm: 2nd -94dB, 3rd -145dB, 4th -194dB, 5th -242dB.

White follower (second C45 is current sink for top triode cathode; 1kohm in lower cathode and grid adjusted for about 40mA; 20k in lower grid with cap coupling to a 3kohm load in the upper triode plate): 1kohm load: 1kHz, 500mV peak in: 2nd -115dB, 3rd -139dB, 4th -143dB, 5th -143dB, 6th -145dB, 7th -147dB. And there is energy beyond that, and it's not falling very much. By comparison the simple follower reaches the numerical generator levels by about the 6th.

At much higher levels, 10V peak in, the simple follower gets to residuals at about 13th, with the 2nd at -68dB and 3rd at -93dB, but White has stuff no lower than -160dB everywhere, with 2nd at -90dB, 3rd at -157dB, 4th at -127dB, 5th at -132dB, 6th at -134dB, 7th at -136dB, 8th at -144dB, 9th at -158dB, 10th at -150dB, etc.
 
Although it is easier said than done, at least without letting sand creep in, if one picks off the plate current of the upper White tube so as to keep the plate voltage constant, the growth of high-order distortion products is drastically reduced according to sims.
 
[quote author="bcarso"]Although it is easier said than done, at least without letting sand creep in, if one picks off the plate current of the upper White tube so as to keep the plate voltage constant, the growth of high-order distortion products is drastically reduced according to sims.[/quote]

You mean, something like adding an emitter follower (or source follower) to drive the anode of the upper tube?
That's what's described in one of the articles that made me ask this question on the forum: http://www.vacuumstate.com/various/SP-15_Article.pdf
(He applies this to a CF with constant current load, not to a WCF. But he sounds very negative about ordinary CF, and totally different about a constant current / constant voltage CF. Allthough, with a heavy load, there's no constant current either, of course ...)

JH.
 
Ah yes that article---in fact I corresponded with Wright when I saw that, and went on to buy his book, quite a while ago.

What I meant with the WCF is just keeping the plate voltage fixed, versus sampling the upper plate current with a resistor and coupling from there to the lower tube grid. PM me with an email that accepts modest-sized attachments and I can send you some schematics (simplified ones---it's not like I've built these latest topologies!).

Allen takes it further by bootstrapping the CF plate to the cathode, approximating constant plate-cathode voltage. This could also be done and might have advantages.

BTW he makes a little blunder when he calculates preamp gain and tacitly assumes that his FET has unlimited transconductance, thus giving him 20dB of change in gain between 150 ohms and 15 ohms. He also doesn't say much about the effect of the FET drain working into the highish load Z of the tube's cathode. Nonetheless, it's probably a pretty good preamp.
 
Here is my room heater (idle current should be half of max output current for hi-fi, or equal to a max output current for high-end):

tower.gif


The same idea as in White's CF, but no feedback loop (only a local feedback in the emitter of the 1'st transistor that converts voltage to current variations)
 
Why not use a choke? If the plate load should be about 1/Gm, the reactance doesn't have to be very high. The plate voltage would be essentially constant due to the low DCR of the choke.
 
[quote author="NewYorkDave"]Why not use a choke? If the plate load should be about 1/Gm, the reactance doesn't have to be very high. The plate voltage would be essentially constant due to the low DCR of the choke.[/quote]

On which exact frequency it should be 1/Gm?
 
[quote author="NewYorkDave"]20Hz. Then use a conjugate load across the choke to make it seem resistive through the audio band.[/quote]

So, where's the benefit of the choke if we come again to a resistor?
 
[quote author="NewYorkDave"]Why not use a choke? If the plate load should be about 1/Gm, the reactance doesn't have to be very high. The plate voltage would be essentially constant due to the low DCR of the choke.[/quote]

If we're talking about the CF plate load for WCF purposes, to the extent that the choke is big enough the current sensing will entail a voltage swing, which affects the distortion.

A "grounded" grid stage to isolate the voltage swing helps a lot, although you can still see the effect of a finite voltage swing.

I do like chokes though when high AC Z's are needed. You can also make a useful simulated inductance with a CF, two Rs and a C. This will reduce the size of voltage rails you'd otherwise need.
 
[quote author="NewYorkDave"]The choke shorts the resistor at DC, which means the DC level at the plate remains essentially constant.[/quote]

In respect to cathode voltage?
 

Latest posts

Back
Top