Re: The Truth about Garters, a heretical view (was: Re: 71A amp)



flipper wrote:

On Thu, 14 May 2009 08:28:03 GMT, Patrick Turner
<info@xxxxxxxxxxxxxxxxxx> wrote:



flipper wrote:

On Wed, 13 May 2009 11:05:17 GMT, Patrick Turner
<info@xxxxxxxxxxxxxxxxxx> wrote:



flipper wrote:

On Tue, 12 May 2009 08:44:08 GMT, Patrick Turner
<info@xxxxxxxxxxxxxxxxxx> wrote:



flipper wrote:

On Mon, 11 May 2009 01:48:15 +0100, "Ian Iveson"
<IanIveson.home@xxxxxxxxxxxxxxxx> wrote:

flipper wrote: A few comments below in reply to a long windy post from Ian.

If Ian had simply enaged with his simulator tool and set up a garter
within it he'd have had something useful to tell us.

If anyone else had pulled out their old breadboard amp which they use
for examining the basics and trying things out then we should have been
told by now just how a real Garter behaves and whatever good or bad
resides in the idea.

Its much clearer and better to cease the blather and take action in ones
hobby shed and solder something up to find outabout.

Patrick Turner.



Well, ok. I took some time and ran simulations for common Rk, separate
Rks, Byrn's circuit, which I call doubled Rk-Fixed Bias, and the
Garter.

I used my original 6GK6 amp as the starting schematic with B+ 315V
for the doubled up Rk case and 304.567V for the singles so idle
current would be identical. In other works, I compensated for the
single vs double Rk bias drop..

Common Rk was 135 Ohms and all the others were 270 Ohms.

The doubled Rk-Fixed Bias idle current came out off by 1uA because of
rounding error when determining the equivalent 'half way' point fixed
bias voltage.

For imbalance I doubled the tubes on one side.

Currents are anode, no screen. (through OPT)

So, here's the data
Nominal Bias with 2 Ideal, Equal, Tubes.

Common Rk 35.560mA each, 71.120mA total
Separate Rk 35.560mA each, 71.120mA total
Doubled Rk-fixed Bias 35.559mA each, 71.118mA total
Garter 35.560mA each, 71.120mA total
'Broken' garter 35.560mA each, 71.120mA total

With double tube imbalance

Normal Tube 'Double' Tube

Common Rk 25.838mA 51.600mA
Separate Rk 35.560mA 40.676mA
Doubled Rk-fixed Bias 35.559mA 38.335mA
Garter 36.950mA 38.938mA
'Broken' garter 35.560mA 40.462mA

Balance Error Total mA

Common Rk 25.762mA 77.438mA
Separate Rk 5.116mA 76.236mA
Doubled Rk-fixed Bias 2.776mA 73.894mA
Garter 1.988mA 75.888mA
'Broken' garter 4.902mA 76.022mA

Knowing how lousy the balance is for a common Rk I was sort of taken
aback when I saw the total current being so close to the others. Makes
sense when you think about it but no wonder a simple PS fuse doesn't
do diddle for a runaway tube. Current shoots through the OPT but
overall current isn't a terribly large bunch more.

Indeed, and the common Rk in Quad-II is one hell of an alchiles heel.

Instead of 70mA per tube, you can have 90mA and 50mA, with one cool tube
and one glowing red, and it can stay like this for years, and the daft
old buggers too lousy to buy new tubes don't seem to notice the terrible
sound.

Well, that's consistent with the simulation I ran so maybe the 'double
tube' approach isn't far off.

Another thing of note is there's not a huge difference in total
current among the lot. The Doubled Rk-fixed Bias is the best in that
regard, as we surmised, but it's only 2mA, less than 3%.

The Garter was the best balance, even beating the double value Rk with
fixed bias by a respectable 28%, but just separate Rks are an
astronomical improvement over the common Rk.

With fixed bias, and adjustable for each side, the balance should be
perfect.

But it might not stay perfect.

If "the daft old buggers" don't buy tubes when glowing red what makes
anyone think they'd bias the thing?

Just saying, that's supposed to be the 'advantage' of self bias.
Albeit virtually self defeating with a common Rk.

'Broken' Garter is the 'starting point' circuit in the qualitative
analysis I did. I.E. It is with the grids connected to the local 'half
way' point so you can see what the tubes do without the garter
feedback but all other parameters the same. Note that, as I described,
the 'low current' tube's current increases when the Garter is
connected while the 'high current' tube's current is less than it was
without the feedback. They 'move towards each other'.

Interesting to see that total current goes down rather than just
exchanging one side for the other.

This is probably good enough for a rough relative comparison but I
wouldn't take it as representative of a real life errant tube because
doubling the tube doubles gm and that's not the case with grid leak.

As a side note, my current mirror keeps them within 5uA, with 'ideal'
resistors and transistors. Or, in other words, the balance will be
determined by the sense resistor precision and, to a lesser degree,
transistor matching (because the emitter sense resistors reduce
transistor Vbe dependency).

The trouble with current sinks for each cathode is that in AB operation
the Ek rises dramatiically.

You do this every time I mention the current mirror: go off into a
discussion of CCS when I keep telling you a current mirror is NOT a
CCS.

I know CCS doesn't work well AB and it's intuitively obvious when one
realizes tube current, unless constrained, is not 'constant' going
into B. That means the CCS forces the tube out of normal bias when
driven past Class A.

The current mirror doesn't work that way. One side is plain ole
cathode bias with a current sense transistor inside. The other side is
a *variable* current sink that sinks the same current as the cathode
bias side.

That means as the cathode bias side's current increases so does the
current sink on the second side and if the tubes are matched it would
be identical. If they are not matched it makes them look matched, as
if both were plain ole cathode bias. Except they're current matched.

Now, we know that cathode bias is not prefect either in AB but it's
better than dual current sinks and has the advantage of no adjustment
by "the daft old buggers" needed. And with the 'cathode bias' current
mirror balance is dern near perfect.

Cathode bias is sure very imperfect when you have low class A and high
class AB amounts in the power character.

Which is why I specifically said "as long as there's sufficient Class
A" down below.

Exactly what constitutes the magic "sufficient" value isn't 100% clear
and it's interesting to observe that a simple idle Ia calculation
underestimates the transition point because the tubes don't go rigidly
into cutoff, and that tends to create an average current near idle
longer than a simple Ia calculation would suggest.

The effect is much more pronounced with triodes.

The usual rule of thumb is to have 50% in pure class A with the rest in
class AB for the rated load.
This means that if you connect a higher load there is less PO total but
more class A and less AB and if you connect a lower load there is more
PO total with less class A and more AB.

I like to set up a PP pair of KT88/6550 in UL to make 60W into 3 ohms,
AB1, absolute max, with only a few watts of class A1. By the time you
get to 8 ohms more than 1/2 the PO is class A, but you only get 34W
total, but the sound is dreamy.

Fuctards at many mass made hi-end factories will never addopt such
quaint ways to set up their amps. Absolute max PO is at 8 ohms when 8
ohms is connected to the 8 ohm OPT tap. Connecting 3 ohms would destroy
their bloody amps.


And when the idle current is low, and class AB amount high with a low RL
value, the tubes are not in their linear region even while in the small
amount of class A. But the current THD is mainly even order and there is
good cancelling, so you get away with it except the THD is 3 times more
than if you stick to the rule of 50% class A out of the possible total.

And the loads each tube see while in class A can be different if the
tubes gn is different, so when you draw the loadline for a single tube
in a class AB amp it is a curved line with a slope starting off at about
1/2 RLa-a then swinging up to 1/4 RELa-a after one tube has cut off. The
transition is a kinked line with beam and pentodes, but more swayed with
UL and triode so the odd order THD due to switching or crossover
distortion becomes worst with pure beam or pentode op in class AB.


And with a current mirror on one tube to ensure Iadc remains equal no
matter what, then the same problem in class AB exists as with separate
Rk and Ck.

Yes, it acts like a 'super duper' separate Rk/Ck.

Its not as bad as with a CCS on one side with a current
mirror, or two CCS.

Quite a bit better, in fact.

Yeah, should be.

I prefer high class A% which means cathode bias is all you'd need, and
then have something to adjust the balance with visual indication.
See the schematics at the page on revised Quad-II amps at
http://turneraudio.com.au/quad2powerampmods.htm
The last effort to instil life into a pair of Quad-II was the one there
with KT90 outputs and a pair of red LED which indicates bias balance and
which also tells you other things.

Why bother with 'indicators' when the current mirror simply 'makes' it
balanced?

I like to give the owner a visual indication of the bias condition, or
when the Ek of each tube begins to wobble unevenly due to high power
being used, or when a wrong load is connected.

In the sample of Quad-II at my site where I used a pair of KT90 instead
of the original KT66, the bias balance indicator LEDs rarely ever
change brightness even when fairly insensitive 4 ohms speakers are
connected to the 8 ohm tap setting. And I have slightly less Ia in each
output tube than originally specified for the KT66.

There's another sample pair of Quad-II with KT88 in triode and with
fixed bias, and I got a very nice 20W.
The balance had hardly moved since 1998 and the same Sovtek tubes have
been used every day by the owner. He now runs fairly sensitive Chinese
speakers so 20W is more than enough.


So one might allow a 10% Ek rise then have zeners limit it. But that
isn't always too good because Ek might vary somewhat with equal Ik in
each tube. And if you have multigrids the Ik consists of Ia and Ig2, and
if Ig2 isn't equal between a pair of tubes then Ia won't be although Ik
will be.
I decided the best way was cathode bias with separate Rk&Ck and with a
high % of class A %.
I thn worked out that Ek could be stabilised by dynamically bypassing
the ac cathode currents and the THD/IMD then measueres just as well as a
fixed bias amp when well into class AB1. The ac bypassing network
including any old power transistor does not work at all while the amp
works in pure class A1. See the page at
http://www.turneraudio.com.au/300w-5-bias-stabilizer.htm

Yes, I know. I've seen it.

I'm just not convinced the extra complexity is needed as long as
there's sufficient Class A because the transient Class B doesn't shift
bias much.

Correct, but with dynamic bias stabilization you can run the tubes with
less Ia, as one can with fixed bias, and still achieve good distortion
figures.

It's been my observation that distortion is less with higher Ia even
when bias is held rigid (I.E. fixed bias) so I'm not convinced the
lower Ia with 'bias stabilization' is an improvement in the Class A
region.

The slightest transient above the class A limit will be bypassed. Ek
then does not move even slightly.

One would think, however, the 'stabilized bias' is 'some degree
better' during Class B transients but 'how much' is difficult to
quantify and depends on how much one small shift vs the other small
shift is. On the other hand, it may be swamped by the lower Ia if the
purpose is to lower idle.

With less bias current, Rk can be a larger value than used for pure
class A, so the larger Rk has more Ia regulating effect than a low value
Rk. Pda in a 6550 need only be 15W instead of 30W.

In my 300W amps there are 12 output tubes.

Having 12 pots for bias adjustment just for one channel is a royal PITA.

But you have to make it fool proof and reliable, so it has to be cathode
bias, and for reliability the Ia is low, and Pda is low.

The tubes last longer this way and the music is excellent.

I'm sure it is. The question would be if there's another means to
achieve essentially the same end results.

There isn't any other way to achieve stable cathode biasing when you run
up the signal to clipping with a sine wave in AB.

I did play with your circuit but found it difficult to adjust the
'current bypass' for minimum shift, and that was with 'ideal'
components (simulations).

Build a real circuit and play with values. Then you'll soon find out how
to change values a bit to allow a slight rise in Ek but of less than 10%
of the idle value, and between idle and the +10% the bias rises almost
as a straight line, regardless of output load.

That doesn't mean it 'fails' but that the
degree of 'improvement' may not, in real life, be as much as one
predicts from the basic theory.

Its possible to set up the bypass transistors so that Ek falls a bit
before rising just before clipping and this means your'e bypassing too
much current, ie, the transistors are letting too much current drain out
of each Ck.
If you examine the wave form at the top of each Ck with LF with normal
cathode biasing, its an ugly picture of non-linear waves and a big phase
shift. But with the dynamic bypasser, the waves at the top of the caps
have far less amplitude. The transistors merely provide a path to bypass
the same ac currents that would otherwise charge up the Ck.
The transistors therefore improove the whole AB operation.



One should also note that my circuit and yours are intended to solve
different problems. I was trying to get near perfect balance
(succeeded) and 'bias stabilization' doesn't help in that regard.
There were also serious problems in trying to meld the two so I opted
for the less complex approach. At least so far.

The bias stablization works with more than 2 output tubes where you wish
to get best AN performance but with no hassles of lots of interactive
bias adjust pots.

As soon as more than two output tubes are used you have to ensure that
Iadc for each tube is fairly well regulated, and the only way to know
all the tubes are OK and happy is to have some visual means where their
Ek across a set Rk is compared to a reference voltage and a read out is
given with a pair of LED of different colour, or a single bi-colour LED.
When I re-wired a bleedin awful VT100 18mths back I got rid of the
terrible direct coupled drive to the 8 output 6550 which had only one
bias voltage per channel of 4 x 6550, and I went to plain RC coupling
from a normal LTP with a 12BH7 each side, then had 8 bias pots. BUT, to
make things easy for the owner there is a green-red bi-colour LED near
each bias pot on each side panel of the box and as long as the tubes
have a very slightly green LED all is well and bias is within +/- 4mA.
This bitch of an amp kept blowing tubes up and blowing mains fuses. ARC
should be ashamed of their idiocy.
But I have not heard from the owner in 18mths so I know all is well
after my drastic mods to entirely re-design these amps.

Someone phoned me the other night about re-enginering a pair of
Reference 600W mono with 16 x 6550 in each channel. Same old story about
terrible reliability and fuses and smoke and bad sound.

If that job comes off, I will dump the entire existing innards into the
bin, and start all over again.

I really fail to see why anyone would ever need to have 600W for each
channel for a domestic hi-fi system.
So it would be prudent of me to try to set up the tubes in cathode bias
which would reduce the effective Ea. The existing loading will become
more suited, and there is provision for various loads, and if anyone
always only ever uses the 4 ohm outlet then the PO total of say 350W
with 16 x 6550 when an 8 ohm speaker is used will have 100W of class A
content at least.
So because so much PO is available, there isn't any need for dynamic
bias stabilization.

Asking 600W from 16 output tubes is the same as asking 75W from just a
pair. McIntosh and others try to do it. Its Bull***. One should never
ask for more than 60W, and that should only be possible at a load
slightly below the lowest one is allowed to use, ie, about 3.2k a-a when
the Ea and Eg2 for UL is at 500V. But I've seen may makers who like to
have 3.2k : 4, 8, and 16 ohms.
One should never ever use the 8 or 16 connections and ONLY ever use the
4 ohm tap which then makes an 8 ohm speaker give a load to the tube of
6.4k a-a, which is far better than 3.2k.

The only garter that I might see will be on the leg of the dancing girl
after I get paid for the work.

Patrick Turner.
.

Loading