[west]

Doesn't a 6DJ8 have a s/n ratio 8db over a 12AX7? If that's true, then why
do many of us (P. Turner) use the 12AX7 in the phono input? I'm not talking
about manufactures with budget constraints, I'm talking about us.

west

[Patrick Turner]

west wrote:

Doesn't a 6DJ8 have a s/n ratio 8db over a 12AX7? If that's true, then why
do many of us (P. Turner) use the 12AX7 in the phono input? I'm not talking
about manufactures with budget constraints, I'm talking about us.


Let us just consider at least a few more facts, shall we?

Take a look at my 10 tube preamp schematic design shown at about 1/2 way
down the page at
http://www.turneraudio.com.au/miscel...chematics1.htm

This design WAS the flagship of my ideas for an integrated preamp with
most bells and whistles for audiophiles wanting good phono replay.

There was only one very serious limitation, the SNR when used with a low
output
moving coil cartridge without a step up tranny of below about 1.5mV of
standard rated output.
I had been using the dsign with my Shure V15 moving magnet cart
and the SNR was not just adequate, but very good, and good when compared
to many other commercial designs
using tubes, bjts, or opamps to suit MM carts of similar input levels.

I will say here that with a step up tranny of 1:10, an MC cart making
only 0.37 mV
such as the Denon 103R will produce 3.7mV at the tranny sec and the
12AX7 input stage will be
utterly adequate. Denon invented the MC and the 103 has been in
production from well before
there were quiet enough devices allowing direct connection without a
step up tranny.
but the tranny works, because the typical MC cart output impedance may
be 15 ohms,
and a 1:10 voltage ratio tranny converts this to 150 ohms,
and still much lower and therefore quieter thana typical MM cart with
Zout in thousands of ohms.
The MC was the choice of bradcasting stations and commercial operations
even though its cost was higher, which didn't matter. Domestic use of MC
was limited, children playing with MC equipped TTs were never to lead to
domestic bliss.


For those wanting a good simple effective cheap MM amp, the 10 tube
preamp with
12AX7 µ-follower input stage is excellent value.
Some MC are highish output, such as the Sumiko Blue Point Special, I
have one,
and it also works OK with 12AX7.

Another schematic was later tried using 6DJ8 and is detailed about 1/3
down the page at
http://www.turneraudio.com.au/quad22preampmods.html

This schematic initially gave SNR performance which was no better than
the 12AX7,
and it also relied on the careful choice of tubes for its SNR.
Even though theory would suggest you should get a 6DJ8 circuit to be
quieter
than the 12AX7, it was actually difficult to achieve in practice.
The paralleled 'DJ8 has gm of around 7 times more than 1/2 an'AX7.
Noise varies as the square root of gm differences, so we see
the DJ8 at 1 / 2.65 times quieter than the AX7, which is -8dB,
ie, the 'DJ8 gives slightly under 1/2 the noise of the AX7.
But only extremely well chosen 'DJ8 would actually be any better than a
well chosen 'AX7,
and 8dB is not a big quantity of anything in electronics.
A 12AT7 or 6J6 also should give better noise results than 12AX7, but
don't count on it.

The revised Quad22 phono amp does outperform the original EF86 Quad
circuit on noise
and gain etc, and I sold this very amp just before Xmas after doing a
few slight
revisions. Removing the zener diode shunt regulation of the 'DJ8 B+
supply
bypassing C1 on the above schematic reduced the zener noise at LF maybe
1 dB,
but since I first built that preamp years ago I have learnt to keep
zener diode
regulation well away from any sensitive input stages. Massive value caps
work quite well enough thank you.
Before selling it I went through my entire stocks of 'DJ8 to find the
quietest pair,
and finally got it to produce -63dB SNR unweighted, relative to 2mV,
and the amp was just barely better than the 12AX7.

After this exercize, I figured there had to be a more predictable way
to get low noise from a phono amp, yet still retain the
use of passive RIAA filtering and the blameless musicality, low noise,
low thd/imd
of triodes following the initial amp stage which always establishes the
SNR
unless you have a seriously faulty following tube set.

Take a look at the schematic at the page of the 'Rocket' at 1/4 the way
down..
http://www.turneraudio.com.au/preamp...hono-2005.html

Here we have a j-fet used as the input device to amplifiy the incoming
phono
cart signal.

The SNR of this type of input stage is about 20 dB better than the -63dB
unweighted
in reverence to 2mV.
I built the Rocket schematic after success with the earlier similar form
which gave the same SNR improvement at
http://www.turneraudio.com.au/preamp...ated-2006.html

This earlier amp remains in use and works perfectly, and is a great
match with a Denon103R MC with
only 0.37mV output.
The nature of the noise of the j-fet is mainly a fine high pitched hiss
with very little LF sputter or rumble so common from tubes, showing that
the
noise bandwidth character of the fet suits the phono amp gain profile
where there is 20dB more gain at 10Hz compared to 1kHz.
So the noise with the fet not only is much lower in measured level but
its less noticeable.
It is entirely smothered by the noise of the cart noise in an
unmodulated groove of
a vinyl recording.

Possibly I could use a cart with output as low as some Ortophon types at
0.1mV,
but I won't tempt fate. Allen Wright introduced me to the idea
I have used in these later j-fet input stages and the noise performance
of mine has been a lot better than in his implememtations in his 1988
FVP
which used NFB to do the RIAA eqing.

A step up tranny might be safer with carts of only 0.1mV.
Even if the tranny is 1:10, such a cart only produces 1mV at the tranny
sec, so you still need a quiet amp.

There are a family of j-fets which give fine input stage quietness
compared to most tubes, except
perhaps the 417A, or russian 6C45pi, the latter which likes to run with
about
30mA before it supposedly has enough gm to be equal to about
5 x 6DJ8 all in parallel.

Using 3 x 2SK369 in parallel for an input stage with 2k2 as the dc load
with 16mA Ia also is nice and quiet, perhaps 0.6 times the noise of one
fet is produced,
which isn't much difference, 4 would give noise reduction factor of 0.5,
and 8 would give 0.35.
One seems quiet enough though.

I have not bothered to try the 6C45pi, and nobody here has yet done the
work
of presenting the the real world results on such a tube's use with
phono.
I wish someone would, so I would not have to damn well do the tests
myself.

But I do know a lot about what works and doesn't work in phono amps, and
what the differences are between 12AX7 and 6DJ8 and j-fets.


I recently improved the Rocket's suseptibilty to stray magnetic fields
by providing well twisted input wires from the unbalanced RCA input
socket
to the 0V rail and gate input. I cannot stress how important it is to
get this right to reduces hum.
I have non ferrous aluminium and brass for the chassis case, and only an
internal
mild steel sheet cover over the j-fets to stop hum fields.
In future, I think I will only use all steel for the chassis case.
I also revised the Rocket 0V rails near the phono input to eleiminate a
loop in it, and found this also notched the
LF noise a bit lower. This gave the present owner greater liberty to set
his Rocket
closer to other bits of his gear without any audible hum.
Even without the latest slight mod, the amp was quieter than a
Sutherland battery powered all SS jobby
with all opamp usage and 2mm thick steel case to block magnetic fields.


In my mind I have a balanced input design with differential j-fets for
real balanced input to try
but so far I am in no hurry to try such a thing because of time, and the
difficulty
of getting better than results obtained so far with all SE topology, no
global NFB and passive RIAA eq.

Patrick Turner.










west

[Patrick Turner]

"François Yves Le Gal" wrote:

On Sat, 17 Feb 2007 09:19:22 GMT, "west" wrote:

Doesn't a 6DJ8 have a s/n ratio 8db over a 12AX7?

It can, as the grid resistor noise can be substantially lower (Rn = 2.5/gm
and Vn= SQR(Rn)) but this is only a single value/parameter, with the actual
result very much depending on the actual circuit.

If that's true, then why
do many of us (P. Turner) use the 12AX7 in the phono input?

Patrick also uses 6DJ8's and/or hybrids. Vide for example
http://www.turneraudio.com.au/preamp...hono-2005.html

The cascode input stage I show at the above schematic
has inherent lack of microphonics, and the noise of the cascode cathode
driven
grounded grid 6DJ8 is supressed because there is effectively lots of
current FB
involved with the DJ8, because the j-fets Rd is about effectively 180k,
and the anode load is 14k approx at 1 kHz.
The Allen Wright circuit basis from which I have derived my amp
works very nicely, with the fet drain current ideally suited to the tube
anode current.

Patrick Turner.

[Ian Bell]

Patrick Turner wrote:
devices allowing direct connection without a
step up tranny.
but the tranny works, because the typical MC cart output impedance may
be 15 ohms,
and a 1:10 voltage ratio tranny converts this to 150 ohms,
and still much lower and therefore quieter thana typical MM cart with
Zout in thousands of ohms.


I think there is a fundamental error here. A tranny with a 10:1 voltage
ratio will transform the source impedance by the turns ratio SQUARED; in
the example given by 100 times not 10. So it converts the MC output
impedance of15 ohms to 1500 ohms.

Ian

[Ian Bell]

west wrote:

Doesn't a 6DJ8 have a s/n ratio 8db over a 12AX7? If that's true, then
why do many of us (P. Turner) use the 12AX7 in the phono input? I'm not
talking about manufactures with budget constraints, I'm talking about us.


Principally because the 8dB improvement is obtained only under the
conditions stated. If source impedances, gains and loads differ from these
then the improvement may well not be achieved.

Ian

[Patrick Turner]

"François Yves Le Gal" wrote:

On Sat, 17 Feb 2007 15:11:23 GMT, Patrick Turner
wrote:

There are a family of j-fets which give fine input stage quietness
compared to most tubes, except
perhaps the 417A, or russian 6C45pi,

Have you tried some German Post locatl tubes such as the C3G? Out of
production but can be found easily at decent prices.

http://www.jogis-roehrenbude.de/Roeh...st/C3g/C3g.pdf

This pentode is almost exactly equal to a 6EJ7 which is nine pin mini
and which is commonly found around the world
as a sharp cut off pentode with highish gm of about 12ma/V, and in
triode the
6EJ7 is like C3g.

The 6EJ7 has i think the german equivalant number EF180, but the rare
EF280 or is it
E280F special quality is 20+mA/V. these suit mini nine pin sockets.
But all these tubes are only ever going to be marginally
quieter/better than a 12AX7.

Triode input resistance is given by 2.5 / gm, ( from RDH4 ) wheras a fet
input noise resistance
= 0.7 / gm, ( taken from the formula in a British Amateur Radio manual
);
all in theory of course. Shot noise and "flicker" noise also make LF
noise in tubes.
One little tiny TO92 package 2SK369 has gm = 40mA/V, EINR = 17.5 ohms,
giving about 0.14uV of noise at 20kHz bw.
Bandwidth is important, and noise varies as the sq.rt of bw, so with an
amp
of 40kHz bw, noise is 1.41 times more. In an RIAA filter, the effective
bw is reduced from
20kHz to 50Hz, so noise is reduced by a factor of 1 / sq.rt (20,000/50)
= 1/20.
Due to the ear's sensitivity to higher F the weighted noise figure is
less...
12AX7 EINR ( equivalent input noise resistance ) = 2.5 / 0.0012 = 2k1
approximately,
so grid input noise is typically 1.5uV, bw = 20kHz.


The noise in j-fets is very compared to tubes.

mosfets on the other hand have bad "popcorn noise".

But where you have a sack full of C3g, then of course one should use
them if possible,
and maybe 4 in parallel in triode would be quiet enough for MC.

6AC7 is another hi gm tube which is octal and has possibilities.
It has a steel case, and was used in old military gear from the 1940s,
and there are a few around. I used one for gain in a wien bridge
oscillator
I re-built last November, with direct coupled 6CM5 white CF.

To test the noise of a any given tube, set the tube up
with a typical operating condition with say supply = 300V,
RC cathode biasing, and say 100k for an R load, and with cap coupled
output using
1 uF to a following quiet amp with gain of say 1,500, and bw from 10Hz
to 20kHz.
This amp may be all tube, or an opamp; its got to be quiet, but
the noise of the tube being tested will swamp any noise produced by a
good avearge quality quiet following amp.

The grid input of the tube under test for noise has its grid grounded by
a lead no more than 25mm long,
and has dc applied to its heaters lest the
noise results are swamped by hum.

Say the tube is 1/2 a 12AX7, and gain is found to be 60 with the 100k
load.

The total gain between AX7 grid and following amp is 60 x 1,500 =
90,000.
Any grid input noise even with the grid grounded may typically be 2uV,
and 90,000 x 2uV = 0.18Vrms at the output of the test rig, and you can
inspect its nature on a CRO and listen to it via a power amp and speaker
connected.
Microphony can be checked.
Sometimes you see 1.8Vrms on the CRO, and thus input noise
must be 1.8 / 90,000 = 20uV, and the tube is quite stuffed, and fit only
for the bin.

After testing many twin triodes this way, you soon get the hang of which
ones of your collection
are worth keeping, and quiet, and you can label them as you test them at
the
amount of input noise measured this way with a felt pen so you know
later what you have when you reach into your
tube bins.

Many old triodes have high noise, and can be microphonic, and NOS they
ain't, they are
trash. Also to be checked is the idle reverse grid current across a 1M
grid biasing R
where you have grounded the grid via say 5uF instead of directly
grounding the grid.
the 5 uF cap needn't be high voltage rated since grid V dc is going to
be low, so the C can be small.
The grids should be slightly negative with respect to 0V but old worn
gassy tubes have a +ve
grid, and the voltage fluctuates, and changes with Ia, and such a tube
is just old junk,
and usable for experiments, but never for a final quality circuit.

Grounding the grid via a C should be done with quite large value C if
this is how you want to do it.
Otherwise at very low F there C has a high Z and the bias resistor noise
of 1M will
produce lots of noise compared to the tube, and all of it low F, since
the C shunts the
R to ground at HF.

1 1M R has roughly 22 times as much noise as the EIN for a 12AX7, or
maybe 44uV and where you have an amp
with only 1M at its input and high gain, such as for a microphone, then
without a low Z microphone
to shunt the 1M, the noise is really bad.
If you have 5uF shunting the grid to 0V, the bw is reduced from 0 to
20kHz to 0 to 0.0318Hz,
and noise is then 44 x sq.rt ( 0.0318 / 20,000 ) = 0.055 uV.
This may seem low, but in phono amp with high gain at LF, it is all
amplified up....
Many ppl here would have noticed how most MM amps with 1khz gain = 100
are noisier without the cart connected to the amp and with gain turned
up.

That's mainly the noise of the 47k load resistor commonly used that you
hear,
and when you connect the MM cart the 47k is shunted, and the amount of
series R
at the input in series with the EINR of the grid reduced to negligible
levels most of the time.
The total volatg of noise amplified = sq.rt of sum of the noise voltages
squared
of input resistances and grid input noise, so 2uV plus 2uV of noise in
both R become a total of
sq rt ( 4 + 4 ) = 2.83uV.

For more info, read the RDH4, twice, then do a few tests on real tubes.

One cannot easily measure 2uV. Not even 20uV. So you must amplify it,
and find out what the noise must be at the input
by calculation, since gear able to read 2uV of noise directly is hard to
find.

Some of my 12AX7 read 0.8uV of EIN noise, and I ain't ever selling them
to anyone
unless they are in a top quality preamp.

Old fashioned tube testers are useless at finding out this critical info
about
preamp tubes.


All these considerations need to be taken seriously for low noise
operation
with tubes, and a very large % of tubes in old hi-fi sets i get given to
repair fail
the noise tests dismally.

In the old days of studios having many bits of gear with lots of 12AX7,
the battles against noise
developing was constant as tubes aged, and the techs of the day would
just have to turn up
all the gains, shunt the inputs to 0V, then plug in tubes which were
quiet and move the noisier of them
further along the gain chain, say to line amps, or power amps where
their noise production
was not audible.
Early bjts had attrocious noise figures, although they improved over
time. J-fets saved the day for SS, and fet inputs
on opamps can give nicely low noise figures.
Those James Bond listening devices buried in concrete walls of embassies
around the world
often had j-fets used as the microphone amps, and gave away secrets
unemcumbered with noise.

Its difficult to hide a tube anywhere.

Patrick Turner.

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:
devices allowing direct connection without a
step up tranny.
but the tranny works, because the typical MC cart output impedance may
be 15 ohms,
and a 1:10 voltage ratio tranny converts this to 150 ohms,
and still much lower and therefore quieter thana typical MM cart with
Zout in thousands of ohms.

I think there is a fundamental error here. A tranny with a 10:1 voltage
ratio will transform the source impedance by the turns ratio SQUARED; in
the example given by 100 times not 10. So it converts the MC output
impedance of15 ohms to 1500 ohms.

Ian

You ARE quite correct, and Zout = 1,500 ohms,
and you are not asleep, so all is well.

1k5 is a lowish Z though.

if the step up was 5:1, you get 2mV if cart was 0.4mV,
and Z = 375 ohms.

Patrick Turner.

[Patrick Turner]

Ian Bell wrote:

west wrote:

Doesn't a 6DJ8 have a s/n ratio 8db over a 12AX7? If that's true, then
why do many of us (P. Turner) use the 12AX7 in the phono input? I'm not
talking about manufactures with budget constraints, I'm talking about us.


Principally because the 8dB improvement is obtained only under the
conditions stated. If source impedances, gains and loads differ from these
then the improvement may well not be achieved.

Ian

The 8dB improvement is based on comparisons of amp noise with grounded
grid
inputs.
With real world carts & loadings, the noise you get is the result of the
noise
from these rather than from the amp.

With high gm j-fets with carts connected directly to gates where the
cart has Z = 15 ohms,
noise is low compared to any MM situation and a tube grid.

Patrick Turner.

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:
devices allowing direct connection without a
step up tranny.
but the tranny works, because the typical MC cart output impedance may
be 15 ohms,
and a 1:10 voltage ratio tranny converts this to 150 ohms,
and still much lower and therefore quieter thana typical MM cart with
Zout in thousands of ohms.

I think there is a fundamental error here. A tranny with a 10:1 voltage
ratio will transform the source impedance by the turns ratio SQUARED; in
the example given by 100 times not 10. So it converts the MC output
impedance of15 ohms to 1500 ohms.

Ian

You ARE quite correct, and Zout = 1,500 ohms,
and you are not asleep, so all is well.


That's a relief ;-)

1k5 is a lowish Z though.


Indeed. The point I intended to make is that the 'free' voltage gain from a
tranny is accompanied by a commensurate increase in the noise source
resistance. This, and a range of other factors, make the selection of a
transformer ratio a non-trivial exercise. I have no experience of low noise
valve design but a lot in BJT low noise design experience. BJT
manufacturers publish NF contours versus source impedance, so there is an
optimum source resistance for minimum BJT noise - typically about 5K, so
for best noise performance you might adjust the turns ratio to achieve
close to this. With care, the transformer can provide useful gain and a NF
of no more then 0.5dB, thus significantly reducing the noise requirements
of the BJT.

You other post on measuring short circuit valve input noise was very
informative - I have saved it for future reference.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:
devices allowing direct connection without a
step up tranny.
but the tranny works, because the typical MC cart output impedance may
be 15 ohms,
and a 1:10 voltage ratio tranny converts this to 150 ohms,
and still much lower and therefore quieter thana typical MM cart with
Zout in thousands of ohms.

I think there is a fundamental error here. A tranny with a 10:1 voltage
ratio will transform the source impedance by the turns ratio SQUARED; in
the example given by 100 times not 10. So it converts the MC output
impedance of15 ohms to 1500 ohms.

Ian

You ARE quite correct, and Zout = 1,500 ohms,
and you are not asleep, so all is well.


That's a relief ;-)

1k5 is a lowish Z though.


Indeed. The point I intended to make is that the 'free' voltage gain from a
tranny is accompanied by a commensurate increase in the noise source
resistance. This, and a range of other factors, make the selection of a
transformer ratio a non-trivial exercise. I have no experience of low noise
valve design but a lot in BJT low noise design experience. BJT
manufacturers publish NF contours versus source impedance, so there is an
optimum source resistance for minimum BJT noise - typically about 5K, so
for best noise performance you might adjust the turns ratio to achieve
close to this. With care, the transformer can provide useful gain and a NF
of no more then 0.5dB, thus significantly reducing the noise requirements
of the BJT.

I thought the problem with the bjt is that its base-emitter current flow
is noisy, and one must run a typical bjt at low collector current of
typically
0.2mA for low noise, and to get higher effective gm which would be low
at such a low Ic, one then parallels
lots of bjts, and the base input resistance becomes even lower than what
it is with one
bjt, so as the R declines, so does noise.

The j-fet such as 2SK369 has gm of 40mA/V at 5mA of Id, and its input Z
is very high like a tube's,
so, but don't ask me exactly why the 2SK369 is so quiet when one might
expect the opposite
since actual Rin is high. The EINR is very low.
The MPF102 is another j-fet used for 1,001 purposes, but its gm is
far less than a 2SK369, and so its not any quiter than a triode.

I have not done too many measurements of cartridge noise and cart +
transformer
combo noise, but in the case where a cart of 15 ohms produces say 0.14uV
of noise,
when a 1:10 tranny is used the noise must become 1.4uV.
If there is a 12AX7 input tube with its input noise of 2uV
then total noise is s.rt 6 = 2.5uV.
A cart of 15 ohms may make 0.3mV of signal
which becomes 3mV at the amp input, and if the total noise of 2.5uV is
present then the unweighted SNR is 2.5uV : 3,000 uV, or about -61.5dB,
and
quite tolerable, considering the effect of the RIAA filtering where
the HF have been boosted, and a lot of the noise is reduced by the
filter along with
the level of HF.

The j-fet input stage allows the direct connection of the MC cart to
amp.
so you may have 0.14uV of noise in series with the fet noise of say
0.14uV, typical,
and total noise is 0.2uV, and SNR at this point in the circuit
is 0.2uV : 0.3mV = -63.52 dB, and quite acceptable.

Using the 0.3mV cart with 0.14uV of noise ahead of the 12AX7
with input noise of 2uV means total noise is 2.004mV; the low noise of
the MC cart
does not increase the tube noise much at all, but here the snr
is 2uV : 0.3mV = -43.5 dB, and this is quite attrocious, and the reason
why
MC carts can't be used directly connected to tubes.
The 1:10 transformer would fix the problem as explained above.




You other post on measuring short circuit valve input noise was very
informative - I have saved it for future reference.

The thing to remember is that if you have a preamp to amplify the noise
output
of a given tube at its anode, that preamp need only have similar noise
performance
of the tube under test, and still the resulots won't change much.
The noise of the tube under test will swamp the measurement of noise
even if the input tube
of the test preamp has the same input noise.
This is how it is if you calculate the noise and all the points along
the circuit.
So noise measurement is quite easy, and my preamp used to amplify
noise is based around 12AT7 gain stages and 12AU7 CF output
and there is a switched input attenuator, and volume pot
ahead of the CF.
This preamp is NEVER anywhere near as noise making as the signal from
the tube under test.

Patrick Turner.


f

s



Ian

[Ian Bell]

Patrick Turner wrote:

I thought the problem with the bjt is that its base-emitter current flow
is noisy, and one must run a typical bjt at low collector current of
typically
0.2mA for low noise, and to get higher effective gm which would be low
at such a low Ic, one then parallels
lots of bjts, and the base input resistance becomes even lower than what
it is with one
bjt, so as the R declines, so does noise.


Low collector current are indeed necessary for best noise performance in a
BJT. gm is not a parameter normally used for BJTs as the are effectively
current controlled current sources rather than voltage controlled current
sources like valves. Paralleling BJTs is rarely necessary these days.


I have not done too many measurements of cartridge noise and cart +
transformer
combo noise, but in the case where a cart of 15 ohms produces say 0.14uV
of noise,
when a 1:10 tranny is used the noise must become 1.4uV.
If there is a 12AX7 input tube with its input noise of 2uV
then total noise is s.rt 6 = 2.5uV.
A cart of 15 ohms may make 0.3mV of signal
which becomes 3mV at the amp input, and if the total noise of 2.5uV is
present then the unweighted SNR is 2.5uV : 3,000 uV, or about -61.5dB,
and
quite tolerable, considering the effect of the RIAA filtering where
the HF have been boosted, and a lot of the noise is reduced by the
filter along with
the level of HF.


Tolerable I would agree but not acceptable IMHO.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:

I thought the problem with the bjt is that its base-emitter current flow
is noisy, and one must run a typical bjt at low collector current of
typically
0.2mA for low noise, and to get higher effective gm which would be low
at such a low Ic, one then parallels
lots of bjts, and the base input resistance becomes even lower than what
it is with one
bjt, so as the R declines, so does noise.


Low collector current are indeed necessary for best noise performance in a
BJT. gm is not a parameter normally used for BJTs as the are effectively
current controlled current sources rather than voltage controlled current
sources like valves. Paralleling BJTs is rarely necessary these days.


But all devices have a gm.

Sure the bjt is a current device, but at the micro class A signal level
the gm is still there,
a certain Vbe yields a certian Vc output signal.
The equivalent model is like a gm device with low R shunt at the input.
the higher the signal, the more non linear the gm becomes, R becomes,
and what remains fairly linear is the hfe or base to collector
current...
The bjt isn't a favourite for me at all...



I have not done too many measurements of cartridge noise and cart +
transformer
combo noise, but in the case where a cart of 15 ohms produces say 0.14uV
of noise,
when a 1:10 tranny is used the noise must become 1.4uV.
If there is a 12AX7 input tube with its input noise of 2uV
then total noise is s.rt 6 = 2.5uV.
A cart of 15 ohms may make 0.3mV of signal
which becomes 3mV at the amp input, and if the total noise of 2.5uV is
present then the unweighted SNR is 2.5uV : 3,000 uV, or about -61.5dB,
and
quite tolerable, considering the effect of the RIAA filtering where
the HF have been boosted, and a lot of the noise is reduced by the
filter along with
the level of HF.


Tolerable I would agree but not acceptable IMHO.

Well fine, but surface noise from the vinyl itself swamps the amp noise
when the amp is below -60db snr, and its no use having -80 or -90 dB amp
noise
relative to the cart input signal.

If the medium conveying the signal is quieter, as CD is supposed to be,
then OK, quiet amps are obviously of benefit because all noise is evil.

But to my ears and to ears of many others, Vinyl sounds fine when its
good, and better than
a CD of the same thing from the same master tape, and vinyl refuses to
go away....
The 78 was quickly dropped in favour of the Vinyl LP.
We did have 1/4 tapes for awhile, and they could be extremely listenable
as well.
The whole idea of analog is OK when its done well, with minimum noise
intrusion.

What is the snr at the concert hall? Isn't the live performance the
measure
standard for hi-fi?

A lone violin playing to 1,200 seated fans can sound breathtaking; are
we trying to listen to the
silence, or the violin?

Patrick Turner.





Ian

[Ian Bell]

Patrick Turner wrote:

But all devices have a gm.

Sure the bjt is a current device, but at the micro class A signal level
the gm is still there,
a certain Vbe yields a certian Vc output signal.

Indeed but the relationship between the two is complex. As collector current
is proportional to base current gm is not an appropriate parameter for BJT
design.

The equivalent model is like a gm device with low R shunt at the input.
the higher the signal, the more non linear the gm becomes, R becomes,
and what remains fairly linear is the hfe or base to collector
current...

Precisely, which is perhaps why the usual small signal model is just a
base/emitter resistor and a collector base current source in parallel with
a resistor.

The bjt isn't a favourite for me at all...


They are just different is all.

Ian

I have not done too many measurements of cartridge noise and cart +
transformer
combo noise, but in the case where a cart of 15 ohms produces say
0.14uV of noise,
when a 1:10 tranny is used the noise must become 1.4uV.
If there is a 12AX7 input tube with its input noise of 2uV
then total noise is s.rt 6 = 2.5uV.
A cart of 15 ohms may make 0.3mV of signal
which becomes 3mV at the amp input, and if the total noise of 2.5uV is
present then the unweighted SNR is 2.5uV : 3,000 uV, or about -61.5dB,
and
quite tolerable, considering the effect of the RIAA filtering where
the HF have been boosted, and a lot of the noise is reduced by the
filter along with
the level of HF.


Tolerable I would agree but not acceptable IMHO.

Well fine, but surface noise from the vinyl itself swamps the amp noise
when the amp is below -60db snr, and its no use having -80 or -90 dB amp
noise
relative to the cart input signal.

If the medium conveying the signal is quieter, as CD is supposed to be,
then OK, quiet amps are obviously of benefit because all noise is evil.

But to my ears and to ears of many others, Vinyl sounds fine when its
good, and better than
a CD of the same thing from the same master tape, and vinyl refuses to
go away....
The 78 was quickly dropped in favour of the Vinyl LP.
We did have 1/4 tapes for awhile, and they could be extremely listenable
as well.
The whole idea of analog is OK when its done well, with minimum noise
intrusion.

What is the snr at the concert hall? Isn't the live performance the
measure
standard for hi-fi?

A lone violin playing to 1,200 seated fans can sound breathtaking; are
we trying to listen to the
silence, or the violin?

Patrick Turner.





Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:

But all devices have a gm.

Sure the bjt is a current device, but at the micro class A signal level
the gm is still there,
a certain Vbe yields a certian Vc output signal.

Indeed but the relationship between the two is complex. As collector current
is proportional to base current gm is not an appropriate parameter for BJT
design.

You are correct, but one still needs to consider the gm of the bjt
when establishing the gain around a circuit to establish
the effectiveness of the NFB etc.
When you measure the signal Vbe, and Vc find open gain = Vc / Vbe,
then usually its a highish figure depending on the RL,
and the gm can be approximately worked with Gm = Gain / RL.
So where you have gain = 1,000, and RL = 5k, gm = 1,000 / 5,000 = 0.2
Amps per volt.
This is handy to know for that particular bjt and its idle condition.

And when a bjt is loaded with a CCS, the Ic current change is almost
zero,
and so base current becomes tiny.
And where a darlington pair is used in a high RL situation, the Rin
becomes *very high*,
and the devices act like a gm driven article like a tube or fet, with
very little current drive required.
Emitter current FB also raises the Rin for any bjt circuit.
Typical gm is 0.15A/V for a signal bjt carrying 5mA of Ic. There is a
typical Rc, often around 20k,
and the gain equation for tubes can be used, Gain = gm x RL, or
if you said the bjt has an amplification factor, µ, ie, a figure of gm x
Rc,
then Gain = µ x RL / ( RL + Rc ).
No consideration need be given to the hfe or input resistances.
A j-fet or mosfet can certainly be considered like a tube with µ, Gm,
Rd,
and all tube formulas will more accurately apply.

The 2SK369 set up with Id quiescent = 5mA has Rd = 80k, Gm = 40mA/V,
and µ = 3,200, rather like a 6AU6 pentode which swallowed a bottle of
steroid pills.

A power mosfet such as the 2SK134 has Rd = 220ohms, Gm = 0.9A/V, and µ =
approx 200,
with Id q = 0.8Adc.

Bjts are more complex to model, and I loathe them mostly, but they seem
to work
best in signal circuits when combined into darlington pairs.
A tube can walk alone, but a bjt needs a mate to hold its hand in the
dark.



The equivalent model is like a gm device with low R shunt at the input.
the higher the signal, the more non linear the gm becomes, R becomes,
and what remains fairly linear is the hfe or base to collector
current...

Precisely, which is perhaps why the usual small signal model is just a
base/emitter resistor and a collector base current source in parallel with
a resistor.

The bjt isn't a favourite for me at all...


They are just different is all.

I still don't like bjts most of the time.

Without bucket fulls of NFB they are hopeless voltage amps.

Patrick Turner

[Arny Krueger]

"Patrick Turner" wrote in message


I still don't like bjts most of the time.

Yeah, due to competition, its hard to make money building BJT amps by hand.

Without bucket fulls of NFB they are hopeless voltage amps.

If you sensibly use enough local feedback (i.e., an emitter resistor) to
reduce a BJT's voltage gain to that of a 12AX7, the BJT will likely be far
more linear.

Sound of a dead horse being whacked in the background.

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

But all devices have a gm.

Sure the bjt is a current device, but at the micro class A signal level
the gm is still there,
a certain Vbe yields a certian Vc output signal.

Indeed but the relationship between the two is complex. As collector
current is proportional to base current gm is not an appropriate
parameter for BJT design.

You are correct, but one still needs to consider the gm of the bjt
when establishing the gain around a circuit to establish
the effectiveness of the NFB etc.
When you measure the signal Vbe, and Vc find open gain = Vc / Vbe,
then usually its a highish figure depending on the RL,
and the gm can be approximately worked with Gm = Gain / RL.
So where you have gain = 1,000, and RL = 5k, gm = 1,000 / 5,000 = 0.2
Amps per volt.
This is handy to know for that particular bjt and its idle condition.

And when a bjt is loaded with a CCS, the Ic current change is almost
zero,
and so base current becomes tiny.
And where a darlington pair is used in a high RL situation, the Rin
becomes *very high*,
and the devices act like a gm driven article like a tube or fet, with
very little current drive required.
Emitter current FB also raises the Rin for any bjt circuit.

An interesting approach, especially if you are confident with valve design
and are moving the BJT design. However the gm analogy breaks down in real
world situations simply because every BJT has emitter current feedback due
the internal emitter resistance re which happens to be inversely
proportional to collector current. The means a BJTs gm varies hugely with
collector current and is therefore far less usefull a measure than it would
otherwise be. As I said before I have very little valve design experience
but I get the impression that a valve's gm is pretty independent of anode
current.

snip

A j-fet or mosfet can certainly be considered like a tube with µ, Gm,
Rd,
and all tube formulas will more accurately apply.


FETs are very similar (in modeling terms) to valves and it is natural to
employ similar parameters and design approaches.

Ian

[Patrick Turner]

Arny Krueger wrote:

"Patrick Turner" wrote in message


I still don't like bjts most of the time.

Yeah, due to competition, its hard to make money building BJT amps by hand.

Without bucket fulls of NFB they are hopeless voltage amps.

If you sensibly use enough local feedback (i.e., an emitter resistor) to
reduce a BJT's voltage gain to that of a 12AX7, the BJT will likely be far
more linear.

Not necessarily so.

And I cannot bring dead cats back to life,
or have a dead horse move anywhere.

Patrick Turner.

Sound of a dead horse being whacked in the background.

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

But all devices have a gm.

Sure the bjt is a current device, but at the micro class A signal level
the gm is still there,
a certain Vbe yields a certian Vc output signal.

Indeed but the relationship between the two is complex. As collector
current is proportional to base current gm is not an appropriate
parameter for BJT design.

You are correct, but one still needs to consider the gm of the bjt
when establishing the gain around a circuit to establish
the effectiveness of the NFB etc.
When you measure the signal Vbe, and Vc find open gain = Vc / Vbe,
then usually its a highish figure depending on the RL,
and the gm can be approximately worked with Gm = Gain / RL.
So where you have gain = 1,000, and RL = 5k, gm = 1,000 / 5,000 = 0.2
Amps per volt.
This is handy to know for that particular bjt and its idle condition.

And when a bjt is loaded with a CCS, the Ic current change is almost
zero,
and so base current becomes tiny.
And where a darlington pair is used in a high RL situation, the Rin
becomes *very high*,
and the devices act like a gm driven article like a tube or fet, with
very little current drive required.
Emitter current FB also raises the Rin for any bjt circuit.

An interesting approach, especially if you are confident with valve design
and are moving the BJT design. However the gm analogy breaks down in real
world situations simply because every BJT has emitter current feedback due
the internal emitter resistance re which happens to be inversely
proportional to collector current. The means a BJTs gm varies hugely with
collector current and is therefore far less usefull a measure than it would
otherwise be. As I said before I have very little valve design experience
but I get the impression that a valve's gm is pretty independent of anode
current.

But for tiny voltage production, the gm remains substantially constant
for a given class A working point.

Tube gm and Ra vary considerably with Ia conditions, and their product
which
equals the amplification factor, µ, is the most constant parameter.
In small signal preamp situations, a triodes gm and Ra vary only a tiny
amount;
it is the variations in these parameters that cause the thd we see.

Say you try to set up an average signal bjt in common emitter gain stage
with rail = 40V, to get say 12Vrms output. THD of the bjt will exceed
that of a 1/2 6SN7,
B+ 300V, Vout =12Vrms, and be maybe 20 times worse.

The use of an emitter resistor and a darlington pair of bjts makes
matters far better,
so there is a large reliance on NFB to compensate for the changes
in gm and Re within the bjts.

J-fets also are attrocious voltage amps when the V rises much above
0.1Vrms output
compared to almost any tube, but they are good enough without NFB for
use
in the input stage for a microphone or phono amp.
Horses for courses.




snip

A j-fet or mosfet can certainly be considered like a tube with µ, Gm,
Rd,
and all tube formulas will more accurately apply.


FETs are very similar (in modeling terms) to valves and it is natural to
employ similar parameters and design approaches.

Its nice to know the numbers, and then to hear good music as well
because at least the numbers look about right.

I like versions of the µ-follower using darlington lower gain block and
darlington
bootstrapped follower pair.

This makes it easy to have an open loop gain of say 1,000 for one class
A bjt signal stage, then use
shunt NFB and local emitter current NFB to straighten out the
open loop result for a gain of say 20, and that's a 50 fold reduction of
thd
But the same can be done using a cascaded pair of 12AX7....

Patrick Turner.

Ian

[Ian Bell]

Patrick Turner wrote:


But for tiny voltage production, the gm remains substantially constant
for a given class A working point.


Of course, true for both valves and BJTs, but not helpful if you don't know
its value at the chosen working point. There's little point being limited
to the working point specified in the manufacturers definition of gm.


Say you try to set up an average signal bjt in common emitter gain stage
with rail = 40V, to get say 12Vrms output. THD of the bjt will exceed
that of a 1/2 6SN7,
B+ 300V, Vout =12Vrms, and be maybe 20 times worse.


An unfair comparison because you are asking the BJT to swing over 90% of its
rail voltage whereas the valve swings just over 10%. The statement about
distortion is meaningless if topology and load are not defined.

The use of an emitter resistor and a darlington pair of bjts makes
matters far better,
so there is a large reliance on NFB to compensate for the changes
in gm and Re within the bjts.


There are a lot of ways to 'make things better'. Both valves and BJTs
benefit from negative feedback.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:


But for tiny voltage production, the gm remains substantially constant
for a given class A working point.


Of course, true for both valves and BJTs, but not helpful if you don't know
its value at the chosen working point. There's little point being limited
to the working point specified in the manufacturers definition of gm.

But you can know the gm. Just set up the circuit and measure it, simple.
There is variation between manufacture data and actual useage
conditions.
The data is a guide only.




Say you try to set up an average signal bjt in common emitter gain stage
with rail = 40V, to get say 12Vrms output. THD of the bjt will exceed
that of a 1/2 6SN7,
B+ 300V, Vout =12Vrms, and be maybe 20 times worse.


An unfair comparison because you are asking the BJT to swing over 90% of its
rail voltage whereas the valve swings just over 10%. The statement about
distortion is meaningless if topology and load are not defined.

Well indeed, but if you want say 1vrms output from a preamp,
a µ-follower 6SN7 stage will give you a gain of 17, Rout = 500 ohms
approx,
and thd 0.01%, and no current NFB or shunt FB, and load = 50k.
Even at 10Vrms, there is only about 0.1% 2H mainly
using an LTP would reduce the 2H and place an extra zero in the thd
measure.
Such stages sound utterly blameless.

It also s possible with a quad of bjts, two darlingtons,
and a 40V supply, but ONLY after lots of NFB has been applied.

Fairness does not become a meaningful issue. Outcome is the meaning.

But hey, have you tried using some KSE340 with B+ = 250V, µ-follower
topology,
two darlington pairs, some shunt NFB to get gain the same at a 6SN7?

This WILL technically outperform the bottle I am sure.
Nobody ever bothers to do this, they'd rather use an opamp and +/-12V
supply.

Which sounds best?

Build it to find out!



The use of an emitter resistor and a darlington pair of bjts makes
matters far better,
so there is a large reliance on NFB to compensate for the changes
in gm and Re within the bjts.


There are a lot of ways to 'make things better'. Both valves and BJTs
benefit from negative feedback.


With a preamp, only a tiny % of the dynamic range is used,
so all that's needed is the common cathode gain stage loaded by very
high RL 30 x Ra,
and you have superb linearity without loop FB. But high Rout, so a CF
buffer is added,
and if you had an uncle, his name would be Bob.

With a 2SK369 j-fet, you get about 1% thd per volt of output without any
NFB,
and its quite ****tily attrocious.
But the RL also makes a difference to the spectra of the thd, and using
just the right
load value will reduce the 2H to a null, leaving only the pentode like
odd-numbered
H products, and these are maybe 20db down, so when you set up a 2SK369,
or its cousins, 2SK170, 2SK147 etc,
you use the RL for least THD; its 6.4k for 2SK369, and gain is 250 with
Id = 5mA,
so if there is local current feedback with a source R to reduce gain to
50.
THD is thus lowish at 1Vo, but in a phono amp stage1, where the output
is less than 200mV,
thd 0.008% and much less than that being generated further along the
signal path
where the signal is much higher at such seldom used high signal levels.
The max Vo may only be 4.5Vrms from such a j-fet input stage, its drain
sits at +9V only,
but that's more than what we ever would need.

I don't bother with bjts for phono input stages - too difficult.

Don't vurry, vee have vays and meenz to make ze amp soun goot!

Patrick Turner.




Ian

[Ian Bell]

Patrick Turner wrote:

But you can know the gm. Just set up the circuit and measure it, simple.
There is variation between manufacture data and actual useage
conditions.
The data is a guide only.



But there is no point when you can design with the manufacturers data as is
(sans gm).


snip


Well indeed, but if you want say 1vrms output from a preamp,
a µ-follower 6SN7 stage will give you a gain of 17, Rout = 500 ohms
approx,
and thd 0.01%, and no current NFB or shunt FB, and load = 50k.
Even at 10Vrms, there is only about 0.1% 2H mainly
using an LTP would reduce the 2H and place an extra zero in the thd
measure.
Such stages sound utterly blameless.


You are going to have to be more specific. I thought we were talking about
single device stages. AFAIK the mu-follower requires three triodes.

It also s possible with a quad of bjts, two darlingtons,
and a 40V supply, but ONLY after lots of NFB has been applied.


You can do it easily with three BJTs and no need for 40V supply either, and
it will drive a load a lot less than 50K.

Fairness does not become a meaningful issue. Outcome is the meaning.


Indeed, and the quantity of NFB is irrelevant too.

But hey, have you tried using some KSE340 with B+ = 250V, µ-follower
topology,
two darlington pairs, some shunt NFB to get gain the same at a 6SN7?

This WILL technically outperform the bottle I am sure.
Nobody ever bothers to do this, they'd rather use an opamp and +/-12V
supply.

Which sounds best?


First you need to define 'best'

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:

But you can know the gm. Just set up the circuit and measure it, simple.
There is variation between manufacture data and actual useage
conditions.
The data is a guide only.



But there is no point when you can design with the manufacturers data as is
(sans gm).

Yes, you could, and I have, and when I am done, I like to know
how much voltage needs to be applied base to emitter to get a collector
current change,
worked out by Vc / RL.
Some of the base voltage applied is used to power the input resistance,
but that is irrelevant.





snip


Well indeed, but if you want say 1vrms output from a preamp,
a µ-follower 6SN7 stage will give you a gain of 17, Rout = 500 ohms
approx,
and thd 0.01%, and no current NFB or shunt FB, and load = 50k.
Even at 10Vrms, there is only about 0.1% 2H mainly
using an LTP would reduce the 2H and place an extra zero in the thd
measure.
Such stages sound utterly blameless.


You are going to have to be more specific. I thought we were talking about
single device stages. AFAIK the mu-follower requires three triodes.

Two triodes are all that's required for µ-follower.
See the schematic of a typical stage with low Rout, gain = approx 42,
and very low thd
at the output of the "Rocket" at

http://www.turneraudio.com.au/preamp...hono-2005.html



It also s possible with a quad of bjts, two darlingtons,
and a 40V supply, but ONLY after lots of NFB has been applied.


You can do it easily with three BJTs and no need for 40V supply either, and
it will drive a load a lot less than 50K.


There are Ziclai Pairs also, and sure they'll drive less than 50k.
But 50k is a nice typical power amp input resistance which doesn't load
down a preamp.


Fairness does not become a meaningful issue. Outcome is the meaning.


Indeed, and the quantity of NFB is irrelevant too.

But hey, have you tried using some KSE340 with B+ = 250V, µ-follower
topology,
two darlington pairs, some shunt NFB to get gain the same at a 6SN7?

This WILL technically outperform the bottle I am sure.
Nobody ever bothers to do this, they'd rather use an opamp and +/-12V
supply.

Which sounds best?


First you need to define 'best'

I am happy with triodes.

They are best for me.

Other ppl's bestnesses are best for them, and I wish them all wellnesses
unto them.

Patrick Turner.

Ian

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

But you can know the gm. Just set up the circuit and measure it,
simple. There is variation between manufacture data and actual useage
conditions.
The data is a guide only.



But there is no point when you can design with the manufacturers data as
is (sans gm).

Yes, you could, and I have, and when I am done, I like to know
how much voltage needs to be applied base to emitter to get a collector
current change,
worked out by Vc / RL.
Some of the base voltage applied is used to power the input resistance,
but that is irrelevant.



Why? The input signal is most likely not connected base/emitter same as in a
tube it is most likely not connected grid/cathode. What you want to know,
in fact what you design for, stage gain
is



snip


There are Ziclai Pairs also, and sure they'll drive less than 50k.
But 50k is a nice typical power amp input resistance which doesn't load
down a preamp.


That's fine if all you are driving is a 50K load over a short distance.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

But you can know the gm. Just set up the circuit and measure it,
simple. There is variation between manufacture data and actual useage
conditions.
The data is a guide only.



But there is no point when you can design with the manufacturers data as
is (sans gm).

Yes, you could, and I have, and when I am done, I like to know
how much voltage needs to be applied base to emitter to get a collector
current change,
worked out by Vc / RL.
Some of the base voltage applied is used to power the input resistance,
but that is irrelevant.



Why? The input signal is most likely not connected base/emitter same as in a
tube it is most likely not connected grid/cathode. What you want to know,
in fact what you design for, stage gain
is

I do design for stage gain, and the amount of NFB applied is monitored.
So I also can estimate gain easily with gain = gm x RL, if I know what
the gm is,
and the data damn well ain't got it there, because there are too many
apps for the devices.

I measure what I am not told.

Patrick Turner.




snip


There are Ziclai Pairs also, and sure they'll drive less than 50k.
But 50k is a nice typical power amp input resistance which doesn't load
down a preamp.


That's fine if all you are driving is a 50K load over a short distance.

Ian

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

I like to know
how much voltage needs to be applied base to emitter to get a collector
current change, worked out by Vc / RL.



Why? The input signal is most likely not connected base/emitter same as
in a tube it is most likely not connected grid/cathode. What you want to
know, in fact what you design for, stage gain
is

I do design for stage gain, and the amount of NFB applied is monitored.
So I also can estimate gain easily with gain = gm x RL, if I know what
the gm is, and the data damn well ain't got it there, because there are
too many apps for the devices.


I still don't understand. Unless you apply signal directly base/emitter or
grid/cathode gm only helps know the open loop stage gain. gm varies so much
from device to device (in BJTS) that is is useless as a parameter. Good
design practice is to make stage gain largely independent of active device
parameters.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:

I like to know
how much voltage needs to be applied base to emitter to get a collector
current change, worked out by Vc / RL.



Why? The input signal is most likely not connected base/emitter same as
in a tube it is most likely not connected grid/cathode. What you want to
know, in fact what you design for, stage gain
is

I do design for stage gain, and the amount of NFB applied is monitored.
So I also can estimate gain easily with gain = gm x RL, if I know what
the gm is, and the data damn well ain't got it there, because there are
too many apps for the devices.


I still don't understand. Unless you apply signal directly base/emitter or
grid/cathode gm only helps know the open loop stage gain.

But knowing what the open loop VOLTAGE gain is EXACTLY what we want to
know
irrespective of hfe, input resistance or any current flow issues.


gm varies so much
from device to device (in BJTS) that is is useless as a parameter.

The variations are not all that huge amoung a given batch
of bjts you may buy to build say 4 gain stages in a two channel
preamp.

And the variations don't matter hugely if they vary say between
say 0.1A/V and 0.2A/V. The gain into say 5k RL will be either 500 or
1,000 approx,
and we just like to know this, and when we have reduced such gain to say
25,
the reduction of the thd will be between 20 and 40 times, and enough,
usually.

Good
design practice is to make stage gain largely independent of active device
parameters.


This means having so much gain it doesn't matter, and this is how opamp
circuits and some discrete bjts/fets and even some tube circuits are
configured.
In a follower config, the open gain at 1kHz may
be 200,000, yet the FB reduces it to just under 1.0, so thd is
reduced by nearly 200,000 times.
But a discrete gain stage with open loop gain = 500,
and closed gain 25, we only have thd reduction by a factor of of about
1/24.
( This is about the amount sometimes used in tube power amps )
So if the bjt had 2.4% at 2.4Vout, ( Typical attrocious BJT voltage amp
behaviour without any FB )
it only falls to 0.1% at the 2.4vo with the FB applied.
But at 0.24vo, thd might be 0.01%, and might be acceptable.

A 6SN7 triode would not need any external loop or current NFB to achieve
the same result or better.

Patrick Turner.


Ian

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:


I still don't understand. Unless you apply signal directly base/emitter
or grid/cathode gm only helps know the open loop stage gain.

But knowing what the open loop VOLTAGE gain is EXACTLY what we want to
know
irrespective of hfe, input resistance or any current flow issues.


That's the point, in a BJT it isn't irrespective of hfe or current flow. As
I mentioned earlier, a BJT always has some negative feeback owning to re.
If hfe is large, then open loop gain is close to RL/re so gm is basically
1/re. Since re depends on collector current (re ~ 25/Ic) where Ic is in mA,
then gm ~ Ic/25


Good
design practice is to make stage gain largely independent of active
device parameters.


This means having so much gain it doesn't matter, and this is how opamp
circuits and some discrete bjts/fets and even some tube circuits are
configured.

No it means having *sufficient* gain you ensure the circuit performs to spec
with any sample of the active device.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:


I still don't understand. Unless you apply signal directly base/emitter
or grid/cathode gm only helps know the open loop stage gain.

But knowing what the open loop VOLTAGE gain is EXACTLY what we want to
know
irrespective of hfe, input resistance or any current flow issues.


That's the point, in a BJT it isn't irrespective of hfe or current flow. As
I mentioned earlier, a BJT always has some negative feeback owning to re.
If hfe is large, then open loop gain is close to RL/re so gm is basically
1/re. Since re depends on collector current (re ~ 25/Ic) where Ic is in mA,
then gm ~ Ic/25

Huh? why drag in current change at the input all the time with BJT
circuits?

Have ye not thought ONLY in terms of Voltage gain = Voltage output /
voltage input ?
Therew are currents running all over the joint in bjts, and sure its
neat to think about
them and make ehatever conclusions of models or formulas you like,
and its nice to think of bjts as current operated devices.

Current cannot flow unless its propelled by voltage change.

With bjts, the voltage gain is voltage gain, and
a certain change between b and e causes whatever Ib to flow, and we
don't
always need to know, but whatever happens we get a change in RL current,
and even if RL = 0 ohms,
there is a current change, and so simply knowing what the gm is very
handy
for rough calcs of gain, and the later amount of applied NFB, in
whatever form is may be.

Triodes have electrodtatic NFB within, a form of slightly non linear
shunt voltage NFB.
But we still can design the tube based on knowing gm, and Ra and RL.
as we can also design gain in a bjt stage based around gm and Rc and RL.

The low input R of bjts forces us to use a darlington pair to try to
increase
Rb input, because in a shunt NFB situation with R network, the low Rb in
takes current and so the R1 of the shunt divider network has I flow
larger than the R2 from the collector
to base. In fact using an emitter R for local current NFB also raises Rb
in, the darlington
connection makes Rb in even lower, and a CCS load on the collector
reduces Ic change, and finally you have Ibin
at a negligible amount, and the darl pair can have a huge gain, and the
gain with shunt FB
then becomes an additional gain reduction and equal flow in R1/R2 of the
shunt FB network flows,
and simple equations apply. The current FB with emitter Re does nothing
to reduce the
output resistance at the collector below the value of its RL.
The shunt voltage NFB with R1/R2 causes a large drop in Rout
and we can tailor it easily, and don't ask me to quote the very simple
formula to
quantify the amount of shunt or series VOLTAGE NFB needed to reduce a
pure current
source or finite RL value to a wanted Rout.
I have it somewhere in my files but where I don't recall now.
Perhaps you'd like to derive it for us and express Rout in terms
of gm and ß, and with Rc, or RL.

Good
design practice is to make stage gain largely independent of active
device parameters.


This means having so much gain it doesn't matter, and this is how opamp
circuits and some discrete bjts/fets and even some tube circuits are
configured.

No it means having *sufficient* gain you ensure the circuit performs to spec
with any sample of the active device.


Sufficient open loop gain generally always means having a huge amount at
500Hz, because it declines at
6dB/octave above 500Hz, or thereabouts.

Specs these days usally allways means 0.001% THD at any level blow
clipping,
which means if the OLG THD = 4%, to get it down to 0.001% the gain
reduction
and THD reduction with NFB must be approx 4 / 0.001 = 4,000 times,
or 72dB of NFB.

This propels the device to sell well because nobody
wants to settle for 0.1% any more.

Sure, thw attrocious linearity of the devices is banished by the immense
amount
of NFB, ie, the "sample of the active device" characteristics are
virtually eliminated.

I leave it to others to decide if an amp made under this recipe sounds
any better
or different to a single triode stage with zero applied loop NFB.

Patrick Turner.

o

Ian

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:


I still don't understand. Unless you apply signal directly
base/emitter or grid/cathode gm only helps know the open loop stage
gain.

But knowing what the open loop VOLTAGE gain is EXACTLY what we want to
know
irrespective of hfe, input resistance or any current flow issues.


That's the point, in a BJT it isn't irrespective of hfe or current flow.
As I mentioned earlier, a BJT always has some negative feeback owning to
re. If hfe is large, then open loop gain is close to RL/re so gm is
basically 1/re. Since re depends on collector current (re ~ 25/Ic) where
Ic is in mA, then gm ~ Ic/25

Huh? why drag in current change at the input all the time with BJT
circuits?


I didn't. I was talking about quiescent collector current and how it affects
gm.


Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:


I still don't understand. Unless you apply signal directly
base/emitter or grid/cathode gm only helps know the open loop stage
gain.

But knowing what the open loop VOLTAGE gain is EXACTLY what we want to
know
irrespective of hfe, input resistance or any current flow issues.


That's the point, in a BJT it isn't irrespective of hfe or current flow.
As I mentioned earlier, a BJT always has some negative feeback owning to
re. If hfe is large, then open loop gain is close to RL/re so gm is
basically 1/re. Since re depends on collector current (re ~ 25/Ic) where
Ic is in mA, then gm ~ Ic/25

Huh? why drag in current change at the input all the time with BJT
circuits?


I didn't. I was talking about quiescent collector current and how it affects
gm.


Well its OK.I agree large changes in gm occur if Ic is varied.

Its the same for tubes, and since the gm changes with Ic, or Ia, or Id
during large collector, anode or drain voltage swings, then distortion
is created because of the gain change within a voltage cycle.
But the less the I change, the lower the distortion, and for a preamp
this is how we operate
a device, in its low current change region, and hence low gm change,
hence
distortion is lower as the output voltage becomes lower.

If CCS load is used, then you don't get any change in collector, anode
or drain current.

One could say then that gm, ie, transconductance becomes unable to
exist, because when we alter input voltage,
a voltage change occurs at collector, anode or drain, but zero current
change. How can this be so?

But certainly in a pentode or triode it is the case, and the device
operates purely
as a voltage device only, no input or output current CHANGE, just the
quiescent idle current flows.
In a typical pentode, the gain becomes = µ, or gm x Ra
and so for say 6AU6, we get 0.004 x 500k so gain = 2,000.
For a 2Sk369 fet, we get 0.04 x 80k = 3,200,
and for a signal maybe 0.15 x 30k = 4,500.
Such huge gains are usually acompanied by gross ditortions,
maybe over 10%, but a shunt FB loop
to reduce the high gain to say 25 will also reduce the distortions
by the amount of gain reductions.

We've drifted right away from the subject of 6DJ8 vs 12AX7

Both can sound well, its how you set them up that matters.

I might probably say the 'DJ8 has a hard cold sound in comparison to the
'AX7.

In preamps, the brand of a given type type makes a diffeence to the
sound;
a NOS Seimans DJ8 probably will sound different, maybe better than
a Sovtek DJ8 of recent manufacture.

Patrick Turner.

[Ian Bell]

Patrick Turner wrote:



Well its OK.I agree large changes in gm occur if Ic is varied.

Its the same for tubes, and since the gm changes with Ic, or Ia, or Id
during large collector, anode or drain voltage swings, then distortion
is created because of the gain change within a voltage cycle.
But the less the I change, the lower the distortion, and for a preamp
this is how we operate
a device, in its low current change region, and hence low gm change,
hence
distortion is lower as the output voltage becomes lower.


Agreed. All I was pointing out was that gm is a function of quiescent
current in a BJT.


We've drifted right away from the subject of 6DJ8 vs 12AX7

Both can sound well, its how you set them up that matters.


Indeed.

Ian

[Ian Bell]

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the start
I have little valve design experience (although I was brought up with
valves and my first construction projects back in the 60s were valve
based), so perhaps you can help me out. I have being playing around with a
6DJ8 stage in switcherCAD using the Rydel_tubes library. With a 50K anode
resistor and a 1K decoupled cathode resistor I only get a stage gain of 14
into a 1Meg load. I tried changing the anode resistor to 500K and the
cathode resistor to 10K and I get the same result. This suggest to me the
model is suspect. What do you think?

Ian

[Ian Bell]

Ian Bell wrote:

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the
start I have little valve design experience (although I was brought up
with valves and my first construction projects back in the 60s were valve
based), so perhaps you can help me out. I have being playing around with a
6DJ8 stage in switcherCAD using the Rydel_tubes library. With a 50K anode
resistor and a 1K decoupled cathode resistor I only get a stage gain of 14
into a 1Meg load. I tried changing the anode resistor to 500K and the
cathode resistor to 10K and I get the same result. This suggest to me the
model is suspect. What do you think?

Ian

Just an update, I mistyped the stage gain, it should be 54 not 14.

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the start
I have little valve design experience (although I was brought up with
valves and my first construction projects back in the 60s were valve
based), so perhaps you can help me out. I have being playing around with a
6DJ8 stage in switcherCAD using the Rydel_tubes library. With a 50K anode
resistor and a 1K decoupled cathode resistor I only get a stage gain of 14
into a 1Meg load. I tried changing the anode resistor to 500K and the
cathode resistor to 10K and I get the same result. This suggest to me the
model is suspect. What do you think?

Depends what Ia is and thus gm and Ra.

As Ia reduces, Ra goes high because the most constant parameter is the
µ.

µ = gm x Ra.

Gain, A, for all tubes with a fully bypassed Rk
is A = µ x RL / ( Ra + RL )

What gain does ruler on paper load line drawn across a copy of the anode
Ra curves show?

I don't use simulation, I use real world experiments with a tube in a
circuit
to really see what gain I get, or load lines.

14 sounds low.

Patrick Turner.


Ian

[Patrick Turner]

Ian Bell wrote:

Ian Bell wrote:

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the
start I have little valve design experience (although I was brought up
with valves and my first construction projects back in the 60s were valve
based), so perhaps you can help me out. I have being playing around with a
6DJ8 stage in switcherCAD using the Rydel_tubes library. With a 50K anode
resistor and a 1K decoupled cathode resistor I only get a stage gain of 14
into a 1Meg load. I tried changing the anode resistor to 500K and the
cathode resistor to 10K and I get the same result. This suggest to me the
model is suspect. What do you think?

Ian

Just an update, I mistyped the stage gain, it should be 54 not 14.

Ian

This is IMPOSSIBLE, because the maximum µ of the 6DJ8 is only 33.

Check, check, and check again.

Patrick Turner.

[Ian Bell]

Patrick Turner wrote:



Ian Bell wrote:

Ian Bell wrote:

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the
start I have little valve design experience (although I was brought up
with valves and my first construction projects back in the 60s were
valve based), so perhaps you can help me out. I have being playing
around with a 6DJ8 stage in switcherCAD using the Rydel_tubes library.
With a 50K anode resistor and a 1K decoupled cathode resistor I only
get a stage gain of 14 into a 1Meg load. I tried changing the anode
resistor to 500K and the cathode resistor to 10K and I get the same
result. This suggest to me the model is suspect. What do you think?

Ian

Just an update, I mistyped the stage gain, it should be 54 not 14.

Ian

This is IMPOSSIBLE, because the maximum µ of the 6DJ8 is only 33.

Check, check, and check again.

Patrick Turner.

My bad, I was dividing a p to p signal by a peak. The real value is 29 which
ties in with you 33 max. SO do I understand correctly, the MAXIMUM open
loop gain from a single stage 6DJ8 is only 33?

Ian

[Patrick Turner]

Ian Bell wrote:

Patrick Turner wrote:



Ian Bell wrote:

Ian Bell wrote:

Patrick Turner wrote:

We've drifted right away from the subject of 6DJ8 vs 12AX7


Drifting back in the general direction of the topic, as I said at the
start I have little valve design experience (although I was brought up
with valves and my first construction projects back in the 60s were
valve based), so perhaps you can help me out. I have being playing
around with a 6DJ8 stage in switcherCAD using the Rydel_tubes library.
With a 50K anode resistor and a 1K decoupled cathode resistor I only
get a stage gain of 14 into a 1Meg load. I tried changing the anode
resistor to 500K and the cathode resistor to 10K and I get the same
result. This suggest to me the model is suspect. What do you think?

Ian

Just an update, I mistyped the stage gain, it should be 54 not 14.

Ian

This is IMPOSSIBLE, because the maximum µ of the 6DJ8 is only 33.

Check, check, and check again.

Patrick Turner.

My bad, I was dividing a p to p signal by a peak. The real value is 29 which
ties in with you 33 max. SO do I understand correctly, the MAXIMUM open
loop gain from a single stage 6DJ8 is only 33?

Yes, max gain from a 6DJ8 is 33, and only avalaible if the load is a
CCS.

If Ra was 5k when Ia = 2mA, RL = 50k, Rk bypassed, Ea about 100V,
we get A = 33 x 50 / ( 50 +5 ) = 30.0

If you get 29, then it might be because my guesitimates are wrong,
and the gm, Ra & µ could be slightly different to what I guessed, or
because I have not included
the following cap coupled biasing R which should be above 150k, and
which is in parallel
with the 50k to make up the ac RL total.
I shouldn't ever guess, or assume anything, and I never do, when I build
a real amp, because I check everything out,
all the working points from the Ra curves on the data sheets. Just a
quick look
on such sheets tells me the Ra at any Ea & Ia and µ.

Tonight's homework for you :- work out a formula for gain including
an unbypassed resistor.


Patrick Turner.


Ian

[Ian Bell]

Patrick Turner wrote:

Yes, max gain from a 6DJ8 is 33, and only avalaible if the load is a
CCS.

If Ra was 5k when Ia = 2mA, RL = 50k, Rk bypassed, Ea about 100V,
we get A = 33 x 50 / ( 50 +5 ) = 30.0

If you get 29, then it might be because my guesitimates are wrong,
and the gm, Ra & µ could be slightly different to what I guessed, or
because I have not included
the following cap coupled biasing R which should be above 150k, and
which is in parallel
with the 50k to make up the ac RL total.

MY cct has:

RL = 25K (50K anode R and 50K load via cap)
Ia = 2.97mA
Ea = 101V

I shouldn't ever guess, or assume anything, and I never do, when I build
a real amp, because I check everything out,
all the working points from the Ra curves on the data sheets. Just a
quick look
on such sheets tells me the Ra at any Ea & Ia and µ.

Tonight's homework for you :- work out a formula for gain including
an unbypassed resistor.


If we define the gain with a bypassed resistor as A = (mu * RL)/(Ra + RL)

The with an unbypassed resistor Rk it will be

A' = 1/(1/A + Rk/RL)

Ian

[Ian Bell]

Ian Bell wrote:


If we define the gain with a bypassed resistor as A = (mu * RL)/(Ra + RL)

The with an unbypassed resistor Rk it will be

A' = 1/(1/A + Rk/RL)

Ian

I should add this is basic current feedback theory and applies to valves and
BJTs alike. In BJTs, A is generally much larger so A' ~ RL/Rk, but one
thing I have learned about valves is that their open loop stage gain is
quite small so this approximation is not normally valid.

Ian

Ian

[Patrick Turner]

Ian Bell wrote:

Ian Bell wrote:


If we define the gain with a bypassed resistor as A = (mu * RL)/(Ra + RL)

The with an unbypassed resistor Rk it will be

A' = 1/(1/A + Rk/RL)

Ian

I should add this is basic current feedback theory and applies to valves and
BJTs alike. In BJTs, A is generally much larger so A' ~ RL/Rk, but one
thing I have learned about valves is that their open loop stage gain is
quite small so this approximation is not normally valid.

Ian

Ian

I always first consider the gain with unbypased RK, so you may have 15V
at the 50k RL
of a 6SN7, so there is say 1V Vgk, A = 15.
if Rk = 2k2, Vk = 2.2 x 15/50 = 0.66V, so the Vg
must then be 1 + 0.66 = 1.66, so A' = 15/1.66 = 9.036.

Or A' = A / ( 1 + [A x ß] )

In this case ß = 2.2 / 50 = 0.044,
so A' = 15 / ( 1 + 15 x [0.044] ) = 9.036.

This equation is proved to be right in the process above.

However, its to be remembered that A or A' are negative if Vg is +ve.

This would make the equation really for A' = -A / (1 + [ -A x ß ] )

This would not work out if the basic ohms law method I began with is
viewed as the practical truth.

So is ß really a -ve value? so that -A x -ß is a positive quantity
because two -ve values are multiplied?


There is a reasonably simple process of adding the formula for A = (mu *
RL)/(Ra + RL)
to what I said above to get one formula which accepts a number on the
bottom line
to be the effect of the Rk in reducing A to A'

Patrick Turner.