[Ian Iveson]

Ian Bell wrote: [below]

Thanks very much, Ian, especially for the model. Sorry for
the delay: I've been fully occupied with the crisis in
Palestine.

When time permits I'll compare the two models and try to
produce a valid composite. In passing, that seems like an
awful lot of decimal places!

Cheers, Ian

in message ...
Ian Iveson wrote:
Ian Bell wrote
I was curious why Jones chose to bias his test rig at
Vg=-3.4Volts but with tubes with mu about 20 and with
20Vrms output you need to bias several volts -ve to be
sure to be away from grid current. This does not mean
that this bias point produces the lowest distortion.
Simulation showed biassing the lower triode also at
about 1.3V should give even lower distortion but of
course the simulation does not simulate grid current.
Of course?

Ah, well, perhaps not. I have tried very many models and
all but one fail to accurately model the normal
operating region where the grid is negative wrt the
cathode. The one that does model the normal operating
region accurately (probably because it is based on The
Audio Designers Tube Register curves which are the
measured curves of actual tubes) does not include a grid
model. Maybe I can make a hybrid and nick the grid
model from the Duncan Amps models.

That would probably be OK but watch out for "LD" that
links the grid model to the anode model. You may need to
rename it, or add it's derivation into your anode model.


Thanks for the tip. I'll watch out for that.

I say probably because in reality grid current has a
relationship to anode current, probably, because I guess
it decreases perveance (less free electrons available so
anode current should fall such that total cathode current
remains the same, all other things being equal). But it
should be a negligible error where grid current is small.


Yes, and re-reading RDH4 on grid current there seems to be
several contributory factors and I doubt the spice model
includes them all.

* Duncan Amplfication Generic Triode Model (Spice 3F4
Implementation)
* Copyright (C)1997-2002 Duncan Amplfication
* Unauthorised Commercial use prohibited
* Please refer to documentation at
http://www.duncanamps.com


.SUBCKT NH6SN7GTB A G K
* ANODE MODEL
BLIM LI 0 V=(URAMP(V(A)-V(K))^ 1 )* 0.0037
BGG GG 0 V=V(G)-V(K)- 0
BRP1 RP1 0 V=URAMP(-V(GG)* 0.02 )
BRP2 RP2 0 V=V(RP1)-URAMP(V(RP1)-0.999)
BRPF RP 0 V=(1-V(RP2)^ 2 )+URAMP(V(GG))* 0.002
BGR GR 0 V=URAMP(V(GG))-URAMP(-(V(GG)*(1+V(GG)*
0.006167 )))
BEM EM 0 V=URAMP(V(A)-V(K)+V(GR)* 19.2642 )
BEP EP 0 V=(V(EM)^ 1.4 )*V(RP)* 0.0000189
BEL1 EL1 0 V=URAMP(V(EP))
BEL EL 0 V=V(EL1)-URAMP(V(EL1)-V(LI))
BLD LD 0 V=URAMP(V(EP)-V(LI))
BAK A K I=V(EL)
* GRID MODEL
BGF GF 0 V=(URAMP(V(G)-V(K)- 0 )^1.5)* 0.000213
BG G K I=V(GF)+V(LD)
* CAPS
CAK A K 0.0000000000007
CGK G K 0.0000000000024
CGA G A 0.0000000000039
.ENDS

So I re-biassed the bottom tube to 1.3V and check the
distortion. With a 120K in series with the oscillator,
grid current distortion began at 6Vrms output. Below
that the distortion was exceptionally good. At 5V rms
is was just 0.04%. Distortion at 2V rms was hard to
measure as it expect it approaches the limit of both
my distortion test set and the oscillator distortion.
I would estimate it to be no more than 0.02% at 2Vrms.
Thanks, Ian. Nice to see actual measurements of real
circuits.

I wonder if you measured at any intermediate operating
points, in between the two you mention? Just wondering
if the change in distortion is a trend or a blip.

Yes I did, I just gave a sample in the summary. I
measured several tubes in 1V increments from 1V to 10V
rms and then in 5V increments thereafter. As expected
the distortion is very close to being directly
proportional to signal level as expected until the onset
of grid current when it rises steeply. Only the lower
level measurements are suspect as they are approximately
double the distortion of the oscillator itself.

Right, I got that bit, and I can see why you expected
what you got. What I was wondering, though, was whether
you had tried a range of bias currents in between the two
you mention, to establish the relationship between
standing current and distortion for the same signal. I
got the impression that you were asserting that lower
standing current is related to lower distortion, but it's
not very clear...perhaps I'm missing something. The
question is slightly ambiguous, because of variation in
mu and therefore in gain for different standing currents,
but that shouldn't be so great as to prevent reasonable
comparison. There is also the problem of ensuring that
the bottom anode remains set at half the supply voltage.
Two points aren't enough to establish a trend. I ask
partly because it seemed to be the question that you set
out with: why did MJ bias at such a level?


What I did do was:

1. Try a lot of tubes at 8mA and -3.4V (bottom triode) and
8mA 1.3V (top triode) and got consistent results between
tube types and also got distortion ruoghly proportional to
output level.

2. Tried a few tubes at 8mA and 1.3V both top and bottom
triodes and got lower distortion at lower levels but grid
current at higher levels. This presumably means that a
lower bias voltage means lower distortion at a given
current though it is by no means conclusive.

3,. Following Patrick's comment I did a quick test
yesterday with both triodes biassed at 5mA and -5V
(actually turned out to be 4.55mA and -4.55V). I got
exactly the same distortion figures at 2V and 20V rms as
in the original experiment (1 above). Frankly I expected
the distortion to go up at the lower current but it did
not - this is however one test on one tube only.

I realise that standing current is also related to
maximum signal capability, but still wonder if I want to
amplify a small signal, where a low standing current
would be sufficient to stay far enough clear of grid
current, will I always get greater distortion by raising
the standing current? Or have I got it the wrong way
round?

I am not sure but I would have expected distortion to
decrease as standing current increases simply because you
move away from the region where the curves bunch. And if
you look at the spacing between curves they seem to me to
be less evenly spaced as the bias voltage gets more
negative which does explain why test #2 gave lower
distortion


Whichever, I'm still wondering about sweet spots.


Me too. Test 3 seems to show that nearly halving the
standing current makes no difference to distortion and
leads to a bias voltage a long way from grid current
problems.

Are you able to vary the HT voltage, to explore other
parts of the safe operating area? Some idea of a trend
would be good there too.

Not at the moment. 320V is as high as I can get with HT
and that's just high enough for the circuit used. I
suspect if I could raise the HT I could get even lower
figures. However I was regularly achieving 0.4% THD at
20Vrms which is at -48dB relative to the fundamental.
For most tubes, Morgan Jones only achieved -50dB second
harmonic with a pentode CCS plate load at 19.5V rms so I
do not expect there is much improvement to be gained by
raising the HT. Personally, I was surprised I got
distortion figures so close to Jones' with such a simple
circuit which speaks volumes for the inherent linearity
of the 6SN7 family.

At present both tube halves sit comfortably within the
SOAR. The bottom tube has a plate/cathode voltage around
140V which at 8mA is less than 1.2W dissipation and the
top one has about 100V across it which is 800mW both of
which are well within the SOAR of the 6CG7. I would be
more concerned about exceeding heater/cathode voltages
than SOAR. At present the heaters are raised to +75V
(6CG7 has max +Ve dc of 100) adn the top cathode is at
about 220V (6CG7 has max -ve dc of 200V which would
allow cathode to go to 275V) so there's no much margin
on either one.

How does simulated total distortion compare with your
real measurements?

As I mentioned above, one model reproduces the
distortion levels very accurately except for grid
current. All the others I have tried are way out.

This comes as a pretty big surprise. Do you mean that you
have measured simulated distortion and compared it to
your real measurements?

Yes, the spice model based on the The Audio Designers Tube
Register curves gives THD results that are extremely close
to the measurements.

This makes a sweep of distortion level v
standing current even more interesting.


Yes, that might be easier to achieve in a simulation than
on the bench.

Could you post a copy of your model, please? AFAIK, the
generic DM model underlying the 6SN7 I posted is the most
accurate and sophisticated available, with one possible
exception that you probably couldn't use on your
simulator.

Considering the DM triode comes with a grid model that
had been omitted from previous, inferior triode models,
and that DM is the originator of nearly all available
models and I would guess *all* in common use, I surmise
that *all* of the improvement in the model you have used
is due to the more appropriate data used to generate it.


That would be my conclusion too.

It is therefore not unlikely that fitting the most recent
DM model to that data would result in a model more
accurate than the one you have, particularly in the BRH
quarter of the SOAR, IIRC.


Possibly. The model I have apparently used quite a
sophisticated curve fitting model so I am not sure why the
DM model should be inherently better.

Which is why I would appreciate a copy of the model text.
If it is similar to the DM generic model, it may be
possible to transfer common parameters, and leave any
extra ones as they are. If not, then the data the model
was fitted to would be useful, although a complete refit
is a tedious business.

No problem and I just realised I told a big fib. The model
is apparently taken from the GE data sheet not the The
Audio Designers Tube Register curves as I stated earlier -
I wonder where I got that from?? - must check that out.
Anyway, here is the 6SN7 portion of the file:


*6SN7 LTSpice model from GE 6SN7 datasheet

.subckt 6sn7 P G K
Bp P K
I=(0.02003791851m)*uramp(V(P,K)*ln(1.0+(-0.07740549711)+exp((4.618036737)+(4.618036737)*((2 0.85288965)+(-110.4389272m)*V(G,K))*V(G,K)/sqrt((28.13407639)**2+(V(P,K)-(7.118597372))**2)))/(4.618036737))**(1.380047579)
Cgp G P 4.0pF
Cgk G K 2.6pF
Cpk P K 0.7pF
.ends 6sn7

Note the preamble to the set of models is as follows:

*Generated by Joel Tunnah using Curve Captor v0.9.1
*UPDATED 02/07/06
*Contents:
*6SN7
*12AX7
*12AU7
*12AT7
*ECC88
*6C4Pi



Cheers

Ian

Ian