Vous êtes sur la page 1sur 39

Adrian

Immler, Oct. 2018

BEST OF HISTORY’S VACUUM


TUBE DESCRIBING ART
MERGED TO NEW SPICE MODELS


^

“Let all things be tested; keep to what is good”
I Thessalonians 5:21

TABLE OF CONTENTS
1 summary 3
2 The background story 3
3 Diodes 4
3.1 The previous efforts 4
3.2 New approach based on Richardson’s law plus grid resistor 4
3.3 The married approach 6
4 Triodes 9
4.1 Reflection of found triode spice models 9
4.1.1 Charles Rydel’s approach 9
4.1.2 Duncan Munro’s approach 11
4.1.3 Norman Koren’s approach 15
4.1.4 Walter Sjursen’s approach 16
4.1.5 Ayumi Nakabayashi’s approach 17
4.1.6 Andre Adrian’s approach 20
4.1.7 Ivan Cohen’s and Thomas Helie’s approach 22
4.2 improved triode approach 25
4.2.1 Modeling the “bill of currents” 25
4.2.1 “µ reduction” @ positive grid voltage 26
4.2.2 modeling the island effect 26
4.2.3 grid current modelling 26
4.2.4 Finetuning 29
4.2.5 Reduction of parameter interactions 30
4.2.6 Equation summary 33
4.2.7 Achieved fit quality 34
4.2.8 The new generic LTspice triode model 37
1 Tetrodes & Pentodes 38
1.1 Do it with passion or leave it 38
1.2 My pentode approach (unevaluated) 38
2 About this paper’s author 39
3 Thanks! 39

2
1 SUMMARY
This paper introduces a set of new tube spice model approaches by merging best ideas
of all approaches found in the web. In addition, some new ideas are introduced which
leads to a better fit quality.
At negative grid voltages, the triode Ia accuracy is similar to Koren or Ayumi models.
At positive grid voltages, the triode Ig and Ia accuracy is on an unrivaled level. This
approach offers new simulation possibilities:
• Bias point setting with grid resistor as often used in battery powered circuits
containing direct heated tubes, or car radio tubes.
• Grid current prediction in low voltage circuits (where grid current is always an
issue).
• Precise simulation of overdriven stages as typically found in guitar amps.
• Pulsed stages (class c and the like)
My new generic pentode model is an enhancement of the triode model. Hence, this
pentode spice model is also suitable in triode connection mode. Furthermore, it is no
longer necessary to use different approaches for power beam / signal pentodes.

2 THE BACKGROUND STORY


An old childhood dream of mine is a self-made tube guitar amplifier. In Dec 2015, I
decided no longer just to dream of it. Of course, it must be something different to the
hundred’s amps found in the web, so my goal was to build a 1W “no danger” low
voltage guitar amp, in a unique dodecahedron housing. Right from the beginning, it
was clear that I would use those fascinating subminiature tubes.
In a low voltage tube stage, the grid current can easy rise to a no longer ignorable
amount. Good spice models for tubes considering this fact were hard to find, so I had
to develop them by my own.
I first failed to find a good generic approach for triodes and pentodes, so I had to
design my amp with a set of non-generic spice models. In addition, they showed
convergence-problems, but this just evoked my ambition to find a satisfying generic
approach...

3
3 DIODES
3.1 The previous efforts

In history, the efforts for vacuum diode spice models had its focus on power
rectifying or large signal behaviour. The found models in the web are either built as a
standard diode model enhanced with a resistor (the most common way for triode grid
current simulation), or are based on Langmuir-Child’s law Id=1/kd*Vaxd as seen for
some vacuum diode models.
Both ways are ok for large signals, but fail at smaller signals, because their anode a.c.
resistance near zero voltage is often far away from reality.

3.2 New approach based on Richardson’s law plus grid resistor


A look at (german) Wiki tells us that for a negative grid voltage, the grid current
follows an exponential equation:
'(
I" = I"$ ∙ e)* (1)

Ig0 means the current flowing at zero voltage, and VT is the voltage that the thermal
emitted electrons are able to overcome due to their (small) kinetic energy. VT is
proportional to the cathodes absolute temperature:

k∙T
V, =
q (2)

While k means the Boltzmann constant, q is standing for the elementary charge. The
thermal voltage VT is about 0.1V as the cathode’s temperature is typically 1200K for
most tubes.
In case a tube spice simulation of an over- or underheated filament is wished, this
seems to be an essential key relation for the development of heating power
considering spice models. However, this paper will not recess this issue as the main
focus is laid on grid current improvements.
In my opinion, the greatest disadvantage of the approach with standard diode model in
series with a resistor is the lack of any grid current at small negative grid voltages. So,
how about replacing the standard diode spice model with the exponential expression
described above?

I"
𝑉" = 𝑉, ln 3 4 + R "8 ∙ I"
I"$ (3)

Due to my limited mathematical skills, I know not how to dissolve an equation


y=ln(x) + x for x, so I have to describe the equation for the grid voltage as a function
of the grid current. Fortunately, this doesn’t matter for spice modeling.
How good will the equation above match to reality?
A screening through my little tube collection didn’t deliver any vacuum diode, so I
decided to measure the grid-to-cathode vacuum diode of a subminiature twin triode
6948. Due to the fact that this triode is dedicated developed for missile applications, it
represents the top of tube designing art, so unwanted effects like secondary emission,
thermal runaway of the tubes parameters, microphony etc. are eliminated at its best. A
perfect tube for research of grid current behavior!
The grid current was measured with a multimeter MetraHit 29s, which has a
resolution of 1nA. The ranges have to be changed manually at currents of about 10%
of the ranges maximum, to avoid too much voltage drop across the multimeter.
To fit the function described above as good as possible to the measured data, the total
sum of all squared relative errors has been minimized. Excel offers with its solver a
suitable tool for this job. It is important to work with relative errors because otherwise
the fit will be poor at smaller grid currents.

Vg [V] grid-diode of a submini Triode 6948 @ Va=0V


1.75

1.5 measured values

exp.function + gridresistor
1.25

0.75

0.5

0.25

0 Ig [A]
0E+00 1E-03 2E-03 3E-03 4E-03 5E-03 6E-03
-0.25

-0.5

-0.75

-1

5
A logarithmic scale for the grid current better shows what’s happening at negative
voltage.

Vg [V] grid-diode of a submini Triode 6948 @ Va=0V


1.75

1.5

1.25 measured values

0.75

0.5

0.25

0 Ig [A]
1E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02
-0.25

-0.5

-0.75

-1

The achievable fit is not so satisfying. While for negative voltage an exponential
expression seems to be the right approach, a simple grid resistor does not a good job
for the positive voltages.

3.3 The married approach

The Langmuir-Childs power function (with the exponent as a variable) seems to be


the better choice at positive voltages, but how to marry these two worlds?
At zero voltage, these two approaches do not touch each other!

6
Id [A]
The two expressions for vacuum diode current,
at negative and positive voltages (exemplified)
1.E-02

1.E-03

1.E-04

1.E-05

1.E-06

1.E-07

1.E-08 Vd [V]
-1 0 1 2 3 4 5

After some days of thinking I got the idea of the term ln(1+ex) which is often used for
soft cutting the variable x. This term shows an exponential behavior at negative values
while it can be simplified to x for positive values. Fortunately, setting an exponent
over this term does not destroy its exponential nature for negative values!
By further developing this promising idea, I got this result expression:
?9
1 '>
I9 = ;V
∙ , ∙ xd ∙ ln 31 + e ∙?9 4@
)*
(4)
k9

As we can see, the variables VT and ex are involved in this equation two respective
three times, so this equation seems to be not as simple as possible. The higher
complexity of this equation has to be accepted to achieve independency for the
variables xg and VT of each other.
The fitting quality can be improved remarkable by adding the contact potential Vct
(introduced by Van der Bijl), with an impressive good result:
?9
1 'A B)CD
I9 = ∙ ;V, ∙ xd ∙ ln 31 + e )* ∙?9 4@ (5)
k9

7
Be aware that the voltage Vd across the diode means in our example the voltage Vg of
the grid hence I took the grid of a triode as a vacuum diode sample.

grid-diode of a submini triode 6948 @ Va=0V


Ig [A]
6E-03

5E-03

measured values
4E-03 married function

3E-03

2E-03

1E-03

0E+00 Vg [V]
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1 1.25 1.5 1.75

grid-diode of a submini triode 6948 @ Va=0V


Vg [V]
1E-02
-1 -0.75 -0.5 -0.25 0 0.25 0.5 0.75 1 1.25 1.5 1.75

1E-03

1E-04

measured values
1E-05
married function

1E-06

1E-07

1E-08
Ig [A]

8
4 TRIODES
4.1 Reflection of found triode spice models

According to my web research, there is much more efforts done to develop spice
models for triodes than for vacuum diodes. Up to now, I found a half-dozen
approaches which have either found a certain propagation or contains interesting ideas
how to describe triode specific effects. Best of all (plus some further improvements)
will be merged to a new approach in chapter 4.2.

4.1.1 Charles Rydel’s approach


Charles Rydel introduces his approach in 1995 with the following set of equations:
O
𝑉H 𝑉L + 𝜈M P
IE = G G1 + K G𝑉H + K
𝜈J 𝜇 (6)

𝛼𝑉H O⁄P 𝑤ℎ𝑒𝑛 𝑉H ≥ 0


IH = Q 𝛽 (7)
𝑤ℎ𝑒𝑛 𝑉H < 0
𝜈[ − 𝑉H
This leads to the following fit to the measured data of my 6948 triode (while the Ia
values @ Vg=2V has to be ignored to achieve a usable fit):

Ia [A]
Rydel-model-fit to measured values of a Sylvania Triode 6948
8E-3

Vg=2V
7E-3

6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

3E-3
Ig@Vg=2V Vg=-4V
Pmax=0.5W
2E-3

Ig@Vg=1V
1E-3

0E+0
Va [V]
0 50 100 150 200 250 300 350 400
As far as I know, Rydel’s equation for the Anode current was the first approach
considering the gm-reduction in the large Va/large negative Vg area, while the µ-
reduction in this area was still not addressed.
Let us give a short explanation to this effect:
The gm- and µ-reduction in the large Va/large negative Vg area is called “island
effect” since the emission of the cathode-parts direct under the grid wires are
already stopped while there are still emitting “islands” of the
cathode surface located between the grid wires. It makes
sense that the influence of the grid wires to the local electric
field right above these islands is reduced since there is more
distance between the grid wires to these islands, and
therefore, the average µ is reduced at large negative grid
voltages. The figure to the right illustrates this situation.

But now back to Rydels approach:
The variable vc (when divided by µ) has the role of the
contact potential Vcon. Unfortunately, when vc is inserted in
the Ia equation as proposed by Rydel, this parameter
interacts with µ (and vice versa), but aside from this, Rydel’s
parameters acts largely independent, so the fitting to given
curves is easy.
In general, Rydel simplifies his equations in that manner that
he keeps the exponent at 3/2 also for not axially symmetric
designed triodes. Most triodes have a more or less
rectangular shape of the plate and therefore an exponent which is different to
3/2 for best fit.
The equation for the grid current is ok for the high Va values, while the cathode
current redistribution from the anode to the grid at low Va values is not
considered at all.
Rydel’s description of the grid current at negative values grid voltages does not
make any sense to me as he uses a hyperbola, while actually the grid current
shows an exponential behavior. Furthermore, the two proposed Ig functions do
not meet each other at Vg=0V. The resulting step can cause convergence
problems. However, Rydels proposal for the grid current leads to the following
fit:
Ig [A] Rydel-model-fit to measured values of a Sylvania Triode 6948
1E-2

1E-3

1E-4

Ig@Va=100V
1E-5
Ig@Va=50V
1E-6 Ig@Va=20V
Ig@Va=10V
1E-7 Ig@Va=5V
Ig_Rydel
1E-8

1E-9
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1Vg [V]

To sum up, the strength of Rydels approach is in its simplicity and fast calculation,
and shows usable results as long as a positive grid voltage is never reached.

10
4.1.2 Duncan Munro’s approach
Duncan Munro introduced his approach in 1997. He used 12 parameters for the static
current prediction. Today, the documentation of his generic triode spice model is no
longer available on his webpage, but can be found elsewhere in the internet.
Unfortunately, his documentation gives no detailed information about the math
behind his triode model and just explained the roles of its parameters. However, by
analysing the spice code found in his documentation, I got the following set of
equations:
A) “Steering voltage” Vem (Barkhausen’s “Steuerspannung”, which has the same role
as the anode voltage in vacuum diodes):
V^_ = max (0, 𝑉L + 𝜇$ 𝑉de ) (8)
Duncan introduced with his approach the worldwide new possibility of a µ-reduction
to better address the island effect. He therefore used a modified grid voltage VGR,
see equation (12).

B) Cathode’s space charge current (modified Langmuir-Child’s law):


I^g = EMC ∙ RP ∙ 𝑉^_ lemBlen∙o8p ($,)q ) (9)

Because the knees of the Ia traces are smoother in island effect area, this can be
described with an exponent raise, controlled by ERI.
C) The (inverted) contact potential of the grid CDO is also considered. To simplify
some following equations, Duncan introduced the voltage VGG. One will find this
voltage VGG also in the later developed approach of Ayumi.
𝑉dd = 𝑉d − 𝐶𝐷𝑂 (10)
D) Raise of the anodes a.c. resistance in the island effect area:
RP = 1 − max(0, min(0.999, (−Vxx RAF){|} )) + 𝑅𝐴𝑃 ∙ max(0, 𝑉HH ) (11)
Duncan also introduced a way to address an anode a.c. resistance increase at positive
grid voltages (if necessary), with the parameter RAP.
E) “µ-reduction” (or VG manipulation) in the island effect area:
𝑉de = max (0, 𝑉dd ) + min (0, 𝑉dd (1 + 𝑀𝑈𝑅 ∙ 𝑉dd )) (12)
F) Ia limitation at small Va values:
𝐼L„…_ = LIF ∙ max (0, 𝑉L )„nm (13)
G) Ia composition from space charge current and Ia limitation:
𝐼L = min (𝐼^g , 𝐼L‡…_ ) (14)
H) Ig composition from the grid-cathode vacuum diode current plus the “gap”
between Ia and the space charge current:
O
𝐼H = GCF ∙ max(0, 𝑉dd )P + 𝐼^g − 𝐼L (15)

According to my web-researches, Duncan was the first who considers the very
important Ig raise when Va is moved toward zero. According to his equation (15) he
has a physical phenomenon in mind that limits Ia similar to the Langmuir-Child’s law,
while the steering voltage is replaced by Va. He called this an “emission limitation”,
although obviously the cathodes electron emission shows no limitation towards zero
when Va approaches zero. However, as long as I cannot find a better name for this
phenomenon, I will keep Duncan’s choice.

11
The emission limitation can be seen in the Ia traces of some triodes capable to handle
high Vg voltages (and also in many Ia traces of pentodes). A good example is the
following triode 6C4 (in red the emission limit):

12
The equations above deliver the following fit to the measured data:
Duncan-model-fit to measured values of a Sylvania Triode 6948
Ia [A]
8E-3

Vg=2V
7E-3

6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

Ig@Vg=2V
3E-3
Vg=-4V
Pmax=0.5W
2E-3

1E-3 Ig@Vg=1V

0E+0 Va [V]
0 50 100 150 200 250 300 350 400
A zoom better shows what happens at lower Va values:
Duncan-model-fit to measured values of a Sylvania Triode 6948
Ia [A]
8E-3

Vg=2V
7E-3

6E-3

5E-3
Vg=1V

4E-3

Pmax=0.5W
3E-3

2E-3
Vg=0V

1E-3

0E+0 Va [V]
0 10 20 30 40 50 60

13
The fit above respects the inventor’s intention for the Ia limit exponent, that this
parameter should be in the range of 1 to 1.5 according to his documentation.
Because there is a further physical phenomenon (first described 1922 by Tank) that
leads to an Ig raise when Va decreases toward zero, the Ig fit quality is still limited.
By ignoring Duncan’s intended range (1 … 1.5) for the exponent of the grid-cathode-
vacuumdiode, this further physical phenomenon can be somehow “considered”, so the
fit quality can be a bit improved (here optimized for best Ia fit @ Eg=2V):
Duncan-model-fit to measured values of a Sylvania Triode 6948,
Ia [A]
LIP=0.72 (out of inventors intended range)
8E-3

Vg=2V
7E-3

6E-3

5E-3
Vg=1V

4E-3

Pmax=0.5W
3E-3

2E-3
Vg=0V

1E-3

0E+0 Va [V]
0 10 20 30 40 50 60
Unfortunately, the 4 parameters RAF, RAS, MUR and ERI for the island effect
modelling influences each other in such a strong amount, that the Solver AddIn from
Excel has troubles to find a least mean square solution. So, the fit has to be done
mainly by hand.
A further improvable point of Duncan’s approach is the sharp transition from the min-
function involved in the Ia composition. This will be addressed in my own proposal,
see chapter 4.2.3.3

14
For the current calculation of the grid-cathode vacuum diode, Duncan involved the contact
potential. Therefore, Ig does not decrease to zero at Va=0, what is a reality. But the grid’s
“Anlauf” current is not described, so the Ig fit at Vg near zero is still way of reality.
Duncan-model-fit to measured values of a Sylvania Triode 6948
Ig [A]
1E-2

1E-3

1E-4

Ig@Va=100V
1E-5
Ig@Va=50V

Ig@Va=20V
1E-6
Ig@Va=10V

Ig@Va=5V
1E-7
Ig_Duncan

1E-8

1E-9 Vg [V]
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

4.1.3 Norman Koren’s approach


The famous Koren triode approach (introduced 1996) shines with excellent anode current
fit while using only 6 parameters (when the contact potential Vct is added). Particular the
fact that island effect is very good describable with each just one (!) parameter is very
impressive. (Duncan needs 4 parameters for the island effect, while achieving a lower fit
quality than Koren)
Koren uses the letter E instead of V for voltages, while it is common to use E for electric
fields. Therefore, I changed this to not confuse others.
Koren proposed the following two equations to describe the anode current:
Vˆ ? (16)
IL = ∙ (1 + 𝑠𝑔𝑛(𝑉ˆ ))
k "ˆ

ˆ )( ’)CD
•• ⎛ ’ ⎞

𝑉E ⎛ “•”• ’ '– ⎞
— (17)
𝑉ˆ = ln 1 + e ⎝ ⎠
k‹ ⎜ ⎟

⎝ ⎠
The first is the classical Langmuir-Child’s law. Koren uses for the steering voltage his own
variable index “g1”.
Koren’s great work was the second equation (original without the important contact
Potential Vct). In its core, it’s still Barkhausen’s equation for the steering voltage Vst = Vg +
D*Va (while D=1/µ), but enhanced with two new variables kp for the triode’s island effect
and kvb for the low Va area.
Korens approach delivers very accurate results for the anode current when the contact
potential of the grid is considered, as the following fit to measured data of my 6948 triode
shows:
Ia [A] Koren-model-fit to measured values of a Sylvania Triode 6948
8E-3

7E-3

Vg=2V
6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

3E-3
Ig@Vg=2V Vg=-4V
Pmax=0.5W
2E-3

Ig@Vg=1V
1E-3

0E+0 Va [V]
0 50 100 150 200 250 300 350 400
The only four disadvantages I can see in Koren’s approach are:
1) Poor diode current model (semiconductor diode with resistor in series). The “Anlauf”
current is missed.
2) Langmuir-Child’s space charge law is not fulfilled for the cathode current at small
values of the anode voltage. In other words: The Ig raise is missed when Va
approaches toward zero.
3) Ia tends to be too high at positive grid voltages
4) While x is intended to form the curvature of the traces, this variable has also big but
unwanted influence to the slope of the curves (as kg1 is intended to set the slope of the
curves).

4.1.4 Walter Sjursen’s approach


Walter Sjursen introduced his triode spice model in Dec 1997. The announcement was
quite promising:

“…provides a plate current that goes to zero at zero plate voltage, has a non-linear grid
current for positive grid voltages, takes the effects of initial electron velocity into account,
and ensures that the sum of the grid and plate currents is equal to the cathode current (total
space current).”

I was then surprised that his paper was not for free, which seems to be rather unusual in the
scene of vacuum tube modeling.
There was also no Sjursen-based triode-model to find in the web, just a fragment of his
grid model mentioned in a blog.

16
I decided to abandon my research on his approach, because of two reasons:
1) It seems to be the intension of the inventor that his approach is a proprietary code.
With a publication in my paper, I would not respect this.
2) The lack of propagation of his approach for triode spice simulations tells me not of a
good overall performance. According to one of the seldom Sjursen–model-users found
in the web, its strength in grid current prediction has to be paid with drawbacks in
terms of Ia traces accuracy.

4.1.5 Ayumi Nakabayashi’s approach


Ayumi introduced his generic triode model in 2001 according to his paper. As far as it is
understandable for me, Ayumi’s triode model seems to be the impressive result of a very
deep understanding of the vacuum tubes internal physics. His Japanese paper about his
generic triode model contains over 100 pages, and describes also terms for the small signal
a.c. characteristics like the transconductance gm or the anode’s a.c. resistance.
His paper is hard to find for not Japanese speakers. I had to “blind” click me through the
several links on his web page, and found it here: http://ayumi.cava.jp/audio/appendix.pdf
The hard to find paper seems to be the reason that the guild of freaks capable of building
an Ayumi triode model is very limited.
Ayumi needs an unbeatable low number of only 4 parameters (for some triodes 6) to
describe the anode current versus anode and grid voltage, including the grid current. A
fundamental principle to achieve this is his assumption that also not axially symmetric
designed tubes (which is true for most tubes except nuvistors) are sufficient describable
with an exponent of 3/2.
As a consequence, the fit quality for triodes with an exponent near 3/2 is very good, while
triodes with an exponent not so near to 3/2 are not so precise describable for grid voltages
near zero. For such cases, Koren’s approach fits better for Ia.
Ayumi considers both physical phenomenons (Tank’s and also the “emission limit”) that
leads to an Ig raise when Va approaches toward zero. Like Duncan, Ayumi uses a simple
min-function for transitions between equations, so the transmissions are not smooth like in
reality, but “sharp-edged”.
However, the following impressive set of equations describes Ayumi’s approach:
VH = 𝑉HH + 𝑉H$ (18)
1
a= (19)
1−𝛼
3 1
b= − (20)
2 1−𝛼

c = 3α − 1 (21)
2
µ_ = 𝜇 (22)
3 − 3α ¡

E © E
⎧𝐺 G 3 K G𝑐 ∙ 𝑉g K G𝑉 + 𝑉g K 𝑖𝑓 𝑉 ≤ 0
⎪ 2a 2 𝜇¡ HH
𝜇¡ HH
I¢ = (24)
⎨ a𝑐 © 𝑉g ˆ.¯
⎪ 𝐺 - ® G𝑉HH + K 𝑖𝑓 𝑉H > 0
⎩ 3 𝜇_

17
1
µ=
1 𝑐 𝑉 (25)
𝜇 − 2 ∙ 𝑉dd
_ g


a𝑐 © 1 ˆ.¯
G‡…_ = 𝐺 - ® G1 + K (26)
3 𝜇
0.5
xH =
1 ˆ.¯ (27)
-1 + 𝜇 ®


𝑉H
𝑥H 𝐺‡…_ 𝑉H ˆ.¯ 31.2 + 0.44 𝑖𝑓 𝑉H ≥ 0
IH = Q 𝑉g + 𝑉H (28)
0, 𝑖𝑓 𝑉H < 0

Ig ‡…_ = ´1 − 𝑥H µ𝐺‡…_ 𝑉g ˆ.¯ (29)

Ig = min (𝐼¢ − 𝐼H , 𝐼g ‡…_ ) (30)




There are only 4 parameters to set:
• G is the overall gain
• µc has a similar role like µ in other’s approaches
• Vg0 is the contact voltage of the grid
• α is a parameter to describe the island effect
For some tubes, Ayumi sets two further parameters Glim and xg not by calculation from
the 4 parameters above, but by best fit to the measured data.
This is the way I choose to for fitting his approach to my 6948 data, as I do not understand
how to calculate Glim and xg from the other parameters. The equations found in his
Japanese paper says that both parameters are dependable from a further (intermediate)
parameter called µ according to equation (26) and (27), while µ seems to be not a fixed
parameter but voltage dependable according to equation (25). This seems to me a conflict.
However, Ayumis approach is a class of its own and by far not understandable for my level
of math and physics.
Although the parameters have a strong influence to each other, they are good optimizable
by the least square method and leads to the following fit (while the measured values below
Va=10V are ignored to optimize the fit quality above this anode voltage):

18
Ayumi-model-fit to measured values of a Sylvania Triode 6948
Ia [A]
8E-3

Vg=2V
7E-3

6E-3
Vg=0V

5E-3

Vg=-2V
4E-3 Pmax=0.5W

3E-3
Vg=-4V

2E-3

1E-3

0E+0 Va [V]
0 50 100 150 200 250 300 350 400

Ayumi-model-fit to measured values of a Sylvania Triode 6948


Ia [A]
8E-3

Vg=2V
7E-3

6E-3

Ig@Vg=2V
5E-3

Vg=1V
4E-3

3E-3

2E-3
Vg=0V
Ig@Vg=1V
1E-3

0E+0 Va [V]
0 10 20 30 40 50
For an approach containing only 6 parameters, the fiting quatity is impressive good for
both the Ia and Ig values.
19
4.1.6 Andre Adrian’s approach
Andre Adrian shows on his webpage andreadrian.de an interesting approach for a generic
triode model. It is the first (but undated) triode model I found in the web that uses the
Richardsons law to describe the “Anlauf” current of the grid.
Furthermore, Andre takes the Tank equations into account to split the cathode current into
Anode current and grid current.
He uses the following equations for his approach (while Va has to be positive):
VLˆ = 𝑉L − 𝐼L ∙ 𝑅L
(31)

ˆ.¯
d’d¶ )q
𝑉Lˆ 𝑉Lˆ ¹
IL = 𝑒 ∙ 𝑚𝑎𝑥 G0, + 𝐷H 𝑉H + 𝑉^ K ∙ (32)
𝜇 𝑉Lˆ ¹ + 𝐵 ∙ 𝑉H ¹

d’d¶ )q
𝑉Lˆ ˆ.¯
𝐵 ∙ 𝑉H ¹
𝑒 ∙ G + 𝐷H 𝑉H + 𝑉^ K ∙ 𝑖𝑓 𝑉𝑔 ≥ 𝑉𝑤
IH = » 𝜇 𝑉Lˆ ¹ + 𝐵 ∙ 𝑉H ¹ (33)
)q ⁄)¼
𝐼$ 𝑒 𝑖𝑓 𝑉𝑔 < 𝑉𝑤

In general Andre’s model is based on some great ideas while unfortunately the
development of his approach seems to be stopped on half way. For example, he uses a not
necessary parameter Dg (Barkhausens “Durchgriff D”, the inverse value of µ). which
seems to me a copy/paste failure from his pentode models.
Like Rydel, Andre mimics the island effect only by an Vg dependable gm modulation and
ignores the µ reduction.
A new and interesting aspect is his way to fine-tune the curves by the means of an internal
anode resistance.
However, Andre Adrian’s triode model delivers the following fit to my 6948 data:
Andre-model-fit to measured values of a Sylvania Triode 6948
Ia [A]
8E-3

Vg=2V
7E-3

6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

Ig@Vg=2V
3E-3
Vg=-4V
Pmax=0.5W
2E-3

Ig@Vg=1V
1E-3

0E+0 Va [V]
0 50 100 150 200 250 300 350 400

20
A very promising idea of Andre is to use one of the Tank’s law to describe the Ig increase
when Va approaches to zero.
Andre-model-fit to measured values of a Sylvania Triode 6948
Ia [A]
8E-3

Vg=2V
7E-3

6E-3

5E-3
Vg=1V

4E-3

Pmax=0.5W
3E-3

2E-3
Vg=0V

1E-3

0E+0 Va [V]
0 10 20 30 40 50
The transition of the grid current calculation from one equation to the other is abrupt, and
the two traces of each equation can touch each other only for one certain anode voltage. As
a consequence, a step occurs for most conditions, and can cause convergence problems.

21
Andre-model-fit to measured values of a Sylvania Triode 6948
Ig [A]
1E-2

1E-3

1E-4

Ig@Va=100V
1E-5 Ig@Va=50V
Ig@Va=20V
1E-6 Ig@Va=10V
Ig@Va=5V
Andre fit @ Va=10V
1E-7
Andre fit @ Va=50V

1E-8

1E-9 Vg [V]
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

4.1.7 Ivan Cohen’s and Thomas Helie’s approach


Cohen and Helie’s 12 parameter approach was introduced in 2012. Their work is an
improvement of Koren’s model and contains the following equations (while I replaced the
letters E through V for voltages):

2 ^¾
𝑉 𝑖𝑓 𝑉P ≥ 0
Ig = Q 𝐾H P (34)
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑒𝑖𝑠𝑒

ˆ )qÇ ’)ÈÉ
ÄÅ ⎛ ’ ⎞
Æ
𝑉g¢ ⎛ “ÄÊË ’)ÅÇ ’ÄÊË— )ÅÇ ⎞

VP = 𝑙𝑛 1 + exp ⎝ ⎠ (35)
𝐾g ⎜ ⎟

⎝ ⎠
1 1
IH = 𝐆 3 + 4 ln (1 + exp (a(𝑉H¢ + 𝑉Î )))Ï (36)
𝑏𝑉g¢ +1 𝑐

The factor 2 in the Ip equation seems to be a copy paste failure since the autors have
replaced Korens term sgn(Va)+1 through a case distinction.

The grid current equation is quite exciting as the grid current starts to flow with an
exponential behavior like the “Anlauf” current, and also ensures a smooth transition to the
classic Langmuir-child law. Compared to my diode equation (5) in chapter 3.3, we can see
that they both softclip Vg with the term ln(1+exp(Vg)).
Unfortunately, the authors seemed not to have the intention to consider the “Anlauf”
current in their model, because they presented the achieved Ig-fit only in linear scaled
diagrams where the small “Anlauf” current is not visible at all. A log scale had been useful.
22
Furthermore, the authors forgot an essential parameter G in their Ig equation of their paper,
to overall scale the Ig axis. I added this further parameter in red to the original equation
because the original equation would not lead to any usable fit.

The first term in brackets is very similar to a term BVg/(Va + BVg) in one of Tanks
current splitting law while BVg is replaced through 1. Ayumi uses also a similar term in
his equation (28) for the grid current.

While Ayumi needs 0 parameters (sometimes 2) to describe the grid current with good fit
quality (except the missed “Anlauf” current), Cohen and Helie uses 5 parameters, and all
makes sense to me. Hence, I expected an excellent Ig fit. Let’s see the result!

Cohen&Helie-model-fit to measured values of a 6948 triode


Ia [A]
8E-3

7E-3

Vg=2V
6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

3E-3
Ig@Vg=2V Vg=-4V
Pmax=0.5W
2E-3

Ig@Vg=1V
1E-3

0E+0 Va [V]
0 50 100 150 200 250 300 350 400

The achieved fit-quality with the 13 parameter Cohen&Helie model is very good. For
negative grid values, the fit quality is not as excellent as Koren’s. The linear term kvb2*Vp
under the root seems to give a small drawback in this region, while it remarkably improves
the Ia fit quality for positive grid values.
For triodes which shows the “emission limitation effect”, the achieved fit quality needs still
improvements at very small Va values and positive grid values.

23
Cohen&Helie-model-fit to measured values of a 6948 triode
Ia [A]
8E-3

7E-3
Vg=2V

6E-3

Ig@Vg=2V
5E-3

Vg=1V
4E-3

3E-3

2E-3
Vg=0V
Ig@Vg=1V

1E-3

0E+0 Va [V]
0 10 20 30 40 50

Cohen&Helie-model-fit to measured values of a 6948 triode


Ig [A]
1E-2

1E-3

1E-4

Ig@Va=100V
1E-5 Ig@Va=50V
Ig@Va=20V

1E-6 Ig@Va=10V
Ig@Va=5V
Ig_fit@Va=50V
1E-7
Ig_fit@Va=10V

1E-8

1E-9 Vg [V]
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1

24
4.2 improved triode approach
The different triode models found in the web have all its strengths and drawbacks. A
summary of this is shown in the table below:

Triode model Island µ-reduction Ig “Anlauf” Ic allocation to Ia emission limit


effect @ pos Vg current @ Ig/Ia @ pos Vg @ pos. Vg and
Vg near small Va
zero
Rydel - -- - - --
Duncan 0 + -- - ++
Koren ++ -- -- 0 (one approach for both effects)
Ayumi + ++ -- + +
Andre - 0 + + --
Cohen&Helie + + ++ ++ --
The aim is now to merge best of all approaches (plus some further ideas) to a new
approach. The green solutions will be chosen.
For the µ-reduction, an own idea will be preferred, as we will see in chapter 4.2.1.

4.2.1 Modeling the “bill of currents”


As well known, we can describe the triodes static behavior for the Va mid range with the
Langmuir-Childs’s law while we replace the vacuum diodes’ anode voltage through the
Barkhausens steering voltage.
Andre Adrian mentioned in his great paper “History of vacuum tube theory” that there is
some confusion about this equation. Which current does this equation mean?
While Barkhausen applied this law to the cathode’s emission current, van de Bijl used the
same equation for the anode current. Who of both was right? In my opinion, neither of
them – it just describes the space charge current Is, which is then split into the anode
current Ia and two reflected grid current parts Igr and Igl, both according to functions of
the electrode voltages. Furthermore, the cathodes emission current Ie is the sum of the
space charge current Is and the grid-to-cathode vacuum diode current Igvd (see equation
5).

𝑉L
VÐÑ = 𝑉H + 𝑉¡Ñ + (37)
𝜇

¾Ð
1
IÐ = Q𝑘Ð ÐÑ 𝑖𝑓 𝑉ÐÑ ≥ 0
𝑉 (38)
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
IL = 𝐼Ð − 𝐼HÓ − 𝐼H‡ (39)

I^ = 𝐼Ð + 𝐼HÔÕ (40)

IH = 𝐼HÔÕ + 𝐼HÓ + 𝐼H‡ (41)

Rydel and Duncan’s approaches are both based on this principle decribed above, but they
have different ways to assign the space charge current to the anode and grid. Rydel always
assign 100% of the space charge current to the anode. This is only true for negative grid
25
voltages. Duncan assigns the space charge current by considering the “emission limit” (but
neglects the Tank’s laws). He reflects the difference between emission limit and Ia to the
grid, as long as the emission limit is smaller than Ia.
Andre Adrian’s approach has a similar current assignment. He splits the emission current
instead of the space charge current, and does not describe a further grid-to-cathode vacuum
diode current Igvd. There is perhaps a misunderstanding of the Tank’s law Andre used:
May it be, that Tank meant the space charge current instead of the emission current,
because Tank proposed an index s (not an index e)?

4.2.1 “µ reduction” @ positive grid voltage


When we have a closer look to the anode current, we discover a correlation between the
anode current reduction at positive grid voltages and the grid current itself. This is
somehow considered by a µ-reduction @ pos Vg like Duncan propose it. However, as an
extra drop at very small Va values is not describable by his approach, we will better choose
a way that is Ig dependent. Hence, we propose an internal grid resistor which reduces the
effective grid voltage for the Langmuir-child’s law, in a linear dependency to the grid
current. So, we define a new equation for an effective grid voltage, used in the steering
voltage expression:

VHH… = 𝑉H + 𝑉¡Ñ − 𝑅H… ∙ 𝐼H (42)

4.2.2 modeling the island effect


The best method to mimic the island effect was created by Koren, so we will adapt his
method to our new triode model. For small Va values, we wish to keep Barkhausens
steering voltage, because the space charge current does not decrease to zero at Va=0 and
positive Vg values.
We can achieve this by choosing the greater value of Barkhausen’s steering voltage and
Korens modified steering voltage:

𝑉L 𝑉L ˆ )qq×
¢Å G ’
VÐÑ = max ;𝑉HH… + ; 𝑙𝑛 31 + exp Æ ˆ’)¶ K 4@ (43)
𝜇 𝑘g

The parameter kvb and the root function introduced by Koren is no longer used, because
the Ia reduction to zero when Va approaches zero will be described by other methods (see
chapter 4.2.3.2). The root will be replaced by the term 1+Va to achieve a faster algorithm.
A minimum voltage of 1V is added to avoid division through zero.

For calculation of the Va independent current part of the grid-cathode vacuum diode, see
chapter 4.2.3.1.

4.2.3 grid current modelling


4.2.3.1 the grid-cathode vacuumdiode (considering the “Anlauf”current)
The grid-cathode vacuum diode can be described very good using the expression we
developed in chapter 3.3. We have to rename the variables to avoid identical names for
different variables.

26
?"
'( B)CD(
1
I"Ø9 = ∙ ÙV, ∙ xg ∙ ln 31 + e )* ∙?" 4Û (44)
k"

4.2.3.2 splitting Is into Ia and Igr


For splitting Is into Ia and Igr, my equation of choose is not from Franz Tank but is
inspired from him, as he did important researches how to describe the ratio between Ig and
Ia as a function of the anode and grid voltage. Tank introduced the following equation
(and some others):
_
𝑉L 𝐵 ∙ 𝑉L ¹
I" = G ∙ G + 𝑉H + 𝑉Ü K ∙ ¹ (45)
𝜇 𝑉L + 𝐵 ∙ 𝑉H ¹

Different to Tanks approach, I found best fit quality by splitting the space charge current
instead of the cathode current. I took over the term 1/(b*Va+1) from Cohen&Helie, and
considered the fact that Igr is equal to Is at zero anode voltage. This leads me to the
following equation:

'((à
1 1
I"Ý ÙV
= ∙ , ∙ xs ∙ ln 31 + e * ∙?Þ 4Û ∙
) (46)
kÞ 1 + 𝑘𝐵 ∙ 𝑉L

4.2.3.3 Ia Emission limit for very small Va values


To model the Ia emission limit, we pick up Duncan’s idea of reflecting the gap between
the emission limit Ial and the Ia value (when there would not be a emission limitation) to
the grid. This means in other words to introduce a new current Igl from the grid to the
anode. For best fit quality, we connect this new current source behind the internal grid
resistor.
In addition, it is helpful to separate the emission-limited anode A from an internal
Anode Ai by a resistor Ra (of any value). This gives the possibility of referring to i(Ra)
when the not emission-limited internal anode current is meant.

1
I"á = i(Ra) − min G ∙ 𝑉 ¾‡ , i(Ra)K (47)
𝑘‡ L


IE = 𝑖 (𝑅𝑎) − 𝑖H‡ (48)

The min-function lead to unnatural sharp transitions from the emission-limited area to the
not emission limited area of Ia. We can see these unwished sharp kinks in Duncan’s and
Ayumi’s models.
I spent dozens of hours to find a smart way how to smooth this sharp kink. As a believer
in the God of the bible, I’m used to share all problems (including the technical one) to him.
I never got disappointed and just can recommend this habit. Certainly there are voices

27
interpreting my experience as a psychological phenomenon. Despite of this voices, I will
keep on doing so, because it just works... However, I got in the end the inspiration of a
solution based on Pythagoras’ theorem for the relationship between the legs of a right-
angled triangle.
While the hypotenuse represents the output of a smooth max-function, the other both legs
stands for the inputs.
ˆ
smax(x, y) = (𝑥 P + 𝑦 P ) äP (49)

By replacing the exponent 2 respectively ½ through n respectively 1/n, we are able to


control the smoothness of the transition. We achieve the desired smooth min-function by
replacing the inputs and the output through its reciprocal values. To avoid divisions though
zero, a small value (for example 1 femto = 1E-15) has to be added to x and y. Be aware
that the achieved smin-function is only working with positive inputs.

Bˆä
smin(x, y, n) = ((𝑥 + 1𝑓)B¹ + (𝑦 + 1𝑓)B¹ ) ¹ (50)


Example of the smin-function, with n=3
1.60E-02

1.40E-02

1.20E-02

1.00E-02 trace x

trace y
8.00E-03
softmin

6.00E-03

4.00E-03

2.00E-03

0.00E+00
0 0.5 1 1.5 2 2.5 3




28
4.2.4 Finetuning
At this point of my triode spice model development, I decided to verify finetuning
improvements with several different triode types found in my small collection: 6021, 6111,
6948, 7963, ECC81, ECC86. A graphical presentation of every single finetuning step with
results of every single triode type would blow up this paper to an unreadable amount, but
the fit quality of my final triode model for this triode types are collected at
adrianimmler.simplesite.com.
4.2.4.1 Keeping Ia positive @ Vg near zero
When we have a closer look to the anode current at grid voltage around zero and small Va,
we discover slightly negative values. The root cause is found in the two different ways how
we calculate the space charge current – one time according Langmuir-Childs law according
equation (38) and one time with the equation (46) considering the “Anlauf-current”. As
the anode current is the difference between Is and Igr, Ia should be zero at Va=0, but
because Igr considers the “Anlauf-current”, Is is overcompensated by Igr @ Va=0.

This offers an easy way to keep Ia positive by adding a compensating current source from
the internal anode to the cathode, while its value has to be calculated as follows:

'((à ¾Ð
1 1
IÞ$ = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û − 𝑉HH… (51)
kÞ 𝑘Ð

4.2.4.2 Avoidance of “Anlauf” current manipulations through the smin-function


When we have a closer look to the grid current in log scale at Vg around zero, we see that
new smin-function introduced in chapter 4.2.3.3 adds an unwished current to the grid’s
“Anlauf” current. Though this effect, the resulting grid current can appear decades too high
at higher Va voltages.
The Igl source has therefore to be “turned off” for this voltage ranges. A hard turn off/on
function is not suitable because of the risk for convergence problems. With several trials
and errors, I finally found a satisfying solution by adding the red terms to the Igl equation:

1
I"á = min 3i(Ra) − min ; ∙ 𝑉L ¾‡ , i(Ra)@ , 𝐼HÔÕ ∙ 𝑒 å)H 4 (52)
𝑘‡

4.2.4.3 Considering the “Anlauf” current slope dependence from Va


When we draw measured Ig data versus Va (with constant Vg between 0 and -1V) we will
find that the grids “Anlauf” current shows a small Va dependency for most tubes. This
shall be considered in the grid current model to improve the general Ig fit quality. I decided
not to introduce extra parameters for this job to keep the triode model as simple as possible
(My triode spice model counts already 14 parameters!). Representing a typical Va
dependency of the grids “Anlauf current”, the modified equation for the grid-to-cathode
vacuum diode no longer just a f(Vg) function:
?"
'( B)CD( Bˆ_∙æ)¶
1
I"Ø9 = ∙ ÙV, ∙ xg ∙ ln ;1 + e )* ∙?" @Û (53)
k"

29
4.2.4.4 Finetuning of the overall gain @ pos. Vg
At pos. Vg, it seems to be a fit quality improvement when the overall gain 1/ks is increased
about 10%. A fine tuning function ft1 is introduced based on the hyperbolic tangent to
avoid sharp steps.
1 + tanh (4𝑉HH… )
ftˆ = 1 + (54)
20

𝑓𝑡ˆ ¾Ð
IÐ = Q 𝑘Ð 𝑉ÐÑ 𝑖𝑓 𝑉ÐÑ ≥ 0 (55)
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

'((à
𝑓𝑡ˆ 1
I"Ý = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û ∙ (56)
kÞ 1 + 𝑘𝐵 ∙ 𝑉L


'((à ¾Ð
ftˆ 𝑓𝑡ˆ
IÞ$ = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û − 𝑉 (57)
kÞ 𝑘Ð HH…

4.2.4.5 Finetuning of Ig @ Vg near -0.5V and small Va


The Ig values seems to be too high in general around grid voltages of -0.5V and smaller Va
values. This behavior can be a bit improved with a further finetuning function ft2 as
follows:
ft P = 7 -1 − tanh ´3(𝑉H + 𝑉¡Ñ )µ® (58)


'((à
𝑓𝑡ˆ 1
I"Ý = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û ∙ (59)
kÞ 1 + 𝑓𝑡P + 𝑘𝐵 ∙ 𝑉L


'((à ¾Ð
ftˆ 𝑓𝑡ˆ
IÞ$ = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û − 𝑉HH… (60)
k Þ (1 + 𝑓𝑡P ) 𝑘Ð

4.2.5 Reduction of parameter interactions


A drawback of almost every triode spice model is the strong parameter interaction between
the exponent (here called xs) and the overall gain (here described with 1/ks). Even a small
change of the exponent makes it necessary to retune the overall gain, so the fitting process
is quite iterative. It would be quite helpful when this to parameters would be more
independent.
To achieve this goal, I introduced a new parameter rad (differential anode resistance)
which has to be set instead of ks, and which can be easy estimated from every triode
datasheet (or measured data of course).

30
The old parameter ks is then automatically calculated considering the exponent xs, while
the Ia-trace slope is retained.
Needless to say that the nonlinear Ia trace has not a constant differential anode resistance,
so we have to define a point Iad where we wish to keep the differential anode resistance
independent of the exponent. A way that sets this point automatically (and also for most
triodes to a suitable value) is the following simple equation:

100
ILÕ = (61)
𝑟LÕ

It took me several evenings to develop this solution for the wished function decribed, so I
am happy and proud to present the following three equations:

1 (62)
kÐ = ¾Ð
𝜇
3 4
𝑥Ð ∙ 𝑟LÕ ∙ 𝐼LÕ ˆBˆ⁄¾ë
Now, a variation of xs does no longer manipulate the Ia slope @ Iad, respectively the xs
influence to the Ia slope is strongly reduced for currents other than Iad. But a variation of
xs still moves the Ia traces, it “shifts” the Ia traces on the Va axis.
A way to eliminate this “shift” at Iad is to replace the contact voltage Vct through a new
voltage Vg0 which has the same value as Vct for xs=1.5.

ì
VH$ = 𝑉¡Ñ + ëæ𝐼LÕ ∙ 𝑘Ð − î.ïæ𝐼LÕ ∙ 𝑘йí_ (63)

1
k йí_ = ˆ.¯ (64)
2𝜇
3 4
3𝑟LÕ ∙ ðæ𝐼LÕ

The two graphics below shows the effect of this approach. The spice simulation was done
twice with xs= 1.5 and 1.2, and Vg= 0V, -1V, -2V.
On the left, ks and Vg0 are hold constant, while on the right, ks and Vg0 are calculated
according equations above.

31
As wished, xs influences now mainly the nonlinearity of the Ia traces, while in traditional
spice models, xs also modifies the average Ia slope (causing a lot of iterative work to
achieve a good fit).
The found way to get independence does not consider the influence of the means for mimic
the island effect, so there remains a small shift in x-direction when the island effect appears
in a severe way.

Beside the space charge current, there are two other currents following the Langmuir-
Childs law: The (more or less) Va independent part of the grid current, called Igvd, and the
emission limit.
While this new concept of determining the differential resistance instead of the 1/k factors
is a real advantage for modeling the Ia traces (@ negative Vg), the advantage turned out to
be not so big for the emission limit, and neglectable for Igvd. Hence, this new concept is
not applied to Igvd.

For the emission limit, the following equations are suggested:

𝐼‡Õ = 1𝑚 ∙ æ𝑟LÕ‡ (65)

1
k‡ = ¾‡
1 (66)
3 4
𝑥‡ ∙ 𝑟LÕ‡ ∙ 𝐼‡Õ ˆBˆ⁄¾ñ

The suggested equation for Ild shows some potential for improvements, since the Il slope is
still dependable from xl for some triodes.

32
4.2.6 Equation summary
Instead of some “”bill of currents” equations, it is better to show the relationship between
the internal currents in apraphical way:

The core of my triode model consists of the following set of equations:


VHH… = 𝑉H + 𝑉H$ − 𝑅H… ∙ 𝐼H (42)

𝑉L 𝑉L ˆ )qq×
¢Å G ’
(43)
VÐÑ = max ;𝑉HH… + ; 𝑙𝑛 31 + exp Æ ˆ’)¶ K 4@
𝜇 𝑘g

𝑓𝑡ˆ ¾Ð
IÐ = Q 𝑘Ð 𝑉ÐÑ 𝑖𝑓 𝑉ÐÑ ≥ 0 (55)
0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
?"
'( B)CD( Bˆ_∙æ)¶ (53)
1
I"Ø9 = ∙ ÙV, ∙ xg ∙ ln ;1 + e )* ∙?" @Û
k"


'((à
𝑓𝑡ˆ 1 (59)
I"Ý = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û ∙
kÞ 1 + 𝑓𝑡P + 𝑘𝐵 ∙ 𝑉L

1 (52)
I"á = min 3i(Ra) − min ; ∙ 𝑉 ¾‡ , i(Ra)@ , 𝐼HÔÕ ∙ 𝑒 å)H 4
𝑘‡ L

Equations for Finetuning:


To be honest, the first equation (60) is actually more than a finetuning function, because a
lack of this equation can lead to an Ia trace crossing, which may cause convergence
problems.

'((à ¾Ð
ftˆ 𝑓𝑡ˆ (60)
IÞ$ = ∙ ÙV, ∙ xs ∙ ln 31 + e)* ∙?Þ 4Û − 𝑉
k Þ (1 + 𝑓𝑡P ) 𝑘Ð HH…

33
1 + tanh (4𝑉HH… ) (54)
ftˆ = 1 +
20
ft P = 7 -1 − tanh ´3(𝑉H + 𝑉¡Ñ )µ® (58)

Equations for a reduced influence of xs to ks:


100
ILÕ = (61)
𝑟LÕ

1
kÐ = ¾Ð
𝜇 (62)
3 4
𝑥Ð ∙ 𝑟LÕ ∙ 𝐼LÕ ˆBˆ⁄¾ë
ì
VH$ = 𝑉¡Ñ + ëæ𝐼LÕ ∙ 𝑘Ð − î.ïæ𝐼LÕ ∙ 𝑘йí_ (63)

1
k йí_ = ˆ.¯ (64)
2𝜇
3 4
3𝑟LÕ ∙ ðæ𝐼LÕ

Equations for a reduced influence of xl to kl:


𝐼‡Õ = 1𝑚 ∙ æ𝑟LÕ‡ (65)

1 (66)
k‡ = ¾‡
1
3 4
𝑥‡ ∙ 𝑟LÕ‡ ∙ 𝐼‡Õ ˆBˆ⁄¾ñ

4.2.7 Achieved fit quality

Ia [A]
Immler-model-fit to measured values of a 6948 triode
8E-3

7E-3

Vg=2V
6E-3
Vg=0V

5E-3

Vg=-2V
4E-3

3E-3
Ig@Vg=2V Vg=-4V
Pmax=0.5W
2E-3

Ig@Vg=1V
1E-3

0E+0
Va [V]
0 50 100 150 200 250 300 350 400

34
Immler-model-fit to measured values of a 6948 triode
Ia [A]
8E-3

7E-3

Vg=2V
6E-3

Ig@Vg=2V
5E-3
Vg=1V

4E-3

3E-3

2E-3
Vg=0V
Ig@Vg=1V

1E-3

0E+0 Va [V]
0 5 10 15 20 25 30 35 40 45 50

Immler-model-fit to measured values of a 6948 triode


Ig [A]
1E-2

1E-3

1E-4

1E-5

Ig@Va=100V
1E-6 Ig@Va=50V
Ig@Va=20V
Ig@Va=10V
1E-7
Ig@Va=5V
Ig_fit@Va=50V
1E-8
Ig_fit@Va=10V

1E-9
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
Vg [V]

35
Immler-model-fit to measured values of a 6948 triode
Ig [A]
1E-2

Ig@Vg=2V
Ig@Vg=1V
1E-3

Ig@Vg=0.5V
1E-4

Ig@Vg=0V
1E-5

1E-6 Ig@Vg=-0.5V

1E-7

1E-8
Ig@Vg=-1V

1E-9 Va [V]
0 20 40 60 80 100 120 140 160 180 200

Parameters:
mu = 90.9
rad = 27575
Vct = 0.545
kp = 322
xs = 1.488
Rgi = 159
kB = 0.35
radl = 2300
tsh = 4.8
xl = 1
rgd = 400
Vctg = -0.19
xg = 1.189
VT = 0.103

The achieved fit quality is satisfying now. But, like in every development, there are still
potentials for improvements:

• The grid current for Vg near zero shows in general too high values when Va is
small (less then 30V) too.
• The need of the finetuning-functions ft1 and ft2 let me suppose that there still is a
structural problem in my approach.

36
4.2.8 The new generic LTspice triode model
*Generic triode model from Adrian Immler, version1, Oct 2018
*History's best of tube describing art (plus some new ideas) is merged to this new approach.
*@ neg. Vg, Ia accuracy is similar to Koren or Ayumi models.
*@ small neg. Vg, the “Anlauf” current is considered.
*@ pos. Vg, Ig and Ia accuracy is on a unrivaled level.
*This offers new simulation possibilities like bias point setting with MOhm grid resistor,
*Audion radio circuits, low voltage amps, guitar distortion stages or pulsed stages.
* grid
* | anode (plate)
* | | cathode
* | | |
.subckt GenTriode_i1 G A K
.params
*Parameters for the space charge current @ Vg <= 0
+ mu = 70 ;Determines the voltage gain @ constant Ia
+ rad = 10k ;Differential anode resistance, set @ Iad and Vg=0V
+ Vct = 0.48 ;Offsets the Ia-traces on the Va axis. Electrode material's contact potential
+ kp = 173 ;Mimics the island effect
+ xs = 1.5 ;Determines the curve of the Ia traces. Typically between 1.2 and 1.8
*
*Parameters for assigning the space charge current to Ia and Ig @ Vg > 0
+ kB = 0.27 ;Describes how fast Ia drops to zero when Va approaches zero.
+ radl = 420 ;Differential resistance for the Ia emission limit @ very small Va and Vg > 0
+ tsh = 8 ;Ia transmission sharpness from 1th to 2nd Ia area. Keep between 3 and 20. Start with 20.
+ xl = 1.2 ;Exponent for the emission limit
*
*Parameters of the grid-cathode vacuum diode
+ Rgi = 80 ;Internal grid resistor. Causes an Ia reduction @ Ig > 0.
+ kg = 500 ;Inverse scaling factor for the Va independent part of Ig (caution - interacts with xg!)
+ Vctg = -0.07 ;Offsets the log Ig-traces on the Vg axis. Electrode material's contact potential
+ xg = 1.3 ;Determines the curve of the Ig slope versus (positive) Vg and Va >> 0
+ VT = 0.104 ;Log(Ig) slope @ Vg<0. VT=k/q*Tk (cathodes absolute temp, typically 1150K)
*
*Parameters for the caps
+ cag = 1p ;From datasheet. May lead to gain-decrease @ highest audible frequencies!
+ cak = 1p ;From datasheet
+ cgk = 1p ;From datasheet
*
*Calculated parameters
+ Iad = 100/rad ;Ia where the anode a.c. resistance is set according to rad.
+ ks = pow(mu/(rad*xs*Iad**(1-1/xs)),-xs) ;Reduces the unwished xs influence to the Ia slope
+ ksnom = pow(mu/(rad*1.5*Iad**(1-1/1.5)),-1.5) ;Sub-equation for calculating Vg0
+ Vg0 = Vct + (Iad*ks)**(1/xs) - (Iad*ksnom)**(2/3) ;Reduces the xs influence to Vct.
+ kl = pow(1/(radl*xl*Ild**(1-1/xl)),-xl) ;Reduces the xl influence to the Ia slope @ small Va
+ Ild = sqrt(radl)*1m ;Current where the Il a.c. resistance is set according to radl.
*
*Space charge current model
Bggi GGi 0 V=v(Gi,K)+Vg0 ;Effective internal grid voltage.
Bahc Ahc 0 V=uramp(v(A,K)) ;Anode voltage, hard cut to zero @ neg. values
Bst St 0 V=max(v(GGi)+v(Ahc)/(mu), v(Ahc)/kp*ln(1+exp(kp*(1/mu+v(GGi)/(1+v(Ahc))))));Steering volt.
Bs Ai K I=ft1()/ks*pow(v(St),xs) ;Langmuir-Childs law for the space charge current Is
.func ft1() {1+(1+tanh(4*v(GGi)))/20} ;Finetuning-function for better overall fit at pos Vg
*
*Anode current limit @ small Va
.func smin(x,y,k) {pow(pow(x+1f, -k)+pow(y+1f, -k), -1/k)} ;Min-function with smooth trans.
Ra A Ai 1
Bgl Gi A I=min(i(Ra)-smin(1/kl*pow(v(Ahc),xl),i(Ra),tsh),i(Bgvd)*exp(4*v(G,K))) ;Ia emission limit
*
*Grid model
Rgi G Gi {Rgi} ;Internal grid resistor for "Ia-reduction" @ Vg > 0
.func Ivd(Vvd, kvd, xvd, VTvd) {1/kvd*pow(VTvd*xvd*ln(1+exp(Vvd/VTvd/xvd)),xvd)} ;Vacuum diode function
Bgvd Gi K I=Ivd(v(G,K)+Vctg-1m*sqrt(v(Ahc)), kg, xg, VT) ;Grid-cathode vacuum diode
.func ft2() {7*(1-tanh(3*(v(G,K)+Vg0)))} ;Finetuning-func. improves ig-fit @ Vg near -0.5V, low Va.
Bgr Gi Ai I=ft1()*ivd(v(GGi),ks, xs, 0.8*VT)/(1+ft2()+kB*v(Ahc));Is reflection to grid when Va appr. 0V
Bs0 Ai K I=ft1()*ivd(v(GGi),ks, xs, 0.8*VT)/(1+ft2()) - ft1()/ks*pow(v(GGi),xs) ;Compensates neg Ia @
;small Va and Vg near zero

*Caps
C1 A G {cag}
C2 A K {cak}
C3 G K {cgk}
.end

1 TETRODES & PENTODES
1.1 Do it with passion or leave it
For the diodes and especially for triodes, I first of all did a thorough web research to explore
and to analyze existing tube spice approaches. And for tetrodes and pentodes, there is a lot
of stuff from Koren, Duncan, Ayumi, Alexander Gurskii, Reefman, André and many others.
But I decided to leave that analysis of all these pentode models, because after my intensive
work on triodes, I feel tired and I can’t continue with passion.
Instead, I just took my triode model and enhanced it to a new tetrode/pentode approach with
the principle found during my triode work. May others do the comparison to the existing
tetrode/pentode spice models ...

1.2 My pentode approach (unevaluated)


*Generic pentode model from Adrian Immler, version1, Oct 2018
*This model is an enhancement of Adrians generic triode model to achieve pentode behaviour.
*Hence, it is also suitable when the pentode is "triode connected".
*Convenient for power beam pentodes as well as for small signal pentodes (just play with radl!).
*
* grid1
* | grid2
* | | plate (in this model, "anode" means the internal virtual triode anode)
* | | | cathode
* | | | |
.subckt GenPentode_i1 G1 G2 P K
.params
*Parameters for the space charge current @ Vg <= 0
+ mu1 = 40 ;Determines the voltage gain @ constant Ia
+ rad = 10k ;Differential anode resistance, set @ Iad and Vg=0V
+ Vct = 0.48 ;Offsets the Ia-traces on the Va axis. Electrode material's contact potential
+ kp = 173 ;Mimics the island effect
+ xs = 1.5 ;Determines the curve of the Ia traces. Typically between 1.2 and 1.8
*
*Parameters for assigning the space charge current to Ia and Ig @ Vg > 0 and small Va
+ kB1 = 0.27 ;Describes how fast Ia drops to zero when Va approaches zero.
+ radl = 700 ;Differential resistance for the Ia emission limit @ very small Va and Vg > 0
+ tsh = 6 ;Ia transmission sharpness from 1th to 2nd Ia area. Keep between 3 and 20. Start with 20.
+ xl = 1.5 ;Exponent for the emission limit
*
*Parameters of the grid-cathode vacuum diode
+ Rg1i = 80 ;Internal grid resistor. Causes an Ia reduction @ Ig > 0.
+ kg1 = 500 ;Inverse scaling factor for the Va independent part of Ig (caution - interacts with xg!)
+ Vctg1 = -0.07 ;Offsets the log Ig-traces on the Vg axis. Electrode material's contact potential
+ xg1 = 1.3 ;Determines the curve of the Ig slope versus (positive) Vg and Va >> 0
+ VT = 0.104 ;Log(Ig) slope @ Vg<0. VT=k/q*Tk (cathodes absolute temp, typically 1150K)
*
*Parameters for the caps
+ cg1p = 1p ;From datasheet. May lead to gain-decrease @ highest audible frequencies!
+ cg1All= 1p ;From datasheet
+ cpAll = 1p ;From datasheet
*
*Parameters to enhance the triode model to a pentode model
+ mu2 = 20
+ kB2 = 0.15
+ Rg2i = 300
*
*Calculated parameters
+ Iad = 100/rad ;Ia where the anode a.c. resistance is set according to rad.
+ ks = pow(mu1/(rad*xs*Iad**(1-1/xs)),-xs) ;Reduces the unwished xs influence to the Ia slope
+ ksnom = pow(mu1/(rad*1.5*Iad**(1-1/1.5)),-1.5) ;Sub-equation for calculating Vg0
+ Vg0 = Vct + (Iad*ks)**(1/xs) - (Iad*ksnom)**(2/3) ;Reduces the xs influence to Vct.
+ kl = pow(1/(radl*xl*Ild**(1-1/xl)),-xl) ;Reduces the xl influence to the Ia slope @ small Va
+ Ild = sqrt(radl)*1m ;Current where the limited anode a.c. resistance is set according to radl.
*
*Space charge current model
Bggi GG1i 0 V=v(G1i,K)+Vg0 ;Effective internal grid voltage.
Bahc Ahc 0 V=uramp(v(P,K)/mu2+v(G2i,K)) ;voltage of the virtual triode anode, hard cut to zero
Bst St 0 V=max(v(GG1i)+v(Ahc)/(mu1), v(Ahc)/kp*ln(1+exp(kp*(1/mu1+v(GG1i)/(1+v(Ahc))))));Steering volt.
Bs Ai K I=ft1()/ks*pow(v(St),xs) ;Langmuir-Childs law for the space charge current Is
.func ft1() {1+(1+tanh(4*v(GG1i)))/20} ;Finetuning-function for better overall fit at pos Vg
*
*Anode current limit @ small Va
.func smin(x,y,n) {pow(pow(x+1f, -n)+pow(y+1f, -n), -1/n)} ;Min-function with smooth trans.
Ra A Ai 1
Bpl G2i P I=i(Rp) - smin(1/kl*pow(v(P,K),xl),i(Rp),tsh);Ia emission limit
*
*Grid model
Rg1i G1 G1i {Rg1i} ;Internal grid resistor for "Ia-reduction" @ Vg > 0

38
.func Ivd(Vvd, kvd, xvd, VTvd) {1/kvd*pow(VTvd*xvd*ln(1+exp(Vvd/VTvd/xvd)),xvd)} ;Vacuum diode function
Bg1vd G1i K I=Ivd(v(G1,K)+Vctg1-1m*sqrt(v(Ahc)), kg1, xg1, VT) ;Grid-cathode vacuum diode
.func ft2() {7*(1-tanh(3*(v(G1,K)+Vg0)))} ;Finetuning-func. improves ig-fit @ Vg near -0.5V, low Va.
Bg1r G1i Ai I=ft1()*ivd(v(GG1i),ks, xs, 0.8*VT)/(1+ft2()+kB1*v(Ahc));Is reflection to grid when Va appr.
zero
Bs0 Ai K I=ft1()*ivd(v(GG1i),ks, xs, 0.8*VT)/(1+ft2()) - ft1()/ks*pow(v(GG1i),xs) ;Compensates neg Ia
*@ small Va and Vg near zero
*
*additional model parts necessary for a pentode
Rg2i G2 G2i {Rg2i}
Rp P A 1
Bg2r G2i A I=i(Ra)/(1+kB2*max(0,v(P,K)))
*
*Caps
C1 G1 P {cg1p} ;from datasheet
C2 G1 K {cg1All/2} ;most datasheets gives a cap "g1 to all except plate". As this model does not
consider the
*heater or the ambient as further electrodes for parasitic caps, best way is to assume this " g1 to all"
cap
*as it would be half to cathode and half to g2.
C3 G1 G2 {cg1All/2}
C4 P K {cpAll/2} ;most datasheets gives a cap "plate to all except g1". As this model does not consider
the
*heater or the ambient as further electrodes for parasitic caps, best way is to assume this " plate to
all" cap
*as it would be half to cathode and half to g2.
C5 P G2 {cpAll/2}
.end

2 ABOUT THIS PAPER’S AUTHOR


Together with my wife, I live at the east boarder of Switzerland. Since 15 years I work as a
hardware development engineer at Brusa Elektronik AG, with the aim to give my part to
electrify the automotive world.
In my spare time, I love to immerge into the not so efficient world of tube circuits, and to
discover fascinating views trough the viewfinder of my Pentax DSLR.
However, at the end of my time here on earth, the only thing that matters will not be any
generic tube spice model of mine - but to be called a child of the Almighty, who is the
source of life and peace and the author of every natural law.

3 THANKS!
To my wife to resign the many ours this paper needed to be written.

To all mentioned authors of vacuum tube models for your ideas. Best of it is merged now!

To my infinitely wise God, inspiring me to find a min-function with smooth transition.

”…there is nothing better, than that a man should rejoice in his own works; for that is his
portion.”
Ecclesiastes 3:22

39

Vous aimerez peut-être aussi