Вы находитесь на странице: 1из 10

Audio Engineering Society

Convention Paper 8507


Presented at the 131st Convention
2011 October 20–23 New York, USA

This Convention paper was selected based on a submitted abstract and 750-word precis that have been peer reviewed
by at least two qualified anonymous reviewers. The complete manuscript was not peer reviewed. This convention
paper has been reproduced from the author’s advance manuscript without editing, corrections, or consideration by the
Review Board. The AES takes no responsibility for the contents. Additional papers may be obtained by sending request
and remittance to Audio Engineering Society, 60 East 42nd Street, New York, New York 10165-2520, USA; also see
www.aes.org. All rights reserved. Reproduction of this paper, or any portion thereof, is not permitted without direct
permission from the Journal of the Audio Engineering Society.

A Triode Model for Guitar Amplifier


Simulation with Individual Parameter Fitting
Kristjan Dempwolf1 , Martin Holters1 , and Udo Zölzer1
1
Helmut Schmidt University, Hamburg, Germany

Correspondence should be addressed to Kristjan Dempwolf (kristjan.dempwolf@hsu-hh.de)

ABSTRACT
A new approach for the modeling of triodes is presented, featuring simple and physically-motivated equations.
The mathematical description includes the replication of the grid current, which is a relevant parameter for
the simulation of overdriven guitar amplifiers. If reference data from measurements of practical triodes is
available, an individual fitting to the reference can be performed, adapting some free parameters. Parameter
sets for individual models are given. To study the suitability for circuit simulations, a SPICE model is
created and tested under various conditions. Results of the model itself and when embedded in SPICE
simulations are presented and compared with measurements. It is shown that the equations characterize the
properties of real tubes in good accordance.

1. INTRODUCTION ical foundations given in the textbooks. Numerous


With the introduction of the transistor, the tube researchers promoted the idea and the formulations
technology rapidly lost importance and managed to got modified and improved again and again. The
survive only in a few special applications. Regard- first models were inspired by the desire for hi-fi cir-
ing the audio domain, tubes are still popular with cuits simulations. The recent activities towards tube
audiophiles and musicians and can be found in (high- modeling consider overdriven amplifiers or real-time
end) hi-fi as well as in guitars amplifiers. The cir- simulations outside the SPICE engine [1].
cuit simulation tool SPICE did not provide any tube
and some people started to create their own simple
models. These were composed of basic components 2. VACUUM TUBE BASICS
and mathematical descriptions related to the phys- In this section we give a very brief introduction to
the fundamentals of vacuum tubes as it was partly
Dempwolf et al. Individual Triode Model

presented in a recent paper [2]. The tube experts 2.2. Triode Characteristics
may skip this part. For detailed exploration we refer Triodes have a third electrode placed near the cath-
to the historic standard literature [3, 4, 5]. ode, the so-called grid (G). Unlike the anode it is
not a solid plane but constructed of fine wire as a
2.1. Diode Characteristics
mesh, ladder or helix. The spacings between the
The simplest tube is the vacuum diode, where two
wires are large compared to the wires’ thickness, so
electrodes, anode and cathode are mounted in a vac-
the electrons can pass the grid on their way from the
uum cylinder. The cathode (K) is heated and hence
cathode to the anode easily.
electrons are emitted by thermionic emission. The
anode (A) or plate is designed to collect the elec- A voltage applied to the grid is able to control the
trons. Electrons can not be emitted by the cold an- flow of electrons from the cathode: a negative grid
ode and in unheated condition, no current flow is voltage causes an electric field that counteracts with
possible. the electric field from the anode, the cathode current
is reduced. A higher negative grid voltage inhibits
2.1.1. Initial Velocity Current
the current flow thoroughly. Small changes in grid
The electrons leave the hot cathode with random ve-
voltage cause high changes in current flow, thus it is
locities, some are slower and some are faster. With-
possible to use triodes for amplification.
out an external voltage applied to the anode, and
even for a slightly negative anode with respect to For clarity we define the following conventions: Va
the cathode, some of the faster electrons reach the is the anode- and Vg the grid voltage, both referred
anode so that a small current is observed. To sup- to the cathode potential. We define Ik = Ig + Ia
press the current flow a small negative voltage must where Ik , Ia and Ig are the cathode, anode and grid
be applied. The current through the diode I follows currents, respectively. Ia and Ig are directed into
V
the device.
I = I 0 · e ET , (1) For the introduction of a grid between anode and
with anode voltage V , thermal voltage ET (in Volt) cathode it is a common prospect to replace the three-
and current I0 at zero voltage. terminal triode by a two-terminal one [4], where an
electrode with the effective voltage is placed at the
2.1.2. Space-Charge Current grids’ position,
When a positive voltage is applied from anode to  
cathode, the emitted electrons are pulled by the an- Va
Veff = Vg + . (3)
ode and the current flow increases. But only a part µ
of the emitted electrons reach the anode. Most of
The amplification factor µ expresses how much
them have only low velocity, they stay near the hot
higher the cathode current is influenced by the grid
cathode and form the space charge, a cloud of nega-
than by the anode voltage. As for the diode we speak
tive charges. These (negative) charges screen the an-
of the space-charge region and the cathode current
ode and obstruct subsequent electrons to continue to
is calculated analog to equation (2),
the anode. The resulting current is self-limited and
depends on the anode voltage. This is expressed by 3
Ik = G · (Veff ) 2 , Veff > 0. (4)
the Langmuir-Child equation
3
I = G · V 2, V > 0, (2) Equation (4) is valid for the “normal” operating con-
dition of triodes: the grid is slightly more negative
with perveance G, a constant that only depends on than the cathode, but the anode voltage is very high
the geometrical construction of the tube. (e.g. Vg = −2 V and Va = 300 V). The grid draws
no current and consequentially anode and cathode
2.1.3. Saturation Current
current are equal, i.e. Ig = 0 and Ia = Ik .
Beyond a certain voltage the current saturates. All
emitted electrons are drawn to the anode and the 2.2.1. Grid Current
current remains nearly constant when the external As a special case we have to consider the opera-
voltage is increased further. tion with positive grid voltage. Since a positive grid

AES 131st Convention, New York, USA, 2011 October 20–23


Page 2 of 10
Dempwolf et al. Individual Triode Model

now likewise attracts electrons, a positive current same manufacturer may show up to 20 % deviation
Ig arises and hence the cathode current divides into from each other. Exemplary measurements are dis-
anode and grid current. In most applications this is cussed e.g. in [9, 10]. Comparing tubes from current
an unwanted effect because the flow of grid current production, new old stock (NOS) or used ones will
introduces additional distortions, but it can be ob- reveal high variations of the electrical characteristics
served in guitar amplifier designs where it leads to - and explain why they “sound” differently.
effects like the blocking distortion [6, 7]. For a cor-
3.1. Measurements
rect simulation the inclusion of the grid current can
To have reliable reference data, numerous measure-
play an important role, especially if coupled stages
ments on common standard triodes (12AX7, 12AU7
in a cascade are examined.
and 12AT7) were performed by the authors. The
The primary-grid-current law [5, 8] gives an approxi- first measurements were done manually, see [2], what
mation of the grid current. The basic concept is that turned out to be a highly time-consuming procedure.
the electron paths inside a tube depend only on the The measurements in this study were executed using
ratio of anode and grid voltages but not on their an automated setup. Grid and anode current were
amplitudes. Thus the division into grid and anode measured at the same time in good resolution for a
current has to be a function of the voltage ratio, a fine mesh of discrete Vg and Va voltages. The meters
consideration that was affirmed by experiments. In were arranged in a way that the voltages were cap-
Spangenberg’s book [5] we find the equation tured correctly. The specifications for 12AX7 triodes
s were: Va =20 V to 300 V in steps of 20 V with an ac-
Ia Va Va curacy of ±100 mV and Vg =−4 V to 3 V in steps of
=δ , > 0.8, Vg > 0 (5) 100 mV with an accuracy of ±10 mV. For other tri-
Ig Vg Vg
odes the Vg range was adapted to capture the signif-
with the constant δ, the current-division factor. For icant area of operation. All currents were measured
the grid current follows with an accuracy of 100 nA. The constraints given
by maximum power consumption Pmax and maxi-
1 mum cathode current Ik,max were respected.
Ig = Ik · q , Vg > 0. (6)
1 + δ VVag
3.2. Observations on the Current Measurements
Figure 1 opposes qualitatively a measurement from
2.2.2. Secondary Emission an old 12AX7 tube (RSD) and Langmuir-Child’s
When an accelerated electron strikes an electrode, law. While the measurement in principle follows
new electrons can be expelled from it. This effect is the formula, two differences are visible: First, the
called secondary emission. Regarding triodes with theoretic 32 -exponent does not fit perfectly for real
negative grid this has marginal consequence, because triodes, see 1(a). In the plotted example, the slope
the anode has the most positive potential and all of the measured curve is lower, representing an ex-
secondary electrons formed at the anode will fly back ponent < 1.5.
to it. For the positive grid, secondary electrons may
The second difference can be found in the lower part
originate from the grid, too, and will be attracted by
of the curve. Equation (4) yields Ik = 0 for Veff =
the anode or even, if the grid is more positive than
0, and is not defined for voltages Veff < 0. The
the anode, the other way around. Deviations for
measurement offers a small current for this case and
practical tubes from equation (6) can be explained
a smooth transition towards zero current. This is
to a large extent by this effect.
depicted in Figure 1(b), where the highlighted region
tags the range Veff < 0.
3. PRACTICAL TUBES
Real tubes depart occasionally clearly from the ide- Equation (6) states that the grid current Ig depends
alized formulations given in Section 2 or the man- both on Vg and Va . Examining the considered work-
ufacturer’s data sheets. Degeneration, aging effects ing range we find that this is not true for all types
or small constructive variations have influence on the of triodes. Figure 2 illustrates the dependency for
characteristics. Even tubes from the same type and 12AX7 and 12AT7 tubes. Plotted are the Ig = f (Va )

AES 131st Convention, New York, USA, 2011 October 20–23


Page 3 of 10
Dempwolf et al. Individual Triode Model

8 3

Ig in mA →
1.5
anode current in mA →

anode current in mA →
12AX7/ECC83
6 1
2 0.5
4 0
1 20 60 100 140 180 220 260 300
2
Va in V →

Ig in mA →
0 4
12AT7/ECC81
0 100 200 300 0 100 200 300
2
Va in V → Va in V →
(a) exponent variation (b) Veff < 0 range 0
20 60 100 140 180 220 260 300
Fig. 1: Anode current calculated with Langmuir- Va in V →
Childs formula (dotted) and measurements from a
RSD 12AX7 triode (solid) in a qualitative plot. Fig. 2: Grid current as a function of Va for different
triode exemplars for Vg = 0.5 V, 1 V and 1.5 V.

curves measured for different exemplars. While the


12AT7 curves reflect the expected behavior (Ig falls cause the polynomial for the perveance, which is the
with increasing Va ), the 12AX7 curves show only a only real constant, has a high excursion.
small dependency on Va . Cohen and Helie [14] improved Koren’s model and
added a more realistic grid current. Ig is expressed
4. TUBE MODELING by a piecewise-defined function with a linear part,
Basically we can distinguish between physical mod- a second-order polynomial and a smooth transition.
els, which are based on the physical tube equations They performed current measurements in a similar
and heuristic or phenomenological models, which manner and identified Koren’s parameters individu-
have no physical foundation. In the following we ally for three triodes.
will briefly review three approaches to get an intro-
duction to the subject. For more information on the 5. INDIVIDUAL MODEL
SPICE simulation of tube circuits we refer to a re- Based on the findings from Section 3 a new descrip-
cent book [10] and the original papers. tion is derived, following the idea that the model in
The phenomenological description suggested by Ko- general has to be physically motivated, i.e. the for-
ren [11] is one of the most popular models, featur- mulations have to regard the traditional equations.
ing a library with common tubes. In his approach Furthermore the model has to be adaptable, so that
the triode is composed from basic circuit elements, simulations can be fitted individually to a selected
namely current sources, resistors and diodes and a tube. The proposed model is designed as .subckt
mathematical description. The parameters were im- for the software tool LTspice1 . This decision was
proved by Konar [12] later. This model gives good made because LTspice is free, easy to use and has
results for simulations with negative grid. no limitations of circuit elements. Another advan-
tage over other SPICE simulators is the possibility
In the approach of Cardarilli et al. [13] the “tube to directly save *.wav-files.
constants” (e.g. µ, G) are exchanged by 3rd-order
polynomials, then the model (with its many un- 5.1. Cathode and Grid Current
knowns) is fitted to measurement data. The pre- From the measurements we made the observation
sented results, containing the simulation of a guitar that the exponent in equation (4) may differ from
amplifier, are very promising. Nevertheless the for- 1 LTspice IV, Linear Technology Corporation,
mula has some lack of physical interpretation, be- www.linear.com/designtools/software/

AES 131st Convention, New York, USA, 2011 October 20–23


Page 4 of 10
Dempwolf et al. Individual Triode Model

the theoretical value 32 . This is also supported by the


literature: there are historic research papers on the 3 3
exponents’ variation, e.g. [15], and in some books
explicitly a parametrized exponent is used [3, 5]. In 2 2
the upcoming model we use the approximation 1 1
Ik ≈ G · (Veff )γ , Veff > 0. (7)
-2 0 2 4 -1 0 1 2
with exponent γ as a free parameter. Note that the
transition for Veff ≤ 0 is at first not considered in Fig. 3: Smoothing with the log/exp combination.
this equation. Original functions fi (solid) and smoothed functions
From equation (6) we expect the grid current to fol- gi (dashed) and hi (dotted) with adaption C = 2.
low Ig = Ik · ∆, where the factor can be roughly as illustrated in Figure 3 (right).
approximated by
r 5.3. New Model Equations
1 Vg To express the cathode current for the new model
∆= q ≈δ . (8)
1 + δ̃ Va Va we will use equation (7) and apply the smooth tran-
Vg
sition as in Section 5.2, leading to
This expression matches the relation Ik /Ig satisfac-
 1 γ
 !

torily, except for low values of Vg (i.e. the voltage Ik = G log 1 + exp Ca Vµa + Vg

ratio is high), as explained in Section 3.2. Ca
5.2. Smooth Transition (15)
To create a smooth transition we decided to use a with adaption factor Ca . Most measurements show
combination of exponential function and logarithm. a slight twist compared to equation (15), the mea-
As introduction we examine a simple function sured Ik current is higher for high and lower for small
( anode voltages and Vg > 0. Better fitting results are
x ,x > 0 obtained when the adaption factor compensates for
f1 (x) = (9) F
0 , x < 0. this. In this study we will simply use Ca = 1+V a
> 0,
but other expressions f (Va ) may work well, too.
Obviously there is a knee at x = 0. To smoothen
this discontinuity we consider a second function A similar expression can be made for Ig . Obviously
it is not possible, to bend equation (8) so that it
g1 (x) = log(1 + ex ) (10) meets the required slope for the low end, see Fig-
showing the same asymptotic behavior as equa- ure 4. To provide for the correct trend in this range
tion (9) for greater values of |x|. With increasing we expand equation (8) to
x the result tends to g1 (x) ≈ log(ex ) = x, while the  2 s  2  − 32
function value approaches zero in −x direction. The Vg Vg Va4 Vg Vg
∆≈δ · 4 =δ ·
curve shape can furthermore be adapted towards the Va Va Vg Va Va
linear function using an additional factor C, giving (16)
h1 (x) = log(1 + eC·x ) · 1
C. (11)
This relation is illustrated in Figure 3 (left), where 0.4
equations (9), (10) and (11) are plotted. The same
0.2
approach can be applied to power functions, e.g.
0
(
x2 , x > 0
f2 (x) = (12) 0 1 2 3
0 ,x < 0
2 Vg in V →
g2 (x) = log (1 + ex ) (13)
  2 Fig. 4: Ratio ∆ of grid and cathode current for
h2 (x) = log 1 + eC·x · C1

(14) Va =20 V and Va =60 V, measurements (solid) and
computed acc. to equation (6) (dashed).

AES 131st Convention, New York, USA, 2011 October 20–23


Page 5 of 10
Dempwolf et al. Individual Triode Model

* CONNECTIONS: ANODE/PLATE * .SUBCKT 12AX7 A G K


* | GRID * *~~ parameter ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
* | | CATHODE * .PARAMS Cgk=2.3P Cag=2.4P Cak=.9P
* | | | * .PARAMS G=0.00224 mu=103.2 C=3.401 y=1.264
.SUBCKT 12AT7 A G K .PARAMS Gg=618U Cg=9.908 xi=1.314 ig0=80.25N
*~~ parameter ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ *~~ subfunctions ~~~~~~~~~~~~~~~~~~~~~~~~~~~
.PARAMS Cgk=2.3P Cag=1.6P Cak=.2P F=231.4 .FUNC LX(va,vg) {LOG(1+EXP(C*(va/mu+vg)))/C}
.PARAMS G=0.005071 y=1.277 mu=89.23 .FUNC LX2(vg) {LOG(1+EXP(Cg*vg))/Cg}
.PARAMS d1=0.8592 Cg=101.3 r=1.623 *~~ grid current Ig ~~~~~~~~~~~~~~~~~~~~~~~~
*~~ subfunctions ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ GIg G K VALUE={Gg*PWR(LX2{V(G,K)},xi)+ig0}
.FUNC LX(va,vg) {LOG(1+EXP(F/(1+va)* *~~ anode current Ia ~~~~~~~~~~~~~~~~~~~~~~~
+(1/mu*va+vg)))*(1+va)/F} GIa A K VALUE={G*PWR(LX{V(A,K),V(G,K)},y)-
.FUNC LX2(va,vg) {LOG(1+EXP(Cg*vg/va))/Cg} +Gg*PWR(LX2{V(G,K)},xi)-ig0}
.FUNC D(va,vg) {d1*PWR(max(0,vg)/va,2)} C1 G K {Cgk}
*~~ grid current Ig ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ C2 G A {Cag}
GIg G K VALUE={G*PWR(LX{V(A,K),V(G,K)},y) * C3 A K {Cak}
+D{V(A,K),V(G,K)}/PWR(LX2{V(A,K),V(G,K)},r)} .ends
*~~ anode current Ia ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
GIa A K VALUE={G*PWR(LX{V(A,K),V(G,K)},y) * Fig. 6: LTspice subcircuit for a 12AX7.
+(1-(D{V(A,K),V(G,K)}/PWR(LX2{V(A,K),V(G,K)},r)))}
C1 G K {Cgk} current can be changed to
C2 G A {Cag}
C3 A K {Cak}  1 ξ
  
.ends Ig = Gg log 1 + exp Cg Vg + Ig0 , (20)
Cg
Fig. 5: LTspice subcircuit for a 12AT7.
with the perveance Gg , exponent ξ and the adap-
Now we apply the log/exp combination to the latter tion factor Cg . A constant Ig0 may be helpful during
part to allow for a smooth transition, giving the fitting procedure. Not surprisingly this formula
!ζ shows similarities to the diode equation (2). Assum-
 2   1 ing that Ik has only a marginal twist regarding Va
Vg 
Vg 
∆=δ log 1 + exp Cg Va we can furthermore approximate the adaption factor
Va Cg
in equation (15) to be constant, Ca = f (Va ) → C,
(17)
where the exponent ζ will be close to − 32 . Examining   !γ

Va
 1
the surface represented by this formula we find that Ik = G log 1 + exp C µ + Vg (21)
C
for very small values of Vg the slope is dominated by
the squared term and then approaches the slope of The anode current is again calculated by equa-
equation (8). This resembles the measured current tion (19). This approach was already presented in
ratio well. Note, that the limitation Vg > 0 still has [2] where more details are given. The SPICE code
to be respected. Anode and grid current are then is shown in Figure 6.
calculated by
5.5. Fitting
Ig = Ik · ∆ (18) The presented equations, namely equations (15)
and (18) or equations (21) and (20), contain sev-
Ia = Ik − Ig . (19) eral free parameters which were adapted to reference
measurements. This was done using the surface fit-
These equations still reveal the well-known physics ting toolbox in Matlab. Example parameter sets
but with some degree of freedom. Furthermore, they are collected in Tables 1 and 2. The variance in pa-
are continuously differentiable and have no discon- rameters reflects the electrical variance of the tubes.
tinuities. Figure 6 gives the corresponding SPICE
code. 6. RESULTS
5.4. Alternative Approach
In some cases Ig has only low dependency on Va . 6.1. Comparison with other Models
This is especially true for the popular 12AX7 high-µ The proposed model has to be compared to the ex-
triode. To keep it simple, the expression for the grid isting approaches. Figure 7 shows the family charac-

AES 131st Convention, New York, USA, 2011 October 20–23


Page 6 of 10
Dempwolf et al. Individual Triode Model

No Label Condition G µ γ F Cg δ ζ
1 Valvo (GER) old, used 2.85E-3 92.67 1.065 189.1 55.16 0.889 -1.534
2 Telefunken (GER) old, used 5.07E-3 89.23 1.227 231.4 101.30 0.859 -1,623
3 Ei (CZ) new 2.69E-3 60.53 1.237 256.9 236.6 0.698 -1,542

No Label Condition G µ γ F Cg δ ζ
1 TAD new 1.77E-3 21.18 1.319 103.1 66.27 0.411 -1.517
2 EGRO old, used 1.52E-3 21.92 1.300 69.0 143.30 0.670 -1.441
3 JJ (CZ) new 8.80E-4 18.49 1.528 93.4 54.05 0.747 -1.388

Table 1: Fitting parameters for the 12AT7 and 12AU7 triodes.

No Label Condition G µ γ C Gg ξ Cg Ig0


1 RSD (GER) old, used 2.14E-3 100.8 1.303 3.04 606E-6 1.354 13.9 185n
2 Neuhaus (GDR) NOS 1.42E-3 111.5 1.341 3.14 439E-6 1.553 27.0 305n
3 JJ (CZ) new 2.49E-3 98.3 1.132 2.91 105E-6 1.047 6.1 309n

Table 2: Fitting parameters for the 12AX7 triodes (alternative approach).

teristics of the new model (black) and the measure- Vb


ments of one 12AT7 (circles). The dotted curves Ra Cout
show the characteristics of Koren’s model with the
parameters suggested by Konar [11, 12]. The curves Rg
have generally the same progression but there are Va
strong differences found for the grid current.
Vg
Vin
This observation is not surprising, because the new Rk Ck
model was individually fitted to the measurement
data, so it should align better than a general ap-
proach. More meaningful is the comparison to other Fig. 8: Common-cathode amplifier stage.
individual models. We decided to examine the one
proposed by Cohen and Helie [14]. To achieve an eq-
uitable comparison their model was fitted in a simi- of the chosen amplifier stage. This common-cathode
lar manner and to the same measurement data. The amplifier (CCA) is a simple but representative cir-
results are displayed in Figure 7 as dashed curves. cuit which can be found in almost all guitar pream-
The improvement due to the individualization is plifiers. We skip a detailed circuit analysis and are
clearly visible and the curves align better with the content with the information, that this is a high-gain
measurements. However, in this comparison the new inverting amplifier. The bias point is a bit negative,
model gives better fitting results, because it can cap- but for large input amplitudes the grid will be tem-
ture the grid current if Ig = f (Va ). porarily positive. The same tubes as those used in
the fitting procedure were operated in this test cir-
Deviations remain for Vg ≈0 V where already a small cuit and the responses to various excitation signals
grid current arises. This is not covered by the model. were measured.
The results for the alternative approach are dis-
cussed in [2]. 6.2.1. Time Signal Waveforms
Example waveforms of voltage Vg and current Ig for
6.2. Simulation of Common Cathode Amplifier a sinusoidal excitation with different amplitudes and
A test circuit was prototyped to check to what extent frequencies are shown in Figure 9. A four-channel
the model replicate real triodes when used within a oscilloscope with 10 GΩ probes was used and the
circuit simulation. Figure 8 depicts the schematic amount of grid current was estimated by the voltage

AES 131st Convention, New York, USA, 2011 October 20–23


Page 7 of 10
Dempwolf et al. Individual Triode Model

14

anode current in mA →
12
10
8
6
4
2
0
-4 -2 0 2 -4 -2 0 2 20 60 100 140 180 220 260 300
grid voltage in V → grid voltage in V → anode voltage in V →

grid current in mA →
Va = 40 V Va = 80 V
6 Vg = −2 V, −1 V, 0 V, 1 V and 2 V
Va = 120 V Va = 160 V

0
-4 -2 0 2 -4 -2 0 2 20 60 100 140 180 220 260 300
grid voltage in V → grid voltage in V → anode voltage in V →

Fig. 7: Comparison of different triode models as grid family (left) and plate family characteristics (right).
Shown are Koren’s approach (dotted), Cohen and Helie (dashed) and the new model (solid). The discrete
measurement points are marked with ◦.

drop across Rg . As the plots illustrate the waveforms given in Figure 10.
are replicated satisfactorily.
7. CONCLUSION
6.2.2. Distortion Analysis A new triode description was presented, featuring a
Time signal comparisons give only limited informa- good replication of the grid current and physically-
tion on how good a simulation performs. A signif- motivated formulations. The equations are mainly
icant measure is the harmonic content of the out- based on the Langmuir-Child’s and the current-
put. The exponential sweep technique [16] was used division law. Free parameters were used to perform
to compute the introduced distortion both for ref- an individual fitting to measurement data of prac-
erence system and simulation of the CCA circuit. tical triodes. It was shown, that the equations are
Figure 10 shows the harmonic distortions of the dis- able to characterize the properties of real tubes in
cussed circuit. The installed tube in this plot was good accordance. Furthermore, an alternative ver-
an old Telefunken 12AT7. As the plots illustrate, sion for linear triodes was proposed. Based on the
the simulation using an individual model achieves a descriptions, SPICE models were implemented.
high similarity to the measurements.
To prove the suitability for the simulation of audio
The same procedure was repeated with a NOS circuits, the models were embedded in a typical tube
“Neuhaus” 12AX7. Reference measurements and preamplifier. Reference measurements and simula-
simulation results using the alternative approach are tion results for time and frequency domain were pre-

AES 131st Convention, New York, USA, 2011 October 20–23


Page 8 of 10
Dempwolf et al. Individual Triode Model
Ig in µA →

Ig
6

Ig

Ig
3 50 150 100
0 25 100
-3 50 50
0
-6 -25 0 0
Vg in V →

Ig

Ig

Ig
0 0 0
-1 -2.5 -5 -4
-2 -5
-10 -8
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4 0 1 2 3 4 5
time in ms → time in ms → time in ms → time in ms →

Fig. 9: Ig (top) and Vg (bottom) for the CCA. From left to right: 4 kHz/1 V, 4 kHz/4 V, 4 kHz/8 V, and
500 Hz/8 V with measurements (blue) and simulations (dashed black). The installed tube was a RSD 12AX7.

sented and showed a good match. [9] M. Zollner, Physik der Elektrogitarre,
(preprint), Regensburg, 2010.
8. REFERENCES [10] A. Potchinkov, Simulation von
Röhrenverstärkern mit SPICE, Vieweg +
[1] J. Pakarinen and D. T. Yeh, “A review of digi-
Teubner, 1st edition, 2009.
tal techniques for modeling vacuum-tube guitar
amplifiers,” Computer Music Journal, vol. 33, [11] N. Koren, “Improved vt models for spice simu-
no. 2, pp. 85–100, Summer 2009. lations,” Glass Audio, vol. 5, pp. 18–27, 1996.
[2] K. Dempwolf and U. Zölzer, “A physically- [12] M. Konar, “Vacuum tube parameter iden-
motivated triode model for circuit simulations,” tification using computer methods,” [online],
in Proc. 14th Int. Conf. Digital Audio Effects http://www.birotechnology.com, 1998.
(DAFx-11), Paris, Sept. 19–23 2011.
[13] G. C. Cardarilli, M. Re, and L. Di Carlo, “Im-
[3] H. Reich, Principles of Electron Tubes, McGraw proved large-signal model for vacuum triodes,”
Hill, 1st edition, 1941. in IEEE International Symposium on Circuits
and Systems (ISCAS), 2009, pp. 3006–3009.
[4] H. Barkhausen, Lehrbuch der Elektronenröhren
und ihrer technischen Anwendungen, Verlag S. [14] I. Cohen and T. Helie, “Measures and parame-
Hirzel, Leipzig, 1945. ter estimation of triodes, for the real-time sim-
ulation of a multi-stage guitar preamplifier,” in
[5] K. Spangenberg, Vacuum Tubes, McGraw Hill, Proc. 129th AES Convention, San Francisco,
1942. USA, Nov 4-7 2010, number 8219.
[6] M. Blencowe, Designing Tube Preamps for Gui- [15] H. Kniekamp, “Die Abweichungen der
tar and Bass, Blencowe, 2009. Verstärkerröhrenkennlinien vom e3/2 -Gesetz,”
[7] R. Aiken, “What is blocking distortion?,” Telegraphen- und Fernsprechtechnik, vol. 20, no.
[online], www.aikenamps.com/ BlockingDistor- 3, pp. 71–76, 1931.
tion.html, 1999. [16] A. Farina, “Simultaneous measurement of im-
pulse response and distortion with a swept-sine
[8] F. Tank, “Zur Kenntnis der Vorgänge in
technique,” in 108th AES Convention, Paris,
Elektronenröhren,” Jahrbuch drahtl. Telegr. u.
France, Feb. 19-24 2000.
Teleph., vol. 20, pp. 80, 1922.

AES 131st Convention, New York, USA, 2011 October 20–23


Page 9 of 10
Dempwolf et al. Individual Triode Model

|H| in dB →

|H| in dB →
0 0

-20 -20

-40 -40
|H| in dB →

|H| in dB →
0 0

-20 -20

-40 -40

100 1k 10k 100 1k 10k


frequency in Hz → frequency in Hz →

Fig. 10: Harmonic distortion analysis of the CCA driven with an old Telefunken 12AT7 triode. Measure-
ments (left) and simulations (right) for amplitudes 1 V (top) and 28 V (bottom). Shown are the fundamental
(black), odd (dashed gray) and even order harmonic responses (dotted) up to 4th order.
|H| in dB →

|H| in dB →

0 0

-20 -20

-40 -40
|H| in dB →

|H| in dB →

0 0

-20 -20

-40 -40

100 1k 10k 100 1k 10k


frequency in Hz → frequency in Hz →

Fig. 11: Harmonic distortion analysis of the CCA driven with a NOS Neuhaus 12AX7 triode. Measurements
(left) and simulations (right) for amplitudes 1 V (top) and 28 V (bottom). Shown are the fundamental (black),
odd (dashed gray) and even order harmonic responses (dotted) up to 4th order.

AES 131st Convention, New York, USA, 2011 October 20–23


Page 10 of 10

Вам также может понравиться