Proc. of the 16th Int.

Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

A DIGITAL MODEL OF THE BUCHLA LOWPASS-GATE

Julian Parker and Stefano D’Angelo∗

Dept. of Signal Processing & Acoustics,
Aalto University,
Espoo, Finland
[email protected]

ABSTRACT The recent explosion of interest in analog modular synthesis,

and particularly the work of ’west coast’ designers such as Don
In recent years there has been an increasing amount of interest in Buchla and Serge Tcherepnin, has lead to a great resurgence in
the style of synthesis implemented by Don Buchla in his instru- interest in the lowpass gate. Contemporary manufacturers such as
ment designs from the early 1960s until the present. A key part Make Noise Music and Doepfer manufacture their own takes on
of the Buchla synthesizer and its characteristic quality is the ’low- the circuit, and many designs are available in the public domain
pass gate’ filter and the acoustic-like plucked sounds that it pro- for those inclined to build themselves. Hence, it is desirable to
vides. In this work we examine the circuit of the low-pass gate, attempt to create a ’virtual analog’ (VA) version of the filter for
both its audio and control portions. We propose a number of digi- use in computer-music situations. VA research has been a popular
tal models of these circuits, as well as a model of the photoresistive topic in recent years, and has examined many of the tools of analog
optoisolator or ’vactrol’ used within them. In the case of the audio synthesis – particularly filters [3, 4, 5, 6, 7, 8, 9, 10] and effects
path of the device, we pay particular attention to maintaining desir- processing [11, 12, 13].
able behavior under time-variation of its parameters. The resulting In this paper we examine a simplified version of the Buchla
digital model retains much of the interesting character of the ana- low-pass gate, constructed from Buchla’s own Model 292 designs1 ,
log system, and is computationally cheap enough to use within a and derivative public domain versions of the circuit2 . This version
standard computer-music setup. of the circuit is meant to be a ’canonical’ version – one that con-
tains all the important behaviours and features of the system, but
1. INTRODUCTION without the many tweaks and alterations that were added for spe-
cific applications or to compensate for certain misbehaving com-
In parallel to Bob Moog’s early work on the analog synthesiser ponents.
[1], work towards a modular analog synthesis system was also In Sec. 2 we describe the audio path circuitry of the filter, and
proceeding on the other side of the USA in California. Driven develop two discretizations of it. The first is a simple approach
by the composers like Morton Subotnick and the San Francisco based on the discretisation of the transfer function. The second
Tape Music Centre, Don Buchla was developing the ’Buchla Mu- takes a more complex approach, which preserves the topology
sic Box’ [2], a modular analog synthesiser with a very different fo- of the system and behaves significantly better under time-varying
cus than Moog’s system. Whereas synthesis in the Moog paradigm conditions. We also propose a method for extending this model
was based primarily on the shaping of harmonically rich but static to the non-linear case. In Sec. 3 we describe the control circuitry
waveforms with a resonant low-pass filter – so-called ’subtractive’ of the device and how it can be modelled digitally, including the
synthesis, the synthesis used in Buchla’s instruments had a con- behaviour of the vactrol. In Sec. 4 we briefly describe an imple-
trasting approach. In the Buchla system, the primary focus was on mentation of the complete model. In Sec. 5, we conclude.
varying timbre at the oscillator level via nonlinear wave-shaping or
frequency modulation techniques. These waveforms would then 2. AUDIO PATH
be processed through the ’low-pass gate’ (LPG), a special kind of
filter invented by Buchla that used photoresistive opto-isolators or Fig. 1 shows a schematic of the simplified version of the Buchla
’vactrols’ as its voltage controlled resistance element. 292 audio path. The filter centre frequency is controlled by vari-
The LPG was a very different tool to the Moog-style filter. ation of Rf , which is the resistive part of a photoresistive opto-
Firstly, it exhibited a fairly gentle roll-off of high frequencies. Sec- isolator or ’vactrol’. The specific vactrol used in this circuit is
ondly, when turned to low cutoff values it began to attenuate the generally the Perkin Elmer VTL5C3 or VTL5C3/2 [14]. The orig-
sound as well as filter it. Lastly, the vactrol used for control exhib- inal implementations of the circuit contain a 3-pole switch. For
ited a very lazy response to modulation – it would track increas- simplicity, this is represented here as two one-pole switches. This
ing values very quickly, but decreasing values much more slowly. switch was used to switch between the three modes of the device
This resulted in a very musical behaviour. The filter’s control in- – ’Both’, ’VCA’ and ’Lowpass’.
put could be hit with an impulse or short envelope, and the filter In ’Both’ mode, the circuit consists of a passive two-pole filter
would open very quickly before decaying slowly away to silence. network. This corresponds to the circuit shown in Fig. 1 with both
The sound produced was very much like that of a struck physical
1 http://rubidium.dyndns.org/~magnus/synths/
object. This property was used extensively in, for example, the
works of Morton Subotnick. companies/buchla/
2 http://modularsynthesis.com/nrm/lopass/lopass.
∗ This work was supported by GETA htm

DAFX-1
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

The component values used to model the circuit in the three

Vactrol ’VCA’ different modes are given in Table 1. The magnitude responses of
If the circuit in each mode, as Rf is varied, are given in Fig. 2.

Table 1: Values of fixed components in different modes of circuit

C1 R↵
from Fig. 1.
-
Vout
Vin + Component Both VCA Lowpass
V+
C3 C1 1nF 1nF 1nF
Vx Vfb + C2 220pF 220pF 220pF
- C3 0 0 4.7nF
’Lowpass’
C2 Rα 5 MΩ 5 kΩ 5 MΩ

Figure 1: Simplified schematic of the Buchla 292 LPG audio path. 2.1. Continuous Model
We can construct the differential equations governing the circuit
switches disengaged. In this mode the system acts as a two-pole given in Fig. 1 by applying Kirchoff’s Current Law (KCL) at the
lowpass filter, with the cutoff frequency controlled by Rf . The nodes. Firstly, assuming ideal op-amp behavior, we can write the
poles of the filter are not-coincident, and so the slope of the filter equations at nodes Vout and Vfb as:
is closer to -6dB/oct than -12db/oct. When Rf gets close to the Vout = V+ (1)
value of Rα , a potential divider is formed and the overall gain of
the system begins to drop. Vfb = aVout (2)
In ’VCA’ mode, an additional low-value resistor is added in where a is a gain coefficient dependent on the configuration of the
parallel to Rα . This is equivalent to only the ’VCA’ switch in resistors in the feedback amplifier.
our schematic being engaged. This is clearly equivalent to simply We then write down the KCL equations at nodes Vx and V+
lowering the value of Rα (and it will be modeled this way during respectively:
the rest of this work). The effect of this change in resistance is
Vin − Vx V+ − Vx dVx d
that the poles of the filter are shifted to a higher frequency and the + − C2 + C3 (Vfb − Vx ) = 0 (3)
potential-divider effect of the combination of Rf and Rα begins at Rf Rf dt dt
lower values of Rf . The result is that variation of Rf now provides Vx − V+ V+ dV+
− − C1 = 0 (4)
reasonably clean attenuation of the input signal, hence the name Rf Rα dt
’VCA’ mode. In the real circuits, the ’VCA’ switch also changes
By substituting in the expressions for Vout and Vfb , we obtain the
the input op-amp configuration (not shown) so that it has a gain of
governing differential equations of the system, in terms of Vx and
4-5dB instead of unity, presumably to normalize loudness between
Vout :
the modes.
In ’Lowpass’ mode, an extra feedback path containing a ca- dVx d
= b1 Vin + b2 Vx + b3 Vout + b4 (d1 Vout + d2 Vx ) (5)
pacitor C3 is switched into the circuit. This changes the topology dt dt
of the circuit to closely resemble a Sallen-Key [15] filter. Com- dVout
= a1 Vx + a2 Vout (6)
pared to a standard S-K filter, this circuit contains an extra ca- dt
pacitor C2 and a shorting resistor Rα . The resulting frequency where
response is steeper than that of ’Both’ mode, with a small resonant
1 1 1 1
bump at the cutoff frequency. We have also added a feedback am- a1 = , a2 = − ( + ), d1 = a, d2 = −1,
plifier to this path in order to provide control over the level of this C1 R f C 1 Rf Rα
resonant peak. This is not present in the original Buchla versions 1 2 1 C3
b1 = , b2 = − , b3 = , b4 = .
of the circuit, but it is present in more recent derivative circuits. C2 R f C 2 Rf C2 R f C2
This mode can be simulated in a model by setting the value of ca-
pacitor C3 . Using a value of C3 = 0 disengages the feedback 2.1.1. Transfer function
path.
By substituting Eq. (6) into (5) and transforming to the Laplace
domain, we can construct the transfer function of the filter:
Both VCA Lowpass
1
0 HLPG (s) = (7)
α1 + α2 s + α3 s2
Mag./dB

−20
−40
where
−60 2Rf
0.1 12 10 0.1 12 10 0.1 12 10 α1 = 1 + (8)
Frequency/kHz Frequency/kHz Frequency/kHz Rα
Rf
α2 = Rf (2C1 + C2 − C3 (a − 1) + (C2 + C3 ) ) (9)
Figure 2: Magnitude response of modes of LPG circuit, for values Rα
of Rf between 1kΩ and 1M Ω. α3 = Rf2 C1 (C2 + C3 ) (10)

DAFX-2
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

From the transfer function, we can derive an expression giv- when we simplified Eqs. (5) and (6) to the transfer function (7),
ing the feedback path amplification, a, necessary to produce the we implicitly discarded the true states of the system (the capacitor
maximum resonant peak gain. We do this by evaluating |HLPG | on voltages), and constructed new states based on the output of the
the imaginary axis, finding the maximum by taking the derivative entire system at previous time-steps. This change of states also
w.r.t. frequency, and then evaluating the function at this point. We obviously includes a change to the coefficients used to update the
can then take the derivative of this function w.r.t. a, and solve the states (i.e. the state transition matrix), which govern the stabil-
resulting expression to find the maximum w.r.t. a: ity of the system. In time-invariant conditions, this change does
not effect the system’s response. However when parameters are
2C1 Rα + (C2 + C3 )(Rα + Rf ) varied, particularly when they are varied quickly, the behaviour of
amax = (11)
C 3 Rα the system can be wildly different. It’s also worth noting that the
original system had three states (one for each capacitor), whereas
When a > amax , the poles of the circuit cross the imaginary
the system in transfer-function form has only two states. This is
axis and the system becomes unstable.
because one of the states of the original system, the voltage associ-
By evaluating HLPG at DC, we can see that the DC gain is ated with capacitor C3 , is dependent purely on the other two states
given by: and hence is degenerate. Collapsing this state into the other states
Rα
HLPG (0) = (12) is fine in the time-invariant case, but not in the time-variant case.
Rα + 2Rf
This expression highlights the potential-divider action that appears 2.2.2. Topology-preserving model
when the value of Rf nears the value of Rα . It can also be used
to normalise the loudness between modes by calculating the dif- An alternative to discretising the system via its transfer function is
ference in DC gain between the modes when Rf is at its minimum to attempt to build a digital model that preserves the topology of
value (which depends on the physical properties of the vactrol). the system described in Eqs. (5) and (6) by simply replacing any
continuous blocks with their discrete equivalents. Fig. 3 shows this
2.2. Discretization topology in domain-agnostic form. Preserving the topology of the
system preserves its states, and hence the resulting system should
A variety of approaches can be taken to the discretization of the be well-behaved in time-variant conditions if the continuous-time
system given in Eqs. (5) and (6). In this section we examine two system was also well behaved. The structure of the system high-
such approaches. lighted by this representation is interesting to note. The system
appears to contain two lowpass filters (consisting of capacitors
2.2.1. Direct form model C1 and C2 ), surrounded by two feedback loops. One feedback
loop contains only a straight multiplier, and represents the lack of
The simplest and most traditional approach to discretization would buffering between and hence interdependence of the two low-pass
be to apply the bilinear transform to transfer function (7), thus filters. The other feedback loop contains a differentiator block, and
obtaining a z-domain transfer function which can be implemented represents the action of the capacitor C3 .
in one of any number of standard digital filter topologies. The
resulting digital transfer function is of the form: d1

βd1 + βd2 z −1 βd3 z −2

 
z−1
HLPG 2Fs = HLPG (z) = (13) a2
z+1 αd1 + αd2 z −1 αd3 z −2
a1
i.e. a two-pole, two-zero digital filter.
Any of the direct form topologies could be used to implement
this transfer function. For the purpose of this work we use direct
form II transposed, which exhibits advantageous numerical prop-
erties in many situations [16]. To mitigate the frequency-warping
effects of the bilinear transform, we recommend oversampling the Figure 3: Block diagram of topology of Eqs. (5) and (6)
filter by a factor of 2. This warping effect is often also addressed
by pre-warping the frequency control of the filter. However, in this The primary problem when discretising a filter via this ap-
case the process of pre-warping is significantly complicated by the proach is the delay-free loops that are formed if the discrete in-
fact that there is no simple cutoff frequency parameter. The posi- tegrator blocks have a delay-free path through them. One solu-
tion of the poles of the filter is a complicated function of Rf , Rα tion to this problem is to use purely explicit integrators, generally
and the capacitor values. As such, oversampling is a better option. Forward-Euler. This is the approach taken in the delta-operator
Magnitude responses for the discretised filter implemented via approach to filter design [17]. However, explicit integrators gener-
this method are shown in Fig. 5. The match with the continuous- ally have poor accuracy and very poor stability unless the system is
time responses given in Fig. 2 is reasonably good for ’VCA’ and substantially oversampled. This is because they map stable poles
’Lowpass’ modes. in the s-plane to unstable poles in the z-plane.
The real drawback of this implementation is encountered when A common approach taken in the virtual analog field has been
modulating the filter. Modulation appears to produce transients to choose more robust implicit integrators such as Backward-Euler
within the structure, which cause the output of the filter to grow. or Trapezoidal, and then insert a z −1 block into any feedback paths
Long-term modulation can cause the filter output to diverge to in- which contain a delay-free loop [3]. This additional delay ele-
finity. This is obviously not desirable for a musical filter which ment usually requires some compensation in order to maintain the
is intended to be modulated. The reason for this behaviour is that correct behaviour of the system. Backward-Euler integrators have

DAFX-3
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

been used in this context [4], but they possess a similar problem the system with any standard technique to find an expression for
to Forward-Euler integrators in that they change the Q-factor of each of the outputs in terms of only the states of the system and the
poles when mapping from the s-plane to the z-plane. In this case, input. With the outputs calculated, we update the states and move
unstable poles are mapped to stable poles, and consequently, whilst on to the next sample. This process is equivalent to re-arranging
stability is maintained, the relationship between the strength of the the system to eliminate the delay-free loops, whilst still maintain-
resonant peak of the filter and cutoff frequency which is present ing all the individual states.
in the continuous system is lost. This is an important perceptual Alternatively, a more efficient approach is to solve the linear
quality in defining the sound of a synthesizer filter, and hence it is system for only one of the outputs and then re-arrange the other
desirable to maintain. Recent work has applied Trapezoidal inte- two output equations to allow them to be calculated consecutively
grators in similar structures, which maintain the Q-factor of poles from the other outputs in addition to the states and input. For ex-
during the s-plane to z-plane mapping [18]. ample, we could write:
Another approach has been to instead solve the delay-free loops
a2 b4 sd + 2Fs (b3 − 2b4 d1 Fs )so − 2a2 Fs sx − a2 b1 yi
by iteration [7, 8, 5, 9], although this is only strictly necessary if yx =
the loop contains a non-linearity. More recently, methods have a1 b3 − a2 b2 − 2a1 b4 d1 Fs + 2a2 b4 d2 Fs
been proposed that allow a simpler solution of the implicit prob- (20)
lem [19]. so + a1 /2Fs yx
yo = (21)
In this work, we apply a method similar to that of Zavalishin 1 − a2 /2Fs
[19]. Firstly, we construct the system by replacing all blocks in
yd = sd + 2Fs (d1 yo + d2 yx ) (22)
Fig. 3 with their digital equivalent. The integrator blocks and dif-
ferentiator block are implemented using the trapezoidal rule, as To improve computational efficiency, many other simplifications
this is the equivalent of the bilinear transform and possesses the of these equations can be performed – including substituting back
same desirable qualities. The trapezoidal integrators and differen- in the original component parameters in an attempt to find a more
tiator are implemented in direct form II transposed (DF2T). Their efficient form. This form is merely presented for concise explana-
block diagrams are given in Fig. 4. Addition blocks and multipliers tory purposes.
are trivially replaced with their digital equivalent. Magnitude responses of the digital filter produced are shown
in Fig. 5. Again, the filter is oversampled by a factor of 2 to reduce
the frequency warping introduced by the trapezoidal integrators
and differentiator. A close correlation with the magnitude response
1 1
of the continuous system from Fig. 2 is seen.
z z
Compared to the DFT2 model, this version performs excel-
lently in terms of time-varying behaviour and stability. The filter
1 Z
d appears to be stable under all physically reasonable parameter val-
ues and under any rate of modulation, without any oversampling.
dt
It becomes unstable when the feedback amplitude a exceeds the
value given by Eq. 11, but this is also the case in the real system.
Figure 4: Block diagrams of direct form II transposed (DF2T) in- Modulation with a smooth signal such as a sinusoid does not pro-
tegrator and differentiator elements. duce any unwanted transients or increase in volume, and sounds
clean.
With these digital blocks inserted, we can write expressions
for the outputs of all of the state-containing blocks (the differen- Both VCA Lowpass
tiator and integrators) in terms of their own state and their inputs. 0
Mag./dB

If we denote the outputs and states of the blocks as yd and sd for −20
the differentiator block, yo and so for the Vout integrator and yx −40
and sx for the Vx integrator, and yi for Vin , we have: −60
0
Mag./dB

yd = sd + 2Fs (d1 yo + d2 yx ) (14) −20

1 −40
yx = sx + (b1 yi + b2 yx + b3 yo + b4 yd ) (15) −60
2Fs 0.1 1 2 10 0.1 1 2 10 0.1 1 2 10
1 Frequency/kHz Frequency/kHz Frequency/kHz
y o = so + (a1 yx + a2 yo ) (16)
2Fs
Figure 5: Magnitude response of modes of digital models of LPG
and circuit, for values of Rf between 1kΩ and 1M Ω. Top: DF2T
sd = −sd − 4Fs (d1 yo + d2 yx ) (17) model. Bottom: Topology-preserving model.
1 2.2.3. Limiting or decoupling resonance
sx = sx + (b1 yi + b2 yx + b3 yo + b4 yd ) (18)
Fs
1 When used in lowpass mode the LPG circuit exhibits an uneven
so = so + (a1 yx + a2 yo ). (19) resonance response with respect to Rf , as can be inferred from
Fs
Eq. (11). In fact, the maximum feedback gain possible before the
The first set of equations can be treated as a system of linear system becomes unstable also varies depending on Rf . This be-
equations, with the outputs, yd , yx and yo , as unknowns and the haviour is naturally also present in the digital models. We can ame-
states and input, sd , sx , so and x as constants. We can then solve liorate the effects of this behaviour in a number of ways. Firstly, if

DAFX-4
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

we only want to make sure that the filter remains stable no matter form of the equation for yd , given by Eq. (14) in the linear case,
the selection of a, we can use (11) as a dynamic upper limit for would be:
the allowable possible values of a. Alternatively, the expression
can be used to decouple the resonance control from Rf and Rα by yd = sd +2Fs (d1 [tanh(xo )+(x−xo )(1−tanh2 (xo ))]+d2 yx ).
defining a new parameter 0 < anorm ≤ 1. We can then use this (25)
value and amax to calculate the actual feedback gain used as: Similarly, the equation for the update of sd , given by Eq. (17) in
the linear case, is modified to:
a = anorm amax . (23)
sd = −sd −4Fs (d1 [tanh(xo )+(x−xo )(1−tanh2 (xo ))]+d2 yx ).
This makes the filter more controllable, but it means that the res- (26)
onance no longer strictly behaves as it does in the real system. The system can then be solved as in the linear case, using these
Whether this is appropriate depends on the application of the model. new expressions and also Eqs. (15), (16), (18) and (19). The re-
Note that using this parameterisation, the filter can be made to self- sulting system looks much like that given in Eqs. (20)-(22), with
oscillate over its entire range by setting anorm = 1. the addition of a few new terms representing the non-linearity.
Fig. 6 shows the normalized magnitude spectrum of the first
2.3. Non-linearities 10 harmonics of this non-linear model, measured using a sine-
sweep technique [23]. The cutoff controlling resistance Rf is set
The models of the circuit presented above are completely linear. to 110kΩ, and the resonance feedback path gain a, is set to the
In reality, some non-linearity may be present in the photoresistive amax corresponding to this value of Rf . 4x oversampling was used.
parts of the vactrol and in the op-amps. In older versions of the cir-
cuit, a transistor buffer is used at the output instead of an op-amp
buffer, which exhibits much more significant non-linear behaviour. 0
In some recent derivative versions of this circuit, feedback limiting Mag./dB
diodes are used in the resonance feedback pathway. This configu- −50
ration would exhibit strongly non-linear behaviour similar to that
−100
seen in the filter of the Korg MS-20 [20]. 1 2 3 4
10 10 10 10
It is not possible to easily introduce non-linear behaviour into Freq./Hz
the direct-form based model described in Sec. 2.2.1. The topol-
ogy preserving model described in Sec. 2.2.2 is more suitable. Figure 6: Normalised magnitude spectra of first 10 harmonics of
However, the introduction of non-linearities complicates the so- the non-linear model, with Rf = 110kΩ and a = amax .
lution of the system of equations that is necessary to resolve the
delay-free loops, especially if the non-linearities are transcenden-
tal functions. In order to solve the system, it is necessary to make 3. CONTROL PATH
an approximation of the non-linearity which leaves the system of
equations in a solvable form. The simplest approximation is to in- The filter described in Sec. 2 takes as its main frequency control the
stantaneously linearise the system around the point of operation of two resistive parts of a photoresistive opto-isolator (vactrol) with
the non-linearity by calculating its tangent. For an arbitrary dif- value Rf . Since much of the interesting behaviour of this circuit
ferentiable function of some input x, this linear approximation is is in its response to control, the connection between the applied
given by: control voltage and the value of Rf must also be modelled.
A simplified version of the control circuitry of a Buchla 292-
f (x) ≈ f (xo ) + f 0 (xo ) · (x − xo ) style filter is given in Fig. 7, and the values of its components in
= f (xo ) − xo f 0 (xo ) + xf 0 (xo ) (24) Tab. 2 Modelling the relationship between applied control voltage
and vactrol resistance can be split into a number of stages. Firstly,
where xo is the operating point, and priming represents differenti- we have to model the control circuitry to produce the correct rela-
ation w.r.t. x. This can also be recognised as the first-order Taylor tionship between the CV input (and offset) and the current through
approximation of the function around the operating point. If we the LED part of the vactrol. Secondly, we have to model the map-
treat the operating point xo as a constant, we can solve the system ping between the LED current and the resistance of the resistive
in the same way as we did in the purely linear case. parts of the vactrol (Rf ). Lastly, we have to model the memory
The question of how to choose xo is a difficult one. The sim- effects of the vactrol, and the filtering of the control value that it
plest approach would be to approximate xo as the value of the produces.
input to the non-linearity at the previous time-step. This approxi-
mation is reasonable when the input to the non-linearity is slowly Vs
changing. If the input is changing quickly, for example if it con-
+
tains strong components near the Nyquist frequency, the error will Rf
CV O↵set -
be significant. This error could be reduced by oversampling, by
If
iterating, by constructing a more complex predictor for the oper-
ating point, or by using a higher-order approximation of the non- Rf
linearity.
Vactrol
As an example of applying this technique, we can consider the Vc
addition of a simple diode-limiter model, represented by a tanh()
function [21, 22], to the resonance feedback path (i.e. the path
connecting the second integrator to the differentiator). The new Figure 7: Simplified schematic of the Buchla 292 control path.

DAFX-5
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

Table 2: Values of fixed elements in circuit from Fig. 7. and G = 2 × 105 .

By equating currents through R9 and the vactrol LED, which
Comp. Value Comp. Value Comp. Value is hereby modeled by the Shockley ideal diode equation, we obtain
Cc 2nF R 1 + R2 10kΩ R3 150kΩ
R4 470kΩ R5 100kΩ R6 20kΩ V4 + G V3 +Ia (R 6 +R7 )
 V3 −V4

α
R7 33kΩ R8 4.7kΩ R9 470Ω = IS e nVT
−1 , (35)
R9
R10 10kΩ Vs 15V
where IS is the reverse saturation current, n is the emission coef-
ficient and VT is the thermal voltage. By fitting the data from the
3.1. Control Circuit datasheet [14] of the device we have extracted values IS = 2.7551
nA and n = 3.9696, and we have chosen the thermal voltage to
The control circuit essentially consists of a first-order shelving fil-
be VT = 26 mV at room temperature (≈ 300 K). V3 can be there-
ter applied to the control input signal Vc and of a subcircuit that
fore explicitly expressed in terms of the system inputs using the
uses the output from the filter to control the amount of current
Lambert W function [24], and further approximated as
flowing through the vactrol LED. The input filter both decreases
high frequencies in the input signal and adds an amount of offset GIa (R6 +R7 − 1 )
!
to it, controlled by the variable resistor modeled by R1 and R2 . α αβ Ia
V3 ≈ − nVT W e αnVT
− . (36)
The current-controlling circuit resembles a logarithmic amplifier G αβ
with the notable difference of the presence of a zener diode con-
nected to ground. When the zener diode operates in its breakdown region, it can
By means of nodal analysis it is immediate to verify that V1 , be modeled as a voltage source of value V3 = −VB in parallel
that is the voltage across R2 , is related to the voltage V2 at the to R8 . Following a similar reasoning as before, we still obtain an
negative input terminal of the operational amplifier by equation of the form of (35), which can be solved in this case for
V4 , leading to
R1
Vs + V
R3 2
V1 = , (27)
 G[I (R +R )−V ] 
a 6 7 B
R1 R1
1+ R2
+R 3
V4 ≈nVT W e αnVT

(37)
and that the current I4 flowing from the input filter towards the G
positive terminal of the control input can be described, in the Laplace − [Ia (R6 + R7 ) − VB ] .
α
domain, as
The resulting current If flowing through the vactrol LED can
V2 (s)−Vc
[1 + sCc (R4 + R5 )] R5
− Cc V2 (0) be then computed as
I4 (s) = . (28)
1 + sCc R4   G[I (R +R )−V ] 
 nVT W e a αnV 6
T
7 B
if V3 = −VB ,
It is also possible to approximate V2 in terms of the voltage V3 If = R9
(38)
across the zener diode and the filtered input signal as 
βV3 + Iαa otherwise,
V3 (s) + (R6 + R7 ) Ia (s)
V2 (s) ≈ , (29) while the current-to-resistance mapping can be modeled by fitting
α the data on the datasheet [14] using
where
A
Vb (s) V Rf = + B, (39)
Ia (s) = +  s , (30) If 1.4
R5 R3 1 + R1
R2
where A = 3.464 ΩA1.4 and B = 1136.212 Ω.
1 + sCc (R4 + R5 ) Since the computation of the Lambert W functions in (36) and
Vb (s) = Vc (s), (31)
1 + sCc R4 (38) can be problematic both because of potentially high computa-
 
1 1 tional cost and the possibility of their arguments being outside of
α = 1 + (R6 + R7 ) + . (32) the range of common numerical data types, we hereby introduce
R3 R5
valid and inexpensive approximations. Firstly, it is worth remem-
In this circuit, the zener diode can be interpreted as as a limiter bering that W (1) = Ω ≈ 0.567 and W (x) ≈ log(x) when x is
that forces V3 into the range delimited by its forward voltage drop sufficiently big. Then, it is possible to notice that the derivatives of
VF and its breakdown voltage −VB . A typical breakdown voltage the arguments of the exponentials in both equations with respect
isVB = 3.9 V, while we can ignore the value of VF since it should to Vb are always greater than 6 · 105 . These two considerations
never be reached in this circuit. Between the two values, the zener suggest that the Lambert W functions in (36) and (38) will have
diode has neglibile current passing through it, therefore it can be a shape, in terms of Vb , and hence Ia , that resembles a piecewise
seen as an open circuit. Under this assumption, the voltage V4 linear function being constantly 0 when the argument of the inner
between the vactrol LED and the resistor R9 , can be expressed by exponential is negative and turning very rapidly to a straight line
  when it becomes positive.
G Ia
V4 = V3 R9 β − + [R9 − G (R6 + R7 )] , (33) Therefore, W (ex ) can be approximated as a first-order poly-
α α nomial for x ≥ xi , 0 when x ≤ −xi and with a not-a-knot cubic
where spline otherwise
1
α
−1 1
β= − , (34) w(x) = k0 + k1 x + k2 x2 + k3 x3 . (40)
R6 + R7 R8

DAFX-6
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

In the case of (36), by using xi = 600, which corresponds to a 3.2. Vactrol Model
maximum variation in terms of Vb of circa ±0.01 V, we obtain
k0 = 1.468 · 102 , k1 = 4.920 · 10−1 , k2 = 4.167 · 10−4 and A vactrol is made from two components, an LED and a light-
k3 = 7.391 · 10−9 , while the linear part can be expressed by dependent resistor (LDR) combined together within a light-proof
wl (x) = x − kL = x − 6.386. This leads to a direct mapping housing. The LDR section is similar to those available as sepa-
between Ia , and thus Vb , and V3 that can be expressed as rate components, consisting of a region of semiconductor mate-
rial separating two conductors, which are in turn connected to the
 Ia
terminals of the component. High resistance semiconductors are
− αβ
 c. 1,
usually used, with the particular substance depending on the fre-
  
 GIa R6 +R7 − αβ1
α Ia
V3 = − G nVT w αnVT
− αβ c. 2, (41) quency of light which the component should respond to. The most


 α common semiconductor used for visible-light LDRs, and in vac-
kL G nVT − Ia (R6 + R7 ) c. 3,
trols, is cadmium sulphide. When photons of sufficient energy hit
where case 1 corresponds to the material, electrons are promoted from the valence band of the
Ia ≤ −xi G(R αnV T
+R − 1 )
, semiconductor to the free conduction band, decreasing the resis-
6 7 αβ tance of the material. The energy (and hence frequency) of light
case 2 to
needed to make the LDR respond depends on the band gap be-
−xi G(R αnV T
1 ) < Ia < xi G(R αnVT
1 ) , and
+R −6 7 αβ 6+R − 7 αβ tween the valence band and the conduction band, which varies be-
case 3 to Ia ≥ xi G(R αnVT
1 . tween semiconductors. Electrons from the conduction band are
6 +R7 − αβ )
constantly decaying back down to the valence band, therefore the
For (38) we chose to simply consider the Lambert W function
overall resistance depends on the balance between the rate of elec-
as a piecewise linear function that is constantly 0 when the expo-
tron promotion due to incident photons and the rate of electron
nential argument is negative and linearly growing when this is pos-
decay. The resistance therefore depends on the intensity of the
itive. Then, in order to better fit SPICE simulation results, we mul-
incident light.
tipy the W approximation in its linear region by a scaling factor
When voltage is applied across the LED portion of the vactrol,
γ = 0.001. Also, in order to guarantee continuity between zener
photons are emitted, which then hit the semiconductor part of the
breakdown and open-circuit regions, the open-circuit If expression
LDR and decrease its resistance. This process results in a useful
is added to the breakdown region approximation. In the end, the
component that acts much like an ideal voltage-controlled resis-
resulting value is constrained between minimum (If,min = 10µA)
tance if we consider only long time-periods. However, the prop-
and maximum limits (If,max = 40mA). If,max is set by the point
erties of the semiconductor in the LDR mean that the behaviour is
at which the op-amp would saturate, and can be varied to change
more complicated on short time-scales, particularly in response to
the control response. The resulting If approximation is therefore
transient variations of illumination (and hence voltage applied to
the LED). Notably, decreases in resistance due to increased illu-

 If,min c. 1,

βV + Ia
 mination happen more quickly than the increase in resistance pro-
3 c. 2,
Iˆf = γG[Ia (Rα6 +R7 )−VB ] Ia (42) duced by a reduction in illumination. A voltage impulse applied

 αR9
− βV B + α
c. 3, over the LED therefore produces a quick drop in resistance in the

If,max c. 4,

LDR, followed by a slower decay back to high resistance. This re-
sistance profile is remarkably similar to the amplitude envelope of
where case 1 corresponds to Ia ≤ α (If,min − βV3 ), a struck physical object, hence the special sound it produces when
case 2 to α (If,min − βV3 ) < Ia ≤ R6V+R B
7
, used as the variable resistance element within a synthesizer filter.
VB γGV B +αR 9 ( B β+If,max )
V Adapting full models of photoconductor transient dynamics
case 3 to R6 +R7 < Ia < γG(R6 +R7 )+R9
,
[25] for musical signal processing use is a significant task, and left
γGVB +αR9 (VB β+If,max )
and case 4 to Ia ≥ γG(R6 +R7 )+R9
. for future work. Instead, we take a heuristic approach to modelling
the particular vactrol used in the LPG. This vactrol is generally
the Perkin Elmer VTL5C3 or VTL5C3/2. We propose an ad-hoc
40
35 Model non-linear filter structure that will behave similarly to the vactrol
SPICE
30 with correct setting of parameter values. This structure is shown in
25 Fig. 9. It consists of a lowpass filter constructed using an integra-
If (mA)

20
15
tor. The cutoff frequency (or response time in this case) is switched
10 between two different values dependent on the sign of the deriva-
5 tive of the input signal. These values are chosen from the datasheet
0 [14] for the VTL5C3/2 as being approx 12ms in the positive-going
-10 -5 0 5 10
Vc (V)
direction and 250ms in the negative-going direction. This chosen
value is then modulated further by the current output value of the
Figure 8: Vc to If relations, produced by model and by SPICE. vactrol model, so that it responds quicker when at high values, as
also indicated on the datasheet.
Fig. 8 shows the form of the relationship between Vc to If , in
the case of the above described model, and in a SPICE simulation.
Values of R1 , R2 = 5kΩ and R6 = 10kΩ are used. The cor-
respondence between the modelled relationship and that derived
from SPICE is very close in the region below the zener breakdown
voltage VB . The fit is less good after this point, but still adequate
for the purpose. Figure 9: Block diagram showing structure of vactrol model.

DAFX-7
 Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-5, 2013
Proc. of the 16th Int. Conference on Digital Audio Effects (DAFx-13), Maynooth, Ireland, September 2-6, 2013

4. IMPLEMENTATION AND RESULTS [8] F. Fontana and M. Civolani, “Modeling of the EMS VCS3
voltage-controlled filter as a nonlinear filter network,” IEEE
The above described models of the audio and control circuitry of
Trans. Audio, Speech and Lang. Process., vol. 18, no. 4, pp.
the LPG were implemented in Max/MSP using the Gen~ exten-
760–772, May 2010.
sion. The controls available in the model are a CV input (which
operates on voltage values), a CV offset, a switch to engage or dis- [9] S. Zambon and F. Fontana, “Efficient polynomial implemen-
engage ’lowpass’ mode, and a resonance control. The ’VCA’ mode tation of the EMS VCS3 filter model,” in Proc. 14th Int.
is implemented by providing a control which varies Rα . This al- Conf. Digital Audio Effects (DAFx-11), Paris, France, Sep.
lows the response to be smoothly varied between filter-like and 2011, pp. 287–290.
VCA-like by the user. These implementations are available at the [10] J. Parker, “A simple digital model of the diode-based ring-
website associated with this paper 3 , along with audio demos. modulator,” in Proc. 14th Intl. Conf. Digital Audio Effects
The resulting model seems to retain much of the character and (DAFx-11), Paris, France, 2011.
dynamics of the original device. In lowpass mode, especially in the [11] S. Arnardottir, J. S. Abel, and J. O. Smith, “A digital model
non-linear version , the model makes an interesting filter for sub- of the Echoplex tape delay,” in Proc. 125th Audio Eng. Soc.
tractive synthesis, with a unique and quite aggressive character. Conv., paper no. 7649, San Francisco, CA, May 2008.
We encourage readers to try the available implementation them-
[12] V. Välimäki, J. Parker, and J. S. Abel, “Parametric spring
selves in order to evaluate the quality of the sound.
reverberation effect,” J. Audio Eng. Soc., vol. 58, no. 7/8, pp.
5. CONCLUSIONS 547–562, 2010.
[13] C. Raffel and J. O. Smith, “Practical modeling of bucket-
In this work, we have examined the structure of the Buchla low- brigade device circuits,” in Proc. 13th Int. Conf. Digital Au-
pass gate filter. A robust digital model of the audio part of the dio Effects (DAFx-10), Graz, Austria, Sep. 2010, pp. 50–56.
circuit is constructed by generating a digital filter structure with
the same topology as the analog system, and solving its delay-free [14] Perkin Elmer Optoelectronics, Photoconductive Cells and
feedback loops. The control portion of the circuit is modelled by Analog Optoisolators (Vactrols), 2001.
approximating its response and applying the output to heuristic [15] R. P. Sallen and E. L. Key, “A practical method of designing
models of the vactrol’s current to resistance mapping and memory RC active filters,” IRE Trans. on Circuit Theory, vol. 2, no.
effect. The resulting combined model was implemented for real- 1, pp. 74–85, 1955.
time use in Max/MSP and Gen~. [16] J. O. Smith, Introduction to digital filters: with audio appli-
cations, Julius Smith, 2007.
6. ACKNOWLEDGMENTS
[17] J. Kauraniemi, T. I. Laakso, I. Hartimo, and S. J. Ovaska,
Many thanks to Dr. Timothy Stinchcombe for some very useful “Delta operator realizations of direct-form IIR filters,” IEEE
correspondance about this circuit and its variations. Trans. on Circuits and Systems II: Analog and Digital Signal
Processing, vol. 45, no. 1, pp. 41–52, 1998.
7. REFERENCES
[18] S. D’Angelo and V. Välimäki, “An improved virtual ana-
[1] R. A. Moog, “A voltage-controlled low-pass high-pass filter log model of the moog ladder filter,” in Proc. IEEE Int.
for audio signal processing,” in 17th Convention of the Audio Conf. Acoustics Speech Signal Proc. (ICASSP), Vancouver,
Eng. Soc., New York, USA, Oct. 1965. BC, Canada, May 2013.
[2] T. J. Pinch and F. Trocco, Analog days: The invention and [19] V. Zavalishin, “Preserving the LTI system topology in s-to
impact of the Moog synthesizer, Harvard University Press, z-plane transforms,” Tech. Rep., Native Instruments, 2008.
2004. [20] T. E. Stinchcombe, “A Study of the Korg MS10 & MS20
[3] T. Stilson and J. O. Smith, “Analyzing the Moog VCF Filters,” 2006.
with considerations for digital implementation,” in Proc. Int. [21] D. T. Yeh, J. S. Abel, and J. O. Smith, “Simulation of the
Computer Music Conf., Hong Kong, Aug. 1996, pp. 398– diode limiter in guitar distortion circuits by numerical so-
401. lution of ordinary differential equations,” in Proc. of the
[4] A. Huovilainen, “Non-linear digital implementation of the 10th Int. Conf. on Digital Audio Effects (DAFx07), Bordeaux,
Moog ladder filter,” in Proc. 7th Intl. Conf. Digital Audio France, 2007, pp. 197–204.
Effects (DAFx-06), Naples, Italy, Oct. 2004, pp. 61–64. [22] A. Huovilainen, “Design of a scalable polyphony-MIDI syn-
[5] F. Fontana, “Preserving the structure of the Moog VCF in thesizer for a low cost DSP,” M.S. thesis, Aalto University,
the digital domain,” in Proc. Int. Computer Music Conf., 2010.
Copenhagen, Denmark, Aug. 2007, pp. 291–294. [23] A. Farina, “Simultaneous measurement of impulse response
[6] V. Välimäki, F. Fontana, J. O. Smith, and U. Zölzer, “In- and distortion with a swept-sine technique,” in Proc. AES
troduction to the special issue on virtual analog audio effects 108th Conv., paper no. 5093, Paris, France, Feb. 2000.
and musical instruments,” IEEE Trans. Audio, Speech and [24] R.C.D. de Paiva, S. D’Angelo, J. Pakarinen, and V. Välimäki,
Lang. Process., vol. 18, no. 4, pp. 713–714, May 2010. “Emulation of operational amplifiers and diodes in audio dis-
[7] M. Civolani and F. Fontana, “A nonlinear digital model of tortion circuits,” IEEE Trans. on Circuits and Systems II:
the EMS VCS3 voltage-controlled filter,” in Proc. 11th Int. Express Briefs, vol. 59, no. 10, pp. 688–692, 2012.
Conf. Digital Audio Effects, Espoo, Finland, Sep. 2008, pp. [25] A. E. Iverson and D. L. Smith, “Mathematical modeling of
35–42. photoconductor transient response,” IEEE Trans. on Electron
3 http://www.acoustics.hut.fi/go/dafx13-lpg/ Devices, vol. 34, no. 10, pp. 2098–2107, 1987.

DAFX-8