Vdocuments - MX BBC Ring Modulator
Vdocuments - MX BBC Ring Modulator
Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Proc. of the 14th International Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Julian Parker
Department of Signal Processing and Acoustics,
Aalto University,
Espoo, Finland
[email protected]
ABSTRACT
1:1 1:1
The analog diode-based ring modulator has a distinctive sound
quality compared to standard digital ring modulation, due to the
non-linear behaviour of the diodes. It would be desirable to be Modulator Output
able to recreate this sound in a digital context, for musical uses.
However, the topology of the standard circuit for a diode-based
ring modulator can make the process of modelling complex and
potentially computationally heavy. In this work, we examine the
behaviour of the standard diode ring modulator circuit, and pro-
pose a number of simplifications that maintain the important be-
haviour but are simpler to analyse. From these simplified circuits, Carrier
we derive a simple and efficient digital model of the diode-based
ring modulator based on a small network of static non-linearities. Figure 1: Schematic showing the layout of a traditional ring-
We propose a model for the non-linearities, along with parameter- modulator circuit.
isations that allow the sound and behaviour to be modified dynam-
ically for musical uses.
type of ring-modulation circuit can be seen in Figure 1. The diodes
1. INTRODUCTION used in the circuit have a strong effect on it’s final behaviour. The
use of silicon diodes results in a hard clipping or ’chopping’ ef-
Ring-modulation (RM) (also known as four-quadrant modulation) fect and hence extremely bright extra harmonics, whilst germa-
is a technique similar to amplitude modulation (AM) and frequency nium diodes produce a softer non-linearity and a ’warmer’ sound
modulation (FM) and was developed, like AM and FM, for radio [1]. Later ring modulators instead employed a structure based on
transmission applications. Ideal RM is a special case of AM (i.e. VCAs, producing an output with far less added harmonics [5].
the multiplication of a carrier and modulator signal), where both The basic idea of ring-modulation is trivial to implement dig-
the carrier signal and the modulator signal are centred around 0V. itally, as it involves only a simple multiplication of two signals.
When this condition is fulfilled, the carrier and modulator signals However, such a digital ring modulator lacks the additional non-
are completely cancelled and the output of the system consists of linear behaviour (beyond the multiplication itself) and extra gener-
only the sum and difference frequencies of these inputs. This effect ated harmonics of a real analog ring modulator, and hence a large
is musically useful, because it allows harmonic sounds to be trans- part of its characteristic sound. It is therefore desirable to produce
formed into clangourous inharmonic sounds whilst still retaining a digital model that can replicate this behaviour to some extent.
some of their original character and articulation. Previous work has derived ordinary differential equations govern-
Ring modulation was first used as musical effect by German ing the ring-modulator circuit, and solved them numerically us-
avant-garde composers [1], notably Karlheinz Stockhausen. Ring ing the Forward Euler method [6]. However, this method requires
modulators were also used by the BBC’s Radiophonic Workshop significant over-sampling (a factor of ×128 is suggested), and is
in the 1950s and 1960s, notably to produce the distinctive voice hence not ideal for real-time usage.
of the ’Daleks’ in the television show ’Doctor Who’ [2]. Estab- In Section 2 of this paper, we propose a simplified circuit
lished as a normal part of the early electronic music studio, ring which behaves analogously to the traditional diode ring-modulator.
modulators were naturally included in early analog modular syn- In Section 3 we discuss how this circuit can be modelled digitally
thesizers. Don Buchla included a ring-modulator as one of the first using a network of static non-linearities. In Section 3.1 we discuss
modules of his System 100 synthesizer, built for Morton Subotnick how the shape of these non-linearities can be derived, and in Sec-
[3]. Robert Moog’s initial complement of modules did not include tion 3.2 we discuss the results produced by this algorithm. Finally,
a ring-modulator, but one was later added via a collaboration with in Section 4, we conclude.
Harald Bode. Thus, the ring-modulator came to be an element
available on a significant proportion of analog synthesizers, both 2. A SIMPLIFIED RING MODULATOR CIRCUIT
modular and non-modular.
The name ’ring-modulation’ refers to the way in which this Previous work has suggested that majority of the distortion charac-
technique was often implemented in its early days, using a config- teristics of the diode ring-modulator can be explained by assuming
uration that employed a ’ring’ of diodes [4] [5]. A schematic of this that only two of the diodes in the ring conduct at any one time, the
DAFX-1
DAFx-163
Proc. of the 14th Int. Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Proc. of the 14th International Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Vin
particular pair selected by the polarity of the carrier voltage [7] [8]. VC +
Rout Rin 2
It is then postulated that distortion of the signal is produced by the
non-linear characteristics of the diodes, and by the finite time it v2 v1 v5 + −
D1
takes to switch between pairs [7] [8]. However, the presence of
transformers in the circuit described in Figure 1 complicates the i1
a)
behaviour of the circuit in a number of ways, primarily by cou-
pling voltage between nodes of the diode ring and causing, in some
Vin
circumstances, three diodes to conduct simultaneously. This be- VC −
Rout Rin 2
haviour can be confirmed easily by measurement or by simulation
in a package such as SPICE or Qucs [9]. v4 v3 v6 + −
D3
We propose a simplified circuit which does not contain a trans-
former. By examining the circuit in Figure 1, we can see that if i2
the modulator waveform is denoted Vin and the carrier waveform
denoted Vc , the voltage at the two input nodes of the diode ring VC +
Vin
should be VC + Vin /2 and VC − Vin /2. We therefore replace the Rout Rin 2
input transformer and Vin and VC voltage sources with two volt- v4 v1 v5 + −
D4
age sources at these nodes providing the combined voltages. We
then replace the output transformer with two resistors connected i1
to ground. Figure 2 shows a schematic of this circuit. By exam- b)
ining this circuit, it should be clear that the diode pair D1 and D3
conduct when the signal Vc is positive, and the diode pair D2 and VC −
Vin
D4 conduct when it is negative. The signal Vin is present at both Rout Rin 2
opposite nodes of the diode-ring, in opposing polarities, but is of v2 v3 v6 + −
smaller amplitude and assumed to not bias the diodes. The output D2
is taken across the two resistors Rout (i.e. voltage v2 − v4), both i2
of which have a high resistance value. The resistors Rin have a
low resistance. Figure 3: Schematic showing the layout of the simplified,
transformer-less ring-modulator circuit when a) VC is positive and
b) VC is negative.
Vin −
VC + i1
2 +
v5 circuit. There are four parallel signal paths, two which process the
Rin combination VC +Vin /2 and two which process VC −Vin /2. Each
D1
v2 of these signal paths consists of a non-linearity representing the
v1
voltage-to-voltage transfer function of the resistor-diode-resistor
chain, plus (as necessary) inversion to represent the direction of
the diode and whether the branch of the circuit ends on v2 or v4.
D4 D2
Vin
A block diagram showing this structure is given in Figure 4. The
Rin VC −
2 next task is then to derive the form of the non-linearity used on
each of the parallel paths of the structure.
v4 v3 v6
D3 + −
3.1. Diode Non-Linearity Model
i2
Rout Rout One possible method of modelling the diode non-linearity present
in each of the branches of the simplified circuit is to use standard
nodal analysis to solve for the voltage at v2 or v4 in one of the
branches. We do this by modelling the diode using Shockley’s
Figure 2: Schematic showing the layout of the simplified, ideal diode equation, which is given by:
transformer-less ring-modulator circuit. vD
iD = IS (e nVT − 1) (1)
We then make the assumption that current only flows through
where iD is the current through the diode, IS is the reverse bias
two diodes simultaneously, controlled by the polarity of Vc [7]
saturation current, vD is the voltage across the diode, VT is the
[8]. We then assume that the closed diodes act as an open circuit
thermal voltage and n is the ideality factor.
and hence produce two substitute circuits representing the state of
The exponential functions introduced by the use of Shockley’s
the circuit when VC is positive, and when it is negative. Figure 3
diode equation produce a relationship between the input and output
shows a schematic of these circuits.
voltage of the diode that is only solvable analytically by applica-
tion of the Lambert W-function. Instead we expand out the expo-
3. MODELLING THE SIMPLIFIED CIRCUIT nentials using Taylor series, and then solve the resulting implicit
relationship. The input-output voltage relationship of the diode
By examining the simplified circuit, it is easy to see what kind of when modelled in this way is given in Figure 5. The parameter
digital signal processing structure could be used to replicate the values used to calculate this curve are given in Table 1, and are
DAFX-2
DAFx-164
Proc. of the 14th Int. Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Proc. of the 14th International Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
VC
−1
Vin −1
0.5 −1
−1
Figure 4: Block diagram showing the signal processing structure used to model the simplified ring modulator circuit.
Output(/V)
1.5
This method of calculating the non-linearity is only suitable Figure 5: Graph showing the input-output relationship of the diode
for producing a look-up table, as it involves a large amount of wave-shaper. The version derived from circuit analysis is given in
computation. More desirable would be a non-linear function that solid grey, whilst the piecewise approximation is given as a dashed
requires few operations to calculate and which lends itself to varia- black line.
tion by the user in order to adjust the sound of the effect. Looking
at the form of the curve, it is clear that it is separated into three clear that whilst not identical, the output of the digital model is rea-
main sections. At low voltage, the output is approximately zero. sonably similar to the measured signal. There is a discrepancy in
At high voltages, the relationship is approximately affine. These the peak voltage between the real and modelled ring-modulators.
sections are connected by a smooth curve. It is therefore possible This is probably caused by the lack of buffering on the outputs
to produce a very similar curve using a piecewise function, con- of the real ring-modulator (as it is a passive circuit), which leads
sisting of a zero section, a polynomial, and an affine section. This to the device having an output impedance that causes some signal
function is given by: loss when connected to a standard audio pre-amplifier.
Figure 8 shows the frequency spectrum of the output of model
v ≤ vb when presented with 50 Hz Vin and 1500 Hz VC . As expected,
8
< 0
>
f (v) = h (v−vb )2
v b < v ≤ vL (2) visible are the sum and difference frequencies at 1550 Hz and
2vL −2vb
: hv − hv + h (vL −vb )2
> 1450 Hz, along with further integer multiples of the modulator
L 2vL −2vb
v > v L frequency above and below the carrier frequency. Also visible are
odd harmonics of the modulated spectrum.
where v is the input voltage, vb is a parameter specifying the equiv-
Informal listening tests show that the output of the algorithm is
alent of the diode forward bias voltage, vL is a parameter giving
sonically satisfying when applied to synthetic material, with much
the voltage beyond which the function is linear, and h is a param-
of the character of a vintage ring-modulator present. When applied
eter specifying the slope of the linear section. The shape of this
to more natural sound sources, for example the human voice, the
function, with parameters adjusted to minimise maximum error
result of the algorithm is harsher and brasher than that of simple
within the region of interest compared to the modelled curve, is
digital multiplication, as would be expected. Variation of the pa-
given in Figure 5.
rameters of the diode wave-shaping function, as given in Equation
3.2. Results 2, allow a variety of distortion characters to be generated - from
soft germanium-diode style saturation to the harsher distortion as-
Figure 7 shows the output of the algorithm when when driven by a sociated with silicon diodes. Sound examples of the algorithm ap-
1V 500 Hz sinusoidal Vin and a 1V 1500 Hz sinusoidal VC . Figure plied to both synthetic and natural sources are available at the web-
6 shows the output of a real diode-based ring modulator, built to a site associated with this work [11]. Evaluation of the exact accu-
design consistent with the schematic given in Figure 1 [10]. It is racy of the algorithm compared to both the analog ring-modulator
DAFX-3
DAFx-165
Proc. of the 14th Int. Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Proc. of the 14th International Conference on Digital Audio Effects (DAFx-11), Paris, France, September 19-23, 2011
Normalised Magnitude(/dB)
further investigation. However, the author would like to emphasise ï20
that aim of the algorithm is to produce a computationally cheap
ï40
vintage ring-modulator style effect, rather than to replicate the be-
haviour of the analog circuit exactly. ï60
DAFX-4
DAFx-166