Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

DIGITAL GREY BOX MODEL OF THE UNI-VIBE EFFECTS PEDAL

Champ Darabundit ∗ Russell Wedelich Pete Bischoff

University of Southern California Eventide Inc. Eventide Inc.
Los Angeles, CA Little Ferry, NJ Little Ferry, NJ
[email protected] [email protected] [email protected]

ABSTRACT
This paper presents a digital grey box model of a late 1960s era
Shin-ei Uni-Vibe 1 R
analog effects foot pedal. As an early phase
shifter, it achieved wide success in popular music as a unique mu-
sical effect, noteworthy for its pulsating and throbbing modula-
tion sounds. The Uni-Vibe is an early series all-pass phaser ef-
fect, where each first-order section is a discrete component phase
splitter (no operational amplifiers). The dynamic sweeping move-
ment of the effect arises from a single LFO-driven incandescent
lamp opto-coupled to the light dependent resistors (LDRs) of each
stage. The proposed method combines digital circuit models with
measured LDR characteristics for the four phase shift stages of
an original Uni-Vibe unit, resulting in an efficient emulation that
preserves the character of the Uni-Vibe. In modeling this iconic
effect, we also aim to offer some historical and technical insight Figure 1: A closeup of the reflective housing (left) and lamp-LDR
into the exact nature of its unique sound. assembly (right)

1. INTRODUCTION
Each pair of single order all-passes, when mixed with the original
The Uni-Vibe was a Shin-ei Companion effects box branded as signal, creates an instantaneous frequency notch where their 180o
the Univox Uni-Vibe and distributed by the Unicord Corporation phase shifts intersect. For this reason, most phasers match the cen-
in the late 1960s. The Uni-Vibe is thought to be the prominent ter frequencies in pairs or use 2nd order filters. Notch locations
guitar effect in popular music recordings such as Jimi Hendrix’s are modulated by a Low Frequency Oscillator (LFO) sweeping the
Woodstock performance in 1969, Robin Trower’s 1974 “Bridge all-pass center frequencies. As a phaser, the Uni-Vibe is partic-
of Sighs,” and Pink Floyd’s 1973 “Breathe.” The sound of the ularly unique in two respects: the instantaneous positions of the
Uni-Vibe is characterized by throbbing pulse, “double beat,” and notches are determined by the complex interaction between a sin-
a lo-fi sweep [1]. It has a simple user interface with potentiometer gle incandescent lamp and four adjacent LDRs housed in a roughly
controls for Volume and Intensity, a foot pedal for varying rate or cube-shaped reflective chamber (shown in Figure 1), and an almost
speed, and a switch for Chorus or Vibrato Mode. arbitrary choice of phasing capacitor values. The capacitor values,
While the Uni-Vibe was marketed as a simulation of a rotating which determine the all-pass center frequencies, result in staggered
“Leslie” style speaker cabinet [2], in a more recent interview the or spread notches unlike other phasers which tend to use matched
inventor, Fumio Saeda, revealed his inspiration drew more from pairs. In addition to a phasing sound this leads to a band limited
Radio Moscow broadcasts modulated and distorted by the iono- tremolo-like effect. The exact reasoning for the capacitor selection
sphere as he listened on short wave radio in Japan [3]. In fact, the remains a mystery to Uni-Vibe aficionados [4].
Uni-Vibe modulation circuit was largely extracted from his first The uniqueness of the Uni-Vibe has thus been the subject of
effects box, the Psychedelic Machine, which was a combination of many attempts to commercially clone the original such as ([5] [6]
fuzz and modulation (called Mood), both of which were directly [7]) and [8] by Dunlop Manufacturing who now own the Uni-Vibe
inspired by the combination of fuzzy distortions and pitch modu- trademark, among others. Many of these clones attempt to emulate
lation of the ionosphere distorted broadcasts. the Uni-Vibe by recreating the original circuit, or by attempting to
The Uni-Vibe is considered to be a phaser or phase shifter sim- replicate a similar lamp and LDR combination.
ilar to the MXR Phase 90 of the time [4]. A phaser mixes an input Previous published work on Uni-Vibe analysis was done in [9]
signal with the same signal’s output from a series chain of all-pass and [10]. Related work in [11] provides a thorough overview for
filters to generate a number of notches in the frequency spectra. direct digital implementation of generalized analog phasers con-
structed with operational transconductance amplifiers and field ef-
∗ For Eventide Inc. fect transistors. Whereas, the more specific modeling of the MXR
1 Uni-Vibe is a registered trademark of Dunlop Manufacturing, Inc. Phase 90 pedal from [12] tabulates the main nonlinearity from a
Copyright: c 2019 Champ Darabundit et al. This is an open-access article dis- JFET used as the notch sweeping variable resistor within a state
tributed under the terms of the Creative Commons Attribution 3.0 Unported License, space discrete model of the circuit. The authors in [13] describe
which permits unrestricted use, distribution, and reproduction in any medium, pro- an ad-hoc method for modeling a vactrol, a single element pack-
vided the original author and source are credited. age combining an LED and LDR, in a Buchla low-pass gate. They

DAFX-1
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

Figure 2: The complete Uni-Vibe circuit schematic

reason that realizing full models of photoconductor transients for and suggestions for future work.
musical signals is a significant task based on the models in [14].
We propose a digital grey box emulation of the Uni-Vibe, sim- 2. UNI-VIBE CIRCUIT
ilar to the phaser modeling work done in [15]. The grey box
model approach combines physical measurement data (a black box The Uni-Vibe can be broken down into five basic blocks (as shown
model) and a physically informed model derived from circuit anal- in Figure 2): the pre-amplification section, the phasing section, the
ysis (a white box model). Our primary motivation for this ap- LFO and lamp section, the output buffer, and the output mixer. The
proach, over complete physical modeling of the entire system, input signal is buffered by the nonlinearity before being passed to
arises from the observed complexity in the coupled interaction be- the phasing section which is a multi-stage phaser with four phase
tween the driving LFO circuit, the single incandescent lamp, the stages. Each phase stage has its own unique LDR and phasing
four adjacent LDRs, and the reflective housing encasing the lamp capacitor, labeled CP in Figure 2. The LDR and R6 combined
and LDRs. We reason that a full model would need to take into in series forms the resistor-capacitor pair determining the center
account the complex physical relationship of at least the lamp, the frequency of the phase-shift contributed by each section.
LDRs, and the additional reflections and shielding provided by the After passing through the phasing sections, the signal is passed
housing. Since our goal is real-time and efficient implementation to the output buffer which is a bipolar junction transistor (BJT)
of this model, we opted for the black-box approach to this complex buffer, before being passed to the output mixer. Within the output
system. Consequently, our emulation uses measurement data taken mixer there is the vibrato/chorus switch, along with the volume
directly from an original Uni-Vibe unit to create three-dimensional control knob. The vibrato/chorus switch switches between two re-
wavetables capturing the behavior of each of the LDRs under the sistive networks to determine whether the signal from the output
influence of the LFO driven incandescent lamp. buffer is passed to the output by itself or mixed evenly with a “dry”
While the behavior of the LFO and LDRs is complex, a model signal originating from the first phase stage. The volume control
of the phasing sections can be derived and discretized via the bi- is a potentiometer adjusting the amplitude of the output. In com-
linear transform for the white box portion. Additionally, we will parison to the phasing section and the LFO, the pre-amplification
show that the discrete transistor phase splitting does not result in section and the output buffer section do not modulate the audio
ideal all-passing, and that this does have a perceivable effect in the path in a significant way at low signal levels and their contribu-
Uni-Vibe’s sound. tions are ignored in our analysis. The effects of these sections at
This paper will first provide an overview of the Uni-Vibe pedal high signal levels due to distortion remain for future work. Volume
and circuit in Section 2. In Section 3 we will examine the phasing and the Chorus/Vibrato switch did not need to be extensively mod-
circuit and the derivation of its continuous-time and discrete-time eled as they can be represented by a multiplication and addition,
models, highlighting some of the perceptually relevant aspects yet respectively.
to be thoroughly covered in previous works. Section 4 will cover
the measurement procedure for evaluating the LDRs and the sig- 2.1. LFO Section
nal analysis done to extract resistance curves from the LDRs. Sec-
tion 5 will discuss the results and real-time implementation of the This section contains the most dominant Uni-Vibe controls, Inten-
preamplifier model of the Uni-Vibe. Section 6 offers conclusions sity and Speed, found in the LFO section of Figure 2.The Intensity

DAFX-2
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

knob, which goes from a value of one to nine, with additional min-
imum and maximum settings, scales the LFO voltage across the
bulb, and thus determines the range of the notches’ frequencies
in the phase-shifted signal.The effect of sweeping over a larger
frequency at the same LFO speed is what causes notches to sound
“deeper.” The Speed, controlled via foot-pedal, determines the rate
of the LFO. From toe-to-heel the Speed traverses approximately 0
Hz to 7.6 Hz. The 0 Hz Speed originates from a cancel switch
located at the heel of the foot pedal that turns off the lamp and
prevents the value of the LDRs from changing.
The LFO is a variation on a phase-shift oscillator [4]. A phase-
shift oscillator uses regenerative feedback from an RC network
from the base of QO1 to the emitter of QO2 to produce a sinusoidal
output. The RC network here is the equivalent resistance of the
two RO1 , RO2 , RO3 legs along with the three CO capacitors. The
foot pedal controls the speed by varying the value of the coupled
potentiometers RO1 , which in turn vary the equivalent resistance
of the two resistor legs.
Before the LFO signal is passed to the BJT buffer in the lamp
section, its amplitude is modulated via the Intensity potentiome-
ter. As the rate of the LFO is dependent on regenerative feedback,
coupling the Intensity pot directly to the output of the LFO results Figure 3: A single stage of the phasing circuit. The inverting side
in rate drifting proportionally with intensity and vice-versa, result- comes comes off the collector and the non-inverting side comes off
ing in a non-orthogonal relationship between Speed and Intensity. the emitter
In order to accurately model this parameter interdependency, our
black box model utilizes a three-dimensional wavetable, with input
phase/sample, speed, and intensity axes. an all-pass phaser by representing the Darlington emitter-follower
as an phase inverter with gain ≈ 1 feeding an RC bridge. This
conclusion assumes, incorrectly, that the amplitude of inverting
2.2. Lamp and LDRs
and non-inverting legs of the transistor originating from the collec-
The lamp section of the Uni-Vibe is rather straightforward and tor and emitter of Q2 are balanced, and that the block capacitor’s
consists of a BJT buffer which supplies the LFO signal to the lamp. (CDC ) contribution can be ignored. Through listening, measure-
Within this section is the cancel switch and a trim pot which was ment, and simulation of the Uni-Vibe it was determined that these
not adjusted during our measurements. When the cancel switch is factors could not be ignored and had to be taken into account, thus
flipped all power to the bulb is cut. The resistance of an LDR is in- a corrected analysis of the circuit is presented below.
versely proportional to brightness therefore in complete darkness
an LDR is at its maximum resistance. As the center frequency of 3.1. Phasing Circuit Analysis
the phase shifter is inversely proportional to the resistance of the
LDR, the cancel switch causes the center frequencies to trend to
DC. The cancel switch thus effectively acts as a bypass.
The lamp and the LDRs were specially sourced for the Uni-
Vibe. The lamp is a fast-switching incandescent bulb (but not as
fast as an LED) and the LDRs are made with cadmium sulfide and
are no longer mass produced [3]. The uniqueness and rarity of
these components is what, in part, is responsible for the unique
tone of the Uni-Vibe.

3. PHASING CIRCUIT ANALYSIS AND MODEL

Figure 3 shows the schematic for phase stages 2, 3, and 4 of the Figure 4: A block diagram describing the corrected phasing cir-
phasing circuit. Each stage consists of a Darlington emitter fol- cuit. R0 = LDR + R6
lower that directly drives a phase inverter whose center frequency
is determined by an LDR and phasing capacitor pair, labeled LDR Figure 4 represents the incoming signal to each stage as Vs ,
and Cp , respectively. Due to the nature of the Darlington am- the phase inversion as a gain of −1, and the inverting and non-
plifier circuit and C1 acting as a “bootstrap” capacitor the input inverting gains as β and α, respectively. The phasing capacitor
impedance of each stage is very high [4]. This allows us to con- and the block capacitor remain unchanged while the value R0 rep-
sider each phase stage individually, instead of as a whole. Al- resents the series resistance of the LDR and R6 .
though phase stage 1 is not driven by a Darlington amplifier, we The transfer function H(ω) of the phasing circuit can be found
assume similar behavior as the other phase stages because of the through the superposition of two complex voltage dividers taken
preamplifier section . A similar analysis of this circuit has been from the inverting and non-inverting legs of the circuit. Analyzing
done by [9], who modeled the response of each phase stage as the phasing circuit in this manner makes a digital implementation

DAFX-3
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

ωo − jω
H(ω) = (11)
ωo + jω
Which matches the analysis done by [9]. If the same equalities are
applied to equation (10) we obtain the standard phase response of
a first-order all pass:
 
ω
φ(ω) = −2tan−1 (12)
ωo

Since the bootstrap capacitor C1 allows us to assume high in-

Figure 5: A further simplification of the phasing block diagram put impedance at each stage the full cascade of the phasing section
where the Vs , α, and β have been simplified as an inverting and is:
non-inverting source Y4
H(ω) = Hcn (ω) + Hen (ω) (13)
n=1

of the filter and application of nonlinearity easier as will be shown Where,

in Sec. 3.2. This simplification is represented in Figure 5, where: κcn ωon + jω
Hcn (ω) = (14)
ωon + jω
Ve = αVs (1) κen ωon
Hen (ω) = (15)
and, ωon + jω
and,
Vc = −βVs (2) Cp n
κcn = (16)
Cpn + CDC
Giving us the complete transfer function:
CDC
H(ω) = Hc (ω) + He (ω) (3) κe n = (17)
Cpn + CDC
Hc (ω) and He (ω) are the transfer functions of the inverting 1 CDC + Cp n
ωon = 0
(18)
and non-inverting legs: Rn Cp n CDC
κc ωo + jω for each phase-stage (n = 1, 2, 3, 4).
Hc (ω) = (4)
ωo + jω
Table 1: Measured capacitance, inverting, and non-inverting
κe ωo gains, and LDR resistances
He (ω) = (5)
ωo + jω
Where: LDR (Ω)
Cp (F) α β
avg (M) min. (k) max. (M)
Cp 1 .015µ 1.01 1.11 0.405 12.7 2.79
κc = (6)
Cp + CDC 2 .22µ .98 1.09 0.233 6.86 2.59
CDC 3 470p .97 1.10 0.29 7.69 3.32
κe = (7) 2 .0047µ .95 1.09 0.240 6.22 4.16
Cp + CDC
are constants, and
Each stage has its own unique Hcn , Hen , κcn , κen , and ωon
1 CDC + Cp as each stage has a unique phasing capacitor, LDR, inverting gain,
ωo = (8)
R0 Cp CDC and non-inverting gain which were found through measurement of
is the center frequency of the filter. If we combine equations (1) each phase stage. Table 1 provides these values. We should note
(2) (3) (4) (5) we have the transfer function of the entire phasing that these values are most likely unique to the particular unit we
circuit: measured, and we expect these values are varied among differing
Uni-Vibe units given the tolerances of the components. Different
   
κe ω o κc ωo + jω units most likely have their own particular sound.
H(ω) = α −β (9) The inclusion of the inverting and non-inverting gains as well
ωo + jω ωo + jω
as the block capacitor in the analysis of the phasing circuit are a
Which has the phase response:
sonically relevant addition to the analysis and modeling of the Uni-
Vibe. As seen in Figure 6 with R0 fixed at LDR1’s mean value
the block capacitor causes the magnitude of the transfer function
   
−βω ω
φ(ω) = tan−1 − tan−1 (10) to exhibit a high shelf response rather a unity response all-pass
ακe ωo − βκc ωo ωo
filter. This high shelf effect is emphasized as the stages stack up.
If we let β = 1, α = 1 (i.e. the inverting and non-inverting Additionally dissimilar inverting and non-inverting gains, along
gains are balanced) and CDC = ∞ (which removes the effect of with the block capacitor cause a shift in the phase response of the
the block-capactior) equation (9) takes the form of a standard first- filter as seen in equation (10). In chorus mode the sweeping high
order all-pass phaser in equation (11), as κc = 0, and κe = 1. shelf causes the low end of the signal to match what the output

DAFX-4
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

1 3.2.1. Non-linear elements

0.8
Magnitude (dB)

0.6 It was determined through measurements that the emitter-follower

0.4 pair in each phase-stage was clipping the input signal asymmetri-
0.2 cally. To approximate this non-linear behavior, a tanh() function
0 with a heuristically determined bias and scaling factor is applied
-0.2
10
0
10
1
10
2
10
3
10
4 in the signal chain before (21). Separating our signal into two sig-
Frequency (rad/s nal paths allows us to apply this nonlinearity. A block-diagram of
how this nonlinearity is incorporated is shown in Figure 7. Includ-
360
ing these non-linearities was necessary in modeling the Uni-Vibe
as the clipping added audible harmonics and the biasing adjusted
Phase (deg)

270
the balance of even and odd harmonics.
The scaling factor g determines the level at which the signal
clips by gaining down the input signal. The biasing factor u offsets
180
the signal to determine the extent of the asymmetry in the clipping.
10 0 10 1 10 2 10 3 10 4 The ideal values for g and u were first determined through mea-
Frequency (rad/s
surement of the original Uni-Vibe pedal, and then tuned heuristi-
Figure 6: Transfer function of the first phase stage. α = 1.0009, cally to match. After the signal is clipped by the tanh() function
β = 1.11, R0 = 409.7kΩ it is passed through a DC-blocking low-pass filter. The inclusion
of this filter is an improvement on the original signal path of the
Uni-Vibe. The biasing factor u, which exists in both the Uni-Vibe
would be in vibrato mode and is responsible for the Uni-Vibe’s and in our model, pushes DC impulses through to the output when
characteristic throbbing pulse. ωo approaches DC. These impulses are audible as clicking in the
We reason that the selection of the phasing capacitor values is original Uni-Vibe. Our DC-blocking filter removes the DC offset
such that only two notches can heard at a time, which creates the before equation (19) (20) are applied. Before being outputted the
Uni-Vibe’s characteristic double beat. As the Uni-Vibe stages are signal is gained back up by g −1 to return the signal level back to
not matched pairs like other phasers, without the inclusion of the its original value. The gain values are fairly low, so aliasing dis-
block capacitor instantaneous notches can “pop” in and out of the tortion due to clipping is also fairly minimal, thus in order to keep
frequency spectra. the model efficient we neglected any oversampling.

3.2. Discrete Model 4. MEASURING THE UNI-VIBE LDRS

4.1. Physical Measurement

Due to the difficulty in physically modeling the transient resistance
of the LDRs in the phase stage, a decision was made to measure the
behavior instead. Unlike [10] which measures a Uni-Vibe clone,
we directly measure the actual Uni-Vibe unit. Additionally [10]
only measures one LDR, while we found it important to measure
all four as we could not ensure the LDRs would have the same
values based on their make and position respective to the lamp.
Figure 7: Block diagram of the applied nonlinearity This decision is supported by Table 1 which demonstrates that each
LDR differed from one another. To measure the behavior of a
phase stage’s LDR the output of the stage was shorted and the
From equations (14) (15) a discrete model of each phase stage input was fed with a 1 kHz sine wave. Voltage measurements were
circuit was derived using the bilinear transform and pre-warping taken across the LDR and R6 and across R6 alone, allowing us
the filter around the frequency ωon giving us the discrete transfer to derive the transient resistance of the LDR as part of a voltage
functions for the inverting and non-inverting legs: divider.
VLDR
(κcn tanωon + 1) + (κcn tanωon − 1)z −1 RLDR = R6 − R6 (22)
Hcn [z] = (19) V6
(tanωon + 1) + (tanωon − 1)z −1
Measurement recordings were taken by playing the 1 kHz in-
κen (1 + z −1 ) put signal while in tandem taking the two aforementioned voltage
Hen [z] = (20) measurements. These measurements were recorded on a digital
(tanωon + 1) + (tanωon − 1)z −1
audio workstation using the Expert Sleepers ES-8 USB-audio in-
in the complete discrete equation:
terface which has DC-coupled inputs and outputs [16]. To model
Hn [z] = Hcn [z] + Hen [z] (21) the full breadth of the pedal these measurements were repeated at
fourteen different LFO frequencies and at eleven different intensi-
for each stage (n = 1, 2, 3, 4). ties settings for each of the four LDRs. Consequently an additional
It is important to note that ωon is dependent on the value of measurement was taken of the LFO signal driving the lamp, as a
the LDR and therefore the bilinear transform and frequency pre- method of recording the speed of the LFO in the case of any in-
warping must be applied in real-time. consistencies in the foot pedal settings during measurement.

DAFX-5
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

4.2. Measurement Analysis LDR at a speed of 4 Hz. Visual analysis of the resistance curves
shows the curves have increasing asymmetry as intensity increases
due to the LDRs having different turn-on and turn-off times. The
increasing asymmetry is a result of lamp varying its brightness
more rapidly as intensity is increased.

5. RESULTS AND EXPERIMENTS

5.1. Real Time Implementation

Figure 8: A complete wavetable for LDR1 at 4 Hz

The process taken to analyze the measurement for each LDR

were as follows. The measurement data was converted to their
nominal RMS values based on the specifications of the USB-audio
interface. The recordings were then upsampled by four so enve-
lope extraction could be performed using a maximum value filter
with a rectangular window with a width of 1 kHz. This allowed
us to extract the effect of the circuitry and remove the 1 kHz in-
put signal. Additional filtering was also done to remove power- Figure 9: Noise Spectrograms. Top: Model, Bottom: Uni-Vibe. A
supply hum. A model of the additive noise in the LDR resistances second notch can be seen in the lower frequencies under the main
was found by deriving the transient resistance of an LDR using notch.
equation (22) at a minimum Intensity for each Speed. This noise
model could then be used in conjunction with a Wiener filter, us-
A real-time implementation of the the Uni-Vibe emulation was
ing Matlab’s wiener function, to remove any noise in transient
implemented in C/C++ both as a VST plugin using the JUCE
resistances at higher Intensity settings.
framework and as a process running in a single core of an em-
At this point, single-cycle resistance curves needed to be ex- bedded Linux ARM Cortex-A9 commercial audio product. The
tracted from the clean resistance signal. To preserve more percep- real-time emulation consists of four univibe_phaser modules each
tible information single-cycle curves were extracted by using local of which replicates a single stage of the phasing circuit by us-
maximums to determine the starting and ending peaks for individ- ing the measurement derived three-dimensional wavetable and the
ual periods in the resistance signal. Unfortunately information is discrete-time model of the Uni-Vibe’s phase-stage.
lost at the loop point from the start to the end of a cycle. To mini-
Depending on the selection of which stage to emulate the uni-
mize this local maximums were chosen as the loop point because
vibe_phaser module converts the wavetable of LDR resistances
ωo values arising from the resistances at local minimums are much
into a wavetable of center frequencies ωo before implementation
more perceptible than ωo values arising from the local maximums,
using the known values of each stages’ Cp and CDC . Using a
which approach DC. Each single-cycle period was averaged to-
graphical user interface the desired Intensity and LFO Speed can
gether to further reduce noise in the resistance signal.
be set to linearly interpolate a value of ωo from the wavetable. This
To prepare these resistance curves for implementation in the
value is then passed to an implementation of equation (13).
wavetable each one-period length curve was then downsampled
Figure 9 shows spectrogram output of our real-time imple-
to 64-samples. Then, using the LFO rate data obtained from an
mentation and the original Uni-Vibe with white noise as the input
FFT of the LFO signal, curves of a similar Intensity were inter-
source. Upon visual inspection, the model retains the dynamic be-
polated to be equally spaced across the entire frequency range of
havior of the original unit. Intensity and Speed settings were 7 and
the LFO. This was done to facilitate indexing across the lookup
2.0 Hz respectively.
table. The first two-samples and last two-samples of each curve
were interpolated together to ensure that the loop point was seam-
less at the expense of data loss. Otherwise, any significant disjoint 5.2. Experimental Methodology
would be perceived as an audible click. The result of this analy-
sis is a three-dimensional wavetable containing 640 unique LDR MUSHRA style listening tests were conducted to quantitatively
resistance curves for each LDR representing every combination assess the accuracy of our real-time implementation of the Uni-
of Speed and Intensity settings on the Uni-Vibe, with a sawtooth Vibe running in the ARM Cortex-A9 commercial audio product.
phase input providing exact sample accurate rate. Figure 8 shows A second original Shin-ei Uni-Vibe, two analog hardware clones,
a small sample of this wavetable containing the curves for the first and two digital emulations in hardware were tested alongside our

DAFX-6
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

implementation for a total of 6 units under test. The original hard-

ware unit measured for our implementation was chosen as the
MUSHRA reference.

5.2.1. MUSHRA Test Setup

Reference and MUSHRA anchor recordings were generated as
follows: Unaffected electric guitar passages were recorded into
a UAD Apollo 8 interface via Direct Injection, or DI. Reference
recordings were created by feeding recorded passages through the
original Uni-Vibe set at specific Intensity and Speed settings. All
recordings were done in Chorus mode with Volume set at 5. Due
to the relatively low frequency spectral quality of the electric gui-
tar, 1.5 kHz low-passed versions of the reference recordings were
included as anchors instead of the usual 7 kHz and 3.5 kHz low-
passed anchors.
For test unit recordings, Intensity settings were visually matched
to the reference through printed markings or numerically via graph-
ical displays where applicable. Test unit LFO Speed settings were Figure 10: Listening test 3 results
precisely matched to the reference through waveform inspection of
their phase cancellation cycle or numerically via graphical displays
where applicable. Trial and error was used to ensure LFO phase
alignment between test unit recordings and the reference record-
ing.

5.2.2. Electric Guitar Passage Selection

A total of three electric guitar passages were recorded for the lis-
tening tests. Passages were disparately styled and chosen to reflect
real-world use cases for the Uni-Vibe in addition to test the full
range of Uni-Vibe sounds and settings.
Passage 1 consisted of a 101 BPM Texas Blues styled riff
played with hard plectrum attack intended for a Uni-Vibe set to
approximate a Leslie in tremolo mode. Reference Intensity and
Speed settings were 5 and 4.85 Hz respectively creating a triplet
modulation feel.
Passage 2 consisted of a 4 bar phrase using whole note chords
followed by an arpeggiated version of the same chord progression
played at 110 BPM. Reference Intensity and Speed settings were
7 and 1.89 Hz respectively for a medium depth modulation that Figure 11: Listening test 2 results
followed the quarter note.
Passage 3 consisted of a 62 BPM, two open-chord arpeggia-
tion. Intensity and Speed settings were set at 10 and 0.99 Hz re-
spectively creating a slow and deeply swept modulation. Settings
were chosen so participants could best judge LFO modulation con-
tours in the test.

5.3. Experimental Results

Listening tests were administered using webMUSHRA software
[17] and a total of 20 participants were included in the test. Fig-
ure 10 shows a box plot of the results of listening test 1 where
our model scored closest to the reference. Of the analog hardware
units tested, scores varied widely with two units averaging below
the digital emulations. The second original Uni-Vibe scored sub-
stantially lower than the reference which seemed to confirm our
initial listening impressions of the high degree of variability be-
tween these specific original units. Given the majority of units
tested closer to the reference, we suspect the second original Uni-
Vibe could be out of factory specification warranting further in-
vestigation. Post-test user feedback for listening test 1 revealed Figure 12: Listening test 3 results
a wide range of criteria for judging similarity and included pick

DAFX-7
Proceedings of the 22nd International Conference on Digital Audio Effects (DAFx-19), Birmingham, UK, September 2–6, 2019

attack, tonal balance (timbre), modulation depth, and noise floor. [4] R. G. Keen, “The technology of the univibe,”
The wide range of judgement criteria may explain the large in- http://www.geofex.com, accessed March 5, 2019.
terquartile sizes and low score outliers in these results. [5] MJM Guitar FX, “Sixties Vibe Classic,”
Figure 11 shows a box plot of the results of listening test 2. http://mjmguitarfx.com/product/sixties-
These results mirrored listening test 1 however, Digital Clone 1 vibe-classic/, accessed March 5, 2019.
scored below the anchor. During our own informal comparisons,
Digital Clone 1’s depth of modulation and overall timbre were [6] Black Cat Pedals, “Black Cat Vibe,”
markedly different than the reference, both most likely playing http://www.blackcatpedals.com/black-cat-
large roles when tests were scored by participants. Anchor scores vibe/, accessed March 5, 2019.
averaged higher than listening test 1 indicating low pass frequency [7] Fulltone, “Mini Dejavibe 3 V2,”
may have been set too high. Post-test user feedback reported lis- https://www.fulltone.com/products/mini-
tening test 2 as harder to judge than listening test 1 which could dejavibe-3-v2, accessed March 05, 2019.
explain the average score increase for most units.
[8] Dunlop Manufacturing Inc., “Univibe Chorus/Vibrator,”
Figure 12 shows a box plot of the results from listening test 3.
https://www.jimdunlop.com/product/m68-7-
As in test 2, Digital Clone 1 scored lowered than the anchor, most
10137-07227-5.do, accessed March 05, 2019.
likely for reasons similar to listening test 2. Our model scored
second closest to the reference with Digital Clone 2 scoring the [9] C. Hahlweg and H. Rothe, “The unique sound of the Uni-
highest. Post-test user consensus deemed listening test 3 hard- Vibe pedal,” in Proc. SPIE 8487, Novel Optical Systems
est to score of the three tests. From our own qualitative listening Design and Optimization XV, San Diego, California, United
sessions at these settings, our implementation exhibited a slightly States, Oct. 19 2012.
deeper and squarer-edged modulation contour than the reference. [10] C. Hahlweg and H. Rothe, “The unique sound of the Uni-
We believe this is related to our choice of wavetable splice point Vibe pedal: Part ii. transient behaviour,” in Proc. SPIE 8842,
and interpolation. Novel Optical Systems Design and Optimization XVI2, San
Diego, California, United States, Sept. 30 2013, pp. 1–8.
6. CONCLUSION [11] A. Huovilainen, “Enhanced digital models for analog mod-
ulation effects,” in Proc. Digital Audio Effects (DAFx-05),
In this work a grey box method was proposed for creation of an Madrid, Spain, Sept. 20–22, 2005, pp. 155–160.
accurate Uni-Vibe model. A white-box model of the Uni-Vibe’s
phasing circuit was created through circuit analysis uncovering as- [12] F. Eichas, M. Fink, M. Holters, and U. Zölzer, “Physical
pects of the Uni-Vibe circuit that contribute to its iconic sound. A modeling of the MXR phase 90 guitar effect pedal,” in Proc.
complementary black-box model was also created through mea- Digital Audio Effects (DAFx-14), Erlangen, Germany, Sept.
surement of the Uni-Vibe’s LFO-LDR-lamp interaction. A real- 1–5, 2014, pp. 153–158.
time implementation of the grey box model of the Uni-Vibe was [13] J. Parker and S. D’Angelo, “A digital model of the buchla
implemented on a single core of a Linux ARM Cortex-A9. In lowpass gate,” in Proc. Digital Audio Effects (DAFx-13),
comparison to an original unit and other Uni-Vibe clones our im- Maynooth, Ireland, Sept. 2–6, 2013, pp. 278–285.
plementation was on-average rated closest to the reference in two
[14] A. E. Iverson and D. L. Smith, “Mathematical modeling of
out of three MUSHRA listening tests, scoring second in the third
photoconductor transient response,” IEEE Trans. on Electron
test.
Devices, vol. 34, no. 10, pp. 2098–2107, 1987.
Further work to improve our model would be deriving and im-
plementing more accurate behavior of the non-linear BJT clipping, [15] R. Kiiski, F. Esqueda, and V. Välimäki, “Time variant grey-
as was done by [18], instead of replicating it heuristically. We box modeling of a phaser pedal,” in Proc. Digital Audio Ef-
also plan to further investigate the ideal method of splicing at the fects (DAFx-16), Brno, Czech Republic, Sept. 5–9, 2016, pp.
wavetable loop point. Conversely, judging from our measurements 31–38.
of the LFO-LDR-lamp interaction, it may be also possible to create [16] Expert Sleepers Ltd., “ES-8 USB Audio Interface,”
a simplified parametric model of that interaction. https://www.expert-sleepers.co.uk/es8.
html, accessed June 19, 2019.
7. ACKNOWLEDGMENTS [17] Michael Schoeffler, Sarah Bartoschek, Fabian-Robert Stöter,
Marlene Roess, Susanne Westphal, Bernd Edler, and Jür-
Special thank you to Chuck Zwicky (www.zmix.net) for making
gen Herre, “webMUSHRA: a comprehensive framework for
Uni-Vibe recordings of his original unit for the listening test com-
web-based listening tests,” Journal of Open Research Soft-
parisons.
ware, vol. 6, no. 1, 2018.
[18] D. T. Yeh, “Automated physical modeling of nonlinear audio
8. REFERENCES
circuits for real-time audio effects - part ii: BJT and vacuum
tube examples,” IEEE Trans. on Speech and Audio Process-
[1] H. Shapiro, M. Heatley, and R. Mayer, Jimi Hendrix Gear,
ing, vol. 18, no. 3, pp. 1207–1216, 2011.
Voyageur Press, Minneapolis, MN, USA, 2014.
[2] Unicord Incorporated, Uni-Vibe Operating Manual, 1968.
[3] M. Yoshiyuki, “What is Uni-Vibe? pt.1,”
https://www.digimart.net/magazine/article/
2018052303189.html, May 5 2018.

DAFX-8