L
R
By Jake Rothman
Noise about noise – a new slant on analogue
noise with tilt control – Part 1
Fig.1. Roland TR909 Snare drum hit – note the large amount of noise in the signal.
I
’ve often described my job as
either eliminating distortion or generating it. It’s the same with electronic
noise. However, this time, instead of minimising noise in a pre-amplifier, we’ll work
on noise generation for audio synthesis.
Noise is an essential component of synthesis, especially for voice, snare drums
and cymbals. Fig.1 shows a typical snare
hit on a storage scope. The little module
in this project can be an ideal addition to
the battery-powered analogue synth (see
the Audio Out series covered in the August
2018 issue of Practical Electronics (actually
EPE back then)). Its tilt tone control circuit
is also useful for Hi-Fi in its own right.
Random fluctuations
Normal electronic noise is truly random,
caused by many effects, the main one being thermal agitation. For musical use
we’re much more interested in more dramatic noise produced by crystal lattice
defects in semiconductor crystals and
chip surface irregularities. Contaminants
and reverse bias leakage currents are also
significant. Manufactures do their best to
avoid making noisy devices; better, more
consistent manufacturing helps to produce
less noisy devices, but they can’t eliminate
these effects completely and some noise
is inevitable. Most sensible engineers and
builders discard noisy devices, but synth
builders hoard them – see Fig.2 for my notso-secret cache of noisy bits and pieces in
their own drawer.
Practical Electronics | April | 2020
The famous Roland TR-808 drum machine used specially selected ‘defective’
2SC828 transistors, which were marked
with a spot of pink nail-varnish! When
the supply of those noisy transistors dried
up Roland ceased production – a decision
that was doubtless reinforced by a general prejudice against analogue at the time.
In an interesting about turn, in 2017 Roland reissued the TR-808 (as the TR-08) but
the original analogue noise is now simulated
in software. This technique was described
in John Clarke’s White Noise Generator project in the September 2019 issue of PE. As a
‘software-wary’ analogue engineer, I’d like
to present my digital-free solution, which,
like many analogue designs came about
through a serendipitous accident.
Voltage regulator problem
Fig.2. Any noisy devices I come across are
hoarded in a special drawer for synthesiser
noise sources, including some 1965
AF114 germanium transistors that are so
noisy they may have radioactive isotopes
in their crystal structure!
when I put the regulator chip in the circuit
the noise level was four-times worse. The
new circuit is shown in Fig.4.
I realised that what had happened was
the regulator was amplifying its own
noise. This can be explained by looking
at the regulator’s internal circuit shown
in Fig.5. The internal voltage reference
was 2.5V and two resistors were used
to set the output voltage. These were effectively a feedback network setting the
gain of the system to almost 10 to get
the required 22V output. Consequently,
the noise from the reference was amplified by this amount. The solution was to
maintain the required DC gain, but reduce the AC gain (to unity) by bypassing
I designed a power amplifier that seemed
to have excessive noise. I tracked down
the problem to the
centre-rail bias genV +
2kΩ
erator shown in
4 5 V
2 .2 µ F
10kΩ
Fig.3. I had decided
V T
3 3 0 0 µ F
Inp u t
–
to use a TL431 shunt
Ou tp u t
regulator to replace
+
2kΩ
4 0 W p o wer
a 22V Zener diode
H al f- r ail
am p l ifer (d iscr ete)
centr e- bias
(which are inherentg ener ato r
+
2 2 V
1 0 µ F
4 0 0 m W
ly noisy due to their
B Z Y 8 8
0 V
0 V
avalanche break0 V
N
o
ise
d
eco
u
p
l
ing
down mechanism);
I simple assumed
the TL431 would be Fig.3. Centre-rail bias generator in a power-amp with a 44V
better. Surprisingly, supply. The 22V Zener was too noisy.
+
n
AUDIO
OUT
AUDIO OUT
59
the feedback resistor with a capacitor, as
shown in Fig.6. This reduced the noise
to about half that of the Zener and I was
pleased to have fixed the circuit.
V +
4 5 V
R
N eg ative
feed back
R E F
a
k
LOAD
R 1
2 kΩ
+
k
R E F
2 2kΩ
a
1 0 µ F
T L 4 3 1
Ad j u stabl e
ze ner /shu nt
r eg u l ato r
Set bias
vo l tag e
R 2
kΩ
0 V
0 V
V OUT = V REF × (1 + R 1 /R 2 )
W her e V REF = 2 . 5 V
Fig.4. Replacing the Zener with a voltage
regulator using this circuit made the noise
unexpectedly worse.
V +
R
k
LOAD
V OUT
Op am p
R 1
R E F
+
R 2
–
R
N eg ative
feed back
r em o ved
V OUT
F u l l AC
g ain
k
R E F
B o o ster
tr ansisto r
(inver ting )
V REF
2 .5 V
1 0 µ F
LOAD
R 1
2kΩ T L 4 3 1 p in view
B ias
o u tp u t
+ 2 0 V
output itself, fixed at 2.5V. If the circuit is
to be used at 9V instead of its intended 5V,
the bias voltage can be changed to 4.5V by
adding R4, which is omitted for normal 5V
operation. To complete the noise generator, a tone control is added, built around
the second half of IC1. We’ll deal with this
part of the circuit later.
V +
5 V
+
T L 4 3 1
a
A short digital noise diversion
R 2
0 V
Fig.7. Bypassing the reference pin to
ground gives maximum AC gain and
maximum amplification of noise.
I was still intrigued by the noise, so I removed all the AC feedback by connecting
the capacitor to ground, as shown in Fig.7,
and got an amazing noise generator. Now
the reference noise was being amplified at
full open-loop. I tried a few different TL431s
and found the noise to be much stronger and
consistent compared to the usual analogue
techniques. Also, the generator worked well
down to about 3.5V, whereas most analogue
systems need at least 9V. This circuit runs
at 5V, the rail chosen for the PE low-power
synthesiser (started in PE, August 2018).
One word of caution, there is always the
risk when using unspecified device characteristics that the manufacturing process
might be improved and the ‘defect’ that is
being exploited will be removed. That said,
I think we are on stable ground with the
TL431 – it was launched in 1978, so Texas
Instruments have had over four decades to
make changes, and the current version is
unlikely to be altered.
Where predictability or a specific spectral
response is needed, we often turn to pseudo-random digital noise generation, where a
string of seemingly random zeros and ones
(see Fig.9) are generated until the whole
sequence repeats again after a long time
(from say 20 seconds to months). Incidentally, the problem of excessive randomness
was exposed by the early shuffle mode on
music players, such as the iPod, where occasionally the same song would be played
consecutively. To stop this happening, the
random number generation algorithm had
to be tweaked to make it appear more ‘random’ to human perception – ironically, by
preventing song repeats and thereby reducing the real randomness.
For sound synthesis a short pseudo-random sequence can be useful, because this
can endow a recognisable repeating pattern within it, effective for simulating
mechanical noises such as engines. For
this, I developed a CMOS hardware-based
circuit built around the usual arrangement
Circuit description
a
G r o u nd
Fig.5. Regulator circuit showing the
internal circuit of the TL431. There is
a reference generator, a series-pass
transistor and an op amp. The resistors
set the gain around the loop.
Fig.8. Shows the noise generator part of
the circuit, consisting of the noisy regulator followed by amplification of 38. This
is provided by non-inverting op-amp stage
IC1a to increase the output level to around
2.2V peak-to-peak (line-level/0dB). No halfrail potential divider network is needed
since this is set by the voltage regulator’s
V +
Fig.9. Digital noise: this screen shot shows
the random string of ‘zeros’ and ‘ones’
which, just like pulse-code modulation
can be integrated by a low-pass filter to
produce analogue noise.
+ 5 V
+ 5 V
LOAD
R 1
2 2kΩ
V OUT
1 0 µ F
M ax im u m
neg ative
feed back
M inim u m
no ise
+
+
3
R 1
k
R E F
T L 4 3 1
a
R 2
R 3
22kΩ
* Om it R 4 fo r
5 V o p er atio n
0 V
Fig.6. Correct bypass capacitor connection
to give minimum noise. Gain at AC is unity,
while the DC gain sets the output voltage
with R2 and R3.
60
+ 2 .5 V
3 0 m V p k -p k
no ise
R 4 *
2 kΩ
(9 V )
+
2
B ias to
IC 1 b
R 2
2 2kΩ
k
R E F
a
IC 2
T L 4 3 1
R 5
1 kΩ
C 1
1 0 0 µ F
0 V
C 2
1 0 µ F
8
IC 1 a*
–
* IC 1 a
T L 0 6 2 o r
M P C 6 0 0 2
1
4
C 3
4 .7 µ F
C 4
1 0 µ F
R 6
P in 1
2 .2 V p k -p k
no ise
+
R
+
Ou tp u t to
til t co ntr o l
kΩ
+
0 V
P in 2
Fig.8. Analogue noise generator with output amplifier. R4 must be added for supply
voltages greater than 5V. R2 limits possible capacitor discharge current into the reference
pin. (This circuit is connected to a tilt tone control which will be covered next month.)
Practical Electronics | April | 2020
V +
1 4
3
1
IC 3 a
4 0 7 0
IC 1 - 4 p o wer r ail s
d eco u p l ed with 1 0 0 nF
d isc cer am ic cap acito r s
1 1
2
4
5
IC 3 a
4 0 7 0
6
G r o u nd al l u nu sed
inver ter inp u ts (IC 4 )
N C
N C
5
7
4
N C
3
N C
1 0
1 3
Q A0 Q A1 Q A2 Q A3
IC 1 a
D A
4 0 1 5
1 2
1 0
1 2
2
‘ R and o m ness’
seq u ence l eng th
8
P u sh to inter r u p t
seq u ence
N C
S2
1 6
5
Q B 0 Q B 1 Q B 2 Q B 3
IC 1 b
D B
4 0 1 5
1 5
S1
9
IC 3 a
4 0 7 0
7
R 2
1MΩ
N C
1 1
5 to 1 5 V
1 3
IC 3 a
4 0 7 0
7
N C
N C
4
3
1 3
1 0
Q A0 Q A1 Q A2 Q A3
IC 2 a
D A
4 0 1 5
1 5
1 2
1 1
2
1 6
Q B 0 Q B 1 Q B 2 Q B 3
IC 2 b
D B
4 0 1 5
C K AG N D
C L R A
C K B
C L R B
C K AG N D
C L R A
C K B
C L R B
9
6
1
1 4
9
6
1
1 4
8
V R 1
220kΩ
Anti- l o g
C l o ck fr eq u ency
p itch
R 1
330Ω 1
C 1
6 8 nF
8
7
2
V +
1 4
C W
3
4
5 to 1 5 V
R 3
1kΩ
IC 4
+
4 0 1 0 6 b C 2
1 0 µ F
IC 4
4 0 1 0 6 a
N o ise
o u tp u t
P o wer - o n
r eset
5
R 4
10kΩ
6
IC 4
4 0 1 0 6 c
0 V
0 V
Fig.10. Digital noise generator circuit for synthesisers. It has a frequency control, sequence length switch and a freeze button.
of shift registers and an exclusive OR gate. As John Clarke said in
his article, the good old MM5837 chip that integrated the whole
lot is now obsolete and expensive. Even the CMOS 4006 18-stage
shift-register used in many synthesisers, such as the Wasp, has
gone. However, there is still the 4015, a four-stage shift-register,
of which two are used in the circuit shown here in Fig.10.
The sound of the digital noise can be changed greatly by varying
the clock frequency with VR1. Slowing it down makes the sound
gradually change from normal white noise to an interesting tonal
noise (‘grey’ noise) with spaced frequency peaks. Eventually, it
just generates random clicks, like a Geiger counter. Modulating
the clock with an envelope generator or low-frequency oscillator
could give even more variation and control. Different tappings
along the shift-register are also employed to give different pattern
repeat lengths set by S1. This can range from about 30 seconds
down to fractions of a second. Interrupting the feedback with a
momentary switch (S2) freezes different bit cycle patterns, giving further weird tonal variations, where you don’t know what
you are going to get. Musicians love that switch and this circuit
is complementary to the analogue circuit.
A breadboard of the circuit is shown in Fig.11. If there is enough
demand I’ll do a PCB. However, if your soldering iron is burning
a hole in your bench, you can at least build the analogue noise
generator PCB for now!
Next month
That’s enough digital – next month we’ll return to strictly analogue when we build the noise generator!
Your best bet since MAPLIN
Chock-a-Block with Stock
Visit: www.cricklewoodelectronics.com
Or p ho ne o u r fr iend l y kn o wl ed g eabl e staff o n 020 8452 0161
Components • Audio • Video • Connectors • Cables
Arduino • Test Equipment etc, etc
Vi s i t o u r Sh o p , Ca l l o r B u y o n l i n e a t :
w w w .c r i c k l e w o o d e l e c t r o n i c s .c o m
Fig.11. Breadboard of digital noise generator (I’ll do a PCB if
enough people want it).
Practical Electronics | April | 2020
020 8452 0161
Vi s i t o u r s h o p a t :
40- 42 Cr i c k l e w o o d B r o a d w a y
Lo n d o n NW2 3ET
61
