This is only a preview of the March 2025 issue of Practical Electronics. You can view 0 of the 80 pages in the full issue. Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Articles in this series:
Items relevant to "Mains Power-Up Sequencer, part two":
|
AUDIO
OUT
AUDIO OUT
L
R
By Jake Rothman
Using Capacitors for HiFi, Part 1
H
aving read the editor’s article
“All About Capacitors” in the
December 2024 issue, I thought
a bit more detail could be added about
their audio aspects, which differ from
other areas of electronics. That took
me down quite a rabbit-hole, as you
shall discover!
Because high-quality audio is a
relatively small area of the industry,
audio engineers have had to develop
their own specific capacitor knowledge
base. This is almost passed down in the
form of folklore. Consequently, there is
a huge amount of “audiofoolery”, ripoffs and falsehoods out there.
It’s a bad idea to ask about capacitors
on audio forums. They tend to be
populated by musicians and DIY HiFi
buffs who will put subtle artefacts
and their own biases before rational
engineering. It’s quite likely you will
find there somebody who suggests recapping your whole system with milspec Teflon capacitors.
Quotes such as “I had my whole mixer
recapped by Quantum Dielectric Energy
Systems.com and it sounds wonderful
and it only cost me $2000” are common.
Often, so-called upgraded PCBs have
mechanical problems. A classic
error is hanging an 80mm-long axial
polypropylene capacitor on the pads
designed for a 2.5mm-pitch electrolytic.
However, within this mass of
falsehoods are some important truths
and circuit techniques. Capacitor
sonics are a subtle second-order
effect. Capacitor distortion was once a
contentious issue; traditional objective
engineers like Peter Baxandall and
Douglas Self didn’t think it was
significant enough to quantify.
It was only when Douglas Self, with
his Blameless amplifier, got the total
harmonic distortion (THD) level down
to 0.001% that the capacitor-induced
tip-ups in distortion at the low and
high frequency ends were noticed and
investigated
Engineers such as Ben Duncan
and Graham Nalty in the 1980s
(who’ve written for PE) also noticed
capacitor effects, but the necessary
instrumentation, such as the Radford
Low Distortion Oscillator (LDO), wasn’t
readily available.
Capacitor non-linearity
The main concern for audio engineers
is avoiding non-linearity, which
manifests itself as harmonic distortion.
This does occur with capacitors, but
they are very low down on the list of
Total Harmonic Distortion (%)
0.5
0.2
0.1
0.05
Dielectric absorption (DA)
0.02
0.01
.005
.002
.001
.0005
.0002
.0001
20
50
100
200
500
1k
Frequency (Hz)
2k
5k
10k
20k
Fig.1: bad capacitor distortion: the typical rapid rise below 1kHz from a 4.7µF 10V
cheap tantalum bead. This was 1V RMS into 600Ω load; the -3dB point is 56Hz.
18
contributors. Loudspeakers are the
worst offenders, with semiconductors
and circuit design next.
Once there are enough active devices
with plenty of negative feedback (NFB)
in a good design, the total harmonic
distortion due to these can drop below
0.001% (sometimes approaching even
0.0001%, which is just 1ppm!). This
is the point where capacitor distortion
can easily dominate that part of the
circuit.
This is because capacitors are usually
not within the negative feedback loop,
so are not linearised. Even if they
are, because an amplifier’s open-loop
gain is finite, the effects of their nonlinearities on signals can’t always be
fully cancelled.
The subjective effects of capacitors
are also overstated, especially since
the distortion produced tends to be
benign second and third harmonics,
as opposed to distortion from op-amps
and other high NFB amplifiers, which
generate a spray of dissonant higher
order harmonics.
I get annoyed when I test an expensive
piece of audio equipment exhibiting
distortion in the region of, say, 0.05%
or more and finding expensive highlinearity capacitors fitted. What’s the
point, besides marketing? They would
have been better off spending that
money on something else!
Known as DA or ‘soakage’, this effect
is a slow release of charge as dielectric
molecules return to their original
random alignment after initially
discharging a capacitor. This is why
a capacitor that has been discharged
can charge itself up a bit again later.
In audio work, this can cause a small
amount of very low-frequency shift
that is subjectively unnoticeable, in
my opinion.
In electrolytic capacitors, the DA
can be as much a 3%, whereas for
plastic-film types, it is much lower. It
can be hidden in electrolytics due to
leakage. The rectification, storage and
Practical Electronics | March | 2025
Bass effects
The distortion from capacitors is
similar to that from magnetic sources,
such as balancing transformers.
It’s said that the distortion from
magnetic tape and transformers is
subjectively enhancing, and yet the
similar capacitor distortion is bad, even
though the capacitor distortion is an
order of magnitude less.
Fig.1 shows a bad case of capacitor
distortion caused by a cheap tantalum
bead. This sort of problem was common
in amplifiers from the 1970s, but
hidden by other distortions.
In my opinion, all distortion is
bad for HiFi and studio monitoring,
because it leads to intermodulation
and a consequent lack of clarity and
muddle. However, for individual
sounds within a mix, such distortion
can be enhancing.
An occasional problem all passive
components can suffer from is bad
internal contacts. These can be
identified as the 30kHz third harmonic
of 10kHz by the equipment shown in
Fig.2. Most manufacturers don’t screen
for this, so the odd rogue capacitor can
be found that has bad distortion across
the whole frequency range. Luckily,
these are rare.
"Capacitosis"
Fig.2: a component linearity tester at Charcroft’s capacitor and resistor plant in
Wales. It screens for bad contacts, which cause distortion.
cone-drift effects in loudspeakers are
much worse.
No distortion harmonics are generated
from DA since it’s a linear effect. On
transient signals, there may be a small
absorption of energy and a release in
the form of a tail. It is this that makes
some engineers such as Jung and Marsh
believe audible distortion must occur.
Maybe it could lead to bias modulation
and a possible variation in THD.
It is a real concern with sample-andhold circuits, though, as used in ADCs
and analogue synthesisers. DA is bad in
paper, mica, glass and Mylar/polyester
capacitors. Non-polar dielectrics, such
as polystyrene, polypropylene and
Teflon/PTFE don’t exhibit it.
A usable model for a polyester
Practical Electronics | March | 2025
capacitor is 1µF in parallel with
two series RC networks: 6nF/1GΩ
and 3nF/200kΩ. I have read some
suggestions that DA is bad for
asymmetrical signals in audio,
but I can’t see how it can
add additional frequency
components/distortion.
Fig.3:
colourful
Phillips/
Mullard C280
capacitors in...
Despite all the evidence that capacitor
distortions are a subtle second-order
effect, capacitor foibles have achieved
a notoriety out of all proportion to
their reality in the audio community,
which is a strange marriage of art and
technology.
I have a theory that it’s psychological,
that people are influenced by what
they see, rather than what they hear
or measure. I’ve christened this
phenomenon “capacitosis”. Maybe it’s
because capacitors are the most visually
diverse and colourful of all electronic
components. A mixed bag of surplus
capacitors from the late
1970s looked like a bag
of sweets!
a
Colorsound
Powerboost
pedal.
19
Fig.4: the NAD 3030 amplifier – a classic, discrete and well-designed HiFi amp.
They certainly provided an initial
attraction for me as a child; I remember
collecting and arranging them. The
Mullard C280 series “liquorice allsorts” (or “tropical fish” in the USA)
capacitors were my favourite (Fig.3).
They are the only capacitor where you
can read the value at any angle two
metres away!
The area of subjective capacitor
analysis is very similar to wine tasting.
I suspect that extremely objective, costdriven engineers would go to a wine
tasting with a pH meter.
I have suffered from “capacitosis”
myself. This happened with a NAD
3030 amplifier (Figs.4 & 5). I put
£30 worth of new pretty ‘audiophile’
capacitors into one channel and left
the other as was.
Upon listening, I was convinced
the modified channel was the best
and then found I was accidentally
listening to the other side, with the
original 5p Chinese grey electrolytics,
green Mylar films and brown ceramic
discs.
When I sold the unit, I hoped the
recipient would notice and say
something like, “I think it sounds better
on the left side or something”. They
never did. However, the NAD circuit
did use the capacitors according to the
basic audio engineering capacitor rules,
ensuring the distortion was minimised.
Those rules are:
1) Bigger is better
This is a rare case of bigger is almost
always better, both physically and
in capacitance. The linearity of a
given type of capacitor depends on
the voltage stress applied across the
dielectric in terms of voltage per metre.
Large physical size is the reason a 250V
polyester capacitor produces five times
less distortion than a 50V component
of the same capacitance.
That is not to say you can’t get good
performance with tiny capacitors – you
can, but you have to be more careful
to choose the right types and use them
appropriately (more on choosing the
type later).
A big old-fashioned capacitor will
have less distortion than its microsized surface-mount equivalent. This
is why designers always ensure (with
coupling capacitors) that there is very
little voltage drop at low frequencies
by making the capacitance very high.
Typically, one makes the -3dB point
2Hz, or 10 times less than the lowest
required frequency.
One problem with this approach is
that the large capacitance values take a
long time for their DC level to stabilise,
causing long settling times when
powering up and thumps on switchoff. This is why 2Hz is the sweet spot;
any higher and you risk introducing
distortion; any lower, and the settling
time becomes noticeable.
2) Low dielectric constant caps
The dielectric constant, K, is the
amount that the dielectric multiplies
the capacitance compared to a vacuum
Fig.7: tough side-cutters
work well to open most
components. The
aluminium can then be
unwound, away from the
rubber bung.
Fig.6 (left): I love demolishing things to see what’s
inside, like this electrolytic capacitor! Notice the coarse,
fibrous separator paper.
20
Practical Electronics | March | 2025
Fig.5: I swapped in expensive
capacitors on one channel of this
NAD 3030 amp and left the other
original. I and a 27-year-old listener
couldn’t hear any audible difference.
The design proved a high-voltage
polymer electrolytic in the power
amp bootstrap circuit was safe, and
tantalum types in certain positions
did not increase distortion.
in its place. The higher the K value,
the worse the audio performance. As
K increases, the smaller the capacitor
is for a given capacitance. A case of
bigger is better again.
Low dielectric constant (K=2-3)
dielectrics, such as polystyrene and
polypropylene, involve non-polar
molecules. That is, the molecule has a
uniform electric charge. Non-polar plastic
films also have a negative temperature
coefficient (tempco) that cancels out with
the positive tempco of metal film resistors
when they are combined (eg, in low-pass
or high-pass filters).
Most other plastic films used in
capacitors, such as polyester, have
polar molecules, with a charged end.
Fig.8: special Black Gate electrolytic
capacitors for audio use by Rubycon.
They are so expensive that I always
reuse them if their ESR is okay.
Practical Electronics | March | 2025
This causes them to become aligned
with the applied electric field, resulting
in higher distortion.
3) Use electrolytics properly
Only use electrolytics (aluminium
or tantalum) when high capacitance
values are required and don’t use them
for audio filtering (eg, avoid using them
in low-pass or high-pass filters). They
are generally only used for AC coupling
or supply bypassing/filtering in HiFi
designs.
Any capacitor up to about 2.2µF can be
replaced with a plastic film type. Wima
makes a very small outline polyester
series, MKS 2, which will outperform
any electrolytic for distortion. The
MKS2B051001N00JS in 10µF/50V is
only 11mm wide and 16mm high with a
5mm lead pitch. However, it costs £3.50.
Using an electrolytic or tantalum of
470nF or even 1uF is silly, except in
SMT, where tantalum capacitors will
give lower distortion than X7R ceramics
(SMD plastic film caps are available if
a little expensive).
Small radial electrolytic capacitors
are also less reliable than larger case
sizes because the surface area and
resulting electrolyte evaporation is
proportionally higher. Fig.6 shows
an unwound electrolytic, while Fig.7
shows how to open them up.
Electrolytic capacitors have a high
dielectric constant, in the region of
eight for aluminium oxide, 28 for
tantalum oxide and 41 for niobium
oxide. These are not particularly high,
but the capacitance-to-size ratio is
also greatly increased by surface area
multiplication from foil etching and, in
the case of tantalums, sintering, where
tantalum powder is compressed into a
porous sponge.
These K values are quite low compared
to high-K ceramics, which can be as
high as 1200, but the electrolytic
oxides are also very thin and have
high voltage stress. This explains
the relatively high distortion (nonlinearity).
Bipolar or non-polarised wet
aluminium electrolytics always give
lower distortion than the polarised
variety, even when used with a
polarising voltage.
T his is becaus e the nat ural l y
oxidised cathode/negative foil (which
breaks down at around 1.5V, causing
extra distortion in normal polarised
capacitors) is formed to the same
voltage of the anode foil, so it can take
the same voltage in either direction.
Some of the high-quality ‘audio’
electrolytics such as the Black Gate
(Fig.8) are asymmetric bipolars, in
that the cathode foil is formed to an
intermediate (eg, 6V) rating, which
gives the best characteristics of both
types.
4) Be careful with tantalum caps
Solid tantalum type capacitors
produce up to 10 times more distortion
than wet aluminium electrolytics
because of the higher K and the solid
manganese dioxide, which has a nonlinear resistance. So this has to be taken
into account in the circuit design.
For a given CV product, eg, 100µF
× 10V = 1000C, they cost six times
more than wet aluminium. However,
they don’t dry out and, if used for
low currents, such as coupling rather
than decoupling, they last for decades,
whereas wet electrolytics are often the
first components to fail.
The leakage is 10 times lower for
tantalum, and they don’t need to
stabilise and reform at power-on after
long periods of disuse. This gives
much lower clicks and thumps, and
equipment can be used immediately.
Tantalum pentoxide is a more
stable material than aluminium oxide
and, once formed, does not degrade.
Therefore, they can live without a
polarising voltage and be stored for
decades.
Tantalum capacitors are an important
part of the sound character of the
famous Neve modules. Fig.9 shows
21
some excellent metal-cased tantalum
types that Neve used.
Some engineers, like Rod Elliott
(https://sound-au.com), disapprove of
tantalums, having experienced a lot of
shorts with tantalum beads. However,
military and aviation engineers love
them.
I suspect the antipathy comes from
having to fix a lot of circuits where
they have been incorrectly used.
When tantalum capacitors fail, they
go short circuit, unlike wet aluminium
type capacitors, which slowly go
open-circuit. Also, the bead type can
catch fire dramatically because the
manganese oxide solid electrolyte acts
like an oxidising agent.
Unlike wet electrolytics, which suffer
no effect from being run at full-rated
voltage, the reliability of tantalums
increases with a reduction in voltage.
Derating, running at no more than 70%
of rated voltage, is good, conservative
design.
It’s best to limit the surge current
with solid tantalum capacitors to 1A,
especially if running them at their full
voltage rating. For example, if you have
a tantalum bead across the output of a
15V regulator, it should ‘see’ a series
resistance of 15Ω. This is rather a
lot, and would reduce its decoupling
ability. Hence my advice not to put
tantalums across power rails.
Many designers did, however, because
National Semiconductor used to
recommend tantalum capacitors
across the outputs of
their regulators.
Their ESR of around
0.7Ω to 2Ω is ideal
for damping the
inductive output of linear regulators.
One way round the surge problem
is to configure the regulator for soft
starting. Another trick is to use a series
decoupling resistor in the power rail.
That seems rather old-fashioned today,
because it reduces the regulation.
It is now possible to get surge-tested
tantalum capacitors, such as the Kemet
T495 series, which require no derating
or current limiting.
It’s difficult to get high-CV products in
solid tantalum. The biggest I use is the
1000µF 10V AVX F721 for mic preamp
gain control blockers (Fig.10). For SMT,
they are the better choice than wet
aluminium, since they are not stressed
by the baking involved. Also, they seem
to solder onto the board better, being
shaped like resistors.
I like tantalum capacitors; used
correctly, they increase reliability and
reduce clicks... at a price.
5) Avoid most ceramics
Don’t use high-K (X7R, Y5V, Y5U,
Z5U) ceramic capacitors for coupling
(or filtering). They distort much
more than tantalum types. Although
excellent for power rail decoupling,
high-K ceramics self-modulate their
capacitance with applied voltage,
causing very high distortion, even if
just used for coupling.
SMT electronics usually use high-K
ceramics, which has led to SMT
assemblies getting a bad name with
regards to audio quality. However, if
you can choose an NP0/C0G capacitor,
it will generally perform very well.
They can be quite expensive and large
for a given capacitance and voltage
rating combination, although they
still can be cheaper than plastic film
in many cases.
For higher capacitances, especially
in coupling roles, you can replace
them with tantalum capacitors. For
100nF and similar values, they come
in similarly sized cases.
High-K ceramics can also act as
piezoelectric transducers and can pick
up or emit music with very strange
results. The construction of piezo
tweeters is basically a lead titanate disc
‘ceramic capacitor’ bonded to a small
paper cone.
Real differences
Generally, capacitor effects don’t
show up in double-blind listening
trials or ABX tests. Where I have heard
definite differences are in the most
highly stressed audio application, the
tweeter section of loudspeaker passive
crossovers (Fig.11).
Just the equivalent series resistance
(ESR) rating of the capacitor can make
a difference. The 0.3–1.0Ω ESR of a
bipolar electrolytic used in cheaper
systems will make a significant
difference in a 4Ω loudspeaker system,
compared to the 0.01Ω ESR of a plastic
film capacitor. This could mean a 25%
reduction in signal level, so everybody
would hear if the electrolytics were
upgraded.
When I designed crossovers, I
included this ESR in the circuit.
When people upgraded my circuits
with plastic-film capacitors, it tonally
unbalanced the sound. I told them to
put series resistors in to fix it.
Interestingly, special bipolar
electrolytics using plain rather than
etched foil are specially made for
crossovers and are called low-loss,
often marked LL, as shown in Fig.12.
The inductor distortion in passive
crossovers is much worse than the
capacitors, especially if a ferrite or
iron core it used (air-cored inductors
are preferred for this reason, but are
rather bulky).
Fig.10: tantalum capacitors don’t get
to very high values. This AVX F721 is
rated at
1000µF
and 10V,
which is
about the
practical
limit.
Fig.9: this Neve output module uses five
tantalum bead capacitors, including one
metal-cased type. I have a huge stock of RAF
surplus tantalums that I use and sell to Neve and
Naim module builders.
22
Practical Electronics | March | 2025
amps and preamps with capacitor
coupling that have THD+N levels
down to 0.0004% and below, with very
little distortion rise at either end of the
frequency spectrum, and without using
any exotic or overly expensive parts. So
total avoidance of any capacitors in a
HiFi circuit seems unwarranted.
Group delay
Fig.11:
the high-pass
tweeter section of
a passive crossover reveals
audible capacitor differences. This
BBC LS3/5A crossover uses plastic-film
capacitors throughout.
The only good capacitor is
a dead no capacitor
This is a more extreme philosophy
popular among some American
and Japanese designers, almost
‘capacitophobic’, where all coupling
capacitors are eliminated and the whole
circuit is DC-coupled end-to-end.
This technique gives flat distortion
performance to DC, but the system may
destroy itself and an expensive pair of
speakers if a DC offset develops and
there is no DC fault protection circuitry
present.
There is no need for audio to go
down to DC; doing so invites a whole
load of extra problems, such as switch
clicks, scratchy pots and thumps. ACcoupling also isolates one stage from
another, generally confining faults to
one stage.
The need for low bias current/offset
voltage audio op amps (generally
single-sourced JFET input types) and
DC servo circuits increases silicon
costs by around five times compared
to using audio bipolar devices, such as
the venerable (and excellent) NE5532.
This chip was designed for AC-coupled
audio, so its DC performance is poor.
Silicon Chip has designed numerous
As well as acting as high-pass filters
and giving rise to bass loss, coupling
capacitors also have an associated
phase shift. The rapid change in phase
can cause a disconnect between the
bass fundamental and associated
transients.
How rapid the phase change is can
be represented as group delay. This
delay becomes noticeable on kick
drums, where the ‘thump’ may seem to
come late, after the ‘click’. It counterintuitively gives a bass boost effect
to male voices, making them sound
‘chesty’. This effect begins to occur at
a 10 times higher frequency than the
cut-off frequency.
Laurie Fincham, KEF’s principal
engineer, became aware of this effect
when developing the loudspeaker CUBE
bass equalisers. He built two systems,
one with coupling capacitors and
one without, and gave a comparative
demonstration at an Audio Engineering
Society meeting in London.
The circuit shown in Figs.13(a) &
(b) gives an example. It is impossible
to remove all phase shifts and highpass filtering from the reproduction
chain. The microphone and sealed-box
DC-COUPLED AMPLIFIERS
TOTAL: 4 POLES
MICROPHONE
(2 POLES)
Fig.12: sometimes, low-loss bipolar
electrolytics like 3.3µF Callins Elcaps
are used in crossovers, such as this
one from a JR149.
Fig.13(a): some audio systems use no
coupling and DC unity-gain capacitors.
This minimises the bass phase shift.
+
SEALED
BOX
SPEAKER
(2 POLES)
–
AC-COUPLED AMPLIFIERS (7 POLES)
MICROPHONE
(2 POLES)
TOTAL: 13 POLES
+
Fig.13(b): most audio systems have lots
of coupling capacitors. If used incorrectly,
they can spoil bass transient response
and increase distortion.
Practical Electronics | March | 2025
BASS
REFLEX
SPEAKER
(4 POLES)
–
23
INPUT C1
POWER
AMP
R1
15kW
R2
470W
+
INPUT 47nF
OUTPUT
47kW
8W
–
15kW
RF * NON-POLARISED WET
ALUMINIUM TYPE IS THE
BEST COMPROMISE
C2*
100mF
RF1
470W
CF1
CF2*
1.5nF
33kW
RF3
* POLYESTER FILM
GAIN = 1 + R2 || R3
R4
RF1.CF1 = RF2.CF2
Fig.15: this shows the Cherry lowfrequency compensation scheme,
which minimises the signal’s phase shift
at low frequencies like 20Hz.
speaker will have inherent high-pass
filtering of second order.
However, it is possible, with a
complementary bass boost to a sealed
box speaker, to give a flat response
down to 5Hz, where the whole system
will roll off with a fourth-order slope
and minimal phase shift.
Contrast this with a normal
reproduction chain, with an additional
seven coupling capacitors and reflex
speaker, shown in Fig.13(b). Now the
total roll-off is 13th order. Add in a
modern sixth-order roll-off equalised
reflex active loudspeaker system and
the total roll-off is of the 15th order.
This leads to a large phase shift,
approaching 1350º, and a consequent
huge group delay, resulting in a
‘warm’ and ‘boomy’ sound. We’re now
accustomed to it; it is typified by Sonos
speakers, Bose Wave radios, Genelec
monitors and American podcasters.
Most people love the effect, but it is
also completely inaccurate, so the
‘capacitophobic’ people may be onto
something.
I was surprised to read a letter raising
all of this by the late mixer designer
Barry Porter in HiFi News and Record
Review magazine, September 1987.
He also said he used non-polarised
electrolytics for coupling. It seems he
was ignored, like many people ahead
of their time.
non-inverting amplifiers, such as most
audio power amps shown in Fig.14 (C2
here), is critical. After all, the linearity
of a feedback system can only be as good
as its feedback network.
Australian power amplifier designer
Edward Cherry investigated the linearity
of electrolytic capacitors back in the
1970s and found them to be much better
than originally believed. He also noticed
that tantalum beads, which were in
fashion at the time, were worse.
He used to compensate for the LF
phase shift in his power amps, such as
in his ETI May 1983 design, using the
circuit shown in Fig.15. The network
CF2, RF3 in series with the feedback
resistor R F 2 reduces the negative
feedback (NFB) in the region across
the same range of frequencies that the
attenuation of the lower feedback arm
due to capacitor CF1 boosts it.
This allows the main feedback capacitor
(CF1) to be just 47µF. Without the added
compensation network, it would need
to be 1000µF. He demonstrated this
using a 20Hz square wave, giving the
same minimal square-wave tilt for the
big capacitor and his circuit with the
lower value.
The voltage attenuation of the feedback
network makes it possible to use lowervoltage capacitors than expected. I have
CAPACITOR
10kW UNDER
TEST
OSCILLATOR
DISTORTION
ANALYSER
RLOAD
600W
Fig.17: I used a 100µF 630V PP
capacitor as a reference and for bias
blocking. Its distortion was below the
resolution of the AP analyser.
24
Total Harmonic Distortion (%)
100mF
630V
POLYPROPYLENE
−3dB = 12Hz
C2
3.3mF
FILM
R3
3.9kW
R4
330W
Fig.16: one way of avoiding an
electrolytic capacitor in a feedback
network.
seen 6.3V used with no ill effect. Due to
NFB action, the maximum signal voltage
across the capacitor is the same as the
input voltage, typically 775mV RMS
maximum.
If the amplifier latches up to one of the
power rails, the feedback resistor limits
the current through the capacitor to a
safe value and leakage currents stop the
voltage building up.
If you want to be extra safe, place 3.3V
back-to-back zener diodes across C2 in
Fig.14, since occasionally the charge
does build up and destroy the capacitor
if the amplifier is offset for some time. I
have seen this with polymer electrolytics.
An alternative circuit that can be
used to reduce the size of the feedback
capacitor is shown in Fig.16. This allows
a polyester capacitor to be used rather
than electrolytic.
Both of these circuits involve the use
of a high-value feedback resistor, which
passes the DC bias current to the input
transistor long-tailed pair. This will
increase the offset voltage.
Biasing
Polarised capacitors give lower
distortion with DC bias, especially solid
tantalum types. Not much is required;
3–5V is usually enough.
I tested the effects of bias by connecting
0.5
Edward Cherry
+VBIAS
47kW
R2
DCFB
ACFB
Fig.14: capacitor C2, drops the DC
gain to unity. A bipolar electrolytic, like
Nichicon’s UVP1E101MPD (100µF
25V), gives low distortion at low cost.
The lower-arm feedback capacitor
used to reduce DC gain to unity in
OUTPUT
R1
47kW
−37V
15kW
RF2
47mF
ALUM.
INPUT
+37V
C1
1mF
0.2
0.1
0.05
0.02
0.01
.005
.002
.001
.0005
NO BIAS
3V BIAS
.0002 9V BIAS
.0001
20
50
100
200
500
1k
Frequency (Hz)
2k
5k
10k
20k
Fig.18: solid tantalum bead capacitor distortion curves, 1V RMS into 600Ω with a
100µF 10V Kemet type. No bias THD = 0.0018%, 3V = 0.0009%, 9V = 0.0007%.
Practical Electronics | March | 2025
MAKES 50mF NP CAPACITOR
Fig.19: some nonpolarised electrolytics
need a DC pulse
around every 10 days
to stop corrosion inside.
Otherwise, internal gas
generation can cause
swelling in storage.
a very expensive 100µF polypropylene
cap in series with the capacitor under
test for DC blocking (Fig.17), which
had much lower distortion than the
capacitor under test. The bias was
applied to the junction of the two
capacitors via a 10kΩ resistor.
I used to worry about using polarised
electrolytics in circuits with no DC
bias, such as op amp systems running
on dual rails. Luckily, the corrosion
inhibitors in most modern electrolytes,
along with the DC pulses that occur on
switch-on and switch-off, are enough
to prevent degradation.
Solid tantalum capacitors don’t
degrade without bias, but the distortion
is higher (Fig.18). Wet electrolytics
do suffer from prolonged storage,
especially the non-polarised (NP)
types. In most NP data sheets, it says
to polarise one way, then the other
every 250 hours. If you don’t do this,
the capacitors can pop in the drawer
after a few years.
I had this problem with some Suntan
CD71 capacitors from Rapid electronics
(Fig.19).
Distortion cancellation
other against reverse polarisation;
second, the second harmonic distortion
(normally the biggest component of
the THD) cancels out. This also occurs
with the input capacitors on differential
amplifier inputs.
Finally, there is a degree of selfbiasing that occurs from the diode
action when each capacitor is reversepolarised. I’ve noticed the effect is
more powerful with solid tantalum
capacitors than wet electrolytics. This
is because of their lower leakage and the
fact that there is no additional cathode
foil capacitor in series.
This self-biasing voltage takes a few
cycles to build up and, with music
signals, it will be unstable. It is easy
to measure it with a scope or digital
multimeter; it can build up to a few
volts. It’s much better to apply a stable
bias to the centre junction, as shown
in Fig.20(b).
To get best distortion cancellation,
the capacitors should be matched in
second-harmonic generation. Using
capacitors from the same batch usually
works. An example of the reduction in
distortion with tantalum capacitors is
shown in Fig.21.
Wet electrolytics can be very linear
when used at high levels back-to-back
with bias. I had to thrash these 100µF
35V Suntan units at the full 26V RMS
output of the Audio Precision to generate
the distortion also shown in Fig.21, with
and without a 5V bias.
I adapted the Cherry circuit by adding
a biased pair of back-to-back 100µF 20V
tantalum bead capacitors, as shown in
Fig.22. The distortion was very low
0.5
0.5
0.2
0.1
0.05
0.2
0.1
0.05
0.02
0.01
.005
.002
.001
.0005
100µF Suntan 35V Al (NO BIAS)
100µF 10V Kemet Ta
.0002
100µF Suntan 35V Al (BIAS)
.0001
5k 10k 20k
50 100 200
500 1k
20
2k
Frequency (Hz)
Total Harmonic Distortion (%)
Total Harmonic Distortion (%)
Putting two polarised capacitors backto-back, as shown in Fig.20(a), can
reduce distortion. It does this in several
ways: first, one capacitor protects the
Fig.20(a):
back-to-back INPUT 100mF 100mF OUTPUT
polarised
OR
capacitors
100mF 100mF
OUTPUT
make a non- INPUT
polarised unit.
+3–8V
Fig.20(b):
applying
a centre
OUTPUT
bias.
10kW (AL)
100kW (TA)
INPUT
100mF 100mF
10V
10V
(see Fig.23). Sometimes it is desirable
to obtain a long lifespan, especially at
studio rack temperatures, by removing
wet electrolytics.
One possible problem with this circuit
is switch on/off thumps and slow
ramping of the power supplies reduces
it. A discharge diode for switch-off may
be a good idea. This can be reversebiased diodes across the base-emitter
junction of the input transistors; a
sensible, cheap precaution against
transistor damage.
Next month
I have plenty more information on
capacitor distortion and how to avoid it.
I will continue on this topic next month,
including information on the effects of
physical capacitor construction on
performance, a discussion of classic
and special types of audio capacitors
and a ranking of capacitors for audio
PE
use.
+15V
4.7nF
47kW
7
3
INPUT
2
NE5534
4
GAIN = 20× (+26dB)
OUTPUT
6
2.2nF
POLYESTER
−15V
15kW
750W
100mF
20V
TANT.
100mF
20V
TANT.
33kW
+15V
1MW
+5V
THD+N ≈ 0.0017%
<at> 1kHz
470kW
Fig.22: an adapted Cherry circuit with
biased 100µF tantalums back-to-back.
0.02
0.01
.005
.002
.001
.0005
.0002
.0001
20
50
100 200
500 1k
2k
Frequency (Hz)
5k
10k 20k
Fig.21: back-to-back tantalum beads, 4V DC bias at 4V RMS Fig.23: the distortion of Fig.22 delivering 20V RMS into 600Ω.
into 600Ω, plus two 100µF Suntan 35V electros back-to-back. It has a slight LF rise but not bad for a normal aluminium cap.
Practical Electronics | March | 2025
25
|