Capacitor Sounds 6 - 10 to 100 µF capacitors and 100 Hz measurements.

Updated & expanded March 2003

Original version Pub. Electronics World Jan 2003 - C. Bateman

Readers of my articles have now seen that many capacitors introduce distortions onto a pure sinewave. In some instances this
distortion results from the unfavourable loading the capacitor imposes onto its valve or semiconductor driver. More often, the
capacitor generates the distortion within itself.

For 1 µF, the lowest distortions are generated by choosing a film capacitor or a Bi-polar electrolytic. Polar aluminium
electrolytics produced considerably larger distortions, even when tested with small AC signals. Ref. 1

While high capacitance, low cost, electrolytic capacitors, can be obtained from distributors. Low cost metallised film capacitors
are restricted typically to 10 µF at 100 volt and 22 µF at 63 volt.

In this final article, which completes last months discussion on electrolytic capacitors, we explore whether a metallised film
capacitor or an electrolytic is our economic, low distortion choice, for capacitors at 10 µF to 100 µF values.

Test frequency.
To avoid overstressing large value electrolytic capacitors, we should reduce our test signal frequency towards 100 Hz. But
sufficiently above or below this frequency, to discriminate between harmonics of the supply mains and the test capacitor.

With minor changes in capacitance values, the PCB used for our 1 kHz oscillator can provide an exceptionally low distortion
100 Hz test signal. Ref. 2 In similar fashion the PCB used for our 1 kHz notch filter and pre-amplifier can also be used at this
frequency. Ref. 3

The AD811 low distortion buffer can output 40 mA. At 100 Hz using a 100Ω series resistor, it can develop a 5 volt test signal
across a 10 µF capacitor. Using a 10Ω resistor, 0.5 volts could be developed across a 100 µF capacitor.

These test voltages are more than sufficient to distortion test any electrolytic capacitor up to 100 µF. However when I designed
the test instruments I decided to provide the ability to measure both values of film capacitors to 5 volts. To produce a larger test
signal with 100 µF capacitors, a more powerful buffer must be used. A low distortion circuit able to drive up to 400 mA has
been designed but needs a different PCB. see Fig. 1

Fig. 1) High power buffer provides low distortion, a gain of two and a 400 mA output. It can develop more than five volts
across a 100 µF capacitor via a 10 Ω current limiting resistor.

1
When testing large value capacitors, a four terminal test system is preferred. Four BNC connectors are provided. which accept
either Hewlett Packard capacitor test jigs or four discrete cables and crock clips. see Fig. 2

Fig. 2) The higher power

100 Hz test system.
Four BNC connectors,
are arranged to accept
Hewlett Packard test jig
fixtures.

The DC bias network

inserts between this
buffer amplifier and
the test jig fixture.

Alternately instead of the

Hewlett Packard test jigs,
this buffer amplifier can
be used with four BNC
test leads fitted with
‘crock’ clips.

Box 100 Hz test equipment.

The oscillator and notch filter printed boards can be used at other frequencies by scaling the values of a few capacitors. Ref. 2

Oscillator board.
For 100 Hz use 100 nF 1% metallised Polypropylene for C1, C2, and C3. Bypass R16 by a wire link. To differentiate between
test capacitor and mains frequency harmonics, replace R23, R24 and R25 with wire links.

Notch filter/pre-amplifier board.

For 100 Hz use 100 nF 1% metallised Polypropylene for C41, C42, C43, C44, C47 and C48. Use 47 nF 1% metallised
Polypropylene for C45 and C46. Use 10 nF 1% metallised Polypropylene for C49.

Output Buffer.
At 100 Hz, 10 µF capacitors can be tested to 5 volts, using the AD811 output buffer amplifier described. Ref. 3 Adding a 10
Ohm current limiting resistor allows 100 µF to be tested to 0.5 volts.

To fully test 100 µF capacitors, a higher power buffer amplifier is needed. It should develop at least 5 volts signal across a
100 µF capacitor via a 10 Ohm current limiting resistor. I designed a buffer amplifier and printed circuit board, able to drive up
to 7 volt or 400 mA, with extremely low distortion. An Elantec EL2099CT output amplifier is used with an input buffer. This
can be an OP295, OPA2134 or an NE 5532A, by connecting one link.. I used an OPA2134 in my prototype. see Fig. 1

Larger decoupling capacitors are used with 1.5 Amp stabilisers. A Perancea 75 by 50 mm PCB case serves as heat sink for the
EL2099CT and the stabilisers. Apart from these changes, the buffer amplifier schematic circuit and the current limiting
resistors/switch follow the approach previously used for my 1 kHz AD811 output buffer.

When testing 100 µF, a four wire test method should be used. Four BNC connectors, two to output the test current and two to
measure the capacitor distortions, are spaced at 22 mm centres to fit Hewlett Packard capacitor test jigs alternately four discrete
BNC cables and crock clips can be used.

To measure capacitors larger than 10 µF with DC bias voltage, a DC blocking buffer circuit as already described but made with
larger capacitors is essential.

Two 50 µF 450 volt metallised Polypropylene motor run capacitors, replaced the 11 µF current carrying capacitors of my
1 kHz design. Three 3.3 µF MKP capacitors provide 10 µF for the voltage measuring circuit. These components were mounted
in a die-cast box and hardwired.

Four BNC connectors, were mounted on opposite sides of this box, to mate with my 100 Hz output buffer amplifier and the
Hewlett Packard capacitor test jigs. see Fig. 2

A selectable DC bias voltage was provided, by mounting 20 AA cells and a range switch, in a second die cast box. This was
used with both DC blocking buffer designs.
2
Tantalum bead capacitors.
Some audio power amplifier designs have used small Tantalum bead capacitors, with apparent success. Initial measurements of
a number of Tantalum capacitors revealed large distortions. Measured at 0.3 volts with and without DC bias, my Tantalum
capacitor stocks produced at least ten times more distortion than found with low cost polar Aluminium electrolytics. I decided
to exclude Tantalum bead capacitors from further tests. see Fig.3

Fig. 3) Distortion of this

Tantalum bead capacitor,
is ten times worse than
found with similar value
and voltage aluminium
electrolytic capacitors.

Distortion does reduce

slightly with application
of DC bias.

Aluminium Electrolytic capacitor myths.

As with other capacitor types, much has previously been written about the sound distortions electrolytics produce. As a result,
many false myths, specific to electrolytics have emerged. Most were discussed in my last article, the remainder in this.

a) High ESR Electrolytics degrade sound quality, low ESR is always best.
b) Electrolytics are highly inductive at audio frequencies.
c) Polar electrolytics should be biased to half rated voltage to reduce distortion.
d) Electrolytic capacitor distortion is mostly third harmonic.

A working knowledge of aluminium electrolytic capacitor construction combined with careful measurements, leads to
somewhat different conclusions.

Is a low ESR/tanδδ capacitor always better ?

Capacitor makers test every production capacitor for four key parameters. Capacitance, tanδ, insulation resistance and voltage
withstand. The need to test for capacitance, insulation resistance and voltage withstand is obvious, but why test tanδ ?

The most nearly perfect capacitor needs conducting electrodes, which inevitably have some resistance. This appears in series
with the capacitive reactance to degrade the theoretical -90° of phase difference to a smaller negative angle. As a result the
complementary angle called δ (delta) increases. Every practical capacitor also incurs losses in its insulators and dielectric
system. These further degrade this phase difference, increasing the angle δ.

The ratio of these resistive losses to the capacitors reactance or tanδ is the simplest way to monitor capacitor quality. As losses
increase so does this ratio and the tangent of the angle δ, usually called tanδ. Consequently a large tanδ implies large resistive
losses. These losses do not exist as discrete resistors so are described as ‘equivalent series resistance’ conveniently abbreviated
to ESR. Tanδ and ESR are not finite values but do vary widely with change of measurement frequency.

The most nearly perfect capacitor would exhibit near zero ESR. Low ESR is essential for use in switched mode power supplies,
but does a low ESR electrolytic ensure low audio distortion ?

Of the 100 µF capacitors I tested, the 10 volt Oscon measured the lowest 100 kHz ESR of all, 0.012 Ω and 100 Hz tanδ of
0.035. It would be unreasonable to compare a 10 volt capacitor with higher voltage types so I also measured 10 volt Rubycon
YXF and Elna RSH types. The YXF ESR measured 0.550 Ω, tanδ 0.091. The RSH ESR was 0.505 Ω and tanδ 0.104.

Tested at 0.5 volt with and without 6 volts DC bias, the Rubycon YXF produced the least distortion, 0.0351% with DC bias and
0.00331% unbiased. The Oscon distorted worst of the three, measuring 0.05321% with DC bias and 0.02499% unbiased.

Clearly low tanδ at 100 Hz and low ESR at 100 kHz does not ensure low audio distortion.

3
Are Electrolytics Inductive at audio frequencies?
Radial lead electrolytics are assembled with their connecting tabs attached towards the centre of their anode and cathode foils.
Wound together this produces a near non-inductive winding. As explained in my last article, the main contribution to the
capacitor’s self inductance then comes from the connecting leadwires and tabs, and not the wound element. Ref. 4

This ‘inductive at audio frequencies’ myth is easily proved to be false. The largest capacitor I measured for distortion, the Nitai
220 µF 63 volt Bi-polar, has a case size 25×16 mm. Apart from in the power supply, this is the largest value commonly used in
an audio system. I mounted one on a test jig, its self resonant frequency was 250 kHz, well above audible frequencies. Ref. 5

At all audio frequencies this capacitor must present a capacitive reactance. Self inductance of a lesser value or smaller case size
radial lead capacitor being even less, self resonance of smaller capacitors will occur at higher frequencies. They cannot become
inductive at audio frequencies.

Exceptionally large value capacitors, as often used in power supplies, may appear as either inductive or capacitive depending
on their capacitance value, case size and their connecting leadwires/tracks. Inductance of the leadwires/circuit tracks used to
connect the capacitor, usually well exceed that of the capacitor’s own self inductance. Due to its internal series resistance or
ESR the capacitor’s phase angle will be much smaller than -90° and the capacitor will appear to the circuit as a series
combination of a resistor with a capacitor or as a DC blocking inductance in series with a resistor.

Using a Wayne Kerr B6425 precision LCR meter fitted with my Hewlett Packard capacitor test jigs, I measured a few
capacitors removed from one of my old bench amplifiers at 10 kHz, as representative of capacitors which may be used in
amplifier power supplies, to illustrate the point:-

Elna 4,700 µF 63 volt Cerafine size 82 mm by 35 mm dia, phase angle +7.5°, ESR 11.05 mΩ Impedance 18.3 mΩ
Marcon 4,700 µF 63 volt size 30 mm by 40 mm dia, phase angle -6.5°, ESR 16.65 mΩ Impedance 16.69 mΩ
Marcon 10,000 µF 63 volt size 42 mm by 65 mm dia, phase angle -14.5°, ESR 9.68 mΩ Impedance 10.0 mΩ

At 20 kHz both Marcon types remained as a capacitive reactance having a negative phase angle. At 100 kHz the Elna measured
as an 83 nH inductor, the 4700 µF Marcon as a 26 nH inductor while the 10,000 µF Marcon measured as a 256 µF capacitor.

This last value shows why it is not possible to estimate capacitor self inductance from published impedance/frequency curves.
As frequency increases, the capacitance of all aluminium electrolytic capacitors reduces, some more quickly than others. As the
capacitive reactance reduces with frequency, the capacitors ESR becomes almost a constant value. Phase angles become small
and the impedance curve becomes ‘flat bottomed’. see Fig. 4

Any calculation of resonant frequency based on using the correct self inductance value together with this capacitors nominal
10,000 µF value, obviously produces a very false result.

Measured using the Wayne Kerr B6425 with Hewlett Packard test jigs as above, this Philips 1000 µF capacitor exhibited a -6°
phase angle and 820 µF capacitance at 30 kHz. At this frequency ESR measured 61.5 mΩ, impedance was 61.6 mΩ.

Phase degrees Impedance

Fig. 4) This 1000 µF 25
100 2
volt capacitor has been
measured using two quite
1.6 different methods.
50
The Wayne Kerr bridge
1.2 up to 300 kHz, its
0 maximum frequency.
0.8
This graph was plotted up
to 10 MHz from results
-50
0.4 using my ‘High-
frequency impedance
meter’ with jigs and
-100 0 methods as in EW
0.1 1 10 100 1,000 10,000
January 2001.
Frequency kHz
Phase v Impedance As can be seen both
methods gave almost
1000 µF 25v Philips 135 Capacitor. identical results.

4
Polar electrolytics.
In my last article we saw that every polar aluminium electrolytic capacitor comprises two polar capacitors in series, back to
back. Ref. 1 Wound with an anode and cathode foil, each foil with the electrolyte, comprises one capacitor. The cathode foil
provides a larger capacitance, lower working voltage, than the anode foil.

With no bias voltage, the capacitor produced predominantly second harmonic distortion. In some instances, application of a
very small optimum DC bias did minimise this second harmonic. Increased bias however resulted in increased second harmonic
distortion.

For the 100 µF 25 volt capacitors tested for this article, optimum bias varied by capacitor, from 1.1 to 4.2 volts. Optimum bias
voltage varies with capacitor rated voltage, capacitance value and even from capacitor to capacitor within a small batch.
However the important point is that with all the polar aluminium electrolytic capacitors I tested, (several hundred in all ) this
optimum low distortion bias with no exceptions, was a small voltage. Not the half rated voltage as commonly suggested.

Bi-polar electrolytics.
A Bi-polar electrolytic is made in exactly the same way as a polar capacitor, with one important difference. In place of the
unformed cathode foil we use a second, formed, anode foil.

We still have two polar capacitances in series, back to back, both now the same value and voltage. To make the desired value,
two anode foils of double capacitance are used.

Aluminium electrolytic capacitor designers are accustomed to mixing and matching their available materials, to suit the
capacitor’s end application. So it should not surprise that some designs are semi Bi-polar, i.e. they are made using a lower
voltage deliberately ‘formed’ anode foil as cathode.

Equivalent circuit.
Using this constructional background, we deduced an equivalent circuit for a polar aluminium electrolytic capacitor. see Fig.5

Fig 5) Simplified
equivalent schematic
shows how a polar
electrolytic capacitor
behaves with AC signals,
with and without DC
bias.

Dielectric Oxide. Aluminium oxide has a ‘k’ of eight, similar to that of COG ceramics or some impregnated paper capacitors.
Ref. 6 It is higher than PET, which at 3.3, has the highest ‘k’ of commonly used film dielectrics. It is a low value compared to
the ‘k’ of some thousands, found in ‘high k’ BX, X7R and Z5U ceramics.

More significant is dielectric thickness. Aluminium electrolytic dielectric is much thinner than used in other capacitors and the
dielectric oxide film has a small but easily measured voltage coefficient of capacitance, typically +0.1% with +18v DC bias, but
this is overshadowed many times by its much larger dielectric absorption. An electrolytic capacitor is exceptionally sensitive to
dielectric absorption effects and the applied AC and DC voltages.

Voltage effects.
When our 1 µF 63 volt polar electrolytic was tested using two 0.7 volt frequencies, its third harmonic was -110 dB or 0.0003%.
It created visible intermodulation distortion. We also noted small capacitors rated at 40 to 63 volt exhibit near optimum quality.
Ref. 1

Many 100 µF capacitors will be made with lower voltage, thinner dielectric oxide, anode foil. This capacitance requires lengthy
anode and cathode foils, housed in a larger diameter can. To generate the test voltage across the capacitor, increased current
must pass through the tab connections into the winding, amplifying the affects of any non-linear resistance. It seems probable
that similar harmonic and intermodulation levels will be found but at smaller test voltages.
5
To allow direct comparison between the low cost 1 µF 63 volt polar capacitor and the physically larger Elna Silmic 100 µF 25
volt, I show its distortions measured at 1 volt. This capacitor provided the best 1 volt, no bias, results of the 100 µF polar types
tested for this article. see Fig. 6

Fig 6) The lowest

distorting 100 µF 25 volt
polar capacitor of those
tested, at 1 volt. It should
be compared with Figure
6 (Fig 7 as published in
EW) of my last article.

Both capacitors produced

similar second harmonics.
Third harmonic of the
100 µF has increased by
4.5 dB indicative of
increasing non-linearity
with increasing
capacitance values.

Lower voltage measurements.

Accurate 100 Hz distortion measurements using test signals smaller than 1 volt become quite difficult, for two main reasons.

Supply mains harmonics intrude everywhere and are difficult to reduce using a computer based system.
The smaller test signal reduces the dynamic range of our measurement, dramatically inflating indicated distortion.

For example, using a 0.1 volt test signal, my noise floor is around -112 dB, hence a perfect capacitor producing no distortion at
all will still register some 0.0005%. However if we compare the measured harmonic levels of our electrolytic with those found
for the identical measurement using a metallised film capacitor. We will see any increase in distortion caused by the electrolytic
capacitor.

To distinguish between harmonics from the mains and the test capacitor, my test frequency was displaced a few Hz away from
100 Hz. The Spectra software then ignores mains harmonics when calculating distortions. To assist visual identification, I used
‘Mains’ markers to identify mains harmonics and amplitude markers to indicate the first four harmonics from the test capacitor.

At 100 µF, a metallised Polypropylene capacitor is both large and costly. I used an assembly of 10 µF Evox Rifa MMK
metallised PET capacitors. This works well for small test voltages as a low distortion ‘reference’ capacitor. see Fig. 7

Fig. 7) Distortion plot of

an assembly of ten Evox
Rifa 10 µF 63 volt MMK
metallised PET
capacitors, tested at
0.5 volt.

This 100 µF assembly

was then used as the
distortion reference for
each 100 Hz low voltage
distortion test.

N.B. Labels ‘Mains’

indicate 150 and 250 Hz
harmonics of the 50 Hz
AC supply mains.

6
Electrolytic distortion.
Despite some marketing claims, capacitors are not categorised for distortion so a distorting capacitor would not be considered
defective by its maker. It is the responsibility of the equipment designer to select the correct capacitor for each circuit.

During this investigation I measured many other polar electrolytic capacitors, rated from 10 to 100 volt and with capacitance
ranging from 1 to 220 µF, produced by several different, major manufacturers.

To illustrate this article, I decided to measure three quite different 100 µF 25 volt polar electrolytic capacitors and my
metallised PET assembly. The low cost Rubycon YXF, typical of a modern miniature low ESR capacitor, the much larger and
more expensive Elna Silmic and the considerably more expensive Black Gate FK, physically larger than the Elna Silmic.

The Black Gate FK is a 21×10 mm semi Bi-polar, built using a low voltage anode as its cathode foil. The Silmic is 17×10 mm
and uses a special separator paper incorporating silk extracts. Both were purchased from Audiocom UK.

The Rubycon YXF is a 12×6.5 mm conventional, miniature, low ESR low cost capacitor purchased from Farnell.

Tests were performed using 0.1 volt AC to 0.5 volt AC in 0.1 volt steps, each using DC bias voltages of 0 volt, 6 volt, 12 volt
and 18 volt, a total of 65 separate distortion measurements.

100 µF 25 volt tests.

With the 0.1 volt test signal, measurement noise floor was reduced to -112 dB. With no bias, distortions for the PET reference
capacitor and the Black Gate FK were lost in this noise. Second harmonic for the Silmic measured -106.1 dB and the YXF
measured -102.8 dB.

With 6 volt DC bias, second harmonic for the Black Gate FK measured -99 dB, the Silmic measured -99.5 dB and the YXF
measured -93.5 dB.

With 12 volt DC bias, second harmonic for the Black Gate FK increased to -95.9 dB, the Silmic was -94.4 dB and the YXF
measured -91.8 dB.

With 18 volt DC bias, second harmonic for all three electrolytic capacitors increased again to -93.6 for the Black Gate, -91.3
for the Silmic and -90.0 for the YXF. Distortions now measured some three times greater than the PET assembly. see Fig.8

Fig. 8) Distortion of a
low cost, very small
Rubycon YXF capacitor,
tested at 0.1 volt with 18
volt DC bias, was less
than 50% worse than the
two larger, more
expensive capacitors.

0.2 Volt Tests.

Using a 0.2 volt test signal, the measurement noise floor improved to -118 dB. With and without bias, all electrolytic distortions
increased more than the change of test signal. With no bias the Silmic performed best of the three polar electrolytics,
outperforming the Black Gate FK by almost 6 dB. Fig. 9 B, C

With 18 volts DC bias, dielectric absorption effects increased the second harmonic of the Silmic by 21.7 dB and it’s distortion
to 0.0054%. The Black Gate was less affected and its distortion increased to 0.0037%. The YXF distorted rather more, at
0.0063%.

Third harmonic distortions were visible above the noise floor, but not sufficient to produce measurable intermodulation
distortion.
7
0.2 volt tests with 0 volt DC bias:-
Fig 9A) With second
harmonic at -93.11 dB,
third at -115.14 dB, this
very low cost, miniature
electrolytic produces near
3 times more distortion
than the more expensive
specialist capacitors.

Fig 9B) This specialist

audio capacitor ‘Silmic’
from Elna provides the
best no bias distortion of
the three types tested at
0.2 volts.

However it produces
much larger distortions
than measured for the
reference capacitor.

Fig 9C) This specialist

audio capacitor the Black
gate-FK, produces almost
6 dB more second
harmonic than the Silmic
when tested at 2 volts
with no DC bias.

Tested using 6 volts DC bias, the Silmic and Black Gate capacitors produced almost identical second harmonic distortions at
-93.6 and 93.8 dB respectively. Measured distortion for both was 0.0021%. The YXF second harmonic was at -90.52 dB.

8
0.2 volt tests with 18 volt DC bias:-

Fig 9D) Measured with

18 volt DC bias, second
harmonic for the Silmic
increased by almost 22
dB to -85.3 dB, third
remained at -114.6 dB.
Distortion is now
0.00541%.

Fig 9E) Measured with

18 volt DC bias, second
harmonic for the Black
Gate-FK increased rather
less, by some 12 dB to
give -88.8 dB, third
remained at -111.5 dB.
Distortion is now
0.00368%.

With 18 volts bias this

BlacK Gate FK has now
become the better
capacitor.

However the reference

capacitor harmonics
remain lost in the noise
floor.

0.3 Volt Tests.

With a 0.3 volt test signal, measurement noise floor improved to -123 dB but the PET reference capacitor harmonics remain
buried in noise. Second and third harmonics of the polar capacitors are now clearly visible, their distortions having increased
much faster than the test signal level.

With a 0.3 volt test signal and no bias, the Silmic, at 0.00098%, produced the least distortion of the three electrolytics. It’s
second harmonic measured -100.6 dB, Black Gate -98.5 dB and YXF -89.1 dB. This is the best electrolytic of those I tested
with no bias, however it still produced more than three times the distortion of the PET assembly. Fig.7

With 6 volt DC bias, the Silmic and Black Gate, with second harmonics around -90 dB, produced similar 0.003% distortion.
The YXF second harmonic was -87.3 dB for 0.0043% distortion.

With 18 volt DC bias the Black Gate develops fifteen times more distortion than the PET assembly but now distorts less than
the other two electrolytics. Its second harmonic at -84.1 dB was some 3 dB better than the Silmic and 4 dB better than the low
cost YXF type. Distortions now measured 0.00637%, 0.00840% and 0.00951% respectively. Fig. 10 I

Third harmonics for all three electrolytics have reached the level for measurable intermodulation, which was confirmed by
more tests, using 18 Hz as the second frequency.

9
0.3 volt tests with 0 volt DC bias:-
Fig 10A) Measured at 0.3
volts but no DC bias,
second harmonic for the
YXF at -89.1 dB was
some 10 dB larger than
for the two specialist
capacitors.
Distortion has increased
to 0.00354%.

Fig 10B) Measured at

0.3 volts but no DC bias,
second harmonic for the
Silmic at -89.1 dB was
again the best polar
capacitor of the three I
measured.
Distortion has increased
to 0.00098%.

Fig 10C) Measured at 0.3

volts but no DC bias,
second harmonic for the
Black Gate-FK was some
2 dB worse at -89.1 dB
than measured for the
Silmic.
Distortion now measures
0.00126%.

10
0.3 volt tests with 6 volt DC bias:-
Fig 10D) Measured at
0.3 volts with 6 volt DC
bias, second harmonic
for the YXF increased
by 2 dB to -87.3 dB, third
remained at -110.7 dB.
Distortion is now
0.00434%.

Fig 10E) Measured at

0.3 volts with 6 volt DC
bias, second harmonic for
the Silmic increased by
near 11 dB to -89.71 dB,
third was at -112.95 dB.
Distortion is now
0.00320%.

Fig 10F) Measured at

0.3 volts with 6 volt DC
bias, second harmonic for
the Black Gate-FK
increased by near 8 dB to
-90.9 dB, third remained
at -115.5 dB.
Distortion is now best of
this three at 0.00293%.

Once more applying just

6 volts DC bias has
narrowed the distortion
gap between the
inexpensive YXF and the
expensive Black Gate-FK
style.

11
0.3 volt tests with 18 volt DC bias:-
Fig 10G) Measured at 0.3
volts with 18 volt DC
bias, second harmonic for
the YXF increased by
nearly 9 dB to -80.5 dB,
third remained unchanged
at -110.9 dB.
Distortion has increased
significantly and is now
0.00951%.

I consider this distortion

is far too high for use in
the signal path of an
audio system.

Fig 10H) Measured at 0.3

volts with 18 volt DC
bias, second harmonic for
the Silmic increased by
some 19 dB to -81.3 dB,
third continues at -113
dB.
Distortion is now
0.00840%.

I consider this distortion

is also far too high for use
in an audio system.

Fig 10I) Measured at 0.3

volts with 18 volt DC
bias, second harmonic for
the Black Gate-FK
increased by some 14 dB
to -84.1 dB, third
remained at -115.2 dB.
Distortion is again best of
the three at 0.00637%.

I consider this distortion

is far too high for use in
an audio system.

Increasing DC bias to
18 volts has had a
disastrous effect on
distortion for all three
types.

All three electrolytics produced significant distortions in these 0.3 volt tests. Almost five times larger with no bias, at least
fifteen times larger with bias, than my PET assembly. I consider distortions from these 100 µF polar capacitors tested at 0.3
volts, far exceed the sensible limit for use in the signal path of high quality audio.
12
Using a Film Shunt.
Some writers advocate using a low distortion film capacitor in parallel with an electrolytic, to reduce distortion. Does it work ?
To find out I made a few measurements on these capacitors using a 1 volt test signal, unbiased then with 18 volt DC bias. As
shunt I used my low distortion 1 µF MKP also a 10 µF bank of three 3.3 µF low distortion metallised PPS capacitors.

With 1 µF shunt, second and third harmonics of the Silmic reduced by just 1 dB. Using the 10 µF, both harmonics reduced by a
further 1 dB. This small reduction is not worth the additional PCB space and extra cost, because even with a 10 µF shunt,
distortions far exceed those of my metallised PET assembly.

Fig 11A) Re-measured at

1 volt without DC bias
but using the 10 µF film
capacitor as shunt, we
find second harmonic has
reduced by only 1.5 dB
Distortion was 0.00184%
is now 0.00151%.

Fig 11B) Measured at 1

volt with 18 volts DC
bias and using the 10 µF
film capacitor as shunt,
we find second harmonic
increased to -68 dB.
Distortion is 0.03973% .

I consider these
distortions are far too
high for use in the signal
path of an audio system.

Perhaps a higher voltage capacitor would measure better, or would its much larger area anode and cathode foils simply make
matters worse ?

100 µF 50 volt tests.

Examination of my earlier distortion plots suggested the only suitable 100 µF electrolytic types I had which might measure
lower distortion were the 22×12.5 mm 50 volt Silmic and the 26×12.5 mm 50 volt Panasonic S Bi-polar, Farnell 218-698.

With 0.3 volt test signal and no bias, the 50 volt Silmic distorted more than the 25 volt version. Because of its much longer and
wider foils, second harmonic increased 2 dB, third increased 7 dB and distortion measured 0.00134%. see Fig.12A

Due to the thicker dielectric used for the 50 volt capacitor, with 18 volt DC bias, second harmonic increased less, now almost
6 dB smaller than the 25 volt version. Distortion at 0.00460% was just over half that of the 25 volt version.

13
Comparison tests, 100 µF 25 volt and 50 volt rated ‘Silmic’ capacitors.
Fig 12A) Measured with
no DC bias at 0.3 volt,
we find second harmonic
has increased by 2 dB and
third harmonic by 7 dB
when compared with the
25 volt capacitor.
Distortion has increased
to 0.00134% for 50 volt
capacitor, from 0.00098%
for the 25 volt version.

Fig 12B) Measured with

6 volt bias at 0.3 volt, we
find second harmonic has
increased by another 3.5
dB but third harmonic at -
105.6 dB has not
changed.
Distortion now
0.00185%.

Fig 12C) Measured with

18 volt bias at 0.3 volt,
we find second harmonic
has increased by almost
12 dB, third harmonic at
-105.9 dB unchanged.
Distortion now
0.00460%.

14
Comparison tests, 100 µF ‘Silmic’ polar and 50 volt rated Panasonic Bi-polar capacitors.
Fig 12D) Measured with
no DC bias at 0.3 volt,
we find second harmonic
for this Panasonic Bi-
polar reduced by nearly
12 dB. Third harmonic is
now near the
measurement noise floor
at -119.6 dB.

Distortion is the lowest

seen so far for an
electrolytic and less than
half that of the 25 volt
Silmic.

Fig 12E) Measured with

6 volt DC bias and 0.3
volt, we find second
harmonic almost identical
to that measured on the
Silmic with no bias, 10
dB smaller than for the
25 volt Silmic with 6 volt
bias.
Third harmonic remains
near the measurement
noise floor.

Distortion is now
0.00106%.

Fig 12F) Measured with

18 volt DC bias at 0.3
volt, while distortion has
increased it is less than
half that measured on the
best polar capacitor
tested.
Third harmonic remains
near the measurement
noise floor at -119.6 dB.

Distortion is now the

lowest seen so far for an
electrolytic with 18 volt
bias, half that of the best
polar type measured with
18 volt DC bias.

15
Bi-polar.
The Panasonic S Bi-polar capacitor at 0.3 volt with no bias, produced less than half the distortion of the 25 volt Silmic. Second
harmonic measured -111.8 dB, third -119.6 dB and distortion 0.00042%.

With 18 volt DC bias, second harmonic increased to -92.7 dB and distortion to 0.00237%, half the distortion of the 50 volt
Silmic.

The Panasonic S Bi-polar produced the lowest distortion of all single 100 µF electrolytic capacitors of those I tested, using a
0.3 volt signal and DC bias from 0 volt to 18 volts. see Fig.12 D/F

In my last article we saw how using two polar capacitors in series could reduce distortion. Let us now explore using two Bi-
polar capacitors in series.

Two better than one?

I already had some 220 µF 63 volt Nitai Bi-polar electrolytics, Farnell 317-4906. Two connected together in series would
approximate 100 µF.

Measured at 0.3 volts with no bias, second harmonic level reduced 6 dB compared to the Panasonic S Bi-polar. With second
and third harmonics buried in the noise floor, distortion at 0.00033% measured the same as the PET assembly.

With 18 volt DC bias, second harmonic measured -105.3 dB and distortion 0.00063%. A near four fold improvement compared
to the Panasonic S Bi-polar, more than seven times better than the best polar capacitor tested.

Fig 13) Series pair of 220

µF 63 volt Nitai Bi-polar
electrolytics, measured
with 18 volt DC bias at
0.3 volt, second harmonic
distortion has reduced
dramatically. It is now
seven times smaller than
measured on the best
polar capacitor tested.

Third harmonic remains

on the measurement noise
floor.

To better compare harmonics I examined performances using a 0.5 volt signal. With no bias, those for my PET assembly can
just be seen emerging from noise. Second harmonic -124.3 dB, third -123.9 dB and distortion 0.00020%. see Fig 7

The double 220 µF 63 volt Bi-polar second harmonic -117.7 dB, third -124.1 dB, and distortion 0.00023%, measured
practically the same distortion as the PET assembly. see Fig.14A

With 18 volt DC bias, second harmonic of the double Bi-polar increased to -100.7 dB and distortion to 0.00093%, slightly
more than double the distortion measured on the PET assembly with this bias. see Fig.14C

This is an excellent performance from a pair of inexpensive electrolytic capacitors, but how does this series pair of Bi-polar
capacitors stack up for size and cost ? Can this Bi-polar series pair still produce low distortion tested with a 1 volt signal ?

At 1 volt with no bias, noise floor improved to -132 dB. Distortion of the PET assembly measured 0.00011%, a single
Panasonic S Bi-polar 0.00054% and the Silmic 25v with 10 µF shunt 0.00151%.

The 220 µF 63 volt Nitai series pair measured 0.00016%, practically equalling that measured on the PET assembly, and ten
times less distortion than the Silmic 25 volt polar capacitor.

With 18 volt DC bias, the 220 µF 63 volt Nitai series pair distortion measured 0.00217%.

16
Double Bi-polar series connected pair of 220 µF 63 volt Nitai capacitors tested at 0.5 volts.

Fig 14A) Series pair of

220 µF 63 volt Nitai
Bipolar electrolytics,
measured with no DC
bias at 0.5 volt.
Second harmonic
distortion now -117.7 dB,
third harmonic remains
near the measurement
noise floor.

Distortion measures
0.00023%, practically the
same distortion as
measured using my PET
assembly.

Fig 14B) Measured at

0.5 volt with 6 volt DC
bias. Second harmonic
distortion now -111.3 dB,
third harmonic remains
near the measurement
noise floor.

Distortion now measures

0.00033%.

Fig 14C) Measured at

0.5 volt with 18 volt DC
bias. Second harmonic
increased to -100.7 dB.
Third harmonic remains
near the measurement
noise floor.

Distortion is 0.00093%,
little more than double
that measured for my
PET assembly.

At this voltage the Silmic

measured 0.01312% the
Black Gate-FK was
0.01041%.

17
Double Bi-polar v alternatives.
The series pair requires less PCB area, is lower cost and dramatically outperforms a polar capacitor with film shunt.

At 1 volt with no bias, noise floor improved to -132 dB. Distortion of the PET assembly measured 0.00011%, a single
Panasonic S Bi-polar 0.00054% and the Silmic 25volt with 10 µF shunt 0.00151%.

The 220 µF 63 volt Nitai series pair measured 0.00016%, practically equalling that measured on the PET assembly, ten times
less distortion than the Silmic 25 volt capacitor.

With 18 volt DC bias, the 220 µF 63 volt Nitai series pair distortion measured 0.00217%. Slightly more than six times that of
the PET assembly but nearly seven times less distortion than using the 50 volt Silmic polar capacitor.

This series pair of 220 µF 63 volt Nitai Bi-polar capacitors costs one eighth and takes just one fifth the PCB area of my PET
assembly.

To explore other double Bi-polar options, I purchased 35 volt and 16 volt 220 µF Nitai Bi-polar capacitors for tests.

Smaller Doubled Bi-polar.

With no bias and tested at 0.5 volt, distortion for all three voltage Bi-polar doubles, measured almost the same as the PET
assembly, but 18 volt DC bias revealed large differences. The 16 volt series pair measured 0.00693%, the 35 volt series pair
0.00230% and the 63 volt series pair 0.00093%.

For the lowest possible distortion when DC blocking/signal coupling, I suggest the 16 volt pair is only used with negligible DC
bias, the 35 volt pair be used to say 6 volt bias and the 63 volt pair to say 12 - 15 volts bias. With such small DC voltages, no
voltage sharing resistors are needed.

Used in a ‘Long Tailed Pair’ amplifier feedback network to ensure unity gain at DC, the 63 volt series pair could be used with
supply rails up to 63 volts, without voltage sharing resistors. For higher voltages use a series pair of 100 volt Bi-polar.

This 63 volt series pair can also benefit local supply rail decoupling, but for this use, voltage sharing resistors, passing a few
mA from the supply to the capacitors central connection and ground, must be used.

Conclusions.
Having measured a considerable number of aluminium electrolytics using test voltages from 0.1 volt to 3 volt, with and without
bias, a single Bi-polar type produced lower distortion than larger, more expensive, specialist polar capacitors..

Much better results were obtained by connecting two double capacitance value Bi-polar electrolytics in series. Using 1 volt or
smaller test voltages and no bias, distortions for a double Bi-polar and the metallised PET assembly were similar.

With increasing bias or with increasing test voltage, the metallised PET assembly produced less distortion than any electrolytic
I tested.

Distortion with voltage.

We have seen how the test voltage used influences various capacitors. With sufficient test signal, most film and all electrolytic
capacitors will distort. It is prudent in any audio design to minimise the level of AC signals which are developed across any
capacitor.

At low frequencies this becomes difficult and may force a trade off between capacitor size and distortion. Equally important is
the level of DC bias voltage the capacitor must sustain. If more than a few volts, then for low distortion a low dielectric
absorption material is essential.

Because distortion results from non-linearities inside the capacitor, inevitably it increases disproportionately both with
capacitance value and applied voltage.

The change in amplitude of second harmonic, when tested at a constant signal with and without DC bias, clearly results from
the DC bias voltage used, dielectric absorption and dielectric thickness.

Regardless of capacitance value, to minimise second and third harmonic distortions with increased AC and DC voltages, such
as found in valve amplifiers, then a foil and Polystyrene, foil and Polypropylene or double metallised foil, two-series, MKP
Polypropylene capacitor, should be used.

18
100 µF choice.
Provided the AC voltage developed across the capacitor at the lowest audio frequencies is 1 volt or less and no significant DC
bias is used, a double Bi-polar series pair provides an economic solution.

When higher AC signal voltages, especially combined with significant DC bias, must be applied, the metallised PET
combination produces less distortion. It costs eight times more and takes five times more PCB area than the double Bi-polar.

For the least practical distortion, an assembly of metallised Polyphenylene Sulphide capacitors might be feasible. It needs
double the board area and is five times more expensive than the PET assembly.

For small AC signals with modest DC bias and for supply rail decoupling, I choose the double Bi-polar 63 volt solution.

10 µF choice.
We have three possibilities. A double Bi-polar using two 22 µF 50/63 volt Bi-polar electrolytics, a 10 µF metallised PET or an
assembly of three 3.3 µF PPS capacitors.

The lowest cost solution for use with signal voltages less than 1 volt and no significant bias, is a double Bi-polar series pair.

A 10 µF MMK metallised PET takes the same PCB area and distorts less with DC bias.

The PPS capacitor assembly ensures lower distortion, especially when used with increased AC signals or DC bias voltage.
However it occupies more board area and is expensive.

An assembly of Polypropylene capacitors, as used in the DC bias network, would provide the lowest possible distortion but
requires a five times larger board area and is most expensive.

For small AC signals and modest DC bias, I choose the 10 µF MMK metallised PET capacitor.

END

References.
1) Capacitor Sound part 5 C.Bateman Electronics World December 2002 p.44

2) Capacitor Sound ? C.Bateman Electronics World July 2002 p.12

3) Capacitor Sound 2 C.Bateman Electronics World September 2002 p.16

4) Understanding capacitors - Aluminium and tantalum Electronics World June 1998 p.495.
C.Bateman
5) Evaluate capacitors for SMPS designs. Electronics World Sept 2000 p.696
C.Bateman
6) Reference Data for Radio Engineers. Howard Sams & Co. Inc. New York

19