Capacitor Distortion Part 2

The purpose of an input coupling capacitor is to block dc but ideally to have no effect on the ac input. To achieve this the voltage across the capacitor needs to remain constant. In practice there will be some signal related voltage across the capacitor, partly the expected low-pass filtered version of the input signal but also any distortion produced by the capacitor. To detect and investigate the signal across the capacitor, Vc, we could subtract the original signal Vs and the signal Vi after the capacitor shown in the first diagram and the result would be the total effect on the signal. Far easier is to reverse the capacitor and its load resistance as in the second diagram, then we can directly observe the signal voltage across the capacitor, Vc, which should be practically identical to the signal voltage across the capacitor in the first diagram. The input impedance of the soundcard has some effect, but we are only concerned with comparing different capacitors, not making accurate voltage measurements, so this is unimportant.

Using a variety of test signals and three different capacitors the plan was to investigate the differences in Vc to try to discover what could be responsible for the differences some listeners say they hear between various capacitors. The test signals I tried were white noise, pink noise, 440Hz square-waves and two different music extracts.
The capacitors tested were a EPCOS - B32560J1225K, 2u2, 100V polyester, a PANASONIC - ECWF2225, 2u2, 250V polypropylene, and a PANASONIC - ECEA1HN100U, 10u, 50V non-polarised electrolytic.
The results were somewhat disappointing, I could identify no significant difference between the 2u2 types, I had hoped for at least some minor effect, but I did confirm my expectation that just using a higher value 10u type, even a supposedly inferior type, was the only change to make a dramatic improvement. In this context an 'improvement' is a lower signal related voltage detected across the capacitor. This signal can be observed both by measurement and by recording and listening. Ideally we would hear nothing, and could conclude that nothing audible is being added or subtracted from the signal by the coupling capacitor. I have only included one of the 'music' extracts, 7sec long, here. The other results add nothing useful, and have lower detected signals which are not so easily heard.

Typical Test Results

The measurements are of the peak levels as a function of frequency, not because this is likely to be particularly revealing but because it is a simple measurement which can be applied to a few seconds of music, and should give repeatable results. This is done using the OscilloMeter spectrum analyser which has a peak hold function, which is the red trace in the following pictures. The first is the original input signal, and the other three are labelled with the capacitor type. Ideally these should be zero level, but that would only be achieved with an infinite capacitance with zero impedance. It can be seen that there is very little difference between the 2u2 types, but the 10u has signal level across it over 12dB lower at all frequencies up to about 10kHz, above which the soundcard noise starts to become dominant.

Listening Test.

Having revealed nothing useful from the measurements a comparison more directly related to sound quality effects is simply to listen to the capacitor voltages. Clicking on the following links will download or play the signals previously measured. The first is the original 'music', actually just some vaguely musical noises produced by a computer synthesiser, designed to have a wide range of frequency components. Having set this to a comfortable listening level the following link is a sequence of the three capacitor voltage signals. The first two are the polyester and polyprop 2u2 types, but I'm not saying which is which. The third is the 10u electrolytic, and is obviously at a lower level, confirming that this capacitor has the least effect on the signal. In the second file only one of the stereo channels is used.



The conclusion is that these tests reveal no reason to prefer either a polypropylene or a polyester of the same 2u2 value, but that using a higher value 10u non-polar electrolytic does make a significant 'improvement', though there is no evidence from this test to suggest that any of the effects could be audible if added or subtracted from the original test signal. I also tried a 100u polarised electrolytic, and again the capacitor voltage fell, becoming almost inaudible. The higher leakage currents of the electrolytics could be a deciding factor, for use with my MJR7 circuit the leakage could adversely affect the operating voltage level in the output stage, and so the 2u2 types are more suitable. The increased interference pickup from a physically larger polyprop type found in previous tests then confirms that the polyester type is to be preferred. A reduction in capacitor signal voltage without a great increase in size or leakage could be achieved by using a 4u7 polyester.

There is undoubtedly some difference between polyester and polyprop capacitors, and a more sensitive experiment would be to apply the signal via two 10k resistors to the two different capacitor types at the same time and subtract the two capacitor voltages to detect the difference. This approach may be capable of extracting small differences, but has one serious disadvantage, that it only tells us if there is a difference, not which of the capacitors compared is better for input coupling applications. It may be tempting to assume that the more expensive, or the one with lower dielectric absorption (DA) must be better, but my previous simulations suggested that adding the usual representation of DA as a series resistor and capacitor in parallel with the main capacitance reduces low frequency phase errors. It certainly reduces the impedance of the capacitor slightly, and so reduces the effect on signal voltage. Listening to the two signals and comparing the level to that of the original input signal it is hard to believe there is any important difference, and the dramatic reduction using 10u suggests that capacitor value is in this case vastly more important than the choice of capacitor type.

In other applications such as input low-pass filter and feedback loop compensation, where there is necessarily a significant signal voltage across the capacitor, the capacitor values are generally lower and polypropylene types are then reasonably small and cheap, and the low distortion at higher voltages makes them a good choice, which is why I specified these for my MJR7-Mk5, apart from the 10p value which is more easily available as a COG/NPO ceramic, which have also been found to have good characteristics.