Your first stop on should be the LDSG crossover paper. It covers a lot of info, agreeing and disagreeing with other text below:
LDSG Crossover component paper
Next, a bunch of papers recommended by Jon Risch:
Pease, Robert A.; "Understand
capacitor soakage to optimize analog systems." Jung, Walt & Marsh,
Richard N.;
"Picking
Capacitors", Marsh, Richard N.,
"Considerations
for a High Performance Capacitor"
Note on Cap D-E curves, by
Steve Bench
Note on Cap D-E curves,
by Steve Bench II
Note on Cap D-E curves,
by Steve Bench III
Karl A.
Weber, "On Capacitor Dielectric Materials - A Chemist's View"
Web site with lots of info on capacitors.
E-book on
Electrolytic capacitors, includes basics of capacitors I.
E-book on Electrolytic
capacitors, includes basics of capacitors II.
Here are a number of capacitor discussions with some Madisound forum regulars. I only took a few of the postings This all started with a capacitor AB test at the DIY Spring Warm Up in Annapolis, 2004. Sorry, when I grabbed them, the names were lost. Most are between Bob Cordell and JohnK
Follow ups:
I just did some more careful, more detailed measurements on a Dayton 10 uF film
cap and also on a GE 10 uF 450V commercial film cap. ESR and self resonance
were measured by applying a 2.0V rms sinewave to a 47 ohm resistor in series
with the capacitor. The other end of the cap went to ground. The a.c. voltage
across the cap was measured. The frequency where the voltage is a minimum (an
easily identifiable dip) is the self-resonant frequency. The amount of voltage
at the self-resonant frequency indicates the ESR. For example, at the
self-resonant frequency of the Dayton, the voltage appearing across the cap was
0.34 mV. Dayton ESR: 0.0080 ohms (80 milli-ohms) GE ESR: 0.0089 ohms Dayton
self resonance: 240 kHz GE self resonance: 205 kHz Adding a 6-inch piece of #18
wire in series with the Dayton cap changes the numbers as follows: ESR: 0.0129
ohms (129 milli-ohms) self resonance: 110 kHz This gives some perspective,
showing that even a 6-inch piece of #18 wire has significant resistance and
inductance compared to that of the Dayton cap. These small-signal measurements
indicate that ESR and self-resonant frequency of these caps (they aren't
special or expensive) cannot possibly be influencing frequency response in any
audible way (e.g., well less than 0.1 dB), at least by conventional, scientific
wisdom. These measurements do not, however address issues of large-signal and
non-linear effects. One large signal effect that some have hypothesized about
is that the electric field resulting from the voltage across the capacitor
changes the capacitance somewhat by inducing physical stress. This is a
large-signal, non-linear effect. One indication of such an effect would be to
measure the capacitance as a function of d.c. bias voltage across the
capacitor. I did this for the Dayton 10 uF cap. Capacitance change with bias
going from 0V to 20V was barely detectable, and certainly no more than 0.001 uF
out of 10 uF, for a percentage change of less than 0.01 percent. I must again
emphasize my use of the term "conventional wisdom". I am not saying
that capcitor differences cannot be heard. I am just saying here that any such
audible differences thought to originate from differences in ESR or self
resonance frequency between two decent film caps cannot be explained by
conventional wisdom. Bob
You're right on all counts. Basically, what you are saying is that we need to get into looking at non-linear, large-signal effects, and I agree. I am pretty much unable to find any small-signal differences among decent film caps that could account for an audible difference (again, by conventional wisdom). Once again, with conventional wisdom, time domain and frequency domain behaviors are fully correlated for such passive components - if you do not have a significant frequency domain anomoly, you cannot have a significant time domain anomoly. This is something that A LOT of people without formal EE training do not recognize. The d.c. bias test I did is of limited relevance to a crossover application, as you point out. However, it is a simple indicator of a capacitor's tendency (or not) to have electric-field-induced nonlinearity via field-dependent capacitance effects. For example, this pretty much rules out capacitance intermodulation distortion, where the large low-frequency signal across a series tweeter cap causes the capacitance to be modulated and thus induce intermodulation distortion into the tweeter path. My plan for the needed large-signal tests is to provide the test signal from a 50-watt very low distortion power amplifier, allowing for capacitor voltages in the tends of volts, and capacitor currents in the amperes. One example would be to drive the capacitor through an 8-ohm series resistor, with the cap shunting to ground, then to measure either THD or IM distortion across the cap, especially at the 3-dB corner frequency of the cap and 8-ohm resistor. This test would also highlight any distortion caused by nonlinearity in the capacitor's lead-attach interface. Finally, another point of interest is in regard to the use of capacitors in line level applications, where the signal environment is largely small-signal. The most common example is that of a coupling capacitor. There is also a large mountain of anecdotal data that differences among coupling caps are audible. This is even more difficult to explain by conventional wisdom, since the signal voltages across the caps and the signal currents through the caps are very small and unlikely to exercise nonlinearities. Indeed, when a coupling capacitor is doing its job, the signal voltage across it is close to zero. Thus, audible effects of line level coupling caps are even more perplexing to explain by conventional measurements and conventional wisdom.
http://members.aol.com/sbench102/caps.html
Thanks for bringing this to my attention. I did read it over. The distortion in the ceramic capacitors was made pretty obvious, with a non-straight line. The film caps all looked like they had a basically straight line, but with slightly varying degrees of single line versus very narrow oval. The author seems to be referring to the slight oval-ness as hysteresis, but my initial impression is that it is from a very slight phase error between I and dv/dt. I believe that even a perfect capacitor would result in a circle on the screen if the plot was of I and V, where they are 90 degrees apart. I'll have to give it some more thought. In any case, I agree with the author's concept of exercising the capacitors with moderately high voltage. By the way, not that in his tests the currents flowing in the caps were quite small, since the cap was 0.1 uF and the frequency was only 600 Hz. Finally, if any capacitor is going non-linear, we should be able to see it more clearly with a harmonic or intermodulation distortion test. I hope to try that in the near future. Thanks again, Bob
For those who missed it in the thread below over the long weekend, JPK posted a link to some interesting reading regarding the audibility of capacitors, and more specifically dielectric absorption. (JPK) In the "nearly linear" caps that exhibit some apparent hysteresis the effect is almost certainly dielectric absorption. I found this page that gives a good explanation of dielectric absorption and release and how the hysteresis comes about. If you read this carefully you can see why cap with different dielectrics can/might sound different. http://www.audience-av.com/on_capacitor_dielectric_material.htm (BOB) The article is written by a chemist, and largely explains in physics terms how dielectric absorption comes about, and why/how different materials have different amounts of DA. If you like terms like "dipole moment" and "polarizability" and phrases like "structural charge density differences over intramolecular distances", I'm sure it can be a real turn-on, and one is inevitably impressed with the guy's knowledge of chemistry and physics. He goes on to tell us that electrolytics have high DA, polyester has medium DA, and polypropylene and polystyrene have low DA, something we already knew. But, does it pass the "so what?" test? No. Like so many of the other pseudo-science writings on capacitors, he fails to connect in any way why a higher DA makes an audible difference. He just takes it as a given, and assumes the lower the better. While there exists a lot of anecdotal data linking good sound with low DA, there is no conventional wisdom explanation that successfully connects the dots (at least that I am aware of). Jung and Marsh in their well-known article in Audio 24 years ago also went to great pains to explain and analyze DA, but also failed to link it in any way other than by anecdotal listening data. Even the DA difference between polyester (Mylar) and polypropylene cannot create an audible frequency response difference of, say, 0.1 dB in a typical application. These articles never talk about DA nonlinearity being an issue in film caps, and the literature seems to suggest that DA is not nonlinear, so it also seems that DA is not creating nonlinear distortion that would account for audible differences either. Therefore, if it is responsible for an audible difference, what is the mechanism? We (or at least I) don't know. It is not explained by conventional wisdom (at least not mine). For those that want to know more about DA from someone who is not a witch doctor, go to the National Semiconductor web page and read the article by Bob Pease. This author also says that he assumes that DA and dielectric constant are directly correlated. If we took that to mean proportional, that would mean that the DA of Mylar (K=3.7) would only be 1.5 times worse than that of Polystyrene (K=2.5). We would be paying an awful lot of money for only a modest difference in DA. In my limited search it looks like nearly all of the audio-grade caps of a value of, say, 10 uF, are of metalized polypropylene construction. This holds even for the less expensive ones, like the Daytons and Solens. Few film/foil polypropylenes or polystyrenes go much above 1.0 uF. I would bet the ranch that the $18.00 10 uF Sonicap and the $33 10 uF Auricap are made of metalized polypropylene, just like the $5 10 uF Dayton. DA is usually a physical property of the dielectric material (just ask our chemist friend referred to here), so I would expect DA of these caps to be similar. But, you never know, maybe there is something in the manufacture or construction that can alter it. By the way, our chemist friend would appear to have a relationship with Auricap, so he is not necessarily an unbiased source. Next I want someone to explain to me why my crossover sounds bad because I put my non-polar film caps in backwards. Yes, the auricap comes with a red lead and a black lead, and detailed instructions on which way to install it, depending on the application.
(JPK) Bob, no connection between DA and audible colorations? I know you are familiar with this page, http://members.aol.com/sbench102/caps.html . what it shows is basically the effect of DA on a steady state sine wave. Forget the ceramic case since the results also show that C is voltage dependent for ceramics. The observed hysteresis is pretty much due to DA. Charge trapped at the interface between the dielectric and the "plates" is accumulated and release with a secondary time constant, you already know about that. When the signal is a sine wave this results in a lagged component in the current which causes a phase shift, provided the accumulation/release is linear. But if the signal is not a sine wave, i.e. not periodic and symmetric, this lagged component of the current can and will distort the output signal. How much and how audible depends of the type of signal and the level of DA. But the accumulation and release of charge in dielectric/conductor interface is going on all the time and there is no reason one shouldn’t expect subtle or not so subtle audible effects arising from it. I agree that I wouldn't expect changes in level by 1dB, or even 1/2dB, but resolution of low level can easily suffer.
John, thanks for your comments. I have avoided stating that there is no correlation between DA and audible effects, and am keeping an open mind on that. What I have said so far is that if there is an audible influence of DA, it is not explained by conventional wisdom, such as by frequency response anomolies or non-linear distortions. As I am sure you agree, the capacitor is a minimum phase device in the circuits where we use them, and that you agree that in such circuits the time (phase) and frequency behaviors are completely linked. This means that if there is any "time smearing" or shifting of musical harmonics with respect to the fundamentals, then that can only be accompanied by a frequency response anomoly that is easily measurable. (BTW, guys that are concerned about a small shifting of the phase relationship between fundamentals and harmonics by a capacitor should really not be using a Linkwitz-Riley crossover:-)). I hope you agree with me on this part, and please let me know if you don't. As far as the referenced article goes, we may largely have to agree to disagree. I don't think you can be sure that the slight elipse on the screen is due to DA any more than I can say it is phase shift in the setup. Take a close look at his methodology and test setup. He senses current and quasi-integrates it by a 100 ohms series resistor and a capacitor, then uses another RC network to make the whole thing look more like a true integrator. If two capacitors he is testing differ by 10 percent in capacitance, this alone could possibly cause enough phase shift in the measurement of one to the other to cause one to be slightly elliptical while the other is straight, if he did not re-adjust his phase correction for each capacitor individually. Also, recognize that this is a high voltage test, where nonlinearities may come into play. That is something I have not looked at yet. My statements so far have been made on the assumption that there is no significant nonlinearity in DA (but there may be - I just don't know yet). Moreover, other nonlinearities could be causing the slight elipse observed. I just haven't gotten to those kinds of tests yet. In the meantime, I urge you to read the excellent article by Bob Pease on the National Semiconductor website. Bob
(JPK) I agree that you won't see gross frequency response abortions. But I'm not so sure about nonlinear distortion. If the accumulation and release of the charge trapped in the dielectric is not linear, then there could be nonlinear distortion introduced. I'm not so sure I can agree with your latter statement. To first order you are correct, but when you start looking at DA models for caps things get a little more complex. Still minimum phase, but there are other effects. One simple model of DA is to have a second cap of smaller values with a series resistor in parallel with the basic, ideal capacitance. Obviously when the leads are shorted the ideal cap will loose all its charge, but the secondary cap, which represents the storage of trapped charge in the dielectric, releases its charge at a much slower rate through the series resistor and when the leads of the open again, this secondary cap starts to charge up the ideal cap again. That's the simple model of that voltage snap back. Now consider that an audio signal comprised of two different frequencies. One frequency is a large signal, the other a very small signal in amplitude, may just barely audible. Wouldn't you think it possible that the release of charge accumulated in the dielectric due to the larger signal could swamp the current associated with the smaller signal and at least alter the perception of the audible result? Again, I'm talking subtle changes here. I also agree with your comment on LR crossovers. I think we are in basic agreement here Bob. As I said below I am not a proponent of high end caps. Poly over Mylar or electrolytic, yes. $50 polys over $5 polys. I also argue that audible difference doesn’t equate to better listening experience over the long term. I'll try to get a look at the Pease article. In the mean time here is a signet you may want to check out as well. http://www.designers-guide.com/Modeling/
John, thanks for the link to the Kundert article. It is excellent. One of the more revealing things in this article and in the Pease article is the multiple RC section Dow model of DA and the component values in the model. First and foremost, this article does not deal with possible nonlinearities associated with DA, so we still must allow that nonlinear effects may cause problems, and I think we both agree on that. If they do, I'm expecting they'll show up in a proper distortion test. Secondly, the Dow model (and the equivalent lumped version of the Cole-Cole model) is a strictly minimum phase model. To the extent that these models are accurate, the linear effects of DA are no more than whatever frequency response effect would be had by paralleling the ideal capacitor with some relaively small parasitics, resulting in relatively minor frequency response deviations from the ideal. It is valuable to look at the values of these parasitics. The model for a 1.0 uF polystyrene has 5 series R-C elements, all in parallel with the main capacitor. The C in every case is on the order of 200 pF, or about 0.02% of the main cap value. The R in each secion are on the order of 3000,000 Meg ohms, 300,000 Meg ohms, 30,000 Meg ohms, 3000 Meg ohms, and 300 Meg ohms, respectively. These are VERY, VERY high resistances, implying VERY, VERY low losses. These also imply extremely long time constants. This implies that most of the effects would be at frequencies below the audio band. Even at 20 Hz the nominal impedance of the capacitor of 8K is 40,000 times lower than the 300 Meg ohm element. It is easy to see that with these values, the effect on frequency response of this capacitor in the audio bad should be way less the 0.1 dB. The total of the DA parasitics is about 1000 pF, suggesting that in the best case with total charge redistribution, we would get a DA charge recovery of about 1000pF/1.0 uF or about 0.1%. A Dayton 10 uF polypropylene measured with the 60-second charge, 5-second discharge, 60-second recovery technique, when charged to 10V, yields a recovery voltage of about 2.0 mV, or about 0.02%. That we have a value here that is 1/5 of that of someone else's hypothetical model is pretty encouraging, given that the 60-second charge and recovery lops off probably half of the time constants (the longer ones), and that the 5-second discharge cuts into some of the energy of the shorter ones (also assuming DA of polystyrene and polypropylene are similar). I can't say that I buy your argument about a large signal and a small signal, the smaller one being somehow swamped out by release of charge from the larger one due to DA. Such an argument does not apply to a linear system, which is what is modeled here. Such an argument might, however, apply to a nonlinear system, such as if we were to ferret out a nonlinearity in DA behavior. Bob
In a capacitor thread below "capslock" wondered about microphonics being partly responsible for audible differences among caps. This is theoretically quite plausible, and I've wondered about it myself. Anybody with a tube preamp knows what we are talking about. Tap the old 12AX7 and you hear it in the speaker. A legitimate concern is that vibrations from the music itself could cause microphonics that would color the sound. Capacitors can have microphonics too. Indeed that is the principle on which some microphones operate. A capacitor conserves charge. If you charge up a capacitor to a certain voltage, and then change the value of the capacitor by moving the plates apart, the voltage will increase in order to conserve the charge. I tested a few capacitors for microphonics. I took the capacitor and charged it up to 10V, hooked a fairly sensitive scope across it, then banged the capacitor with a pair of pliers. Not a very sophisticated experiment, but almost certainly valid in giving a hint as to microphonic susceptibility. Banging the cap hard and directly with a pair of steel pliers probably creates an acoustical vibration in the cap that is many tens of dB higher than any acoustical vibration, even for a cap in a crossover in a speaker box. Out of five caps, only one showed any evidence of a microphonic response - an old GE mylar cap, which showed a 2 mV p-p response (0.02% of the initial charge of 10V). All others, which included a Dayton metalized polypropylene, showed no discernable response. I was quite impressed by how quiet they were. One could do an even more sensitive test. Do the same thing, but instead of hooking a scope across the cap, connect the cap wires to the microphone input of a cassette recorder (though an appropriate dc block), then later listen to it. Although one will always be able to find poor or outright defective capacitors that display microphonics, this preliminary test suggests that microphonics are unlikely to be much of a problem. Note that the amplitude of microphonics from a capacitor will tend to be larger for larger voltages across the capacitors. Thus, a d.c. blocking cap in a tube amp with 250V across it will likely be a bit more prone to microphonics.
Copyright: Peter J Smith 2004, Return to helarc.com