|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
68.166.237.49
I found a great article at the National Semi-conductor website, which explains capacitor soakage really well.This article and model supports some ideas about why different types of capacitors work better than others, but it really does not support the use of by-pass capacitors as a solution.
Let me explain. If you read the article, the thing to notice closely is that Bob Pease proposes a model for real capacitors is one of several RC networks in parallel with the primary capacitor.
This is in effect the model for what is also called "capacitor memory" where it seems to remember how it was charged, and could very easily account for time smear.
Unfortunately this article does NOT support the idea that bypass capacitors will help this particular trait. Think about it this way. You have a 4uF capacitor, which in reality is composed of 3.9uF which is in parallel with an RC network consisting of a 10MOhm resistor and 0.1uF capacitor.
Adding any amount of bypass capacitors, even if theoretically perfect, won't prevent the 0.1uF capacitor from re-charging the primary section, INCLUDING the bypass cap.
The benefits you probably ARE getting from using a small high-quality capacitor in parallel with a larger one is that you are bypassing the inductance of the larger capacitor, giving you a much more linear reactance across the frequency band. This is why bypass caps have been used in manufacturing in other industries.
If you are really interested in reducing time smear, according to the way this article models capacitors, you are going to be much better off using parallel capacitors instead. That is, if you need a coupling cap of 4uF, use 4 x 1uF capacitors or 2 x 2uF. This should reduce the memory effects substantially.
What do you fine folks think?
Regards,
Erik
Follow Ups:
Erik
I can relate to what your saying..The big bypass issue came in the mid 1970s and people would add a capacitor bank to a big ass solid state amp like a GAS Ampzilla and then bypass it with a load of mallory .1 caps and they claimed it would walk the high frequencies across the computer grade electrolytics so they wouldnt get absorbed in big caps therefore not losing upper range treble.
After doing this for a couple years,they took the bypasses back off and claimed it sounded better.It makes me wonder where a lot of these crazy ideas are initiated.
Im not against bypassing if you use a big enough cap to where its going to be fucntional or in the case of the cathode bypass on a non electrolytic.I do however thing that a .1 or .22 on an electrolytic is a farce.
Hi!I think it's important to understand that there are two issues, one of which is frequency domain, and the other time domain.
Bypassing a large cap with a very small cap is a great idea for frequency domain issues. When you are doing this though, what you are bypassing is not the capacitor, but the inductance in the capacitor. :)
Think of a real capacitor as being a series LC network. Larger, electrolytic caps tend to have the most L. Now if you draw this in paper, you can see how the bypass effects work. You aren't interested in adding capacitance, so much as creating a low L path. Very much like a crossover.
Say you have a big ass 4,700uF electrolytic capacitor with 120 mH of parasitic, or unintended inductance. What this inductance is going to do is eventually cause a roll off of high frequencies. So, if you can add any capacitor around it, with say 10 mH of inductance, you have just increased your high frequency response significantly. BUT, even with this system, you have to choose your bypass cap well. You can end up with a "hole" or "hump" in the response where the bypass cap's effects doesn't match well with the main. This would definitely cause the sound to be worse than before.
Where small bypass caps don't work, at all, is in the time smear, or Dielectric Absorption problem. For this, it seems you are stuck with either banks of caps, or using higher quality caps to begin with, or maybe new caps with multiple leads will work best. Or, if you want to get really really really outrageously tweaky, you can design an op-amp based circuit with a model of the coupling cap, which outputs an inverse signal to that produced by the DA. Hah. :) That would be a never ending story. :)
Regards,
In your last paragraph you state "If you are really interested in reducing time smear, according to the way this article models capacitors, you are going to be much better off using parallel capacitors instead. That is, if you need a coupling cap of 4uF, use 4 x 1uF capacitors or 2 x 2uF. This should reduce the memory effects substantially." This is not the conclusion I come to.DA is an inherent property of the dielectric. If you look at the model, reducing DA would require either greatly increasing the Rs or reducing the Cs of the "parasitic" network. I don't see how parallelling "macro" caps would accomplish this.
Pease really only covers two means of getting around the issue: 1. create a compensating network and, in effect insert it into a corrective feedback loop. 2. Use better quality caps to begin with. Number 1 will be difficult to accomplish in an audio circuit. To my way of thinking, you'd be better off designing a direct coupled circuit not requiring caps. Number 2 is kind of obvious. What might be surprising to some is that the NPO/COG ceramic cap is pretty darn good at least WRT DA. This also is consistent with my expereince where NPO ceramics are sometimes found in good sounding vintage audio equipment.
Note that the article is written in the context of instrumentation and D to A apps where timers and sample & holds are involved. Presumably while in the hold mode, the cap is being "measured" by an amp with very high input impedance. In audio coupling applications the relative impedance will be much lower and the memory effect much less pronounced. Also, there are other characteristics of real world caps that may have a greater affect on what we hear. Bottom line for me is simple: use quality PP caps wherever possible and don't worry about DA.
As always, YMMV. And if you think my interpretation of the article is flawed feel free to point it out.
DA is an inherent property of the dielectric AND construction.As i understand it, if you took the capacitor and unrolled it, what you would end up with is are two long strips, separated by the dielectric, with the electrodes connected either at one end, or the middle.
So, the problem with DA really has to do with the fact that some of the material is physically and electrically closer to the electrodes than the rest. You can eliminate DA by making all of the dielectric be equally spaced from the conductor. This is unfortunately, impossible.
So, one way to come closer to this is to parallel multiple caps. For instance, a 4uF cap may have a 4" wide strip (I'm making this up). With the maximum distance from the electrode being 2".
By using 2uF caps instead, you get the same capacitance as before, but the film strips are only 2" wide for each capacitor, in effect getting closer to this ideal.
But you know, this is all highly theoretical. I wish I had the time to construct Pease's test harness, I would love to try this with some inexpensive Solen's and see if there is any merit to what I'm saying or not.
Regards,
"DA is an inherent property of the dielectric AND construction."
Construction plays little role in DA. Refer to the articles by Pease and now Sencore courtesy of Jim McShane. Nowhere do they consider ANYTHING except dielectric materials as the primary factor in DA."As i understand it, if you took the capacitor and unrolled it, what you would end up with is are two long strips, separated by the dielectric, with the electrodes connected either at one end, or the middle."
Although many caps consist of "rolled" construction, end terminations are almost never done as you describe: inductance would be problematic."So, the problem with DA really has to do with the fact that some of the material is physically and electrically closer to the electrodes than the rest. You can eliminate DA by making all of the dielectric be equally spaced from the conductor. This is unfortunately, impossible."
If I understand you assertion correctly (and I may not) caps employing your "ideal" construction exist and are readily available and still have DA issues. One term for them is "stacked film" IIRC. In some respects these caps could be considered a physically large # of small caps in parallel...and they still possess significant DA if made of something like polyester. So much for parallelling to get rid of DA.Capacitors have been around a very long time and like most objects of human endeavor, have reached a very high level of development. If parallelling caps could reduce DA, commercial designs would have incorporated some version of it decades ago...but they haven't and DA is still with us.
Well, actually the Multicap line does exactly this, so there are in fact commercial designs which use this principle. However, whether they intended for it to reduce DA or inductance I do not know.Regards,
"So, the problem with DA really has to do with the fact that some of the material is physically and electrically closer to the electrodes than the rest. You can eliminate DA by making all of the dielectric be equally spaced from the conductor."The geometry is not the only factor, in fact it's not even the primary factor. Otherwise why would there be a large difference between various materials used as the dielectric?
Here's an excellent definition of dielectric absorption:
"That property of an imperfect dielectric whereby there is an accumulation of electric charges within the body of the material when it is placed in an electric field." (my italics)
And if you want to know more try the link below. It is a great bit of work.
I'm not denying that the materials matter, at all.I'm just saying that given the same material, having a capacitor which uses geometry with there is less distance between the furthest reaches of the film and the electrodes, you would get less soakage.
I will read your link, perhaps I'm not understanding it well enough yet. :)
"If you are really interested in reducing time smear, according to the way this article models capacitors, you are going to be much better off using parallel capacitors instead. That is, if you need a coupling cap of 4uF, use 4 x 1uF capacitors or 2 x 2uF. This should reduce the memory effects substantially."This is exactly the construction that has been touted by Dick Marsh for his "self-bypassing" Multi-Cap line. Multiple elements paralleled to build a larger value cap in one package seems to fit right in with the points made in the article. It would also seem to be quite advantageous to have those (4) 1 uf caps you mention be as precisely matched as possible. Obviously if you had four caps of different uf values and were to discharge or charge them at the same rate, then at any given time the state of charge of each cap will be different. So unless the four caps you describe (or the Multi-Cap) have tight matches, could the accumulation of the smaller errors by multiple capacative elements be more harmful than one error from a single element??
Bob Pease's writings correspond closely to the famous Jung/Marsh "Picking Capacitors" article from Audio in 1980. Prior to that article being published, the "hot" caps for audio were tantalum electrolytics!
The one danger in this is negelcting to take a "wholistic" approach to it. A single cap of any type/configuration has to be evaluated as part of the circuit/device as a whole. There are other factors to be considered. For example, does the presence of a high DC voltage across the cap change the results? In their article Jung/Marsh reported that the application of a "polarizing" voltage improved the performance of some caps. It would be interesting to see the results in Pease's figure 7 with 200-400 volts DC applied along with the AC signal. Etc., etc...
And in the case of audio gear, what is impact of the caps' D/A on tone? Many here love oil caps - which could lead you to think that either:
1. They are poor judges of tone quality (not likely in any event)
2. D/A is irrelevant (or even beneficial??) in audio frequency circuitry (I don't think so...)Or my personal conclusion:
3. D/A performance is one of MANY variables that contribute to the tone quality of a given unit.
This post is a great thought (re) starter! Thanks for bringing this all to mind again.
Hi Jim,From what I read of the article, if I understand the model correctly, matching caps isn't that big of a deal if you use parallel capacitors. For a couple of reasons. First, the more capacitors in parallel you use, the more likely you are to get an overall capacitance very close to the average of the manufacturer. So, if you are matching the coupling caps for 2 different channels, for instance, and use 10 random parallel capacitors for each channel, your likely to have much better channel to channel matching than if you used 1 randomly chosen capacitor per channel. In this case, random distribution and statistics work in our favor.
The other thing is, as I understand the model, even if you have large variations in the individual capacitor, say +- 10%, the amount of the total that contributes to the time smear is much smaller, so if you cut down a 4uF cap to individual 1uF caps, you will be cutting the latent storage amount as well, so the amount of variance in the latent storage should be small too. Again, if anything, random variations would work in our favor.
However, when dealing with only 2-4 parallel capacitors, it's probably not a bad idea to match them anyway. You could always have a cap that is well out of whack which can throw the whole thing off, especially if only using 2 per channel.
What I'd be curious to know now is whether anyone wants to take up the challenge of actually measuring the relative time smear of some high-end capacitors, say a Solen, Auricap and Multicap, so we can compare. Also, to see how well say a single Solen works compared to 4 parallel, or one in parallel with a teflon cap and why?
If I wasn't in school all the time, I'd consider doing it myself. I wonder how many "high-end" capacitors would actually come out as being better than some mid-end electrolytics or worse? ;)
Take care,
"From what I read of the article, if I understand the model correctly, matching caps isn't that big of a deal if you use parallel capacitors. For a couple of reasons."Hmm, okay, let's continue.
"First, the more capacitors in parallel you use, the more likely you are to get an overall capacitance very close to the average of the manufacturer."
Sure, that's absolutely true. Although that may be a moot point in audio, as rarely do we need such precision as relates to the absolute value.
"So, if you are matching the coupling caps for 2 different channels, for instance, and use 10 random parallel capacitors for each channel, your likely to have much better channel to channel matching than if you used 1 randomly chosen capacitor per channel. In this case, random distribution and statistics work in our favor."
Again, that's true - but does it matter? If one channel uses a .4699 uf coupling cap and the other channel uses a .4582 uf cap (just picking numbers out of the air) the difference is insignificant. It may also be swamped by the differences in the resistance of the next stage, etc.
"The other thing is, as I understand the model, even if you have large variations in the individual capacitor, say +- 10%, the amount of the total that contributes to the time smear is much smaller, so if you cut down a 4uF cap to individual 1uF caps, you will be cutting the latent storage amount as well, so the amount of variance in the latent storage should be small too. Again, if anything, random variations would work in our favor. "
I'm not sure I agree. Dielectric absorption is measured in percentages. As a result, the distortion caused by D/A would be additive in multiple caps, not the average. Since the D/A % is always positive, there is no cancellation it seems to me. The net effect would be that it is entirely possible that the four caps could increase the "smear".
"However, when dealing with only 2-4 parallel capacitors, it's probably not a bad idea to match them anyway. You could always have a cap that is well out of whack which can throw the whole thing off"
My point exactly. My concern (and I don't really know the answer, although I suspect I do) is that small differences in D/A among individual paralleled caps may be worse than the single cap D/A.
Somebody a lot smarter than me (that means almost everyone reading this!) care to comment?
And thanks for your reply to my post too - you got my brain in gear again!!
If less "time smear" is the goal I can't see how more caps are better. But maybe we could make an argument for how quickly the multiple smaller caps would recover from blocking?For coupling, I’d use one cap and deal with blocking by other means. But I can see the wisdom of unequal value power supply caps in parallel. Maybe like a 1/3 and 2/3 size to equal desired amount?
Funny how we get so wrapped up in this in a tube asylum. Loads of stuff is forced to use something like a 100uF/25V electrolytic coupling cap. In a push pull amp we are blessed with the ability to use great caps if we have the room.
Sorry Russ, what is blocking?
Soakage can be nasty: my day job includes sample-holds and analog integrators using op amps. Physically, soakage is a symptom of the electrostatic equivalent of magnetic hysteresis in ferromagnetic materials. Dielectric effects, including the dielectric constant, are due to polarization of molecules inside the dielectric material. Not surprisingly, as a rule, soakage is worse with higher dielectric-constant materials. [Mylar] > [polypropylene] and [X7R ceramic] > [NP0 ceramic].
Fantastic article. Makes me want to replace those mylar coupling caps in my amp with Russian teflons. I wonder if this means that those old PIO motor run caps are not as good as their modern metalized polystyrene replacements. I guess that I should measure and see.
Well, the good point of this article is that you actually CAN measure the amount of latent storage in a capacitor.So we could conduct experiments for instance on different "high end" caps, and see how well they perform with bypass caps and without, and then compare that to parallel main caps instead.
Regards,
This post is made possible by the generous support of people like you and our sponsors: