A lot of valuable "food for thought" , if I may say so.
I've had some good results with using OSCONS through the years. (Conrad Hofmann's case proven ?! )
What makes me wonder, though, is that nobody mentioned
Paul Stamler's essay in AudioXPress, whose essences are to be found here :
http://www.audioxpress.com/magsdirx/ax/addenda/media/505StamlerTables.pdf
Opinions, anybody ?
Thanx & Greez
Siggi K.
I've had some good results with using OSCONS through the years. (Conrad Hofmann's case proven ?! )
What makes me wonder, though, is that nobody mentioned
Paul Stamler's essay in AudioXPress, whose essences are to be found here :
http://www.audioxpress.com/magsdirx/ax/addenda/media/505StamlerTables.pdf
Opinions, anybody ?
Thanx & Greez
Siggi K.
Conrad Hoffman said:Here's a spreadsheet with my measurments of various capacitors at different frequencies. Possibly that can confirm the model.
Conrads Caps
This is great stuff, and should either prove revealing, or put a lot of capacitor myths to bed.
edit- typo
Thanks, Conrad.
The example model that I gave can't really be confirmed, because I already know it's not accurate.
It was just a quick example to try, for a sort of "proof of concept, and maybe also to try to motivate other people to try to help find better methods, or refine that method, for modeling the frequency-dependent characteristics of capacitors, in Spice.
In one sense, though, the modeling seems confirmed: Apparently, some type or topology of ladder network of ideal components should be able to be synthesized that models the frequency-dependence of at least the C, ESR, and L of any real capacitor, to any desired degree of accuracy.
The main problem is, every single capacitor would need its own special model. For example, if I had a good ladder-network model for the Nichicon UHE-series 220 uF/50V cap that I used for the example I posted, it wouldn't be accurate for the Nichicon UHE-series 220 uF/10V capacitor. Their ESRs, alone, are very different at each frequency (although, if we're lucky, they're related by a factor, or something).
So somebody with some nice measuring equipment will probably need to measure capacitors of every value and voltage, of every series, from every manufacturer, eventually, taking data for at least five or more frequencies for each cap (and preferably for "several" samples of each cap).
It sure would be nice if the manufacturers would do this. I'd guess that they probably HAVE done so. Are there any (besides Kemet) who have enough data actually published to make a decent model from? Impedance and ESR curves, with mutipliers (DF?) for the different voltage ratings within a series, or something like that, might do. (The one paper I referenced, by Istvan Novak, the "...BlackBox..." one, has a section near the end where he discusses which frquencies would be best, to take measurements at, or data from.)
If we could then JUST find an easy way to fit a ladder network to the data, we could crank out models like crazy!
I, for one, would absolutely LOVE to have a large library of "real" capacitor models, for Spice, even if it only included ONE manufacturer's whole series, for each type of dielectric.
I think that such a library could be widely sold, actually. "Yet-another business idea...."
Let's start a company!What might ALSO be nice would be to come up with some Spice .param statements that could take a user's relatively-simple measurements and automatically calculate the model coefficients from them (sort of like I did for the transformer model, at http://www.fullnet.com/~tomg/gooteesp.htm ). But I think that's probably being too wishful, although, it depends on the complexity of the ladder-synthesis algorithm, I guess, and on whether or not the measurements or data could be gotten in a practical manner, etc. Then again, SOME improvement in accuracy would probably be better than none. So maybe a "simplified" modeling technique should also be investigated.
At any rate, we (or I, at least) still need to find, or find out about:
1) Existing sources of capacitor data, or methods of measurement for capacitors, to get data that's sufficient for making the models.
2) An automated method of synthesizing the ladder networks, or finding the values for a predetermined ladder-network topology, based on the data collected.
Anyone?
- Tom Gootee
I consulted the oracle of Agilent 4294A and its adapter 16047E:
First they confirmed to me that most caps behave capacitive below 100kHz, which is fitting to Conrad's measurements.
Then they whispered some secrets to me:
E-caps first....
Nichicon Gold Tune, 8200uF/56V:
f_res=8.6kHz (low Q)
Zmin=15mOhms
Above 8.6kHz inductive. Phase angle at 10MHz: 85deg
Lelon A406(M), 100uF/100V:
f_res=223kHz (low Q)
Zmin=112mOhms
Above 223kHz inductive. Phase angle at 1MHz: 29deg. Phase angle at 10MHz: 79deg
Epcos B43858, 22uF/250V:
f_res=3.7MHz (low Q)
Zmin=327mOhms
Above 3.7MHz inductive. Phase angle at 10MHz: 67deg
Epcos B43857, 33uF/450V:
f_res=2MHz (low Q)
Zmin=309mOhms
Above 2MHz inductive. Phase angle at 10MHz: 78deg
AND now some film caps...
Epcos Silvercap (Polyester), 2.2uF/250V:
f_res=1.2 MHz (high Q)
Zmin=16mOhms
Above 1.2MHz inductive. Phase angle at 10MHz: 82deg
Epcos Silvercap (Polyester), 1uF/100V:
f_res=1.9 MHz (high Q)
Zmin=21mOhms
Above 1.9MHz inductive. Phase angle at 10MHz: 82deg
Noname local Chinese Mylar type, 1uF/100V:
f_res=1.9 MHz (moderate high Q)
Zmin=106mOhms
Above 1.9MHz inductive. Phase angle at 10MHz: 79deg
Noname local Chinese Mylar type, 0.1uF/100V:
f_res=5.8MHz (moderate high Q)
Zmin=187mOhms
Above 5.8MHz inductive. Phase angle at 10MHz: 61deg
WIMA MKS4, 1.5uF/630V:
f_res=970kHz (high Q)
Zmin=26mOhms
Above 970kHz inductive. Phase angle at 10MHz: 83deg
Finally a ceramic cap...
Local Chinese Keramik type, 0.1uF/50V:
f_res=5.9MHz (high Q)
Zmin=63mOhms
Above 5.9MHz inductive. Phase angle at 10MHz: 78deg
All displayed graphics of the AC sweeps were looking typical for a simple C-L-R series connection. Only few minor deviations, were visible at some types. So the simple C-L-R series model seems to be Ok to simulate the dominating behavior.
But frankly speaking you should forget to simulation to determine the values of L and R. The differences between the brands and types are to giant. You have to measure your cap and then you can put a C-L-R model in your simualtion (but don't forget to model the PCB inductances and capacitances...) and with this you might be able to predict roughly the interactions of your circuit, your PCB & your cap.
Usually I would rate the behavior below 100kHz in SMPS applications as boring. Did you ever observe issues with the behavior of your supply rails at such lowish frequencies? Usually the headache is coming from the switching transients which cause peaks & notches or ringing in the MHz range.
For me especially the epcos e-caps were stunning! If we calculate the coresponding inductance of the B43858 22uF/250V, then we get 80pH... Nobody can tell if this is the cap or the adaptor or some miscalibration. But you can be sure that your PCB will cause by far more series inductance than this cap.
First they confirmed to me that most caps behave capacitive below 100kHz, which is fitting to Conrad's measurements.
Then they whispered some secrets to me:
E-caps first....
Nichicon Gold Tune, 8200uF/56V:
f_res=8.6kHz (low Q)
Zmin=15mOhms
Above 8.6kHz inductive. Phase angle at 10MHz: 85deg
Lelon A406(M), 100uF/100V:
f_res=223kHz (low Q)
Zmin=112mOhms
Above 223kHz inductive. Phase angle at 1MHz: 29deg. Phase angle at 10MHz: 79deg
Epcos B43858, 22uF/250V:
f_res=3.7MHz (low Q)
Zmin=327mOhms
Above 3.7MHz inductive. Phase angle at 10MHz: 67deg
Epcos B43857, 33uF/450V:
f_res=2MHz (low Q)
Zmin=309mOhms
Above 2MHz inductive. Phase angle at 10MHz: 78deg
AND now some film caps...
Epcos Silvercap (Polyester), 2.2uF/250V:
f_res=1.2 MHz (high Q)
Zmin=16mOhms
Above 1.2MHz inductive. Phase angle at 10MHz: 82deg
Epcos Silvercap (Polyester), 1uF/100V:
f_res=1.9 MHz (high Q)
Zmin=21mOhms
Above 1.9MHz inductive. Phase angle at 10MHz: 82deg
Noname local Chinese Mylar type, 1uF/100V:
f_res=1.9 MHz (moderate high Q)
Zmin=106mOhms
Above 1.9MHz inductive. Phase angle at 10MHz: 79deg
Noname local Chinese Mylar type, 0.1uF/100V:
f_res=5.8MHz (moderate high Q)
Zmin=187mOhms
Above 5.8MHz inductive. Phase angle at 10MHz: 61deg
WIMA MKS4, 1.5uF/630V:
f_res=970kHz (high Q)
Zmin=26mOhms
Above 970kHz inductive. Phase angle at 10MHz: 83deg
Finally a ceramic cap...
Local Chinese Keramik type, 0.1uF/50V:
f_res=5.9MHz (high Q)
Zmin=63mOhms
Above 5.9MHz inductive. Phase angle at 10MHz: 78deg
All displayed graphics of the AC sweeps were looking typical for a simple C-L-R series connection. Only few minor deviations, were visible at some types. So the simple C-L-R series model seems to be Ok to simulate the dominating behavior.
But frankly speaking you should forget to simulation to determine the values of L and R. The differences between the brands and types are to giant. You have to measure your cap and then you can put a C-L-R model in your simualtion (but don't forget to model the PCB inductances and capacitances...) and with this you might be able to predict roughly the interactions of your circuit, your PCB & your cap.
Usually I would rate the behavior below 100kHz in SMPS applications as boring. Did you ever observe issues with the behavior of your supply rails at such lowish frequencies? Usually the headache is coming from the switching transients which cause peaks & notches or ringing in the MHz range.
For me especially the epcos e-caps were stunning! If we calculate the coresponding inductance of the B43858 22uF/250V, then we get 80pH... Nobody can tell if this is the cap or the adaptor or some miscalibration. But you can be sure that your PCB will cause by far more series inductance than this cap.
Add on:
Of course the losses a not fully independent form the frequency, but I never observed issues with this. And if you look to the overall behavior between 100Hz and 10MHz then this frequency dependency is is not a dominant effect. At least not for the electrical behavior.
Furtheron this frequency dependency might differ from what most people are assuming. Please note usually the losses at 100kHz are lower than at 100Hz. This is also fitting nicely to the fact that the allowed AC ripple currents at 100kHz are usually specified higher than the allowed ripple currents at 100Hz.
Last but not least, my statement that I never observed issues with the supply rails at lowish frequencies below 100kHz is valid for room temperature only.
At low temperatures the ESR of e-caps can dramatically increase and can cause really bad headach at low frequencies. For car applications it is worth to have a look this....
Of course the losses a not fully independent form the frequency, but I never observed issues with this. And if you look to the overall behavior between 100Hz and 10MHz then this frequency dependency is is not a dominant effect. At least not for the electrical behavior.
Furtheron this frequency dependency might differ from what most people are assuming. Please note usually the losses at 100kHz are lower than at 100Hz. This is also fitting nicely to the fact that the allowed AC ripple currents at 100kHz are usually specified higher than the allowed ripple currents at 100Hz.
Last but not least, my statement that I never observed issues with the supply rails at lowish frequencies below 100kHz is valid for room temperature only.
At low temperatures the ESR of e-caps can dramatically increase and can cause really bad headach at low frequencies. For car applications it is worth to have a look this....
...I am a little bit trouble to get the screen shots in proper electronic format.... ...would have to program visual basic or test point or similar in order to read out the IEEE interface.... sorry, I simply cannot invest so much time in this topic... Also I do not have this Analyser here at home, some drive to get there...
But the impedance shape was following mostly what you find from simple CLR series connection, - means at low frequencies dominated by Xc= 1/(2pi*f*C) then droping at the resonance frequency to the Zmin value and then increasing again and in the high frequencies dominated by XL=2pi*f*L.
Phase shift zero was also always fitting to the frequency where also the capacitance trace shows the step from pos peak to neg peak, which also is the frequency where we get Zmin.
Phase shift was moving from -90deg at low frequencies up to +60deg...+82deg at 10MHz. In the high Q caps this change in phase shift is looking very steep like step, for the low Q caps it is a softer change.
Mostly looking like in a school book...
But the impedance shape was following mostly what you find from simple CLR series connection, - means at low frequencies dominated by Xc= 1/(2pi*f*C) then droping at the resonance frequency to the Zmin value and then increasing again and in the high frequencies dominated by XL=2pi*f*L.
Phase shift zero was also always fitting to the frequency where also the capacitance trace shows the step from pos peak to neg peak, which also is the frequency where we get Zmin.
Phase shift was moving from -90deg at low frequencies up to +60deg...+82deg at 10MHz. In the high Q caps this change in phase shift is looking very steep like step, for the low Q caps it is a softer change.
Mostly looking like in a school book...
That agrees with what I've seen, though I question the value of measurements at 10Mhz. I have a big gap in my home measurements (for caps > 1uF) between about 20khz and 500khz. For small caps, the old 716-C is good to about 1Mhz. Other than those high priced Agilent/HP units, I don't know of any easy solution for large caps and high frequencies. At 500khz I can switch over to the vector impedance analyzer, which goes to 110Mhz. My experience has been that for anything over a few Mhz, lead length is fatal. Small film and ceramic caps are pretty good up there, but even short leads give an early self resonance and false indication of inductance. Little SMT caps should do far better. BTW, those are very impressive caps you listed. My run-of-the-mill parts over a few uF don't even come close.
Hi Conrad!
...yes I also have some doubts at the highest frequencies. That's why I stoped at 10MHz and did not make any use of the analyzer's ability. Even if it is one of these high price HP/Agilent beasts... you cannot avoid the influence of the outside geometry, which is never exactly the same like during calibration.
Actually I did have almost no leads. The adapter 16047E is directly coupled onto the four BNC plugs of the Agilent 4294A and offers flat massive pressing plates to connect the caps. I used the short calibration and the open calibration and activated both corrections.
So the adapter should be mostly neutral.
And in fact you can see a change in the resonance frequency when you do not connect the caps completey close. Already allowing 2mm of the component leads change the results completely , because the loop area increases and with this the series inductance. So I always inserted the leads complete inbetween the connectors plates, making the caps sitting directly on the adaptor.
In most real applications the layout of our PCB will have more influence than the inductance of the caps.
We are really talking about a few single nH or even less!
...yes I also have some doubts at the highest frequencies. That's why I stoped at 10MHz and did not make any use of the analyzer's ability. Even if it is one of these high price HP/Agilent beasts... you cannot avoid the influence of the outside geometry, which is never exactly the same like during calibration.
Actually I did have almost no leads. The adapter 16047E is directly coupled onto the four BNC plugs of the Agilent 4294A and offers flat massive pressing plates to connect the caps. I used the short calibration and the open calibration and activated both corrections.
So the adapter should be mostly neutral.
And in fact you can see a change in the resonance frequency when you do not connect the caps completey close. Already allowing 2mm of the component leads change the results completely , because the loop area increases and with this the series inductance. So I always inserted the leads complete inbetween the connectors plates, making the caps sitting directly on the adaptor.
In most real applications the layout of our PCB will have more influence than the inductance of the caps.
We are really talking about a few single nH or even less!
...here you can see a pic of the impedance analyzer:
http://www.home.agilent.com/agilent/product.jspx?nid=-536902439.536879654.00&cc=US&lc=eng
...and here a pic of the fixture, which allows to connect leaded components with almost zero effective lead length:
http://www.home.agilent.com/agilent/product.jspx?nid=-536902496.536880747.00&cc=US&lc=eng
http://www.home.agilent.com/agilent/product.jspx?nid=-536902439.536879654.00&cc=US&lc=eng
...and here a pic of the fixture, which allows to connect leaded components with almost zero effective lead length:
http://www.home.agilent.com/agilent/product.jspx?nid=-536902496.536880747.00&cc=US&lc=eng
Have you all seen the measurments here?
http://www.diyhifi.org/forums/viewt...urements&sid=7fbca9d605900ca877c92d2c124d04ec
http://www.diyhifi.org/forums/viewt...urements&sid=7fbca9d605900ca877c92d2c124d04ec
rdf said:
...thanks! This link is really not bad.
What do I learn from these threads and my own measurements?
1. My standard habit to place a ceramic 100nF at OP amp Vcc pins is not always the best idea...
2. Make everything as small as possible & hope that the resulting resonances will happen at a frequencies which do not harm to much.
3. Caps with low Q can be fortunate.
4. New problem: The more I learn, - the more difficult my next layout for my Gen2 class D amp will be.
5. Better stop learning?
ChocoHolic said:
...thanks! This link is really not bad.
What do I learn from these threads and my own measurements?
1. My standard habit to place a ceramic 100nF at OP amp Vcc pins is not always the best idea...
2. Make everything as small as possible & hope that the resulting resonances will happen at a frequencies which do not harm to much.
3. Caps with low Q can be fortunate.
4. New problem: The more I learn, - the more difficult my next layout for my Gen2 class D amp will be.
5. Better stop learning?
1) Only a start that is.
2) Hoping leads to the dark side. Only knowledge can save you.
3) A true Jedi Master how to use all tools knows.
4) Problems exist not. Learning does.
5) Obi-wan, this one is not the one we seek. Clown is he, to Las Vegas he must go.
became strong these days .FWIW, I remembered a simple way to expand my home measurement capability. My old GR 716-C bridge normally tops out at 1.1uF, but is good to 300khz, or a Mhz with somewhat reduced accuracy. But, by plugging a decent decade cap box in parallel with the internal standard, it's easy to extend the range to 1000uF or more. Just tried it, and it works great. I have a polystyrene decade box that will get me to 1000uF (bridge multiplies box by 1000), and if I can find one with another decade, that will get me decent HF measurements to 10000uF, plenty for now. There's no shortage of old 716-C bridges out there- you just need the table space to set it up.
It is said that low ESR types will improve bass control.jnb said:
But I do neither have any measurements nor any comparative listening tests to this.
My impression is more that a fat bank of caps is acting like an ideal capacitor at low audio frequencies.
The most dominant effect that I observed is voltage sagging at high currents. The observed voltage changes were mostly fitting to what one would expect just from load current and ideal cap. I should dig more detailed into this, because this is a very massive contradiction to what multiple knowledgeable people are telling me.
Any hints for non religious threads or papers are highly welcome.
Back to the HF behavior:
I tried to search the resonances of the Wima MKS4 and the local 100nF mylar type just with simple equipment.
I connected a 47Ohm resistor in series to the cap and supplied this series connection with my signal generator.
The scope was connected with GND to the center tap of cap and resistor, allowing a proper voltage and current measurement.
I resonances are so dominant that you easily find them this way (at least for film/foil/etc caps). In the resonance the voltage across the cap is becoming really small, while the voltage across the resistor remains comparably stable.
For the MKS I figured out 890kHz, well not precise but better than nothing.
For the 100nF mylar I found 2MHz.
Furtheron I looked to small caps and I can confirm the statement that small SMDs offer quite good HF behavior.
A 0805 sized 1nF cap with X7R dielectricum showed quite nice capacitor behavior up to my measurement limit (110MHz).
About all I can suggest is this- low esr is no guarantee of anything, since people constantly say polyester (Mylar) is bad, yet it's properties aren't really that bad at all. OTOH, I've never heard anybody say anything good about high esr caps. There's general agreement that electrolytics, tantalums, and soft ceramics are bad, when it comes to coupling. The numbers for PSRR have to be stretched to the breaking point to consider power supply caps part of the signal chain, but if you do that, the same logic should apply to power supply caps. My opinion at the moment is that, like many "x-factor" reasons for sound differences, if there's any actual difference at all, it's extremely minor compared to things that do matter, like good local bypassing, RF elimination, correct grounding to keep AC currents out of the system, stability, and possibly a "good" distortion spectrum for whatever distortion does exist, as opposed to a "bad" one.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.