Scott: When somebody stands up and actually presents documentation of something, and i dont mean this kind of 'documentation' :
'I have given considarable thought to it, and it doesnt matter'.
It's always easy for somebody to say:
I know all about this, and your results are wrong.
What is maybe a little harder, but a whole lot more interesting is: Can you present some kind of documentation to prove my measurements wrong? Since you have such vast experiences in this field, you probably have in your drawer some of your numerous measurements to show ?
'I have given considarable thought to it, and it doesnt matter'.
It's always easy for somebody to say:
I know all about this, and your results are wrong.
What is maybe a little harder, but a whole lot more interesting is: Can you present some kind of documentation to prove my measurements wrong? Since you have such vast experiences in this field, you probably have in your drawer some of your numerous measurements to show ?
Scott: I know what my measurements show, and what they mean in a real life Audio environment.
Whether you agree with my theory or not is really up to you.
I'm not trying to sell you these results, just want to show all you guys one way to improve your DIY amplifiers, by eliminating power supply resonances. And all it takes is a simple measurement with a Tone Generator, power amplifier and a MilliVolt Meter.
If you don't have access to these instruments, you can also avoid these resonances by taking a few precautions in your power supply design:
1..Don't parallel multible capacitors directly.
2..Use only the nessescary amount of uF, bigger is only better up to a certain level. How much is nessescary varies from amplifier to amplifier and i some amplifiers the loudspeaker load also becomes a factor for the power supply size. Can be found with listening tests.
3..If you have to parallel capacitors, do it intelligently, in a way to prevent the capacitors from interacting. The V4P scheme above is one way of doing it, there may very well be other ways too.
Whether you agree with my theory or not is really up to you.
I'm not trying to sell you these results, just want to show all you guys one way to improve your DIY amplifiers, by eliminating power supply resonances. And all it takes is a simple measurement with a Tone Generator, power amplifier and a MilliVolt Meter.
If you don't have access to these instruments, you can also avoid these resonances by taking a few precautions in your power supply design:
1..Don't parallel multible capacitors directly.
2..Use only the nessescary amount of uF, bigger is only better up to a certain level. How much is nessescary varies from amplifier to amplifier and i some amplifiers the loudspeaker load also becomes a factor for the power supply size. Can be found with listening tests.
3..If you have to parallel capacitors, do it intelligently, in a way to prevent the capacitors from interacting. The V4P scheme above is one way of doing it, there may very well be other ways too.
Lars Clausen said:
Lars,
I'm not sure why you've taken offense to my last post. I'm very interested in this thread and would like to quantify the issue you have brought up if it's possible. I apologize if you thought my tone was out of line, I certainly didn't mean my questions as an insult in any way. My sincerest apologies.
I have measurements that I can post when I can get back to my desk in San Antonio...which is an hour and a half away, unfortunately I won't be there until next week.
You've posted some measurements of impedance and phase of capacitors and are trying to post I don't know if these were taken with a Vector Network Analyzer or not, but I will assume for the time being that they were.
I don't know the following things from your post:
1. Drive level, how much signal were you using 0dBm? 0dBV? 0dBmV?
2. Test circuit, were you connected single ended to the S11 port of your instrument?
3. Calibration, cable zeroing? Did you have calibrated test loads of an open circuit, short circuit and some nominal impedance for the frequency range of the measurement. Were these devices used at the end leads of the test interface, eliminating impedance transformation from the test instrument leads?
The magnitude and phase posts you showed in your plots shows a lot of spikes in the phase portion (blue right?) The first cap is a 47uF Black Gate. The impedance is at the 4'th division from the bottom (counting the bottom as 1) The phase line is very clean.
The next two caps, 100uF and 470uF show the impedance to be almost at the bottom. If the bottom is 0 ohms and the top is 1 ohms, how come the black gate capacitor even has the trace on the screen at the 100Hz frequency. 1/(2*3.14*f*c) yields an impedance magnitude of 33 ohms. ~15 ohms for the 100uF and 3.4 ohms for the 470uF.
Now, if your instrument is measuring phase by comparing voltage and current at the output port, and using a PLL to lock on to them both it looks to me like your instrument is having a tough time with low impedances. In the second measurement, the magnitude starts to rise drastically about half way accross. When it goes above the second line, the phase line gets very clean. This indicates to me where the minimum impedance that the instrument can measure the phase between it's output voltage and current. Notice that the Black gate capacitor spends the entire measurement above that impedance point, so the phase can be accurately determined and doesn't show the noise. Notice that the 470uF cap spends the entire measurement below that impedance point, and the instrument is having trouble.
If you are driving a very low impedance the voltage on the output port will be limited because of the internal impedance of the instrument will limit output current. The instrument will then try to lock onto that low voltage, very often locking onto noise, and therefore the reason for the spikes in the phase response.
Scott
What you're missing to show that this isn't random baseline noise is several replications, with all the peaks and valleys showing up in the same places to show that whatever phenomenon you're seeing, it's systematic. To do that properly, you need to connect and disconnect the caps for the replication runs.
Additionally, you need to show the same curves with n times as many averages; if the noise reduces by a factor of the square root of n, it's detector noise. If it doesn't, it's source noise of some sort; it might be "capacitor resonances", it might be something else, but at least you've eliminated half of the possibilities.
You also need to run true baselines with the same signal levels and number of averages so that the intrinsic baseline noise can be known.
You've made a start here, but I'd strongly suggest that you don't have nearly enough data for anyone to conclude that "capacitor resonances" are the source of the choppy-looking baseline.
Additionally, you need to show the same curves with n times as many averages; if the noise reduces by a factor of the square root of n, it's detector noise. If it doesn't, it's source noise of some sort; it might be "capacitor resonances", it might be something else, but at least you've eliminated half of the possibilities.
You also need to run true baselines with the same signal levels and number of averages so that the intrinsic baseline noise can be known.
You've made a start here, but I'd strongly suggest that you don't have nearly enough data for anyone to conclude that "capacitor resonances" are the source of the choppy-looking baseline.
Scott: Happy tone again 
Sorry, i did not make these measurements personally, so i really can't answer your questions. But you are right they were made with a Vector Network Analyser.
SY: Good points ... I am not sure about the noise floor of this analyser, so maybe it's better to take some measurements with another set of instruments where i know the noise floor. However it will only affect the ESR part, as i think the phase noise is not in conflict with any noise floor.
What i will do is go to my office and repeat the actual measurements i presented, and also find the noise floor of the instruments involved. This way i can also present documentation of the real claim here by taking similar measurements of parallelled caps to compare.
Sorry, i did not make these measurements personally, so i really can't answer your questions. But you are right they were made with a Vector Network Analyser.SY: Good points ... I am not sure about the noise floor of this analyser, so maybe it's better to take some measurements with another set of instruments where i know the noise floor. However it will only affect the ESR part, as i think the phase noise is not in conflict with any noise floor.
What i will do is go to my office and repeat the actual measurements i presented, and also find the noise floor of the instruments involved. This way i can also present documentation of the real claim here by taking similar measurements of parallelled caps to compare.
Lars, thanks. Be aware that a phase plot will ALWAYS look spikier than the magnitude plot from which it's derived. That's because of the differentiation nature of their inter-relationship. This is a minimum-phase device, so the two are related by a simple derivative, i.e., the phase is proportional to the derivative of the magnitude with frequency. I won't say "Hilbert transform" because that sounds so geeky.
Even if you know the noise floor of an instrument, you still should be doing (at minimum) the sort of control runs that Scott and I have suggested. And be prepared to be disappointed if the data don't support your hypothesis; nature is unkind in that way all too often, as proved to me by the number of my own seemingly-wonderful ideas that have ended up in the trash bin.
Even if you know the noise floor of an instrument, you still should be doing (at minimum) the sort of control runs that Scott and I have suggested. And be prepared to be disappointed if the data don't support your hypothesis; nature is unkind in that way all too often, as proved to me by the number of my own seemingly-wonderful ideas that have ended up in the trash bin.
Well i have to admit that with the instruments available on my workbench i have not been able to measure the effects i had hoped for, so far. The Chemicon's behave nicely through the entire audio band with low and stable ESR, and the problems don't start until way above the audio band. The resonances i have been able to measure did not seem to change significantly when parallelling several caps.
But in my setup (shown in this thread) i can only measure magnitude NOT phase continuance.
Also i don't have the network analyser at hand right now.
In the next few days i will try to rethink how i can measure the effects using normal benchtop instruments. Maybe it's possible to measure the current flow between two parallel capacitors without disrupting the high Q connection.
For now it's saturday night here in Denmark, and my favourite pub is calling ... Have a good saturday everyone
But in my setup (shown in this thread) i can only measure magnitude NOT phase continuance.
Also i don't have the network analyser at hand right now.
In the next few days i will try to rethink how i can measure the effects using normal benchtop instruments. Maybe it's possible to measure the current flow between two parallel capacitors without disrupting the high Q connection.
For now it's saturday night here in Denmark, and my favourite pub is calling ... Have a good saturday everyone
Time to throw my hat into the ring. I’ve learned an immense amount of material over the past years, mostly from long-standing “experts” on the internet. For example, I have been an apprentice of Linkwitz in speaker design. In amplifier design, I have become a student of John Broskie, Steve Bench, Dr. Gizmo, Lynn Olsen, and Nelson Pass.
Each “master of the subject” has their own personal bias, and presents the apprentice with confounding and conflicting advice. One must pick and choose each opinion carefully. For example, combining Broskie’s hatred for noise, Pass’s desire for simplicity, and Olsen’s preference to coupling transformers.
Lars has some interesting ideas regarding capacitors, which he has back up with real data, which makes it worth looking at. I believe that Lynn Olsen also thinks a similar thing (Quote below from nutshellhifi regarding priorities in amplifier design. Interesting that Lynn chose filter capacitors to be the fifth most important priority):
Linkwitz http://www.linkwitzlab.com/
John Broskie http://www.tubecad.com/
Steve Bench http://members.aol.com/sbench102/aboutme.html
Dr. Harvey "Gizmo" Rosenberg http://www.meta-gizmo.com/Tri/index-1.html
Lynn Olsen http://www.nutshellhifi.com/triode1.html
Each “master of the subject” has their own personal bias, and presents the apprentice with confounding and conflicting advice. One must pick and choose each opinion carefully. For example, combining Broskie’s hatred for noise, Pass’s desire for simplicity, and Olsen’s preference to coupling transformers.
Lars has some interesting ideas regarding capacitors, which he has back up with real data, which makes it worth looking at. I believe that Lynn Olsen also thinks a similar thing (Quote below from nutshellhifi regarding priorities in amplifier design. Interesting that Lynn chose filter capacitors to be the fifth most important priority):
Linkwitz http://www.linkwitzlab.com/
John Broskie http://www.tubecad.com/
Steve Bench http://members.aol.com/sbench102/aboutme.html
Dr. Harvey "Gizmo" Rosenberg http://www.meta-gizmo.com/Tri/index-1.html
Lynn Olsen http://www.nutshellhifi.com/triode1.html
Ach! Why does one worry about current transients in a Class-A amplifier?
Properly designed, the Class-A amplifier power supply would operate in steady-state. Low frequencies would not demand any more power than high frequencies. Therefore, why would the “speed” of a PSU capacitor matter?
The only reason capacitors are used in a Class-A amplifier is to filter out the rectified DC. This is usually 100Hz or 120Hz (corresponding to 50Hz and 60Hz line voltage). This is why “capacitor multipliers” work with Zen amplifiers.
Properly designed, the Class-A amplifier power supply would operate in steady-state. Low frequencies would not demand any more power than high frequencies. Therefore, why would the “speed” of a PSU capacitor matter?
The only reason capacitors are used in a Class-A amplifier is to filter out the rectified DC. This is usually 100Hz or 120Hz (corresponding to 50Hz and 60Hz line voltage). This is why “capacitor multipliers” work with Zen amplifiers.
I just found the reason why my 2 hour attempt last saturday didn't give any usable measurements of current wobbling. I was looking the wrong place, by measuring the spectrum of the voltage resosnances across the terminals of the capacitor. However the point is that this is exactly where the current exchange products null themselves out, and so gives no usable result.
Right now i'm working on another project, but before too long i will device a usable test on how to measure and prove the existance of current wobbling. Then i will publish in this thread.
I know someone once said: 'I have made an
allowance for the actual existence of such a thing. I merely
dismiss it as not a real problem.'
However in this case we have found several years ago that it IS a real problem, in Audio Amplifiers, otherwise we would never have known about this effect in the first place.
I have described the effects soundwise earlier in this thread. Can be recreated by anyone.
After that we devised a solution to address the theoretical removal of current wobbling. (The V4P interconnection). And the solution worked. It solved the (audible) symptoms of the wobbling effect. Now we are working to also devise a test that will actually show the effect on instruments. May be a harder task to get a picture of the beast, but i think definitely worthwhile.
So i would say: We have established that it IS a real problem, we know (from one of the worlds greates capacitor makers) the thing exist, and we will eventually show you the hard evidence too.
Right now i'm working on another project, but before too long i will device a usable test on how to measure and prove the existance of current wobbling. Then i will publish in this thread.
I know someone once said: 'I have made an
allowance for the actual existence of such a thing. I merely
dismiss it as not a real problem.'
However in this case we have found several years ago that it IS a real problem, in Audio Amplifiers, otherwise we would never have known about this effect in the first place.
I have described the effects soundwise earlier in this thread. Can be recreated by anyone.
After that we devised a solution to address the theoretical removal of current wobbling. (The V4P interconnection). And the solution worked. It solved the (audible) symptoms of the wobbling effect. Now we are working to also devise a test that will actually show the effect on instruments. May be a harder task to get a picture of the beast, but i think definitely worthwhile.
So i would say: We have established that it IS a real problem, we know (from one of the worlds greates capacitor makers) the thing exist, and we will eventually show you the hard evidence too.
Lars Clausen said:
If I remember correctly the manufacturer who alerted you to this was Sprague/Chemi-Con? Do you remember the power supply application they were targeting? By any chance was it the Telecom/Computer industry where there are much higher dI/dT demands placed on power supplies due to all of the high-speed logic and memory?
I've looked at the Chemi-Con site and could not quickly find any application notes. If anybody has a link to the paper, it would be appreciated.
Scott
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.