Bruno, I get a real good laugh out of these suggestions that low ESR caps might be a good idea on a SMPS, I guess that stability criterion are not well understood by the ClassDIY forums.
IMHO One of the best tests for a SMPS is the transient load test, if you have $20K or so you can buy some good transient load testers. otherwise get a few big fets and a handful of suitable sized load resistors add a little micro to run the whole show and you have a home brew transient load tester.
As the name implies they change the Load between two set resistive loads and analysis the resultant response of the SPMS. From my experience there are many situations when an otherwise well performing SMPS will exhibit undesirable traits under transient load changes. I had an engineer tell me a few years ago that he had no stability problems with a buck converter that was in production in very high volume. well I ran spice and didn’t like what I saw so dragged in my transient load tester (you'll understand why you need to drag it if you get one) and found that with the right combination of load changes I could make the buck converter go completely unstable. When I found the engineer and took him back to the lab I almost fell of my stool when he decided that everything was OK because the ripple was within spec, he didn't care that the feedback loop was oscillating.
From my own limited experience with ClassD audio amps I have seen the same issues arise and go completely unnoticed except that it did not sound good, so when we looked closely at the amplifier output for a variety of dynamic signal changes we found that there was a lot of ringing on the audio signals. Also when the control loop went unstable we found that the characteristics of the noise shaping completely changed, the modulator basically changed from being synchronous to being hysteretic and with it there was a substantial increase in IMD that was related to the frequency that control loop oscillation.
The point of this little story is that the route cause was that a well intentioned engineer had specified a low ESR cap (good for Audio) on the output of the classD amp. When we replaced the cap with a cheap electrolytic or added a little series resistance the problem was fixed and the soundstage went from being muddy to very clean. ERS in Caps is not always a bad thing!!
ClassDunce
IMHO One of the best tests for a SMPS is the transient load test, if you have $20K or so you can buy some good transient load testers. otherwise get a few big fets and a handful of suitable sized load resistors add a little micro to run the whole show and you have a home brew transient load tester.
As the name implies they change the Load between two set resistive loads and analysis the resultant response of the SPMS. From my experience there are many situations when an otherwise well performing SMPS will exhibit undesirable traits under transient load changes. I had an engineer tell me a few years ago that he had no stability problems with a buck converter that was in production in very high volume. well I ran spice and didn’t like what I saw so dragged in my transient load tester (you'll understand why you need to drag it if you get one) and found that with the right combination of load changes I could make the buck converter go completely unstable. When I found the engineer and took him back to the lab I almost fell of my stool when he decided that everything was OK because the ripple was within spec, he didn't care that the feedback loop was oscillating.
From my own limited experience with ClassD audio amps I have seen the same issues arise and go completely unnoticed except that it did not sound good, so when we looked closely at the amplifier output for a variety of dynamic signal changes we found that there was a lot of ringing on the audio signals. Also when the control loop went unstable we found that the characteristics of the noise shaping completely changed, the modulator basically changed from being synchronous to being hysteretic and with it there was a substantial increase in IMD that was related to the frequency that control loop oscillation.
The point of this little story is that the route cause was that a well intentioned engineer had specified a low ESR cap (good for Audio) on the output of the classD amp. When we replaced the cap with a cheap electrolytic or added a little series resistance the problem was fixed and the soundstage went from being muddy to very clean. ERS in Caps is not always a bad thing!!
ClassDunce
Hi,
from my limited experience i would say that any control loop design that relies on output capacitor ESR zero to close the loop in a stable manner is a bad idea from the very beginning. What do you do when the purchasing department tells you that they can't get your favorite caps? Redesign the control loop?
Otherwise I agree, ESR sometimes is not a bad thing.
Best regards,
Jaka Racman
from my limited experience i would say that any control loop design that relies on output capacitor ESR zero to close the loop in a stable manner is a bad idea from the very beginning. What do you do when the purchasing department tells you that they can't get your favorite caps? Redesign the control loop?
Otherwise I agree, ESR sometimes is not a bad thing.
Best regards,
Jaka Racman
Low ESR Caps
Hi
I think when the word low ESR is used most people take this to mean LOW ESR eletrolytics used for switch mode power supplies. I believe these capacitors sound bad because of their construction and not because their low ESR . Other caps which also have low ESR as one of their characteristics can also sound good like the OSCONs with copper legs.
Regards
AP
Hi
I think when the word low ESR is used most people take this to mean LOW ESR eletrolytics used for switch mode power supplies. I believe these capacitors sound bad because of their construction and not because their low ESR . Other caps which also have low ESR as one of their characteristics can also sound good like the OSCONs with copper legs.
Regards
AP
LESR bashing!
I would tend to use Low ESR on the Bridge to hold up a local voltage. If I had them in the output filter they would be used in a notch application.
When designing a contol loop one assumes all supply voltages will be stable - external to the loop. Perhaps you could add some extra parameters to the transfer function regarding rail voltage instability and make a really smart product!! You probably redesigned that well intentioned engineers product with these upgrades ;-)
With regard to ringing on the audio signals, a classd amplifier employs an output LC filter which is load dependant, we can not expect the LC filter to work well with every load - even if like Mueta the loop is closed after this filter. If the load is specified for 4 ohms we might see an underdamped response at 8 ohms, and an overdamped response at 2 ohms - critically damped at 4.
I would tend to use Low ESR on the Bridge to hold up a local voltage. If I had them in the output filter they would be used in a notch application.
When designing a contol loop one assumes all supply voltages will be stable - external to the loop. Perhaps you could add some extra parameters to the transfer function regarding rail voltage instability and make a really smart product!! You probably redesigned that well intentioned engineers product with these upgrades ;-)
With regard to ringing on the audio signals, a classd amplifier employs an output LC filter which is load dependant, we can not expect the LC filter to work well with every load - even if like Mueta the loop is closed after this filter. If the load is specified for 4 ohms we might see an underdamped response at 8 ohms, and an overdamped response at 2 ohms - critically damped at 4.
Hi,
I am sure Bruno will disagree strongly. But active damping of filter resonance is nothing new. All that is needed is one R and one C and it works with any modulation, fixed frequency or self oscillating. Basically one feeds replica of the filter capacitor current back to the input of the modulator. You can see quick simulaton results here
http://www.diyaudio.com/forums/showthread.php?postid=213929#post213929
http://www.diyaudio.com/forums/showthread.php?postid=213932#post213932
Best regards,
Jaka Racman
I am sure Bruno will disagree strongly. But active damping of filter resonance is nothing new. All that is needed is one R and one C and it works with any modulation, fixed frequency or self oscillating. Basically one feeds replica of the filter capacitor current back to the input of the modulator. You can see quick simulaton results here
http://www.diyaudio.com/forums/showthread.php?postid=213929#post213929
http://www.diyaudio.com/forums/showthread.php?postid=213932#post213932
Best regards,
Jaka Racman
That’s the beauty of these forums; everyone is entitled to an opinion. Logically thinking, for anyone to have the correct answer someone else must be wrong, the mystery is who's who. Fortunately when things get a little hot we can always resort to techno-babble, so maybe that is how this post should start but nahhh..... bring on the insults after all I've just suggested that cheaper components can do a better job than their expensive counterparts, where would the world of "high end audio" be if this were correct.
The very concept is worse than socialism, next thing you know there will be rave reviews for that "home theater in a box" system I just bought at Sam's, for under $100, complete. How would my favorite 10000uF caps cope with the insult, it is probably more stress than their dielectrics can stand, never mind progressive dielectric breakdown we are probably talking Nuclear fusion. "It’s the end of the world as we know it......
ClassDunce
The very concept is worse than socialism, next thing you know there will be rave reviews for that "home theater in a box" system I just bought at Sam's, for under $100, complete. How would my favorite 10000uF caps cope with the insult, it is probably more stress than their dielectrics can stand, never mind progressive dielectric breakdown we are probably talking Nuclear fusion. "It’s the end of the world as we know it......
ClassDunce
You might want to test or at least simulate this first before making any such statements. Your insight is valid for amplifiers having mixed feedback (two takeoff points), but not for UcD and Mueta. Both UcD and Mueta have completely load-independent frequency (and hence impulse) responses. The loop is designed such that the physical frequency/impulse response of the filter is not even part of the frequency response equation! Only when the amp is clipped (ie when the control loop bails out), and the filter gets free reign, does the filter's own response become apparent.Dibley said:
Non-clocked without hysteresis
An alternative mechanism for achieving very similar ends is to introduce a fixed delay rather than a hysteresis. Consider a modulator loop that is an integrator with two inputs: the input analog signal and the output switching waveform. This integrator (loop amplifier) drives a comparator which drives the switching output (inverting with respect to DC input). This general description would theoretically find an analog value of stable output unless the comparator is given either hysteresis or a delay. With a delay, the switching output will drive the error amp beyond the nominal switch point before the output changes state, looking similar to the hysteretic case. The variable frequency, where frequency drops in a parabola with zero frequency as the output reaches the rail, is identical in form to the hysteretic case. But the nominal frequency is simply 1/(4*delay).
Commercial systems such as Class-T use this mechanism for self-oscillation (though Dr. Tripathi's patents fail to describe the loop in this way - rather looking at the whole loop as a sigma-delta sort of architecture, which is also a very appropriate analtical description). One advantage to a delay is that the delay is in any case inevitable (e.g. deadtime delays). Modulating this dely is not only possible but has a role in some of Tripath's patents.
I find this delay-limited loop to be intellectually desirable as it contains only "linear" elements; I find implementing precise and well controlled hysteresis more difficult. One disadvantage of a first-order delay-limited loop is that if you look at the output of the integrator with varying signal, you will find that the integrator ramps between two values that are signal dependent. In the hysteretic case, this error band is always symmetrical.
An alternative mechanism for achieving very similar ends is to introduce a fixed delay rather than a hysteresis. Consider a modulator loop that is an integrator with two inputs: the input analog signal and the output switching waveform. This integrator (loop amplifier) drives a comparator which drives the switching output (inverting with respect to DC input). This general description would theoretically find an analog value of stable output unless the comparator is given either hysteresis or a delay. With a delay, the switching output will drive the error amp beyond the nominal switch point before the output changes state, looking similar to the hysteretic case. The variable frequency, where frequency drops in a parabola with zero frequency as the output reaches the rail, is identical in form to the hysteretic case. But the nominal frequency is simply 1/(4*delay).
Commercial systems such as Class-T use this mechanism for self-oscillation (though Dr. Tripathi's patents fail to describe the loop in this way - rather looking at the whole loop as a sigma-delta sort of architecture, which is also a very appropriate analtical description). One advantage to a delay is that the delay is in any case inevitable (e.g. deadtime delays). Modulating this dely is not only possible but has a role in some of Tripath's patents.
I find this delay-limited loop to be intellectually desirable as it contains only "linear" elements; I find implementing precise and well controlled hysteresis more difficult. One disadvantage of a first-order delay-limited loop is that if you look at the output of the integrator with varying signal, you will find that the integrator ramps between two values that are signal dependent. In the hysteretic case, this error band is always symmetrical.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.