A and T labs K6 SMPS for Class D amp use

[ghemink]

A LC filter will always have a phase shift of -180 degrees and will cause instalillity of osccilation if the gain in a closed loop is close to unity at this -180 degree freqyency. If the Q factor is large, then the -180 degree frequency will be the same as the roll off frequency of the filter. If the Q value is smaller then the phase shift will reach -180 degrees at a higher frequency. The point is that It'll will always reach -180 degrees.
A differential network will give you +90 degrees of phase shift and give you a 1order rising amplitude slope. Adding such a network after the LC filter, will reduce it to a first order (-180+90 degrees=-90), which will always be stable in a closed loop.
The SMPS here is not a special case. PID control is possible and will work much better than a PI controller. A PI controller removes steady state errors but increese instability.
I've attached a schematic of the PID controller I'm using in my Class D design. I've got the abillity to add 20dB more feedback at 10kHz and increese stability by adding R53 and C24 to the feedback network. I've found the values by trail and error in pspice.

Hi Sovadk,

Thanks for your clarification. Your theory that a PID is the best controller in case of an ideal LC output filter is correct of course and I fully agree. Fact is however that when you add large electrolytic caps after the LC filter, these caps will have a parasitic series resistance in the order of at least 10mOhm or so, strongly depending on the type of cap. As a result, at higher frequencies, the output filter effectively becomes an LR filter with a phase shift at high frequencies of only -90 degrees. In the ideal case a proper designed PID controllor with +90 degree phase shift will give you a first order roll-of in combination with an ideal LC filter which rolls of with 12dB/Oct. However, in case of a low Q-factor roll-of and series resistance of the caps or externally added resistors, the phase shift will never reach -180 degrees and will converge to -90 degrees. In such a case, in my opinion and limited experience, a PID controller can not be used and one should use a PI controller to reach the desired open-loop roll-of of 6dB/Oct with -90degrees phase shift.

Best regards

Gertjan

 

[ghemink]

I mainly chose to use SMPS over a transformer because of the regulation and also because I purchased all four of them for the same price of a good toroid.

My final solution for my stability issue is 0.1ohm resistors in series with +- and ground, and an 880ohm dummy load that draws 3.5 watts placed across the rails.

Using a 600 watt load of halogen lights, the voltage drop is less than 1v, better than a trafo would do I guess. The 3.5 watts of heat from the dummy load is much more reasonable than the 10 watts from before, so I'm happy with my fix for now.

There doesn't seem to be any other problems with the setup, it works perfect from low volume all the way to full power now with no more strange clicking and tickings from the supply.

Another benefit of the dummy load is it helps discharge the 130,000uf of capacitance after I turn the amp off.

Happy to hear that you got it working, would be fun to compare a modded SMPS (meaning one with extra caps) with a conventional transformer based supply with the same amount of caps. Maybe I'll do that sometime. I expect the SMPS to win of course

Using lights as a load is indeed nice. I did that initially during SMPS testing, however, the lights have a large peak current during startup (I used conventional bulbs, maybe halogen lights are different). Then I started using resistors as loads and I added some switches so that I could quickly add more load. This allowed me to check the SMPS behavior when switching abruptly from one load to another. I used a scope to monitor the output voltage during switching of the loads so that the dynamic behavior of the SMPS could be monitored. I tried it upto about 5 amps which is about 600W load for the SMPS. The SMPS stayed cool, needless to say, the resistors became very hot

Best regards

Gertjan

 

[sovadk]

This is the amplitude and phase responce of a 350uF 20mF LC filter with and without ESR. With ESR the phase goes towards -90degrees, but it still peaks arround -180degrees with will make your closed loop unstable if the gain is too close to unity.

 

[sovadk]

I couldn't help it, so I designed the PID controller. Output filter is 350uH and 10mF. I persume that the SMPS controller has a gain of 130 (42dB).
As you can see the resonance peak of the output filter is removed and the whole thing is very stable even with no load.
If someone can give me the values of the PI controller you're using, I would very much like to do a comparison.

 

[sovadk]

Here are the orcad files if anybody are interested.

 

[Jan-Peter]

Nice work!

Did we chat with each other during the AES Conference in Denmark?

Regards,

Jan-Peter

 

[sovadk]

No I wasn't there. I'm a student from DTU at thrid semester, so the conference is way too expensive for me to participate in. Maby you'll see me there in a year or two when I've joined the Oersted institute and they'll pay

 

[sovadk]

You wanted to add a larger capasitor bank to the SMPS to improve it's transient behavior. With the right feedback network you'll actually be better off with a smaller capasitor bank.
I've made a PID controller for a system with 10mF and 470uF output capasitor (see the attacthed picture).
The graph to the left shows the output behavior to a step reference voltage af 1V. The graph to the left is the output voltage ripple when the system is loaded with 1A sinus at different frequencys.
The small capasitor bank gives more than 20dB low frequency load rejection. The reason is that you can actually add more feedback when the roll of frequency of the LC filter is larger.
The only drawback is that the ripple from the switchin stage will be larger. Still 350uH with 470uF gives you -53dB (factor 450) of supression of the 75kHz carrier.

By the way. I've optimized the PID controller for the LC filter with a 10mF output capasitor, compared to the previous post.

 

[ghemink]

You wanted to add a larger capasitor bank to the SMPS to improve it's transient behavior. With the right feedback network you'll actually be better off with a smaller capasitor bank.
I've made a PID controller for a system with 10mF and 470uF output capasitor (see the attacthed picture).
The graph to the left shows the output behavior to a step reference voltage af 1V. The graph to the left is the output voltage ripple when the system is loaded with 1A sinus at different frequencys.
The small capasitor bank gives more than 20dB low frequency load rejection. The reason is that you can actually add more feedback when the roll of frequency of the LC filter is larger.
The only drawback is that the ripple from the switchin stage will be larger. Still 350uH with 470uF gives you -53dB (factor 450) of supression of the 75kHz carrier.

By the way. I've optimized the PID controller for the LC filter with a 10mF output capasitor, compared to the previous post.

Hello Sovadk,

Thanks for all your explanations and trying to convince that PID is better. I wanted to go to PID as well but found out that for my application it was not the right way to go. So this is what I wanted to do:

The default SMPS has 1000uF caps in the output filter. I wanted to increase them since when I use them with Class D amps, they will be too small to prevent power supply pumping. Making them bigger is the only (at least simple) solution to prevent that. So that is why I need big caps.

Then I ran a bunch of simulations using switchcad to look at phase margin etc to optimize the feedback network. Then I quickly found out that using a PID with all those extra caps was becoming difficult. I actually tried a PID and it oscillated because of the too slow roll of of the LC filter. The extra caps that I have added now are 8200uf caps with 0.1Ohm resistors in a RCR configuration, the L is 300uF or 350uF and directly after the L are 2 x 470uF caps (on the SMPS PCB), the extra caps are added offboard. With the 0.1 Ohm resistors, the roll-of becomes even smoother (of course). I`m at work so don`t have the simulation results at hand. I have a couple of those RCRs in parallel, at the moment 4, so a total capacitance of 4x8200uF and series resistance reduced to 0.05Ohm because of the 4 RCRs in parallel. After thise 4 RCRs in parallel I have 1 10000uF cap, so essentially it is an RCRC with the the first RCR being 4 RCR circuits in parallel.

Another issue why I gave up on PID is that the gain at high frequencies increases and I was afraid for high frequency noise pickup as the SMPS that I have is quite large and I have the feeling it maybe a bit sensitive to HF disturbing the feedback loop when I cranck up the feedback loop gain too much as can happen with PID.

So, for now my strategy is to use PI and use a combination of passive brute force (large caps) and active force (PI with larger gain than in the original unmodded SMPS). I think that in case of the unmodded SMPS (with only the 1000uF caps) a PID may work better than the original PI. However, with the current improved PI and the extra caps, the performance I get is very good. As an audiphile and even as an engineer, it feels better to me to use more caps as it is a passive way of making things stable instead of compeletly relying on active means to improve the supply. Many people here would already argue that a regulated SMPS is no good and that the only acceptable solution is huge transformers with huge caps. I try to combine both, regulated and reasonable amount of caps to achieve a stable supply.

You can download the schematics of the SMPS at the A and T labs web-site, maybe you can find a way to get a good PID feedback network for it but I think it will be tough to design a good PID network for it when large caps are added. By the way, an SMPS does behave different under different loads. The output resistance of the SMPS depends on the current load. I have measured that and it varies largely. This has a large influence on the Q-factor of the LC filter and makes the behavior very dependent on the current pulled from the SMPS, this is also a complicating factor in the PID design and was probably one of the reason why I got oscillations. I actually measured the open-loop response of the SMPS switching stage by opening the feedback loop. The frequency response that I got was far less peaky (lower Q) than I expected based on calculations, this turned out to be in agreement with the output resistance that was strongly load dependent. So because of all these issues depending on load, designing/modifying the feedback network was actually a lot more difficult then I thought.

Have you tried your theory on a real SMPS? Would be nice to see what expected or unexpected problems you will see and what solutions you can find for the unexpected problems.

Note that this is not an attack or saying that you are wrong and I`m right. I just sum up part of my experiences with an actual SMPS and it was tougher than I initially thought.

Thanks and best regards

Gertjan

 

[sovadk]

I've haven't tryed it on a SMPS but I'm using it on my self osscilating Class D designs to improve THD specs and it works very well.
I've just designed two PID controls for you, one with the big caps and one with the small. I can se on the schematic that they attenuate the output signal with IC3. You just have to take this into account when using the design I've showed you.
Anyway. I don't think you did you PID control right, when you tryed to make it work the last time. The one's I posted here are very tollerent to changes. A 5mF cap insted of a 10mF dosn't make much difference. Also changed Q factor of the output filter dons't change anything either. The P term ensures that the phase never goes below -90 degrees. The total system phase might become +90 for some frequencys if the LC Q factor is very low, but that's stable too. It can't go wrong.

Now you're affraid that the PID controller might pick up HF noise. I understand why. I've attached at graph showing how much gain the two PID controllers have (I've changed to small C bank PID controller a bit). One have unity gain at 6.3kHz and the other at 24kHz. This is well below the carrier frequency. And have in mind that the LC filter also attenuates everything a lot.

 

[classd4sure]

sovadk,

You're either highly gifted, or schools here suck ***.

Probably both, but I'm certain of one.

 

[ghemink]

I've haven't tryed it on a SMPS but I'm using it on my self osscilating Class D designs to improve THD specs and it works very well.
I've just designed two PID controls for you, one with the big caps and one with the small. I can se on the schematic that they attenuate the output signal with IC3. You just have to take this into account when using the design I've showed you.
Anyway. I don't think you did you PID control right, when you tryed to make it work the last time. The one's I posted here are very tollerent to changes. A 5mF cap insted of a 10mF dosn't make much difference. Also changed Q factor of the output filter dons't change anything either. The P term ensures that the phase never goes below -90 degrees. The total system phase might become +90 for some frequencys if the LC Q factor is very low, but that's stable too. It can't go wrong.

Now you're affraid that the PID controller might pick up HF noise. I understand why. I've attached at graph showing how much gain the two PID controllers have (I've changed to small C bank PID controller a bit). One have unity gain at 6.3kHz and the other at 24kHz. This is well below the carrier frequency. And have in mind that the LC filter also attenuates everything a lot.

Hi Sovadk,

Thanks for your suggestions.

Attached is the circuit I used for my simulations. This attachment shows the output filter with the additional caps and the compensation network. The output of 2 for the AC source represents the actual gain of the SMPS taking into account the attenuation of the output signal. This value of 2 for the gain is the small signal DC gain that I measured with the feedback loop cut open and nominal output voltage. The series resistance of 0.5Ohm is what I expect when maximum output current of about 10A is delivered. For smaller output currents it is a lot higher.

In following posts, my simulation results are shown.

 

[ghemink]

Hi Sovadk,

Thanks for your suggestions.

Attached is the circuit I used for my simulations. This attachment shows the output filter with the additional caps and the compensation network. The output of 2 for the AC source represents the actual gain of the SMPS taking into account the attenuation of the output signal. This value of 2 for the gain is the small signal DC gain that I measured with the feedback loop cut open and nominal output voltage. The series resistance of 0.5Ohm is what I expect when maximum output current of about 10A is delivered. For smaller output currents it is a lot higher.

In following posts, my simulation results are shown.

And here is the frequency and phase response taken after the output filter at the 470uF caps (blue lines) and after the compensation network (opamp output) (green lines).

You can see that with this compensation I get a perfectly 6dB/Oct roll of and 0dB gain at 100Hz and a very comfortable phase margin. The roll-of of the LC filter itself is also 6dB/Oct in the region of interest.

 

[ghemink]

And here is the frequency and phase response taken after the output filter at the 470uF caps (blue lines) and after the compensation network (opamp output) (green lines).

You can see that with this compensation I get a perfectly 6dB/Oct roll of and 0dB gain at 100Hz and a very comfortable phase margin. The roll-of of the LC filter itself is also 6dB/Oct in the region of interest.

And the last picture is the response of the PI compensation network only.

This network has the maximum phase shift at about 100Hz so that I get the optimum phase margin at 100Hz where I have my 0dB loop gain.

I think the way I made this compensation is not too bad and done according to the textbooks I would say.

I don't think it can be done much better but I'm open for suggestions of course.

Best regards

Gertjan

 

[ashok]

....It also has an AWG (arbitrary waveforg generator) ....

From a post buy Ghemink post #23.

It's a typo but a great one. I twist it a bit and read it as ....
arbitrary wavefrog generator .......
Nice one.
Yes it looks even nicer because I'm having a ...
Cheers.

 

[ghemink]

From a post buy Ghemink post #23.

It's a typo but a great one. I twist it a bit and read it as ....
arbitrary wavefrog generator .......
Nice one.
Yes it looks even nicer because I'm having a ...
Cheers.

What did you do looking at such an old post? yes, I have some typos every now and then as I`m one of those guys typing with only two fingers.

Cheers

Gertjan

 

[Pierre]

Only a question about A&T Labs K6 supply:

Is that BIG snubber resistor really required? In affirmative case, why does it need to dissipate so much power?
Is that related to a very high primary leakage inductance?
Or to the wide duty-cycle operating range the supply is designed for? (in order to accomodate a variety of input/output voltages?

Thanks for the clarifications

 

[ghemink]

Only a question about A&T Labs K6 supply:

Is that BIG snubber resistor really required? In affirmative case, why does it need to dissipate so much power?
Is that related to a very high primary leakage inductance?
Or to the wide duty-cycle operating range the supply is designed for? (in order to accomodate a variety of input/output voltages?

Thanks for the clarifications

I don't know if really needed, probably reduces EMI???

I can check the temperature of that resistor some time (SMPS is turned of at the moment and it is already late).

Good night

Gertjan

 

[Pierre]

Yes, EMI reduction is what snubbers are used for most of the times. But typical SMPS use up to about 1nF in series with a 5 Watt resistor, maximum, that's why I got surprised when I saw that mega-resistor in the A&T labs design.
Perhaps leakage inductance is so high that the peaks without snubber are huge and a very dissipative one is needed.

 

[ghemink]

Yes, EMI reduction is what snubbers are used for most of the times. But typical SMPS use up to about 1nF in series with a 5 Watt resistor, maximum, that's why I got surprised when I saw that mega-resistor in the A&T labs design.
Perhaps leakage inductance is so high that the peaks without snubber are huge and a very dissipative one is needed.

What is a typical value for the snubber resistor? This one is 50 Ohm. A larger one would dissipate less and could therefore be smaller.

Best regards

Gertjan