current mode control

[classd4sure]

Why isn't it used more?

I'm thinking of a sort of Mueta style amp.. with single cycle error correction.

I think the biggest issue would be with generating a good reference, I don't like Mueta's method though it's probably very good, some other genius patented the use of a DSP for it, what else might work?

Simplicity, in this case, should even overule the "hi-fi" aspect.

All thoughts /ideas are welcome, certainly if anyone's tried it I'd like to hear from you.

Regards,
Chris

 

[Jaka Racman]

Hi Chris,

maybe you can start with the Leapfrog thread. The main problem I see is easy current sense pickup, although one proprosed by analogspiceman should work. I have tried current mode control and it worked reasonably good, although implementation was a bit impractical. I used LEM closed loop hall sensor module for current feedback and hysteretic (bang-bang) control for current loop.

With full bridge power stage current sense is much easier to implement (one curent sense resistor in source lead of each bottom Mosfet and diff amp).

Best regards,

Jaka Racman

 

[analogspiceman]

Hi Chris, maybe you can start with the Leapfrog thread. The main problem I see is easy current sense pickup, although one proposed by analogspiceman should work.

With full bridge power stage current sense is much easier to implement (one current sense resistor in source lead of each bottom Mosfet and diff amp).

Thanks for the reference, Jaka. I having been hoping someone would use that schematic to attempt to build a working circuit and report back. I am very busy working in a non audio field at the moment so I haven't had the time to do anything myself.

If there is a DIYer here willing to try building it, I am willing to draw up a complete schematic and perhaps lay out the circuit board as well (in order to share files, I was thinking of using one of those online layout board houses - ideas?).

With regard to the full bridge, two MOSFET, current sensing idea, I built and tested that very circuit for a product several years ago. It works but suffers troubling distortion proportional to dead time. The problem occurs when the bottom MOSFETs are either both simultaneously on or off. This causes the sensed feedback signal to either erroneously double (looking like shoot through) or drop out during dead time. To achieve audiophile fidelity, sensing all four MOSFETs really is required (per the half bridge sense circuit in the thread you referred to).

Best regards -- analogspiceman

 

[classd4sure]

Hi Jaka,

Thank you for that reply it's very much what I was after.

I've spent a good hour already this morning on the Leap Frog thread, just hunting for ideas. It's part of a loooong list I'm going through.

That particular current sense method seems...... interesting.

Renee (sp?) from Mueta recommended a few coils of coax. I guess with the shield as the sensing coil. It wouldn't saturate anyway if no core is used right, but I see now what 'spiceman was saying about wire resistance drifting with temp is likely to "very" much apply with coax, since the wire and the shield are very different, and I highly doubt their temp co-efficients are anywhere near matched, perhaps it could still be alright for low power though?? It would be nice to use coax since it's so easily found.

I've seen various other sense methods as well from patents which Mueta's reference, from Bose and Carver, mainly the Carver patent though.

I think I'd prefer doing it passively for a few reasons.

I also think the Leap Frog thread died a little early in life.

In particular your Q#3 was never really answered clearly it seems.

Here it is again:

"-do you think that LF method could be improved by summing output of voltage error amplifier with differentiated input signal (capacitor current feedforward like in Mueta)"

To which the reply was:

" 3) ??? Leapfrog already feeds forward the output capacitor's load current. Did you mean something else?"

I kind of have the impression you did in fact mean something else, thought maybe it's not so different as it seemed?

http://v3.espacenet.com/origdoc?DB=EPODOC&IDX=WO0042702&F=0&QPN=US5606289&RPN=US5606289

Page 14 shows a few ways of sensing output cap current, one of which is your hall sensor. I loved the idea at first (found that one yesterday), but when I said quality could suffer I just didn't expect to strike gold first time out, I do want a full bandwidth audio amp though, which from what I've seen rules out hall effect sensors due to their limited bandwidth.

I suppose it would be integrating cap current at an Fs ~300Khz+. Low pass filtering it .. and the sensors bandwidth is still well above the audio band, perhaps it could be used anyway? What frequencies were you switching at with it?

Thanks,
Chris

 

[classd4sure]

Thanks for the reference, Jaka. I having been hoping someone would use that schematic to attempt to build a working circuit and report back. I am very busy working in a non audio field at the moment so I haven't had the time to do anything myself.

If there is a DIYer here willing to try building it, I am willing to draw up a complete schematic and perhaps lay out the circuit board as well (in order to share files, I was thinking of using one of those online layout board houses - ideas?).

With regard to the full bridge, two MOSFET, current sensing idea, I built and tested that very circuit for a product several years ago. It works but suffers troubling distortion proportional to dead time. The problem occurs when the bottom MOSFETs are either both simultaneously on or off. This causes the sensed feedback signal to either erroneously double (looking like shoot through) or drop out during dead time. To achieve audiophile fidelity, sensing all four MOSFETs really is required (per the half bridge sense circuit in the thread you referred to).

Best regards -- analogspiceman

Hi analogspiceman,

I've been thinking about tweaking my little protoamp with your mods just for kicks, but I really can't bring myself to mess with what is my only working amplifier channel (PC speakers discounted).

Even if no one builds it, it's a great thread and I thank you for it as I'm sure many do, it continues to be extremely informative.

Something I like about the UCD is the fully differential implementation. I suppose it wouldnt' take very much at all to modify your current sensing technique to accomadate that, say four more transistors?

Best Regards,
Chris

 

[Jaka Racman]

Hi analogspiceman,

I do intend to build leapfrog amplifier, but an ampliverter design, since current sensing can be done with simple resistor at centertap of secondary winding. I will have more questions in leapfrog thread in a few days. I hope you will be willing to discuss further.

Thanks for pointing out trouble with fullbridge current sensing. I never used it with class BD (three state) modulation. With class AD modulation (diagonal switches on) it performed quite well for aplication it was intended for (microstepping drive for steppers).

Chris,

if sensing capacitor current (Mueta style) is your problem, then integrating capacitor voltage seems to me the best way to proceed. I also have experience with pasively integrating inductor voltage to sense inductor current (directly, not through auxiliary winding). It works quite well when LR and integrating CR time constants are mached. But it is not very practical, because it must be referenced to one of inductor nodes.

Regarding Hall sensors, I used LEM LA-50 very similar to LEM LA100-P . I was positively surprised, since it is very conservatively rated. Delay of current waveform was 300ns or even less IIRC and bandwidth over 1MHz, although it was rated for 1us delay and 150kHz bandwidth. It was only proof of concept circuit, but i was able to produce 16kHz sinusoid at the output. I can't remember switching frequency, but it was probably around 250kHz.

Best regards,

Jaka Racman

p.s. One another thing, the best way to link US patents is http://www.freepatentsonline.com/ . There you can link to whole US 5606289 PDF file

 

[Pabo]

I have built several self oscillating inductor current controlled amplifiers with a voltage loop applied around it. The result was pretty good but not as good as I hoped it would be. The return current was controlled giving the advantage that the signal was referenced to ground.

The great advantage with current mode control is that the output filter is transformed into a first order filter. The bandwidth of the output filter is unfortunately determined by the load in paralell with the filter capacitor and the low frequency gain of the current loop is determined by the load.

The gain of the current loop should be set equal to the closed loop gain of the amplifier giving you a smooth frequency characteristic with nominal load.

The great phase margin of the current loop gives the possibiliy to use up to 20dB of voltage feedback at 20kHz which gives fairly reasonable output impedance.

Unfortunately, combining a hysteretically controlled loop with a higher order integrated output signal does not perform very well. The THD+N is ot redued by the factor of feedback which is applied.

The output signal also contains a large amount o common mode noise as the the negative terminal is a triwave at 500kHz with inductive spikes.

I did not achieve a linear curent loop either. I bought the most expensive low inductive shunts but nothing helped, THD+N did not reach below -70dB at 1kHz.

 

[analogspiceman]

I have built several self oscillating inductor current controlled amplifiers with a voltage loop applied around it. The result was pretty good but not as good as I hoped it would be. The return current was controlled giving the advantage that the signal was referenced to ground.

Due to the high currents and low sense voltages it is very difficult to get a clean signal by this method. Even though one end of the sense resistor is "grounded", current is best sensed differentially. Perhaps you did not take this seemingly redundant step and suffered unwanted distortion.

The great advantage with current mode control is that the output filter is transformed into a first order filter. The bandwidth of the output filter is unfortunately determined by the load in parallel with the filter capacitor and the low frequency gain of the current loop is determined by the load.

The gain of the current loop should be set equal to the closed loop gain of the amplifier giving you a smooth frequency characteristic with nominal load.

The great phase margin of the current loop gives the possibility to use up to 20dB of voltage feedback at 20kHz which gives fairly reasonable output impedance.

Designing with the leapfrog method gives the best possible performance because it combines optimum feedforward with optimum feedback.

Unfortunately, combining a hysteretically controlled loop with a higher order integrated output signal does not perform very well. The THD+N is not reduced by the factor of feedback which is applied.

There is no magic here. No doubt your current feedback signal was distorted to begin with so large amounts of gain would be fruitless.

The output signal also contains a large amount o common mode noise as the the negative terminal is a triwave at 500kHz with inductive spikes.

I did not achieve a linear current loop either. I bought the most expensive low inductive shunts but nothing helped, THD+N did not reach below -70dB at 1kHz.

Common mode impulse noise is very problematic for active circuitry. Sometimes it is best to just give in and add a small common mode inductor in the signal path.

Regards -- analogspiceman

 

[analogspiceman]

I do intend to build leapfrog amplifier, but an ampliverter design, since current sensing can be done with a simple resistor at the center tap of the secondary winding. I will have more questions in the leapfrog thread in a few days. I hope you will be willing to discuss it further.

Thanks for pointing out trouble with full bridge current sensing. I never used it with class BD (three state) modulation. With class AD modulation (diagonal switches on) it performed quite well for the application it was intended for (microstepping drive for steppers).

For a relatively distortion tolerant application such as most motor control, the small residual deadtime in class AD modulation does not lead to problematic distortion when reconstructing output current from just the very convenient bottom legs of the output bridge. Even in my application, which was a 1.5kW subwoofer amplifier, there was enough loop gain to yield acceptably low distortion. But for a full range application with any deadtime at all, one would certainly need to sense current from all four bridge legs to avoid distortion from the doubling/dropout problem.

I would be very cautious about trusting the integrity of a so called ground referenced current sense. My experience is that it is much better to sense "ground" referenced current differentially rather than try to star signal ground from a high current sense resistor. Your layout person will thank you.

Regards -- analogspiceman

PS: Please excuse my adding in the articles to your text. In a past life I worked with a couple of brilliant Polish guys who, in spite of their nearly perfect spoken English, would have me check all of their many conference papers for the proper use of English articles (these, I am told, are entirely absent in most Slavic languages).

 

[subwo1]

I like the idea of sensing the current in the lower half of each bridge leg. I originally was thinking that this way would impose uneven action since only half of the transistors would be controlled.

But since the same current flows through the lower leg as though the catacorner upper MOSFET, I hope to try out sensing each lower leg, but interrupting the associated upper one also based on the lower.

Many do not like sensing current through the MOSFET channels, but I still favor it for one of the very reasons others do not like it. I like how it also provides temperature protection since the current cutoff point goes down as the temperature rises.

Now, admittedly, I am more interested in SMPS design. Maybe this way is less conducive to low distortion. It also is less precise because the trigger points of the gates vary somewhat. The last drawback that the appropriate MOSFET channels can be monitored only when the relevant gate is driven high is also not a problem when using the Schmitt triggers since diodes like the 1N4148 can solve that problem.

 

[Ouroboros]

I use my own design of self-oscillating current-mode control Class-D amp which uses a current transformer to sense the output inductor current.
(I use a Coilcraft CS4100V-01 current transformer intended for SMPSU use). The use of the wideband current transformer is more expensive than the 'virtual earth differentiator' capacitive sensing method used by Mueta, but has the advantage that it is easier to provide cycle-by-cycle current limiting of the output.

 

[analogspiceman]

Something I like about the UCD is the fully differential implementation. I suppose it wouldn't take very much at all to modify your current sensing technique to accommodate that, say four more transistors?

????? As far as I know all UcD amps are of a half bridge topology, but extending the high side current source / current mirror sense technique to a full bridge topology should just be a matter of adding more parts and summing signals.

Regards -- analogspiceman

 

[phase_accurate]

Yes they are half-bridge but they are still built in a differential fashion. The idea is to get rid of the effects of ground bounce (affecting the NFB) and therefore taking feedback differentially.

Regards

Charles

 

[classd4sure]

Hi,

I'm impressed with all the great info everyone has been sharing, thank you all!

Pabo, I believe capacitor current feedback would address just about every problem you experienced, what do you think?


Hi analogspiceman,

As far as UCD the Hypex modules as sold have always been fully differential. It's easily and has been adapted to full bridge as well. However it isn't only fully differential but uses an instrumentation amplifier like layout which improves cmrr.

I'm also under the suspicion that in a higher powered design it couldn't hurt to have a mirrored layout where ever possible to allow EMI to become common mode across the board, certainly across the input and feedback signals.

I just modified your leap frog simulation to a cascoded output stage on the comparator and it worked fine though I'm a total novice with LTspice so the going is slow.

What you currently have as your guestimate is nothing like the product however it does match what is known from the patent and is closer to how my little prototype is actually wired.

What my circuit includes is what's known from the forum and other litterature, nothing more. I will say LP seems to do the trick very well and I'm curious to try it. I'll link you to my version of a UCD~180 so you can see the cascoded output along with the fully differential setup.


http://www.diyaudio.com/forums/showthread.php?s=&threadid=55385


I don't mind if this thread takes me in twenty different directions but I plan on trying a number of things out already.

Best Regards,
Chris

 

[analogspiceman]

I like the idea of sensing the current in the lower half of each bridge leg. I originally was thinking that this way would impose uneven action since only half of the transistors would be controlled.

But since the same current flows through the lower leg as though the catacorner upper MOSFET, I hope to try out sensing each lower leg, but interrupting the associated upper one also based on the lower.

Many do not like sensing current through the MOSFET channels, but I still favor it for one of the very reasons others do not like it. I like how it also provides temperature protection since the current cutoff point goes down as the temperature rises.

Now, admittedly, I am more interested in SMPS design.

Unless your output bridge drive signals are very precise, there will be timing differences between upper and lower diagonal switches. The result is that, for brief periods, either both upper or both lower switches will be on, contradicting the basic assumption of your current sensing scheme (diagonal operation). When both bottom switches are on simultaneously, current is sensed twice such that the reconstructed feedback signal is doubled. Obviously when both top switches are on simultaneously, current is not sensed at all. For current mode control of a power supply, this not a show stopper, but if the signal is to be part of a high fidelity audio feedback scheme, it is unacceptable. Of course, high fidelity sensing of current via the imprecise channel resistance would be right out as well.

Regards -- analogspiceman

 

[classd4sure]

Unless your output bridge drive signals are very precise, there will be timing differences between upper and lower diagonal switches. The result is that, for brief periods, either both upper or both lower switches will be on, contradicting the basic assumption of your current sensing scheme (diagonal operation). When both bottom switches are on simultaneously, current is sensed twice such that the reconstructed feedback signal is doubled. Obviously when both top switches are on simultaneously, current is not sensed at all. For current mode control of a power supply, this not a show stopper, but if the signal is to be part of a high fidelity audio feedback scheme, it is unacceptable. Of course, high fidelity sensing of current via the imprecise channel resistance would be right out as well.

Regards -- analogspiceman

Hi,

It's funny that I never found the lower FET source resistor sense method appealing at all exactly because of what you describe above.

I'd seen in simulation that with a self oscillating amp like this one where everything in the modulator, drivers, output stage.. share strong interractions via their additive delays, such a sense methode would lead to even worse timing errors, whereas if it isn't tight enough the devolopping cross bridge conduction is apt to be gross and rather destructive to your output stage.

Regards,
Chris

 

[subwo1]

Unless your output bridge drive signals are very precise, there will be timing differences between upper and lower diagonal switches. The result is that, for brief periods, either both upper or both lower switches will be on, contradicting the basic assumption of your current sensing scheme (diagonal operation). When both bottom switches are on simultaneously, current is sensed twice such that the reconstructed feedback signal is doubled. Obviously when both top switches are on simultaneously, current is not sensed at all. For current mode control of a power supply, this not a show stopper, but if the signal is to be part of a high fidelity audio feedback scheme, it is unacceptable. Of course, high fidelity sensing of current via the imprecise channel resistance would be right out as well.

Regards -- analogspiceman

Thanks for your thoughts. I can't picture both upper or both lower MOSFETs being on simultaneously with the circuit I am trying since it is a non 100% duty cycle USMPS, but I did foresee a slight timing inaccuracy, nonetheless. Even my USMPS concept is slightly unbalanced enough to cause non gapped toroid heating when used in push-pull configuration. So for full bridge, a coupling capacitor connecting the transformer to the MOSFET totem poles is very helpful when using an non gapped transformer.

I wonder if all cycle-by-cycle protection including MOSFET channel voltage drop detection tends to lower fidelity a lot anyway. It seems that it would throw a wrench in the operation of the voltage feedback loop, but especially add timing errors.

 

[Kenshin]

Some old, self-osc SMPS put the current sampleing resistor between output L and output C.

 

[subwo1]

Some old, self-osc SMPS put the current sampleing resistor between output L and output C.

That way makes sense.

 

[Ouroboros]

It's just as easy to wind a 1:100 ratio wide band current transformer on a small ferrite toroid and put a 100 Ohm (or similar) resistor across the secondary to give you a nice isolated voltage as a function of the primary current. Put the primary of the current transformer in series with the output filter L.

You do need to make a current transformer that has low phase shift at the switching frequency, but it's not too difficult to do this for oscillation frequencies of 300 to 400kHz.

I've even thought of trying a Hall-effect current sensor but perhaps that's a bit over-the-top!