Hello guys, i have a few questions about class D feedback:
I'm thinking of designing/building a high power class D amplifier, for use in a portable audio system. I do a lot of power electronics(currently writing my thesis in this field), but i haven't done any audio, and i have a few questions..
I've read up on various application notes, and when a half bridge is pictured, there is always a feedback from the PWM output from the power stage, to the input stage. Why is the feedback done before the output filter? It makes no sense to compare a square wave with the sinusoidal wave of the input signal, does it? I've seen some circuits that include a LPF in the feedback, but others do not.
Second, what if you have a PWM input? As far as i can tell, feedback in a half-bridge is done to solve problems with variations in the supply voltage. That is, you offset the sinusoidal signal in order to compensate for any offset at the power stage. But if you have a digital input, you really can't do that kind of offset, can you?
Regarding half-bridge vs full bridge: A half bridge supplied by a bipolar supply will not need a dc blocking cap at the output(assuming feedback), but a cap is required if you use a unipolar supply. A full bridge will provide more power to the load, as you can vary the voltage across what is essentially two times the supply voltage. But how about feedback in a full bridge? As far as i can tell, most full bridge systems are open-loop, which i guess is OK, since the voltage variations on the supply will be applied as a common mode voltage to the speaker.
Finally, regarding the driver ICs: I am a bit confused, since the only driver ICs i can find, have integrated MOSFETs. At the power levels i want(500-1000W), this is of cause not an option. I looked at the TAS5012 from TI, which is basically a PCM-to-PWM converter. But how about volume levels and eq settings? Is that totally handled by the device supplying the PCM signal? And what about gate drivers for the full bridge? I am unable to find any ICs with built-in drivers that can handle dead-time and supply the right amount of current to the gates. What requirements do i need?
To summarize:
1) Why is the squarewave PWM fed back to the input, and not the filtered output signal? Is there a lowpass filter in the feedback loop?
2) What if you want to do a power stage that accecpts only PWM input(that is, no analog sinusoidal, but just the PWM)? How do you do the feedback then? Is it even possible
3) How do you do feedback in a full bridge? Is it even needed? Some appnotes says that it is not needed.
4) What drivers to look for? What i am looking for is either just a gate-driver that takes in a PWM signal and outputs the full bridge signals(with level shifting), or something that accecpts PCM, converts it to PWM and outputs the full bridge signals.
Bonus question: Why do full-bridge systems cancel odd harmonics?
Thank you in advance. This seem to be a great forum, and i hope you'll be able to answer my questions.
I'm thinking of designing/building a high power class D amplifier, for use in a portable audio system. I do a lot of power electronics(currently writing my thesis in this field), but i haven't done any audio, and i have a few questions..
I've read up on various application notes, and when a half bridge is pictured, there is always a feedback from the PWM output from the power stage, to the input stage. Why is the feedback done before the output filter? It makes no sense to compare a square wave with the sinusoidal wave of the input signal, does it? I've seen some circuits that include a LPF in the feedback, but others do not.
Second, what if you have a PWM input? As far as i can tell, feedback in a half-bridge is done to solve problems with variations in the supply voltage. That is, you offset the sinusoidal signal in order to compensate for any offset at the power stage. But if you have a digital input, you really can't do that kind of offset, can you?
Regarding half-bridge vs full bridge: A half bridge supplied by a bipolar supply will not need a dc blocking cap at the output(assuming feedback), but a cap is required if you use a unipolar supply. A full bridge will provide more power to the load, as you can vary the voltage across what is essentially two times the supply voltage. But how about feedback in a full bridge? As far as i can tell, most full bridge systems are open-loop, which i guess is OK, since the voltage variations on the supply will be applied as a common mode voltage to the speaker.
Finally, regarding the driver ICs: I am a bit confused, since the only driver ICs i can find, have integrated MOSFETs. At the power levels i want(500-1000W), this is of cause not an option. I looked at the TAS5012 from TI, which is basically a PCM-to-PWM converter. But how about volume levels and eq settings? Is that totally handled by the device supplying the PCM signal? And what about gate drivers for the full bridge? I am unable to find any ICs with built-in drivers that can handle dead-time and supply the right amount of current to the gates. What requirements do i need?
To summarize:
1) Why is the squarewave PWM fed back to the input, and not the filtered output signal? Is there a lowpass filter in the feedback loop?
2) What if you want to do a power stage that accecpts only PWM input(that is, no analog sinusoidal, but just the PWM)? How do you do the feedback then? Is it even possible
3) How do you do feedback in a full bridge? Is it even needed? Some appnotes says that it is not needed.
4) What drivers to look for? What i am looking for is either just a gate-driver that takes in a PWM signal and outputs the full bridge signals(with level shifting), or something that accecpts PCM, converts it to PWM and outputs the full bridge signals.
Bonus question: Why do full-bridge systems cancel odd harmonics?
Thank you in advance. This seem to be a great forum, and i hope you'll be able to answer my questions.
Last edited:
2/ you can feedback to an ealrier stage than the PWM.
4/ Look for a irs2113.
You dont need output coupling capacitors for half or full bridge.
You can do highish power with an IRS2092 but beware it can be hard to get decoupling and pcb layout right for this IC.
nigelwright7557 asked:
1) Why is the squarewave PWM fed back to the input, and not the filtered output signal? Is there a lowpass filter in the feedback loop?
The problem here is the phase shift caused by the output filter. Unlike SMPSU design, where the speed of the NFB loop can be quite slow to keep the stability of the regulator, a class-D amp (basically a four-quadrant buck regulator, using the input audio signal as the set-point), has to have a fast acting NFB loop. The second order output filter will produce a 180 degree phase shift over the frequency range that the NFB has to be operative, given severe stability problems.
NFB taken from the switching node will always have added filtering with a first-order network, which is much easier to compensate for. This filtering may be inherent in the dominant-pole compensation in the input stage however.
The well-known 'UCD' class-D architecture does take the NFB after the output filter, as it has a simple (but very clever) way of utilising the output filter phase-shift characteristic as part of the phase-shift oscillator used to generate the switching signal of the amplifer.
1) Why is the squarewave PWM fed back to the input, and not the filtered output signal? Is there a lowpass filter in the feedback loop?
The problem here is the phase shift caused by the output filter. Unlike SMPSU design, where the speed of the NFB loop can be quite slow to keep the stability of the regulator, a class-D amp (basically a four-quadrant buck regulator, using the input audio signal as the set-point), has to have a fast acting NFB loop. The second order output filter will produce a 180 degree phase shift over the frequency range that the NFB has to be operative, given severe stability problems.
NFB taken from the switching node will always have added filtering with a first-order network, which is much easier to compensate for. This filtering may be inherent in the dominant-pole compensation in the input stage however.
The well-known 'UCD' class-D architecture does take the NFB after the output filter, as it has a simple (but very clever) way of utilising the output filter phase-shift characteristic as part of the phase-shift oscillator used to generate the switching signal of the amplifer.
I think you try to analyze class D amplifiers in terms of DCDC converters.
First accept some facts:
-in DCDC load response (a response to load current step) is essential, in class D it is no issue, a speaker does not generate rapid current steps
- in DCDC you can expect a filter capacitor (often aluminium electrolytic) to have some ESR which simplifies compensation of voltage mode
- regulation of supply voltage variations is just a part of the distortion story, in class D the main reason for distortion is gross nonlinearity in disontinous to continous mode change, since it is highly dependant on dead time settings, it is most often called dead time distortion
- you'd love the distortion of the amplifier to be in order of CD quality, i.e. 44.1kHz and 16 bit. Now calculate yourself a time resolution for 16bit accuracy for 22050Hz sine. (tip: 22050*2^16Hz timebase). Do you now get why most good modulators are purely analog?
- in full bridge the voltage across speaker is made of two out of phase curves and therefore symmetrically distorted if half bridges are identical. Symmetrically distorted sine contains only odd harmonics (Fourier theory)
First accept some facts:
-in DCDC load response (a response to load current step) is essential, in class D it is no issue, a speaker does not generate rapid current steps
- in DCDC you can expect a filter capacitor (often aluminium electrolytic) to have some ESR which simplifies compensation of voltage mode
- regulation of supply voltage variations is just a part of the distortion story, in class D the main reason for distortion is gross nonlinearity in disontinous to continous mode change, since it is highly dependant on dead time settings, it is most often called dead time distortion
- you'd love the distortion of the amplifier to be in order of CD quality, i.e. 44.1kHz and 16 bit. Now calculate yourself a time resolution for 16bit accuracy for 22050Hz sine. (tip: 22050*2^16Hz timebase). Do you now get why most good modulators are purely analog?
- in full bridge the voltage across speaker is made of two out of phase curves and therefore symmetrically distorted if half bridges are identical. Symmetrically distorted sine contains only odd harmonics (Fourier theory)
1) Why is the squarewave PWM fed back to the input, and not the filtered output signal? Is there a lowpass filter in the feedback loop?
A) Filtering the square and feeding it back will help attenuate the nonlinearities of the filter elements. However, stabilizing the loop becomes difficult.
In a NFB loop, a square wave can be compared with a sine wave as it equates only the low frequency components of the sine wave and the square wave due to the loop's finite UGB.
2) What if you want to do a power stage that accecpts only PWM input(that is, no analog sinusoidal, but just the PWM)? How do you do the feedback then? Is it even possible
A) The same above technique applies. However you would have to be careful of the high-frequency components of the PWM input from getting PWM-modulated again down into your signal band. Remember that the loop would be dead to the high-frequency carrier components of you input PWM.
3) How do you do feedback in a full bridge? Is it even needed? Some appnotes says that it is not needed.
A) In my humble opinion, it is needed. It help attenuate dead-time distortion, supply noise, nonlinearities in carrier generation and other non-idealities in the driver stage. Can anyone here mention about the disadvantages of feedback?
A) Filtering the square and feeding it back will help attenuate the nonlinearities of the filter elements. However, stabilizing the loop becomes difficult.
In a NFB loop, a square wave can be compared with a sine wave as it equates only the low frequency components of the sine wave and the square wave due to the loop's finite UGB.
2) What if you want to do a power stage that accecpts only PWM input(that is, no analog sinusoidal, but just the PWM)? How do you do the feedback then? Is it even possible
A) The same above technique applies. However you would have to be careful of the high-frequency components of the PWM input from getting PWM-modulated again down into your signal band. Remember that the loop would be dead to the high-frequency carrier components of you input PWM.
3) How do you do feedback in a full bridge? Is it even needed? Some appnotes says that it is not needed.
A) In my humble opinion, it is needed. It help attenuate dead-time distortion, supply noise, nonlinearities in carrier generation and other non-idealities in the driver stage. Can anyone here mention about the disadvantages of feedback?
1) The PWM has to be low-pass filtered in some way in order to use it for feedback, but this may be done with an integrator as part of the loop, resulting in no visible RC filter at first glance. However, in general, only class D designers with not enough understanding about stabilizing non-trivial feedback loops use pre-filter feedback. The rest use post-filter feedback. For example I have been using post filter-feedback in every design since I learned how to make it work.
A simple method to close a feedback loop around a LC filter is just to add a zero-pole pair to the feedback path to produce >45deg phase margin at the crossover frequency. Some phase lead from crossover frequency down to the filter resonance frequency is recommended to avoid conditional stability problems (namely brief oscillation when coming out of clipping, because the loop may "run out of gain" prematurely at a frequency where phase is >180 deg).
2) The PWM would have to be either synchronized with the amplifier or lowpass filtered to the point where aliasing no longer becomes a problem. The integrator in the feedback loop can do one of the poles, but more would be required. Anyway, as it was pointed out, the time resolution required for making PWM of an audio signal makes digital PWM far from practical.
3) Class D with proper feedback for power supply rejection, output filter control, low output impedance and overall linearization is Hi-Fi. Open loop class D with no feedback is Lo-Fi, it only makes some sense with a tightly regulated power supply. Remember that open-loop PSRR is 0dB both for half and full bridge. Full bridge benefits from PS cancellation only when both outputs sit exactly in the middle (50% PWM). Pre-filter feedback with good PSRR but no output filter control is "Mid-Fi".
4) Bear in mind that the gate driver and the modulator are different things. There are a few suitable gate driver ICs with low propagation times, like IRS2011, IRS20124 or IRS20957, but you would have to design the modulator and the protections on your own if you go that way.
At the power levels that you want there are almost no ICs with integrated modulator that could save you the effort of learning to do "precision arbitrary-waveform power conversion", which is an order of magnitude more complex than conventional SMPS. Take a look at TAS5630, it seems the most powerful class D chip-amp available now.
Full bridge systems don't cancel odd harmonics ("symmetrical" distortion). Both half and full bridge systems cancel even harmonics (asymmetric).
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.