20bit PWM class D amplifier idea

[Kokoriantz]

To transform 16bit into PWM 200khz, it requires less than 80ps definition not possible these days more than 11 bits from commonly found DSPs. How about splitting the word into two 10bits MSB and LSB and treat them in parallel. With 10MSB, the amp powered by 50volt BTL, only +/- 50mv is necessary to be complimented by the remaining lower bits. A 5 volt powered PWM with a 50:1 ratio transformer adds to the output the remaining lower bits. The addition takes place after low pass filters so that the transformer's mutual inductance and a capacitor provides the same low pass to the additive low bits. Here is a schematic of the idea.   

What can possibly go wrong? please comment with most pessimistic arguments.

[T3sl4co1l]

You still need the 80ps timing on the power DAC.

I don't see any arrangement for feedback, I'm assuming this is all open loop right?  Far easier to do it with a modest-order D-S DAC, with a precision comparator to sense the error and drive it towards zero.  You get full ENOB at DC up to some frequency, then noise takes over (ripple + quantization).  Using a higher order D-S or "noise shaping" helps push the spectral content up towards Fs, yielding higher BW at ENOB for the same Fs.

Tim

[SiliconWizard]

Far easier to do it with a modest-order D-S DAC, with a precision comparator to sense the error and drive it towards zero.  You get full ENOB at DC up to some frequency, then noise takes over (ripple + quantization).  Using a higher order D-S or "noise shaping" helps push the spectral content up towards Fs, yielding higher BW at ENOB for the same Fs.

Yes. That's how it's done these days. Modern audio class-D amplifiers do not implement simple PWM. They use delta-sigma (or sigma-delta, I tend to use the latter term while it's the same thing) modulators.

[Berni]

The potential issue here is matching the two precisely enough. When you are working with 10bit that last bit is worth 0,1%. So the ration between the two amplifiers has to be within that, otherwise the smallest bits end up turned into just extra noise.

The actual solution is better methods of modulation. There is no strict rule to use PWM, anything that switches well out of the audio range will work. A single pulse does not need the full output bit depth since what counts is the sum of everything within the sample time. Things like sigma delta use this to distribute the error out by keeping track of the error and driving it towards 0. Most of the modern audio DACs you end up listening to are actually just 1 to 5 bits running at 1 to 50 MSPS, it's the sigma delta modulation placed in front of them that makes them capable of playing 24bit audio with noise figures in excess of -100dB