Minimizing the class D distortion

[Kokoriantz]

Rectification: The above curves are for BD modulation with the bias resistor, without it, the pulses remain in 0 VCC domain.

[Kokoriantz]

This circuit provides the required feedback signal without filtering. I applied with it 16db NFB and see the result in step and 1khz. To keep in mind that my simulated PWM is with triangular wave, if the TPA3116 uses sawtooth ramp, lower NFB to be used as it samples once instead of twice with triangle.

[Kokoriantz]

I want at all cost keep the magnificent bass of this amp all by giving extra NFB. For that I made a transconductance buffer to feed the chip with 500k impedance. I wired the circuit bellow that will acquire the feedback in the other input. As you notice the input transistor's behavior is not effected by the other one but the differential current is obtained in current mirror subtraction.
First I had to use the mute to start up, to hear bright sound. I can see on scope the low frequencies  attenuated by feedback from sagging power supply, a problem to resolve but what concerns the requiem of Verdi, even that of Mozart doesn't sound garbage any more, the voices are distinct.
The input stage has a voltage gain with 9.1k chips input impedance of -1db giving a total of 35db gain to be decreased by feedback. On simulator, the input stage has 0.0002% THD and the feedback 0.0025%.     

[Kokoriantz]

No, I couldn't get back the magnificent low mid sound back. I will try another input buffer using gyrator as some call which is simulated inductor. The output impedance seen by the chip ia more than 1Mohm which can give extra 3db NFB. The input transistor has 0.008% THD full scale.

[Kokoriantz]

I wired the second circuit with post feedback. The low mid particular sound didn't came back except when I reduce the gain back to 20db. It looks like that it is an anomaly that the TPA3116 is creating , and when I apply more NFB, it is getting corrected. In tape recorders it is also the case where only 9.5cm/s has such sound. In 80's, I tried to give the cassette tapes 4.75cm/s the same sound in vain.
The feedback signal extractor worked without flaw. As it is the first time I am able to visualize the output, I want to see how my biasing technique is really working. The post extra feedback is useful at high power to compensate the non linearity of the output inductors. With my 97db efficiency speakers, I will not hear much post feedback effect as the sound at this stage is already far from what I can expect from the best classA .
I thank to all who followed this thread, even in silence, you were encouraging me.     

[Kokoriantz]

I bought a TPA3110 2×15W module for $1.5 to try it in class abD, The loop function betrayed by PSRR is shown bellow to be more linear than that of 3116. The sound is simply high fidelity although the chip is fake with apple logo instead of TI. Unlike the 3116 it reproduces the most difficult requiems of Verdi and Mozart with perfection. The high frequencies are precise and ultra clean, aah my class abD, I love it. Unfortunately, the 8 track style low mids of 3116 is absent, and the 3116 was artificially over sharpening the extreme high frequencies, nothing such with 3110.
I ordered STA516 150W output stage, to use the 3110 or 3118 as modulators, for that I will add a dead time eraser shown in the circuit. May be I will use both modulators selectable to have it flavored or neutral.     
 

[Kokoriantz]

Unlucky for me, the tpa3116 I was using that sounded as 8 track style is defective. I replaced with an identical board and the sound doesn't have any flavor. The input of this, is  differential, unlike the defective the one. If only I can know how to make this also defective.

[strawberry]

a bit of salt for better flavor
D stands for distortion

[Kokoriantz]

Examining the defective nice sounding one, it looks to be fake as the writings are printed were the original is laser graved. The working one didn't pass the test of the two requiems, that is, it is as bad as the fake and without flavor whereas the 3110 has astonishing fine sound best to be used either 2×10W or as modulator for STA516, of course in class abD

[Kokoriantz]

a bit of salt for better flavor
D stands for distortion

In late 70's, a colleague engineer built a classD amp as portable guitar amplifier, a circuit from Hitachi databook. The sound was so horrible, that as you, I said D stands for distortion. As engineers it is our duty to make this type of amplifiers sound good, as analog is already dead. This what I accomplished with the class abD. The TPA3110 circuit plus a pair of 220nF and a 3W 50 ohm resistor cost less than $2.5 try it out, it works with laptop PSU.   

[Kokoriantz]

The DS says:
. For good transient performance, the impedance seen at each of the two differential inputs should be the same.
As my defective chip has non symmetrical inputs, I made the good one also dissymmetrical by giving a gain of 36db on negative with 9.1k internal impedance and 16db on positive with 110k impedance. Now I have the good one also flavored with 8 track sound, sharpened high frequencies and on bonus extra bass. As it is just input impedance matter, it can be switched with balanced and unbalanced to have it neutral/flavored mode. I will apply this now on 3110. 

[strawberry]

placing D class amplifier inside OPamp feedback loop, could compensate LC network nonliniarities (with some limitations)

[Kokoriantz]

The TPA 311X series have max gain of 36db, it is possible to use the 16db of it to post filter feedback. TI website gives a passive solution for that. The circuit I tried on #29 does very good job. I prefer to use rather higher quality inductors. For STA516, I will use these :https://www.sagami-elec.co.jp/en/product/detail.php?type=7G14D, it costs $2 each but I must buy minimum 10 pieces.
For lower power, a Chinese Copy of Sagami 2×10uH in a single package 9.5A, I paid $3.43 :

https://www.aliexpress.com/item/1005003536543610.html?spm=a2g0o.productlist.0.0.2d1d2d7eH6DRcH&algo_pvid=4d9a45b3-d110-4916-aa6e-f35735c3c987&algo_exp_id=4d9a45b3-d110-4916-aa6e-f35735c3c987-12&pdp_ext_f=%7B%22sku_id%22%3A%2212000026212313835%22%7D&pdp_npi=2%40dis%21USD%214.1%213.2%21%21%21%21%21%402103255b16603614128058381e9c67%2112000026212313835%21sea&curPageLogUid=LSrbrPXOVlDj
is a better solution than adding opamp or my circuit.
The TPA3128 looks very similar to 3110 except has 90mhom instead of 240mohm to output 30W@8 ohm.
NXP has also class BD intended chips up to 400W, but their feedback are conceived as analog chip amps with 20db NFB @ 20khz. I will still try it out with TDA8932, a board costs less than $2.