Note the parts in the red ellipses. Fit them and the resonance peak goes away. Specifically the 3R3 + 1uF.
Yep, it seems the designer was focussing more on reducing emitted RF than with peaking.
To those not understanding what Thorsten meant here I will explain.
We see 'nice' behavior with a 4 ohm load. We see less nice behavior with an 8 ohm load. Namely peaking.
So the goal is to ensure a (close to) 4 ohm load for high frequencies even if there is no load or a high impedance load.
This is exactly what the 3.3Ω and 1μF does. In fact, it already 'works' a bit with a 4Ω load as well.
That RC filter (also called Boucherot filter often in combination with a small inductance) provides an extra load for frequencies above 50kHz.
Even when there are none in the music because of the amp being a switching amp.
Some (very basic) calculations at 20kHz , 50kHz and 100kHz will show what the 'real world' (seen by the amplifier) is with a 4Ω(resistive) and 16Ω (resistive) load will be.
Why 16Ω and not 8Ω... well at 100kHz the impedance of a tweeter will be higher (and mostly inductive) so even 16Ω may well be conservative for 100kHz.
A 4ohm resistive load would seen by the amp as:
20kHz 3Ω
50kHz 2.4Ω
100kHz 4.5Ω
Note that in music there will not be a voltage level at 20kHz and most certainly not at 50kHz or 100kHz so while the 'seen' impedance is low the currents (with music) won't but will using a sine-sweep test at full power. One could even make a 1W resistor (that would be used) to go up in smoke.
A 16ohm resistive load would be seen by the amp as:
20kHz 6.5Ω
50kHz 4.4Ω
100kHz 3.5Ω
So... lower than 8Ω and would show hardly any peaking. Problem solved.
Another problem then arises... and that is 'testing' that might occur at say 40kHz at full power.
You see there already is a 1μF in parallel with the output and that Boucherot would just be a second one in parallel. But the rub lies in the resistor.
A fairly large portion of the current will be passing through that resistor. Let's assume a 2W resistor would be used in the Boucherot.
At 40kHz the output is down -2dB opposite 20V (in 8Ω) so about 16V.
At 40kHz about 1/3 of the current will flow through the resistor so about 2A and that resistor will heat up as it gets way more than 2W. In fact... if that testing would be done a bit too long in duration the magic smoke would come out and a burning smell would come out. Don't worry the amp will still work afterwards though.
So... that's probably why no Boucherot is placed.
In practice, when using music even at loud levels, the resistor would not blow up ever but with testing using a slow sweep or tones near full power above 20kHz a burning smell would be there.
Now... is the peaking at 20kHz an issue. With Pavels load we see about +1dB at 14kHz. Let's assume most budget audiophiles can hear this. At 18kHz were up +2dB.
At the same time there will be higher harmonics (inaudible) and some IM for sure but again... with music those levels that will be small as well.
I don't think that is not so much as an audible issue. Certainly not for 'budget hifi'.
Any sane audiophile with higher end speakers will have bought a decent amp anyway and won't be using <$100.- amps.
So is all of this a technical 'issue' (from a design standpoint) ... sure. It does not get our hands together.
Should/would such a cheap amp be inserted in a high-end system ? I don't think so... but who knows.
Is this a non-issue for home and DT speaker usage... no of course not. In fact 1dB extra a 14kHz will not be an issue.
Even +15dB at 40kHz (with music which will have virtually no signal there anyway) will not blow up tweeters and probably still sound 'good enough' for most (IM products will be higher in level) people with most music.
Is this desirable from an engineering P.O.V. ? No... is it of practical concern in the majority of real world cases ? I don't think so.
When you want to be safe... buy a 'better' class-D or AB. When you don't need top notch 'certainty' and are on a budget... buy these devices and enjoy them for as long as they work and buy the newest 'best budget' thingy a few laters later (as an 'upgrade') or when it inexplicably stops working after a few years.