Question regarding measuring filterless class D amp

[Gruesome]

I’m not sure how much detail to put here. I’m adding the graphs at the end, to keep this more manageable. If something looks completely crazy, I’d appreciate feedback, but I don’t expect anybody to go in detail through all this.

First I spent some time using the AD2 (100 MHz USB scope), with no transformer in the signal path:

• at low signal levels the computed difference between the channels taken from the ends of the 8 Ohm load looks truly atrocious. This is with injecting a 1kHz signal from the AD2 generator into the amplifier.
• With a faster time base, I see 10 MHz base frequency pulses, with bursts/chirps/pulse trains repeating either every 16 or 32 pulses, which I assume correspond to the two PWM frequencies of 330 and 660 kHz. The signals I see do not look like rectangular PWM pulses to me.

Switching back to the QA403, and trying to approximate Matt’s setup for the MA12070 (1.46 Hz FFT bin width, 20Hz to 20 kHz measurement window, -60 dBV input signal, 20 dB gain, 5x averaging, A weighting), but using the small audio transformer and no other filters in the loop except what might be built into the A30, I get 100 microVolt noise (-80dBV), not 65 microVolt. I can not get completely rid of the line frequency peaks, despite protective earth grounding the QA403 (via the bottom left BNC shield), the A30 (via the subwoofer out RCA shield) and the shields of the probe cables from the 8 Ohm load to the transformer. Transformer and QA403 orientation (in the local field) do not seem to play a role, but I can induce a few dB of damping at 60 Hz by touching the BNC shields.

Frequency response (gain) looks reasonably flat, when referenced to the transformer.

I then did some THD measurements, which at lower frequencies (<1kHz) are limited by the transformer distortion. I can get below -60dB above 1 kHz for amp input and output (gain = 1) levels between 0.5V and 2V, so maybe this is not so bad.

In light of all this I also think the higher bandwidth harmonics data from yesterday are ok as well, but I’m questioning whether looking at harmonics in the 10kHz to 50 kHz range is really the best way to assess/characterize the audio distortion of this amplifier.

Therefore I did some intermodulation tests, and some of those, at higher gains and power levels, are not looking so bad either. I’m thinking those might turn out to be most useful to compare different amplifiers.

Signals on both sides of the 8 Ohm load, while injecting 1 kHz into the amplifier:

Amp output into 8 Ohm, seen through the transformer, at 1V input:

The AD2 FFT does not look anything like I expected; there is a narrow peak around 250 kHz though:

Back to the QA403, looking at noise with no input into the transformer:

Trying to reproduce Matt’s settings, now with signal (details in text above):

Interestingly, there is a peak at 17.5 kHz, but not at 9 kHz like in Matt’s data.

Making the connection to my higher sampling frequency measurements from yesterday; the noise number is comparable/same when restricting the bandwidth:

amp frequency response (blue) vs transformer response (red):

Plotted as ratio:

THD seems high, but improves at higher signal levels and higher gain. The next three plots are still at gain of 1, red is the transformer reference channel:

THD vs power:

Abandoning the direct reference comparison, intermodulation seems not so bad at maximum gain:

19kHz & 20 kHz 1:1 at 1W:

250 Hz and 8 kHz 4:1 at 5W:

The amplifier is not much worse than the transformer, but obviously at different voltage levels.

[matt]

@Gruesome

Interestingly, there is a peak at 17.5 kHz, but not at 9 kHz like in Matt’s data.

Hi @Gruesome, I wonder if the the artifact is possibly a software setting for the Class D silicon? Seems very possible its a 2X diff (18k versus 9k).

Making the connection to my higher sampling frequency measurements from yesterday; the noise number is comparable/same when restricting the bandwidth:

Yes, so I measured -83.7 dBV and you measured -79.8 dBV, which is pretty close. Your degradation is likely due to power line contribution at -82. There’s no way to clean that up via grounding or relocating the QA403 or DUT? Or, asked another way, can you make it worse by grounding/relocating?

Are you running the amp from a lab supply or wall adapter?

[Gruesome]

Good point, Matt. I did look at the MA12070 data sheet, but did not go through all the gory detail. It can do or be told to do mode switching, and that I think also switches the frequency.

Regarding my sad attempts at scope measurements:

a) the AD2 is of course not a 100MHz scope; it can do 100 Megasamples per second, but the BNC inputs are 3dB down at 30Mhz, or something like that. If one uses them correctly….

b) It’s embarrassing how quickly one can forget everything one knew about doing measurements once retired. I don’t know why I thought I could hook up a scope without any signal return path. Then I did connect a DC grounding path, but that must of course have a truly atrocious impedance for AC. (A quick calculation for a 0.5m loop with 2mm wire gets 3.5 microHenry, or 35 Ohm at 10MHz.) I’ll try again later to look at the amplifier outputs with a proper scope probe, and the probe ground wire hooked up to the amplifier ground.

c) My differential connection to the QA403 input is also wrong, or I should say made without regard for minimizing the area of the current loop. I positioned the 8 Ohm load resistance close to the amplifier, and ran 4 ft (1.2m) sense wires to the load, paying no attention to keeping these wires close to each other. Shielding the individual wires does not help reducing the loop area, since the currents are not flowing on the shields. Unfortunately (for reproducibility), the 60Hz lines are gone today, and minimizing the loop area between the 8 Ohm sense wires has a sub-dB effect on noise. Disconnecting the BNC protective earth ground to the QA403 though reproducibly (plugging/unplugging the grounding BNC) buys 6dB! -80dB with the connection, -86dB (A weighted) without. Grounding both the QA403 and the amplifier must introduce a ground loop via the return shield of the QA403 generator cable to the amplifier.

Sadly, after a QA403 power cycle (USB disconnect) the noise is again 6dB higher, and now reconnecting the (2nd) QA403 PE ground makes noise another 3dB worse. Must be some odd ground loop effect at the 0.1mV level.

[Gruesome]

I can’t say I’m 100% proficient with the AD2 scope. I bought the AD2 because it’s easier to store in a drawer than even a small Tektronix box, and the integration with notekeeping on a tablet seemed - and actually is - a good thing.

Anyways, the output waveforms from the A30 amplifier still do not look like what I expected, now probing with the (30MHz?) probes that came with the AD2, with ground clips attached to the subwoofer out shield, which I confirmed is DC chassis ground of the A30.

1 kHz 100mV input signal, A30 maximum gain, probes set to 10x, timebase 100 microsecond/division:

Looking deeper into the ‘pulses’ with a faster timebase, I see short 660 kHz out-of-phase pulses on the two speaker output channels, and the resulting 1.3 MHz pulses on the M1 difference channel; these pulses get longer as you increase the amplitude of the input signal, or the gain of the amplifier:

This is what shorter pulses look like at 10mV input level; the green difference channel does not look like the 330 or 660 kHz PWM pulses I expected:

[matt]

Hi @Gruesome, those scope traces look like they make sense to me and your bandwidth is fine for the task. In your first plot, you have 5 divisions above zero, and 5 divisions below zero. You can see the class D modulator doing its thing: It’s forcing a series of pulses with a greater than 0 magnitude for 500uS and the forcing a series of pulse with a less than magnitude for 500uS. When you are looking at the shorter traces, you need to be careful with triggering because in auto mode it will re-arm and show you anywhere in the pulse train. You might try normal triggering on a positive edge, and then 250uS of holdoff. That will then show you at the same place in the sine (near a max).

It also looks like this is an h-bridge output. And like @JP-Huijser mentioned above for single-supply, these will idle at your supply voltage divided by two. I used to do a lot of work with scooter sine drive motor controllers at several kilowatt, and I’ll note H-bridge outputs can be hard to make sense of without diff probes.

PS. Your last QA40x plot you shared looks really good with N+D at 86 = 50uVrms A weighting. I’d focus more on radiated power-line emissions if you are seeing variability. Invert the QA403 and see how it changes. Stand it on its side, rotate in the air. If you are seeing changes to the powerline levels, then it’s probably AC that is sneaking in through the air, which means proximity to magnetic fields. But if you are measuring total noise at 50uVrms on this amp, then congrats, I think you are probably there unless the spec sheet is telling you there’s a lot more to be had.

[Gruesome]

Thanks, @matt ! I should learn how to use the trigger holdoff feature on the AD2. I’m sure it can do the same kind of ‘zoomed’ look, where you take a long data sample, and then ‘magnify’ (faster display time base) a section of it on the screen, just like a regular digital scope (or even some old analog scopes, with a pull-out knob for a second time base). It’s just nicer if it is a physical knob to twist.

[kawal]

I think the transformer needs proper loading versus the load of the QA403 of 100k Ohm . Try 1k ohm resistor as load and use the transformer as a differential source. Connect plus input to one side and - input to the other side of the transformer.

[JP-Huijser]

Hi kawal,

Interesting but it is not really necessary to go out differential. If output is single ended you can use it to show the signal on 1 scope channel as well.

Furthermore I would add an low pass filter with pass band upto 25kHz.

Myself i sometime uses the transformer but most of the time I use(d) and active filter with 6th order filtering and common mode isolation.

[Gruesome]

@kawal Maciej, thanks. Wouldn’t this reduce the signal seen at the QA403 input, since a larger fraction of the signal voltage is now going to drop over the DC resistance of the secondary (about 490 Ohm)?