Class D - remains of switching frequency in measurements

[chebum]

All Class-D amps incorporate some sorts of low pass filtering to remove switching power components from the signal. However, I see a lot of switching power traces in published measurements. For example,
Hypex NCore:
https://www.audiosciencereview.com/...ifier-broadband-noise-measurements-png.19469/

Amazon branded amp:
https://www.audiosciencereview.com/...khz-fft-spectrum-audio-meaurements-png.24879/

Crown XLS:
https://www.audiosciencereview.com/...plifier-broadband-fft-measurements-png.19991/

If Class D amps incorporate a steep low-pass filter at around 20-30KHz, why do we see so high noise pikes at 300, 500 and 600KHz in measurements? Why low-pass filters aren't effective against them?

 

[daftcombo]

All Class-D amps incorporate some sorts of low pass filtering to remove switching power components from the signal. However, I see a lot of switching power traces in published measurements. For example,
Hypex NCore:
https://www.audiosciencereview.com/...ifier-broadband-noise-measurements-png.19469/

Amazon branded amp:
https://www.audiosciencereview.com/...khz-fft-spectrum-audio-meaurements-png.24879/

Crown XLS:
https://www.audiosciencereview.com/...plifier-broadband-fft-measurements-png.19991/

If Class D amps incorporate a steep low-pass filter at around 20-30KHz, why do we see so high noise pikes at 300, 500 and 600KHz in measurements? Why low-pass filters aren't effective against them?

I don't think this has anything to do with the audio signal. Those artefacts are produced after the low-pass filter.

 

[chebum]

Those artefacts are produced after the low-pass filter.

Isn't a low-pass filter the last part of the audio chain? If not, what modules do these amps have after a low-pass filter?

 

[andreasmaaan]

All Class-D amps incorporate some sorts of low pass filtering to remove switching power components from the signal. However, I see a lot of switching power traces in published measurements. For example,
Hypex NCore:
https://www.audiosciencereview.com/...ifier-broadband-noise-measurements-png.19469/

Amazon branded amp:
https://www.audiosciencereview.com/...khz-fft-spectrum-audio-meaurements-png.24879/

Crown XLS:
https://www.audiosciencereview.com/...plifier-broadband-fft-measurements-png.19991/

If Class D amps incorporate a steep low-pass filter at around 20-30KHz, why do we see so high noise pikes at 300, 500 and 600KHz in measurements? Why low-pass filters aren't effective against them?

I'm not sure specifically for each amp, but I presume that the low-pass filter would typically be provided by an LC network giving a 12dB/octave roll-off. That would put the output down approximately 50-60dB at typical switching frequencies of around 400-600kHz.

 

[DonH56]

The output (low-pass) filter is fairly low-order of necessity -- high-order, and you probably could not create a stable amplifier with the switching frequencies used. Higher switching frequencies are much harder to design since high-power, high-frequency transistors are expensive. Being able to slew 100 V or more in say 100 ns is tough. This is not like the transistors in your CPU that are operating at under 1 V, and remember (or note) that for switching circuits power goes as P = f * C * V^2. Power transistors are large to handle the current, so capacitance (C) is large, the voltage V is squared, so even at relatively low (compared to a CPU) switching frequency f there can still be a lot of power to handle. So it's very hard to suppress that switching spike. You can also see that in the simple class D circuit I simulated (https://www.audiosciencereview.com/forum/index.php?threads/class-d-amplifiers-101.7355/ ). Note the output is essentially switching rail-to-rail at the switching frequency and that creates a big signal to suppress.

To put in context, for a 100 W amplifier, a spurious tone at -60 dB is 0.0001 W or 0.1 mW. That is, in power -60 dB is 1/1,000,000 the power of a 0 dB signal. That is unlikely to cause problems, and chances are good most crossovers and speakers (driver) will attenuate that much further or just reject it since drivers tend to look inductive.

HTH - Don

 

[Matias]

No tweeter goes that high. In theory the high frequency switching may modulate lower but still it is very clean THD on the nCore amps.

Plus the new Eigentakt modules by Bruno use 2nd order roll-off, so should have even lower switching noise.
https://6moons.com/audioreview_articles/purifi/

 

[chebum]

Higher switching frequencies are much harder to design since high-power, high-frequency transistors are expensive... So it's very hard to suppress that switching spike.

Thank you for the detailed explanation Don. Do I understand correctly that passive low-pass filter (e.g. several resistors and capacitors, not transistors) after the amp itself will make the amp's performance worse and hence not used in production?

No tweeter goes that high.

I have active speakers with RCF ND 950 compression drivers and there is definitely more noise when powered with a class D Pioneer amp (International Rectifier module). The noise is so loud, I had to make an 20dB L-Pad. 3 out of 4 A/B amps I tried were very quiet and didn't need an L-Pad.
I suppose that high frequence noise induce sub-harmonics of some kind that we can easily hear.

 

[DonH56]

Thank you for the detailed explanation Don. Do I understand correctly that passive low-pass filter (e.g. several resistors and capacitors, not transistors) after the amp itself will make the amp's performance worse and hence not used in production?

It would increase the output impedance and interact with the crossover in the speakers (which is also a passive filter unless you are connecting directly to the drivers). "Worse", who knows, but I would not do it. I don't think it is needed and would just add more reactance to the load. The power level of a noise spike 60 dB below full scale is very low and the frequency is way beyond what any speaker is going emit. I would guess parasitic capacitance in the crossover and voice coil, plus inductance of the voice coil itself, will knock that waaay down so even less power reaches the speaker's cone. I.e. negligible.

Note there is already a passive LC low-pass filter inside the amp; otherwise, that switching spur would be much higher. FWIWFM, the output filters in all the class-D amps I have seen, and any additional filter I might add, are LC designs with any resistance (R) being parasitic. Look at the one in the the thread I linked here on AVS; that simple design uses a second-order LC filter at the output as that is what I have seen most often in the few commercial products I have seen.

I have active speakers with RCF ND 950 compression drivers and there is definitely more noise when powered with a class D Pioneer amp (International Rectifier module). The noise is so loud, I had to make an 20dB L-Pad. 3 out of 4 A/B amps I tried were very quiet and didn't need an L-Pad.
I suppose that high frequence noise induce sub-harmonics of some kind that we can easily hear.

I strongly suspect the extra in-band noise is just a reflection of the overall amplifier design and performance rather than switching noise mixing down someplace in the speaker. It could be getting coupled and mixed/rectified inside the amp; that would be a poor design and/or layout. There are plenty of very quiet (high SNR) class-D amplifiers around. That said, high-sensitivity compression drivers are where low SNR is going to create noise, be it class A, AB, D, or anything else. Most often when a fairly high-powered amplifier drives a high-sensitivity driver; that is a common mismatch leading to hiss. An L-pad or similar is a common fix, or even a small series R to attenuate the signal to the driver. Or as you imply just use a different amplifier.

 

[RayDunzl]

I suppose that high frequence noise induce sub-harmonics of some kind that we can easily hear.

An in-room measurement should reveal what you hear.

 

[andreasmaaan]

I suppose that high frequence noise induce sub-harmonics of some kind that we can easily hear.

I think this is highly unlikely for the reasons Don mentioned in the first paragraph of his most recent post.

 

[March Audio]

Thank you for the detailed explanation Don. Do I understand correctly that passive low-pass filter (e.g. several resistors and capacitors, not transistors) after the amp itself will make the amp's performance worse and hence not used in production?


I have active speakers with RCF ND 950 compression drivers and there is definitely more noise when powered with a class D Pioneer amp (International Rectifier module). The noise is so loud, I had to make an 20dB L-Pad. 3 out of 4 A/B amps I tried were very quiet and didn't need an L-Pad.
I suppose that high frequence noise induce sub-harmonics of some kind that we can easily hear.

Hands Up first - I sell class D Hypex amps

The ND950 is an extremely sensitive driver, 110dB (presumably @ nominal 1 watt although the specs dont say). You will find many A/B amps to not be silent at that sensitivity. As with all things there are good and bad class D amps. The hypex ones are quiet. The P701 (NC1200) is 20uV output noise unweighted and the P252 (NC252MP) is 30uV unweighted.

Intermodulation would only be a potential problem if you have a significant input noise signal close to the switching frequency (circa 450kHz for Hypex). However I RF input filter to minimise the chances of this happening. Never experienced an actual problem.

 

[JohnYang1997]

First that's -90db to 5W output into 4ohm load. It's going to be relatively lower for 50w output for 4 even 8ohm. Millivolts or lower at the output.
Second, if it doesn't bother the measurements of the amp, it's not a problem.
Third, maybe you can add a coil to the output to tame the high switching frequency even further but it's really not necessary.

 

[March Audio]

First that's -90db to 5W output into 4ohm load. It's going to be relatively lower for 50w output for 4 even 8ohm. Millivolts or lower at the output.
Second, if it doesn't bother the measurements of the amp, it's not a problem.
Third, maybe you can add a coil to the output to tame the high switching frequency even further but it's really not necessary.

An output coil wont help, any IM causing noise would need to be filter at the input.

 

[JohnYang1997]

An output coil wont help, any IM causing noise would need to be filter at the input.

How? If the point is to not harm the tweeter, a coil will do the job even though the voice coil would have been big enough to not react to the high frequency. If the point is not effecting the amplifier performance at audio frequency then just measure it at audio frequency if the measurements are good then it's ok.
So it's better to have lower noise at that frequency but it's safe to say it's not a big deal.

 

[March Audio]

How? If the point is to not harm the tweeter, a coil will do the job even though the voice coil would have been big enough to not react to the high frequency. If the point is not effecting the amplifier performance at audio frequency then just measure it at audio frequency if the measurements are good then it's ok.
So it's better to have lower noise at that frequency but it's safe to say it's not a big deal.

The potential IM is created in the amp not at the speaker.

The tweeeter is not harmed by way out of band low level risidual switching frequency.

 

[JohnYang1997]

The potential IM is created in the amp not at the speaker.

Of course. But the end of it, it's not a big deal for an amp if there isn't stability issue or too high of amplitude that consumes voltage swing and/ or current capability.
Two things that matters are first doesn't destroy tweeter, second doesn't hurt measurement. And they don't seem to the case here.

 

[March Audio]

We are talking about potential in band (audible) IM. Not filtering of out of band signals for measurement purposes, which yes can be a requirement.

An output inductor to filter out of band signals will not get rid of in band audible IM signals created in the amp.