• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

On Class D Amplifiers Measurements

pma

Master Contributor
Joined
Feb 23, 2019
Messages
5,389
Likes
13,270
Location
Prague
Class D amplifiers have defined a new discipline in audio amplifier design and parameters, similarly as digital CD audio after its arrival in 1980-ties. After some time after its arrival, it was apparent that CD measurement methods would need something else and more than previous, purely analog audio. Similar situation is, IMO, now, with class D amplifiers. I appreciate greatly Amir's work on standard audio measurements in audio band, like SINAD, THD vs. power etc., and also HF spectrum. However, I think that class D amplifiers need another approach as well. This is based both on theoretical assumptions and measurements on real amplifiers.

A year and half ago I borrowed HYPEX NC400 amplifier from a friend and I had it several days here on my test bench. Standard audio measurements are (after the 50kHz LPF is added at the output), excellent and confirm datasheet parameters and other independent tests. But class D amplifiers are basically different from linear amplifiers and this should be taken into account. They operate at high frequency switching frequency with full swing square impulses, these impulses are modulated and after LPF demodulation and removal of the carrier frequency we get the resulting analog signal. This is in a way similar to sampling process in such regard that we have to care about rejection of aliases when input signal frequency is above half of carrier frequency.

To make long story short, I found that HYPEX NC400 is sensitive to even low level input signals with frequencies close to Fc (carrier) and 2Fc.
When the input frequency approaches to Fc (about 460kHz) and 2Fc (about 920kHz), one can see aliases, difference frequencies Fc – Fin (or 2Fc – Fin) appearing at the output of the amplifier. I measured about 56mVp-p output swing of the alias signals (my input signal was only 21mV), in the audio band. I have used a 2xRC analog low pass filter (27kHz) at the scope input to remove 460 kHz switching frequency of the NC400, to see the aliases. So, the input signal containing HF components may, under certain circumstances, create audible alias frequencies at the NC400 output. This is not very good and I guess that a better input anti-alias low pass analog filter would cure the situation. I had some correspondence with Bruno P. and he has confirmed my findings and said that NC400 was a closed chapter, because of design work on Purifi project. It might be interesting if Purifi is free of this issue.

This may seem unimportant, but especially 920kHz frequency is close to those of AM MW transmitters and RFI pickup in mV order is not impossible into common poorly shielded and unshielded link cables and loops and it would create audible artefacts.

1573979613669.png

HYPEX input driven with 915kHz 10mV sine wave. Output alias frequency is 5kHz.

1573979792505.png

HYPEX alias wave on the scope screen

1573979892447.png

HYPEX carrier frequency at the amplifier output binding posts
 
Firstly, yes this is a potential issue.

However don't you think your test signal was a little unrealistically high? Why is that level of signal at those frequencies going to be getting into the amp input?

Secondly, if this is a real world problem we would see these spurious signals popping up in our in band measurements. However we don't.

BTW I do include input RF filtering on my amps.
 
Last edited:
HYPEX NC400 has some kind of input LPF filter as well, only it is not efficient enough. My first plot shows measurement with a 10mV input signal - I do not think it is that high, taking into account what we can sometimes measure at the output of some digital players. The input LPF, in the case of the object measured, should have been more sophisticated, IMO, a simple RC would not do the job taking into account that it should not affect the phase response in the audio band.
 
HYPEX NC400 has some kind of input LPF filter as well, only it is not efficient enough. My first plot shows measurement with a 10mV input signal - I do not think it is that high, taking into account what we can sometimes measure at the output of some digital players. The input LPF, in the case of the object measured, should have been more sophisticated, IMO, a simple RC would not do the job taking into account that it should not affect the phase response in the audio band.
No 10 mV is very high.

So what have you measured at 460kHz at the outputs of a dac or cd player?

Also, do you have measurements of spurious signals at any appreciable level in band on the amp output when in normal use ie not when injecting HF test signals?

What Im saying is can you demonstrate this is a real world problem?
 
Last edited:
Picking up radiowaves is not possible, the ncore input is terminated by a low impedance source. What can be a problem is a poorly designed source spitting out noise at those frequencies. If so, you can hear a warning sound at 4K.;)
 
For my curiosity,

1. Do all Class D amps use the same carrier frequency and if so is there a reason for the selection of that frequency?

2. Not related to this thread but triggered by it, for the different classes of amps, how are the distortion ratios likely to vary as one sweeps down from the reference level through a typical listening range say all the way down to -30db? Ideally, what should be expected? Is there reason to believe most amps meet that ideal?
 
Not all class-D amps use the same frequency, also the frequencies aren't fixed or standard.

The frequencies must be high enough to filter easily using simple LC filters and low enough for the switching devices to not get in trouble.
Technical trade-off between voltages/currents and speed.
 
Last edited:
Do ClassD amps swing rail to rail, net 0V after the choke, with no input, or do they go quiet?
 
they keep on switching and switching and switching till they can't no more ... or the power is switched off.
 
At first thought it's interesting. Then you have to think about it.
A better approach is to use pink spectrum NID multitone input signals to test this. First limit the bandwidth to 20khz then increase to 35khz, 45khz,80khz to see at what point it starts to affect the spectrum under 20khz.
One thing to point out, instead of using sampling as analogy, it's actually more like AM signal. That's how you get to see the signal being "demodulated" into audio band.
Rod Elliott talked about this before, directing the actual rf interference present in the environment. http://sound-au.com/amp_design.htm#s12
 
Do ClassD amps swing rail to rail, net 0V after the choke, with no input, or do they go quiet?
0 is half the rail so 50% duty cycle assuming pwm operation. Other operations should be similar.
 
For my curiosity,

1. Do all Class D amps use the same carrier frequency and if so is there a reason for the selection of that frequency?

2. Not related to this thread but triggered by it, for the different classes of amps, how are the distortion ratios likely to vary as one sweeps down from the reference level through a typical listening range say all the way down to -30db? Ideally, what should be expected? Is there reason to believe most amps meet that ideal?
They don't have the same frequency. Normally in the x00khz range.
As AM radio operates at 535khz-1605khz, you need to either run below that or over that. TI has made class d amp ic operates at over 1.6Mhz instead of below. That's the way they avoid it.
High power RF devices are very rare and are expensive. So generally good class d amps, they run around 400khz. But if components allow, higher frequency can give lower distortion and more room to feedback and other ways to further optimise the performance.
To test the severity of this finding, just use multitone signal that simulates the energy spectrum of music and see if there is ever an issue.
 
One thing to point out, instead of using sampling as analogy, it's actually more like AM signal. That's how you get to see the signal being "demodulated" into audio band.

IMO best approximation is PWM modulation of a square wave at almost constant carrier frequency. Input signal modulates duty cycle of the square wave. 0 is 50% (1:1), Maximum is close to 100% (at 100% the carrier frequency would disappear and modulator would stop function). Input signal may be demodulated by a LPF steep enough with transfer function as requested (Butterworth, Bessel-Thompson). Power supply may be single rail or +/-.
I was working with PWM based optical fibre transmission systems for measurement of fast transient signals in HV laboratories during eighties. The demodulation principle by LPF is simple, however if you get in deep it is not that simple and that's why I started this thread.

The carrier frequency usually does not remain constant for the whole range of input signal amplitudes, depending on modulator design, so the effect of HF interferences and aliases created is not constant as well. IMO, effect of interferences and aliases should be investigated for every class D amplifier, individually. It may show behaviour that is not disclosed by a standard set of measurements.
 
Very interesting OP, @pma!

But you would be able to break it down even more into a simpler version for non-techie dummies like me?

For actual use cases: What kind of input would be needed for this to become a concern, and how would it affect the output from the loudspeakers? Can any input from a preamp or dac include the frequencies which would be needed to trigger this?

EDIT: TBH I'm so non-technical that I don't understand what this means: "AM MW transmitters and RFI pickup in mV order is not impossible into common poorly shielded and unshielded link cables and loops and it would create audible artefacts"
 
Very interesting OP, @pma!

Can any input from a preamp or dac include the frequencies which would be needed to trigger this?

Yes it could and to answer you need to investigate thoroughly the output of the preceding component.
 
Another thing - is it OK if the power amp sends 460kHz rubbish like this into the tweeter? Are there any possible nonlinear effects in the tweeter driver? Or we just keep saying the only issue is negligible amount of added heat? Any serious explorations on this?

1573998811179.png
 
This may seem unimportant, but especially 920kHz frequency is close to those of AM MW transmitters and RFI pickup in mV order is not impossible into common poorly shielded and unshielded link cables and loops and it would create audible artefacts.

:)

Obvious. The problem is that the EE does not acquire knowledge about RF / EMI, wave transmission, electronic fields, signal processing, antennas ... at least in Spain. Only superficial knowledge about some of those fields of knowledge. And so it seems in other educational systems.
 
Another thing - is it OK if the power amp sends 460kHz rubbish like this into the tweeter? Are there any possible nonlinear effects in the tweeter driver? Or we just keep saying the only issue is negligible amount of added heat? Any serious explorations on this?

View attachment 39112
What's the tweeter inductance?That goes a long way toward reducing the current. Here's the idle of some Class D amps I have on hand, showing about 750mV of ultrasonic. Amount of tweeter heating and detected in-band intermod: below my measurement limits.

idle spectrum.png
 
Firstly, yes this is a potential issue.

However don't you think your test signal was a little unrealistically high? Why is that level of signal at those frequencies going to be getting into the amp input?
.
The only source of such a high frequency I can think of just now would be the noise shaped stuff coming out of an SACD player or a DSD DAC. I seem to remember having seen worst case levels up to -30dBFS ( I may be wrong though) which could lead to spurious signals within the audible range.
Secondly, if this is a real world problem we would see these spurious signals popping up in our in band measurements. However we don't.

BTW I do include input RF filtering on my amps.
Yep, that should be done always and fixes many potential problems.
 
Back
Top Bottom