Yes, please
Thanks. Great to see you back again...+10
-
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
- Thread starter pma
- Start date
Yes it was a snark due to your implication which I find offensive. So quit with the insinuations.
The caveat is due to the presumption that's been made that DAC's do produce an amount of RF garbage significant enough to cause a problem.
The "let's test all amps" and create additional work for an already stretched Amir is pointless if this is not the case.
It's about identifying real problems and understanding the real situation instead of leaping because you are spreading FUD without understanding the nature of the actual problem.
I clearly stated "let's start with" to do precisely this. Establish the nature of the problem first.
In terms of external RF tests you have seen Amir's tests of class D. Do you see any indication of weird in band problems? Nope. I am assuming there isn't radio silence where he lives.
Above you see PMAs measurement of RF signals induced in an unshielded speaker wire - an open antenna. These are at around - 80dBV. That's about 100uV. Amplifier input cables however are shielded. This will significantly reduce the signal level. There are a lot of variables, but if its reduced by 30dB that signal would drop to 3 uV. The intermodulation that could be create in the audio band would therefore be lower again. Considering that the noise floor of an amp might already be 20 to 30 uV, you can see that this is a non issue.
Perhaps @DonH56 would be able to provide some comment on the effectiveness of cable shielding. Talking up to 1 MHz in this example.
The real problem you have here is the unnecessary creation of FUD.
The caveat is due to the presumption that's been made that DAC's do produce an amount of RF garbage significant enough to cause a problem.
The "let's test all amps" and create additional work for an already stretched Amir is pointless if this is not the case.
It's about identifying real problems and understanding the real situation instead of leaping because you are spreading FUD without understanding the nature of the actual problem.
I clearly stated "let's start with" to do precisely this. Establish the nature of the problem first.
In terms of external RF tests you have seen Amir's tests of class D. Do you see any indication of weird in band problems? Nope. I am assuming there isn't radio silence where he lives.
Above you see PMAs measurement of RF signals induced in an unshielded speaker wire - an open antenna. These are at around - 80dBV. That's about 100uV. Amplifier input cables however are shielded. This will significantly reduce the signal level. There are a lot of variables, but if its reduced by 30dB that signal would drop to 3 uV. The intermodulation that could be create in the audio band would therefore be lower again. Considering that the noise floor of an amp might already be 20 to 30 uV, you can see that this is a non issue.
Perhaps @DonH56 would be able to provide some comment on the effectiveness of cable shielding. Talking up to 1 MHz in this example.
The real problem you have here is the unnecessary creation of FUD.
Last edited:
Not getting personal here but before Amir jumps in with cold water, you do seem to be doing so when getting defensive and just lashing out.
I don’t even get the logic being used here. This thread is about audible effects when there is RF in the input. Amir’s tests don’t do that kind of input as far as I know. So saying no audible effect found in those tests isn’t relevant here at all. You still have not given any specific test or example of how other classes of amps behave as you claimed. I would be curious to know. Have no skin in the game.
But moving on to more constructive things...
What might be more interesting than making this a test of every amp from a purely scientific curiosity perspective is an experiment to test the bounds of the claim of the OP with one or two Class D amps at different price points (presumably the care they have given to RF filtering including not having any is affected by that).
The experiment would be something like feeding input in the RF range at fairly abnormal levels just to find a frequency where there may be a measurable output in the audible range like the OP’s experiment. If there is one, then narrow down to that frequency and vary the input level for that frequency from very low to some high level and see how the level of spurious signal in the audible range changes and when it is likely to become distortive to content in the audible range. This should be very interesting to see.
Note that the spike in the audible range is more like spurious harmonics in testing devices where its effects are considered bad because of what it adds to the content in those frequencies, not necessarily on its own. Using the same logic if we consider this spurious “harmonic” as an issue we can see if the level of RF needed to create that is wholly unrealistic or not and whether different Class D amps differ on this based on their RF rejection.
DAC manufacturers have been criticized for not having the right filtering that lead to bad measurements. So, I think this is fair game.
But, of course, I am not the one with the measuring instruments!
I don’t even get the logic being used here. This thread is about audible effects when there is RF in the input. Amir’s tests don’t do that kind of input as far as I know. So saying no audible effect found in those tests isn’t relevant here at all. You still have not given any specific test or example of how other classes of amps behave as you claimed. I would be curious to know. Have no skin in the game.
But moving on to more constructive things...
What might be more interesting than making this a test of every amp from a purely scientific curiosity perspective is an experiment to test the bounds of the claim of the OP with one or two Class D amps at different price points (presumably the care they have given to RF filtering including not having any is affected by that).
The experiment would be something like feeding input in the RF range at fairly abnormal levels just to find a frequency where there may be a measurable output in the audible range like the OP’s experiment. If there is one, then narrow down to that frequency and vary the input level for that frequency from very low to some high level and see how the level of spurious signal in the audible range changes and when it is likely to become distortive to content in the audible range. This should be very interesting to see.
Note that the spike in the audible range is more like spurious harmonics in testing devices where its effects are considered bad because of what it adds to the content in those frequencies, not necessarily on its own. Using the same logic if we consider this spurious “harmonic” as an issue we can see if the level of RF needed to create that is wholly unrealistic or not and whether different Class D amps differ on this based on their RF rejection.
DAC manufacturers have been criticized for not having the right filtering that lead to bad measurements. So, I think this is fair game.
But, of course, I am not the one with the measuring instruments!
No, you are just ignoring the factual technical points, or you just dont understand them.
I think your second paragraph demonstrates you clearly dont understand.
Yes this is about audible effects. For the effects to be audible they have to reside in the audible band - ie 20Hz to 20kHz. Amir absolutely measures this. If external RF (radio TX) in his area caused problems we would see it.
Also to further my point about the ultrasonic output level of DACs/CD players, PMA provided a measurement of his old Technics player in another thread.
So we see here that the broadband of noise is no more than about -80dBV which is about 100uV. So we come back to my point in response to PMAs original post. Inputting 10mV or more (100 x greater), as he did to create a problem, is not representative of the real world.
In fact at the Hypex switching frequency of 450kHz the level is around -95dBV which is only 18uV (550 x lower than the original test)!
So, first establish what is typical, and maybe the outliers, for a current selection dacs and then you can develop a representative and meaningful test to apply to amplifiers.
I think your second paragraph demonstrates you clearly dont understand.
Yes this is about audible effects. For the effects to be audible they have to reside in the audible band - ie 20Hz to 20kHz. Amir absolutely measures this. If external RF (radio TX) in his area caused problems we would see it.
Also to further my point about the ultrasonic output level of DACs/CD players, PMA provided a measurement of his old Technics player in another thread.
So we see here that the broadband of noise is no more than about -80dBV which is about 100uV. So we come back to my point in response to PMAs original post. Inputting 10mV or more (100 x greater), as he did to create a problem, is not representative of the real world.
In fact at the Hypex switching frequency of 450kHz the level is around -95dBV which is only 18uV (550 x lower than the original test)!
So, first establish what is typical, and maybe the outliers, for a current selection dacs and then you can develop a representative and meaningful test to apply to amplifiers.
Last edited:
- Thread Starter
- #125
I am sure you know that your assumption that the broadband noise is about -80dBV is simply incorrect. Spectral analysis "noise floor" is related to spectral density and does not directly tell the noise. The noise is calculated by integration over a defined bandwidth. E.g., if you see a spectrum of noise measured with 1nV/rt(Hz) resolution (bandwidth), which would be equal to 48kHz sampling and 48000 samples in the FFT record, and you measure -130dB noise floor, you need to add 20log(rt(20000)) dB, which makes 43dB, to calculate noise over 20kHz bandwidth. So the noise RMS over 20kHz is -87dB, not -130dB. Similarly here. Noise integrated from 50kHz to 1MHz will be definitely much higher than -80dBV, it will rather be at least of 20dB higher. Spectral plot only speaks about noise at one discrete frequency measured with the analysis bandwidth used, so it is a narrow band noise. Only spectrum peaks (lines) may be read directly, regarding their amplitude.
- Thread Starter
- #126
As the interpretation of "noise floor" spectrum and noise is a usual pain, please let me add the noise measurement at the output of the phono preamp. Noise over 20kHz band is in the dashed line.
maty
Major Contributor
About cables, you know, it is a very good idea with star-quad geometry. Power, speaker... Even inside the loudspeakers, with Canare 4S11.
https://www.google.com/search?q=star-quad+geometry
https://www.google.com/search?q=star-quad+geometry
Nope.
The point is we are only interested in the levels at specific frequencies not broad band. Specifically (with Hypex) within 20kHz the of 450KHz switching frequency. Otherwise the alleged IM will fall outside of the audio band.You measured over a 10MHz bandwidth. What was your FFT bin width? Eyeballing the plot it looks like maybe 5 to 10KHz, so we are not going to be that far out.
So on your plot at 450kHz we probably need to revise from 18uV to 27uV.
Last edited:
- Thread Starter
- #129
OK, I went through my archive and found measurements I did 5 years ago, in 2014, on HYPEX NC400. I was interested in output impedance of the HYPEX amplifiers and how it behaves under transient conditions. To measure output impedance, I use the topology of forcing the signal current to the amplifier output
Ideal amplifier should keep zero voltage at its output. If there is a non-zero output impedance, voltage at terminals of the amplifier under test will be Vo = Itest x Zo. Zo would be the output impedance of the tested amplifier and Itest is the forced current. Let's see NC400 behaviour. Blue trace is output voltage of the Generator-amplifier, yellow trace shows voltage at NC400 terminals. 20kHz double-RC filter is used to decrease amplitude of the carrier frequency, which would otherwise completely mess the measurement.
This shows 28Vp-p 1kHz sine forced through 10 ohm to NC400 output. We can see 1kHz envelope on the carrier frequency (yellow) and this envelope enables to calculate Zo(1kHz).
Same measurement at 10kHz.
And at 50kHz, now we can see dramatic rise of output impedance.
Plots here above show that Zo is frequency dependent, as expected. So, let's force a square wave into NC400 output.
2kHz square, 24Vp-p through 10 ohm was forced to NC400 output. NC400 has hard times to control this and needs 100us to settle, to get on its normally low Zo.
Ideal amplifier should keep zero voltage at its output. If there is a non-zero output impedance, voltage at terminals of the amplifier under test will be Vo = Itest x Zo. Zo would be the output impedance of the tested amplifier and Itest is the forced current. Let's see NC400 behaviour. Blue trace is output voltage of the Generator-amplifier, yellow trace shows voltage at NC400 terminals. 20kHz double-RC filter is used to decrease amplitude of the carrier frequency, which would otherwise completely mess the measurement.
This shows 28Vp-p 1kHz sine forced through 10 ohm to NC400 output. We can see 1kHz envelope on the carrier frequency (yellow) and this envelope enables to calculate Zo(1kHz).
Same measurement at 10kHz.
And at 50kHz, now we can see dramatic rise of output impedance.
Plots here above show that Zo is frequency dependent, as expected. So, let's force a square wave into NC400 output.
2kHz square, 24Vp-p through 10 ohm was forced to NC400 output. NC400 has hard times to control this and needs 100us to settle, to get on its normally low Zo.
Last edited:
Pavel, if I understand this correctly, the 50kHz forcing results in a 55mV (or so) error signal, which translates to 0.2 ohms? That doesn't seem very alarming.
In your last scope plot, how do you remove the time constants of your double RC filter?
In your last scope plot, how do you remove the time constants of your double RC filter?
- Thread Starter
- #131
Stuart, at 50kHz, my filter has -7dB attenuation. Output amplitude (yellow) is 80mVp with the filter, so it would be 179mVp without the filter. Test current amplitude is 1.366Ap, which makes Zout = 0.13 ohm. Yes, the value is not dramatic, but the increase of the value compared to 1kHz Zout is.
To your second question, yes there is an affect (known affect) of the filter response, which makes the transient lower than it is in reality and changes the shape. It might be removed by deconvolution, I agree. However, the oscillating response of the Zout is not because of the 2RC filter. This filter has aperiodic exponential-like step response. These oscillations of some 60kHz are created by NC400 output filter. You know why I am posting that - would be interested in NC400 interaction with complex speaker crossovers, rather than in usual 1kHz measurements into purely resistive load.
NC400 output impedance
Nc500 output impedance
@pma It's interesting, but a few questions.
First question, why are you concerned about output impedance at 50kHz?
Btw all the ncore modules have their —3dB response point at 50kHz.
Second question what is the real world relevance of applying a non bandwidth limited square wave?
Nc500 output impedance
@pma It's interesting, but a few questions.
First question, why are you concerned about output impedance at 50kHz?
Btw all the ncore modules have their —3dB response point at 50kHz.
Second question what is the real world relevance of applying a non bandwidth limited square wave?
Last edited:
- Thread Starter
- #133
Because I ask for the product, being it instrumentation or audio, to work flawlessly in the frequency range at least one order higher (10x) than is the supposed bandwidth of the signal processed. And this is not fulfilled.
Re your question of squares, I have already explained that the step response completely describes transfer function of the linear system, because impulse response is a derivative of step response and amplitude and phase response can be calculated by Hilbert transformation of Fourier transformation of impulse response. This forum calls itself "science", right? Step response quickly shows possible issues and possible stability issues, as it does here. 20kHz amplitude spectrum shows a little.
Re your question of squares, I have already explained that the step response completely describes transfer function of the linear system, because impulse response is a derivative of step response and amplitude and phase response can be calculated by Hilbert transformation of Fourier transformation of impulse response. This forum calls itself "science", right? Step response quickly shows possible issues and possible stability issues, as it does here. 20kHz amplitude spectrum shows a little.
Last edited:
"Because I ask for the product, being it instrumentation or audio, to work flawlessly in the frequency range at least one order higher (10x) than is the supposed bandwidth of the signal processed"
No. Rule of thumb doesn't do shit in front of actual relevant measurements. Filtration is external, the internal circuit can work at very high frequency as you know. Using rule of thumb every where will result in shooting yourself in the foot.
Square and step
Step response needs to be infinitely long to describe a linear system. And impulse response usually contains more information for the same length.
Square wave is used for stability issue is not for this reason. It's because square wave contains high frequency content. You can use high frequency sine wave sweep as well.
As I post earlier, Rod Elliott already discussed on this before class D is relevant. Proper input filtering is very hard to achieve in certain environment and is useless in environment that doesn't have much rf interference.
The whole thing is just two aspects:
High frequency content in music
RF interference in the environment
First one it's really small, and you only need to use test signal that mimics the spectrum energy of music signal for testing.
Second one is input filtering, which is hard to do properly in contaminated environment and is not needed at all in non-contaminated environment.
- Thread Starter
- #135
Yes, however even short step reveals system stability and behaviour at higher frequencies. Do not forget it covers both amplitude and phase. 20 kHz spectrum plot does not reveal stability issues. Sine sweep response does not cover phase. Long unit step is needed only to show low frequency behaviour. However it is interesting as well, to show high pass filter response.
I find all this defense of narrow band 20kHz measurements as manufacturer's protection only and a part of marketing strategy. This is a must for contemporary class D technology.
solderdude
Grand Contributor
A measurement suite should cover all (at least important) aspects of the DUT.
There are certain things that need to be tested that differs between say vinyl/tape and digital.
Where vinyl/tape is analog and digital is well.. digital.
I would argue the same is true for amplifiers.
Some tests for analog amps may well be pointless on switching amps and vice versa.
I don't see anything weird about that.
Switching amps may need more or different tests than analog amps.
There are certain things that need to be tested that differs between say vinyl/tape and digital.
Where vinyl/tape is analog and digital is well.. digital.
I would argue the same is true for amplifiers.
Some tests for analog amps may well be pointless on switching amps and vice versa.
I don't see anything weird about that.
Switching amps may need more or different tests than analog amps.
You can measure amplitude and phase with sine sweep easily.
And high frequency I was talking about is from khz to mhz range. 20khz wouldn't show anything.
And it needs infinity long step to describe the complete system. The point was what you said "describe the complete system".
On the defense of 20khz band. It is what is actually needed to be good audio amplifier period. If you want to trade off for other performance sure you can. It's not the point that class D technology is not good enough. Why don't you ask for 90% efficiency from class B amplifiers? It's not necessary and class B can't do it. Class D needs to operate higher than AM frequency to be able to include more feedback to get better performance over 20khz. And that needs to be 3 times faster than current class D amplifiers.
restorer-john
Grand Contributor
What are you going on about??
And I'm sorry, this is just plain rubbish.
What, you want to make it really easy for any DUT?
Should we apply the same dubious requirements to D/A converters too? Don't worry about high frequency, high level THD? Throw out the THD vs Frequency plots altogether and come up with a "mimic of spectral energy of music"? What should we use, Kumbayah?
This is right on the money. It's a concerted attempt to control the narrative and undermine many decades of superlative design and countless luminary engineers' research and development. You will meet opposition here by fanbois, vested interests and sonsumers of the koolaid, but stay the course.
Last edited:
His not talking about your measurements. But that manufacturers only cares about distortion lies in 20khz BW and output impedance under 20khz and omits rf rejection performance etc.What 20 kHz? My frequency response measurements go up to 200 kHz:
Now you get to explain what you want to do here when the amp itself has a roll off filter. What good does it do to feed it infinite bandwidth signal?
Similar threads
- Poll
