Actually, the THD20 of Bob's design is well below 0.001%
Sorry, my bad. I'm sure I've often read Bob quoting his amp's performance as 0.001% THD - I keep forgetting that the measured performance from the paper is much better than that!
So, yes, Bob's amp does outperform the NCore - at 50 watts. But the NCore goes up to 400 watts into 4 ohm and 580 watts into 2 ohms.
I hope we can now agree that the NCore does indeed approach the level of performance of the best linear amplifiers.
I have used Multisim; there is no reason why, if you run a Fourier analysis, you cannot obtain THD readings of far lower than 0.001. Stop using the distortion analyser: it's worse than useless.
Thanks, Stein. Great idea to have a shoot-out. A wide battery of tests would be helpful, perhaps some less conventional like maybe TIM and PIM
. Also things like slew rate, phase nonlinearity, squarewave performance, PSRR, etc, etc. Could be very interesting. I would also look at DF vs frequency. Also, evidence of noise modulation of any kind as a function of signal voltage.Measuring bridge-based class D amplifiers vs non-bridged can also require careful thought.
Cheers,
Bob
Hi Harry,
With regard to clipping into low-Z loads, check my JAES paper on that amp on my site at CordellAudio.com - Home and look at Figure 20, where my amplifier is putting a 2-cycle 20kHz tone burst into a 1-ohm load, 22V peak and 22A peak. Bear in mind, this is just a 50-watt/8-ohm amplifier with a single pair of output MOSFETs.
To be fair, of course, my amplifier did not incorporate V-I limiting protection.
Cheers,
Bob
What's the Hypex's THD20 into 2Ohms then?
From the NC400 datasheet:
Attachments
Indeed. If your amplifier design were to be used in a commercial stand-alone amplifier product (i.e. as opposed to in an active or powered speaker where the load is known), it would need to have such protection circuits added, agreed?
Last edited:
THD wars! Deja vue all over again ...
jan
Let the fun begin!
But seriously, as I stressed above, its not just about THD, be it THD-1 or THD-20.
I think there are many ways to make measurements that bring out misbehavior and weaknesses in both linear and class D amplifiers (but we must bear in mind that those revealing measurements may not be the same type of measurements for linear and switching amplifiers).
Moreover, there are many very different kinds of class D amplifiers, only one of which is the very good Hypex. PWM vs PDM is but one very simple example. These different types of class D amplifiers likely have different shortcomings.
In some of the material in my book's class D chapters there are probably some hints as to what might bring out some weaknesses in some class D designs.
Ultimately, I believe that the PDM designs with very high-speed logic and DSP will win out. As CMOS logic speeds increase, and on-the-fly adaptive circuits are used, things should get even better.
BTW, the new Euro EMI standards that I believe go into effect in January may pose a challenge for some class D amplifiers. The 2-6 GHz range is posing some challenges for some telecom equipment.
Finally, and perhaps most importantly, we must have lots of subjective listening data. In the lifecycle of class D, we may yet only be where we were with solid state amplifiers back in 1972 or thereabouts.
Cheers,
Bob
Hi Harry,
These are really impressive numbers. Bruno has done a great job on this amplifier and has slayed a lot of dragons along the way.
The very low level of IM products above the 3rd is especially impressive. Many class D amplifiers I have seen measurements on do not fare will in this test.
Cheers,
Bob
Its my current pet hypothesis that noise modulation is destined to become the 'Holy Grail' measurement of amplifier (and DAC) subjective SQ. I believe it explains why classA sounds better than classAB (noise modulation induced by PSU noise) and why classD hasn't got there yet, subjectively.
Do you have any ideas on how best to test for it?
That's easy (to say but not to do).
Double Blind Listening Tests at least to Lipsh*tz & Vanderkooy standards. I can advise on ABC tests which give statistical significance faster than their ABX tests. Also on the controls necessary to meet some of the criticisms of Blind Tests.
There are Noise modulation musical (listening) tests which were devised to check out evil Digital in da old days that can be applied here.
Be prepared to find many self declared Golden Pinnae to be deaf .. or certainly less able to reliably detect small differences compared to the man (or woman) in the street.
A simple test for a DAC is to reduce a properly dithered sine wave or music recording to -20dB below the properly dithered noise floor and play this back with loadsa amplification (careful in case a FS signal comes next
)Just listen to it. Piano is particularly good for noise modulation.
Did you work on a Class D amp for B&W?
Last edited:
Don't follow what you're suggesting here, Richard: you don't mean attenuate to -120dB?A simple test for a DAC is to reduce a properly dithered sine wave or music recording to -20dB below the properly dithered noise floor and play this back with loadsa amplification (careful in case a FS signal comes next
Just listen to it. Piano is particularly good for noise modulation.
Yes. But this needs to be done digitally to create a signal to test a DAC.
If you have a ADC/DAC back to back, you can do this with accurate attenuators on the ADC input and accurate high gain amps on the DAC output. At Calrec, we had just such an arrangement for testing mixing desks at stupid gains and stupidly small signal levels. We could play music through it too.
For a 16b DAC, you only need the signal to have its max level at -113dB FS. (The noise of a correctly dithered 16b transmission chain is -93dB wrt to a FS sine wave. see Lipsh*tz & Vanderkooy's many papers on dither for chapter & verse.)
I've done such tests in Jurassic times.
It's a moot point testing 24b DACs like this cos the best ones are juu.ust about -120dB FS noise. Their 'analogue' noise dominates. The 'true' properly dithered noise floor on a 24b chain is at -141dB FS
Actually it would be still worth doing this as a signal near +/- 1 bit would still exercise 'noise modulation'. But the aural effect might be masked by the 20dB greater 'analogue' noise.
Last edited:
These are truly impressive figures - and they are right in the sweet zone of current linear amp designs. Of course, with all out effort, I think linear can give better distortion figures and bandwidth - but that will no doubt change moving forward. Its clear that this product category is undergoing rapid change.
Interesting how the distotion at HF is lower than LF - obviously the action of the output filter.
Last edited:
(bad) modulation of a power supply rail:
An externally hosted image should be here but it was not working when we last tested it.
good (absence of) modulation of a power supply rail:
An externally hosted image should be here but it was not working when we last tested it.
I tend to agree. I have not given enough thought to how to measure noise modulation, but measurement techniques and approaches for class D amplifiers is one area where there will be more material in my second edition.
Cheers,
Bob
These plots are quite revealing. Can you say a bit more about how they came about?
Cheers,
Bob