is THD+N in class D amps HD or noise limited?

I haven't followed D-amps for a couple of years, but quick browsing of manufacturer datasheets and this forum leads me to believe that even with feedback, D-amps are still much worse than average class AB amps, i.e. a good D-amp will have roughly 0.02 % or - 74 dB THD+N, no matter whether analog input or digital. If I have overlooked any designs that perform significantly better, I'd be glad for a pointer.


Datasheets will ususally give THD+D vs. power and sometimes vs. frequency, but I am not aware of any FFT data with a sine input. Again, I'd be glad for any pointers.

The best thing I found was the TAS5630 data sheet that shows about 0.02% THD+N over much of the operating range. It gives an FFT of the noise floor, which seems to sit at about - 135 dB. This would imply that THD+N is dominated by HD, contrary to what I expected. Comments?


Thanks
 
I am not an engineer, but I want to point out that 41Hz's implementations of Tripath TK2050 and TA3020 amps have been measured with 0.007-0.008% THD+N:

41Hz Amp4
41Hz Amp15
Amp15-PS measurements

When the Amp15-PS BBM timing was lowered from 80nS to 40nS, "the THD+N at 1KHz went down from around 0.008% to around 0.005%." (source). This heats up the output FETs a great deal, to the point of blowing them up in some cases, so that the 0.005% THD+N distortion figure may not be realistically attainable in all configurations.
 
Thanks. Especially the last link with the measurements was very helpful. While it shows that the noise floor is excellent, it seems to indicate that THD+N is mostly defined by a -80 dB second harmonic (ok, maybe -82 dB if you take into account signal generator noise).

This is a number you would expect from a poorly designed AB amp with high feedback, or a well-designed AB amp with little or no global feedback.

Are the distortion mechanisms understood in class D?
 

AP2

Banned
2009-12-30 7:52 am
Italy
Hi all,
This FFT is also an excellent outcome reported to an amplifier
in class D.:)
 

Attachments

  • FFT-44K-8R-20w.PNG
    FFT-44K-8R-20w.PNG
    18.6 KB · Views: 667

AP2

Banned
2009-12-30 7:52 am
Italy
Yes, it might be good for class D, but -86 dB 2nd harmonic at 1 kHz is at best low end for class AB.

Hi,
yes, this is not very good for audiophile performances but... see full spectrum please. this is a good commercial audio amplifier class D,300-400w(4R) with PSU on board. you can see FFT at 1KHz from annother amplifier, spectrum is full of harmonics. very clean output has to work very hard.
many harmonics affecting the upper audio frequencies to listen, the voice is as shrill.
I not see clean FFT of others Amp, as the A-B class.:)
 
capslock

There are two basic families of class d. Clocked and self oscillating.

Clocked versions have a theoretical loop gain that is constant vs output power (given that the power supply is firm) which makes THD fall all the way to clipping similarly to a class B amp. The big problem is that their loop gain is low giving a high THD to start with. Adding integrating loops around a clocked amp gives a big increase in high order harmonics unless the fsw and harmonics are filtered out of the fed back signal.

Self oscillating amps have a much higher loop gain at low output powers but the loop gain falls as output signal is increased (due to dropping fsw) giving a rising THD at higher powers. It is somewhat easier to add integrating loops to a self oscillating amp but one has to make sure that the amp does not start switching at the resonance frequency of the output filter.

There are of course many more factors arising from the practical implementations.

One can achieve good audio performance with both clocked and self oscillating amps but it is much more complicated and expensive to do it with a clocked amp.
 
The best thing I found was the TAS5630 data sheet that shows about 0.02% THD+N over much of the operating range. It gives an FFT of the noise floor, which seems to sit at about - 135 dB. This would imply that THD+N is dominated by HD, contrary to what I expected. Comments?

Its a common mistake to read off the axis of an FFT plot without paying attention to the bin size. The FFT is actually not a continuous graph, its the contents of (in this case) 16,384 discrete frequency bins. If the number of points taken were changed, that -135dB figure would also change - making the bins narrower in frequency means each one collects less noise.

So what that graph, together with its annotations is telling us is for each of the 16k frequency points there's a narrow band filter whose output reads -135dB. The bandwidth of each filter is 24000/16384 - roughly 1.5Hz. Since the line looks flat we can predict the total (broadband) noise as being sqrt (16384) higher in level, which comes out at 128X. 128X is 42dB, giving a figure of -93dB. Interestingly, the datasheet comes out with the same 93dB figure for signal/noise ratio, A-weighted.
 

AP2

Banned
2009-12-30 7:52 am
Italy
Hi,
Your reasoning is correct, I agree.
The fft which I put on this thread is produced by the instrument hardware (National Instruments) with on-board DSP big.
128K FFT (I can set up to 512K/point)
24bit AD / DA. up to 200Khz.
But at this point after seeing many datasheets on amplifiers (they all seem cool), I only believe in what I measure myself.
example, if TA2030 has 113dB SN / R ... why? I hear white noise at 30cm from twiter?

regards
 
Tech datasheets often seem unduly influenced by marketing, so yes certainly be skeptical of their claims. I'll often cross-check one claim against another to see how well they stand up.

If you were meaning the TA3020, the datasheet I found said just 102dB relative to 275W/4R which I would expect to give rise to audible hiss at 30cm in a very quiet room. The point is not so much the SNR as the residual in-band noise - this shows as a typical 260uV. If the tweeter is 92dB efficient, it'll put out 92dBA at 1m for 2.83V in. The output noise is 81dB down on this, giving 11dBA at 1m. How this translates to intensity at 30cm is beyond me, but I'd stab in the dark at 10dB louder. I doubt my own room is quiet enough to hear this except perhaps at 4am on a night when my neighbours' ACs aren't running...:p
 

iand

Member
2007-02-05 10:58 pm
London
Tech datasheets often seem unduly influenced by marketing, so yes certainly be skeptical of their claims. I'll often cross-check one claim against another to see how well they stand up.

If you were meaning the TA3020, the datasheet I found said just 102dB relative to 275W/4R which I would expect to give rise to audible hiss at 30cm in a very quiet room. The point is not so much the SNR as the residual in-band noise - this shows as a typical 260uV. If the tweeter is 92dB efficient, it'll put out 92dBA at 1m for 2.83V in. The output noise is 81dB down on this, giving 11dBA at 1m. How this translates to intensity at 30cm is beyond me, but I'd stab in the dark at 10dB louder. I doubt my own room is quiet enough to hear this except perhaps at 4am on a night when my neighbours' ACs aren't running...:p

The best performance I know of is Zetex's DDFA which gets THD down to below -90dB and noise down to below -120dB, which is therefore still THD-dominated.

Whether you can hear noise or distortion at these levels under any circumstances -- let alone normal listening levels -- is debatable to say the least, but it does show what can be achieved using good engineering.

Ian

(disclaimer -- I don't work for Zetex, but did work on the modulator chip design to get picosecond jitter levels -- and I've measured it so I can vouch for their claims)
 

AP2

Banned
2009-12-30 7:52 am
Italy
Certainly listening to white noise on twiter is not a scientific measurement. I am surprised he has heard at night in the laboratory, some amplifiers with the same loudspeaker huge difference in noise.
I have no doubt on the levels of some technology companies.

Regards
 
The best performance I know of is Zetex's DDFA which gets THD down to below -90dB and noise down to below -120dB, which is therefore still THD-dominated.

Whether you can hear noise or distortion at these levels under any circumstances -- let alone normal listening levels -- is debatable to say the least, but it does show what can be achieved using good engineering.

Ian

(disclaimer -- I don't work for Zetex, but did work on the modulator chip design to get picosecond jitter levels -- and I've measured it so I can vouch for their claims)

Ian, thanks for the pointer that made me read up on DDFA. Are any affordable commercial implementations or evaluation boards around by now?

90 dB harmonics are about the level of lower midrange DACs. Audible or not, good multi-level sigma delta DACs can do a lot better, as can traditional class AB power amps.

Any idea what the limiting factor is? Output stage nonlinearity? Lack of feedback bandwidth?
 

iand

Member
2007-02-05 10:58 pm
London
Ian, thanks for the pointer that made me read up on DDFA. Are any affordable commercial implementations or evaluation boards around by now?

90 dB harmonics are about the level of lower midrange DACs. Audible or not, good multi-level sigma delta DACs can do a lot better, as can traditional class AB power amps.

Any idea what the limiting factor is? Output stage nonlinearity? Lack of feedback bandwidth?

The only commercial product I'm aware of -- though I don't know anything more about this than the public -- is the NAD one. There are eval boards from Zetex but only for commercial customers, not the hobbyist market -- it's impossible for most semiconductor companies to support this market, the sales volumes don't justify the effort.

I'd guess that the distortion -- which is still only around 0.0025% at full power -- is limited by either the output stage distortion (greatly reduced by negative feedback, of course) or the distortion generated by the PWM reference DAC which generates the "ideal" output signal used in the feedback loop. It's certainly not limited by feedback bandwidth given the 108MHz internal sampling rate or the 850kHz PWM frequency.

More relevant to audio applications is probably the overall dynamic range and also the low-level linearity, which is good down to around -140dBFS -- if this is ever audible, which is unlikely given the SPL you'd have to be listening at...

Either way the overall THD/noise/linearity is better than any other DAC+amp combination that I'm aware of. The argument about whether such performance is needed for audio I leave to the subjectivist vs objectivist camps, though those in both camps who've heard the DDFA seem to like it.

Ian
 
I guess there are now several implentations around that are significantly better than -86 dB harmonic distortion, all of them self-oscillating:

1. the Hypex nCore with - 103 dB THD+N at 200 W - various improvements such as improved discrete comparator, better understanding of large scale non-linearity of the modulator, improved gate drive and vastly improved overall feedback loop

2. the IRS20955 reference design IRAUDAMP4 with -98 dB THD+N at 20 - 50 W - this uses a humble TLC081 as integrator and no post-filter feedback (though it could probably be added)

(interestingly, the IRAUDAMP6 with IRS20957 achieves only -94 dB which I suspect is due to higher voltage, higher capacitance and resitance FETs being used)

3. the IRS2092 reference design IRAUDAMP5 that has -86 dB THD+N at 60 W and a pretty clean spectrum (- 90 dB 2nd HD and then essentially nothing up to the 6th harmonic), this is similar to IRAUDAMP4, but with an on-board integrator

any more examples that we overlooked back then or have come out recently?
 
I am not convinced at all that THD+N or harmonics are the key factors for a good sounding amp as long as it is reasonably low - say 0.05% at half power.
Otherwise most tube amps could not sound well.
And no speakers could sound acceptable at all.

But I also enjoy to measure harmonics - always making me feel like quantifying sound quality. Extremely satisfying illusion. ;)
 
I am not convinced at all that THD+N or harmonics are the key factors for a good sounding amp as long as it is reasonably low - say 0.05% at half power.
Otherwise most tube amps could not sound well.
And no speakers could sound acceptable at all.

But I also enjoy to measure harmonics - always making me feel like quantifying sound quality. Extremely satisfying illusion. ;)

It seems to be the same problem that we have had for many years; we have no good way of measuring sound quality with other equipment than our ears.

I have not seen anyone quantify "good sound" to specs that we can measure as designers... we will test stuff like THD+N, IMD, TIM, crosstalk and so on, but only to see that nothing is completely wrong before we go disturb the golden ear that has to approve our design.

\\\Jens :)
 
Last edited: