System Clock Oscillator .... does it reduce THD?

Jump to Latest
Baldin

Baldin

Member
Joined 2005
sensibleaudio.dk/
#1
Working on different DSPs, with ADCs and DACs, I was wondering what happened to the ultimate clock.
Remember 10 - 15 years ago, every body wanted to change the clock in digital sources like CDs to something fancy from better discrete Xtal oscillators, over TCXOs to OCSOs.

Have anyone done some THD test of such changes?

I'm planning on using X1E0000210427 https://jlcpcb.com/partdetail/SeikoEpson-X1E0000210427/C841675 which is a 10ppm Xtal, in conjunction with a ADAU1452

But I could go for a Crystek CCHD-575 https://www.audiophonics.fr/en/comp...p-12445.html?search_query=osci&fast_search=fs
Promising very low phase noise.

Or one could go crazy with a real OCSO https://www.golledge.com/products/sine-output-oven-controlled-oscillator-with-5v-supply/c-26/p-285

Or a NeutronStar!?!
http://www.newclassd.com/index.php?page=36


The big question is whether it has any impact at all on performance, or it is just an impact on the wallet 😉 .....
 
#2
Low close-in phase noise reduces noise sidebands around the signal to some extent, although those are often dominated by voltage reference noise rather than close-in phase noise.

Depending on the type of DAC, a low phase noise floor may or may not reduce the DAC noise floor measurably.

THD has nothing to do with it.
 
#3
Hi Baldin-San,
I guess the system clock phase noise is related to the spurious/sideband of the output sound spectrum. On the other hand, the frequency variance is not so a matter for the tone characteristics.
The below link is useful information but written in Japanese. So let's try the machine translation.

(Orignal Link)
EDN Japan Article
(Machine Translation)
google translate

I usually use RTL-SDR & SDR# or Analog Discovery2 to see the short-term jitter of the system clock.
スクリーンショット 2023-02-19 3.32.28.png

スクリーンショット 2023-02-19 3.37.52.png

CyberPit
 
Last edited:
#6
If you ask a 44,1 Fs system to convert a 11,024 kHz signal and you move the 3rd "time tick" 10% forward but still convert the indicated integer sample value to the correct voltage level - and, this particular sample (in the middle of second half of the sine) is surrounded by perfectly converted samples - both in level and time - what you get is a wave form of the negative "sinc" that represent something else than was intended - its somehow a bit skewed... this mean that THD will be affected - because the wave form indicates something else than a pure sine of one frequency - very little, but still - I cant see anything else happen really... the scewed sine defamation will show up in a FTT as new energy at an other frequency -> THD...

No?

//
 
#7
In the old days of multibit DACs, the signal was latched or switched to the output on the clock. If that clock transition came at the wrong time, the change from the previous sample would come too early or too late. In sigma delta, due to their very high oversampling, the time chunks that could be misplaced became much smaller.

Just look at tests of current DACs by Amir at audiosciencereview. Most are near perfect with jitter sidebands at -130 dB with quite modest oscillators.
 
#8
@TNT Sure, systematic jitter with a very specific relation to the sample rate and signal frequency might cause harmonic distortion, but the question relates to crystal oscillators. Those normally produce random jitter that randomly phase modulates the signal, leading to sideband skirts rather than harmonics.
 
  • Like
Reactions: Markw4 and TNT
#10
Baldin said:
...the fact that it only seems to affects the noise floor...
Click to expand...
Not exactly. There is what is sometimes referred to as jitter, and what is referred as close-in phase noise. Jitter is also known as far-out phase noise. They have rather different effects. Jitter affects noise floor (not that the noise floor of a dac is a constant, it can follow the audio signal but that's another subject). Close-in phase noise may affect signal detection of closely adjacent audio frequencies. IOW, a FFT spectral line, if examined in a very high resolution FFT has noise skirts around the base of the spectral line. On an FFT one can't tell the difference between phase noise and amplitude noise when visualizing the skirts, but the phenomena that produce skirts are more orthogonal in time, at least for some purposes. That is, unless one supposes humans hearing is virtually the same as a FFT analyzer that discards phase information. Well then, it turns out to be not so simple. It is uncontroversial that humans can hear phase at LF, thus making LF-transient time-domain waveform-fidelity of potential interest. Also there is some evidence humans can be sensitive to phase in some cases at higher frequencies but it depends very much on the actual time-domain waveform.

Where close-in phase noise has been hypothesized to affect sound a lot is in the 1Hz to 10Hz band. Of course music pitch perception in humans doesn't go that low. However a lot of transient events have amplitude changes in that frequency band. Maybe consider a timpani drum's attack, sustain, decay changing over time. To represent that information in the frequency domain would appear to require corresponding frequencies to exist. Either that or they are beat note effects (interference patterns that don't involve their own FFT spectral lines).

Anyway, with better clocks some people have reported improved sound, including a more accurate stereo illusion of sound stage imaging in L/R and depth dimensions. Accurate reproduction of such effects may also require a system which has characteristics not necessarily well represented in typical published measurements. I will not go into more detail at this time on that subject, but it might be something to consider later.

All the above having been said, once clocks and phase noise are 'good enough' there may be a point of diminishing return. Often close-in phase noise is not the biggest problem in a system. That can be true even with a system that measures exceptionally well using standard 'figure of merit' measurements. Of course, some people will disagree with much or all of what I just said. All I can say is that if you really want to know what is attainable, the best thing to do is to come and listen to a system pretty close to as good as it gets. Its because people can only know the best they heard so far. Beyond that they have no way knowing what is actually encoded into a digital recording or even an analog one. Sometimes a visitor will stop by here in Auburn to see for themselves. I promise it would interesting 🙂
 
Last edited:
  • Like
Reactions: fedde
#16
For measuring close-in phase noise with FFT the FFT size should be very large (at least 4M). Zooming to +/-10Hz around the fundamental would show the close-in phase noise better.
See e.g. here.

Whether or not close-in phase noise is audible is another matter. Low close-in phase noise indicates that Vref and clock circuitry are well implemented so it is a "figure of merit".
 
You must log in or register to reply here.
Top Bottom