Excellent post Tony, although l have to Re Read it to digest it properly.

Damien

l guess it would be good to show one good measurement and then a bad one, with some arrows to briefly explain what we should be looking for.

I have a series of them planned, test equipment with how and why we use next though

After that some practical experiments of what we see and how it relates to the real world.

Tomorrow if I get time, I'll put up about a well known TAS Gustard Dac A22 before measurments and explain what they mean and you could interpet them but right now I have a Nagra HD dac thats requires some TLC

That's some kit there using, and a interesting videos Tony.
l get the point, although some terminology is lost on me, but it's really interesting.That graphics overlay on the noisy power rail is a great tool.

Wow...
Thank you Tony,for taking the time to explain this in basic terms, l am sure other TAS members will find this interesting.
l know your a busy guy, so thanks again.

Thanks, Tony. Interesting. A small question though. Is Num the number of samples taken to produce the mean and standard deviation? If so it is interesting that the standard deviation of the measurements of voltage taken with 5,500 samples (first reading) is larger than when 301 samples was used. As these measurements are taken over time, does that then suggest that there is more variation in voltage, but with a longer periodicity, than indicated during the shorter sampling 'window' (I assume that the sampling frequency was the same) - voltage drift, etc?

Actually I think I might have just answered the question myself if I am reading the screens correctly: the first reading was done using 10 MS/s whereas the second was done with 2. If that is the case then does that mean that the variation observed (albeit small in range) is at a shorter periodicity than indicated by the first set of observations?


As an aside, a friend has a fairly large collection of old Avometers (he is a broadcast engineer by training and so I guess it is a bit like me with cameras and microscopes!). He gave me one a while back that came in a batch he'd bought as it was marked as having come from one of the old ICI plants and has their old orange livery (and I worked for ICI for 20 years). I don't use it, but it does sit on display in the workshop!

It’s been a wee while since I last placed some information on my blog, apologies we have been insanely busy of late, however today I'm going to cover a popular topic on TAS, aftermarket clocks.

Today I'm going to give you pointers on how to look for the more quality sounding clocks and what exactly you should take into account when doing your research into selecting one for your system

So are the stated specifications for various products REALLY delivering what they state, and looking for ultimate low phase noise @ 'X' and 'Y' frequency is this the holy grail or are they other factors at play here that determine / generate a more realistic indicator of how this will change the sound quality of your system.

For example does a more straighter and cleaner looking rise time leading edge of the wave form change the perceived sound quality in any way? Conversely is having minimal ' Gibbs Ears' or overshoot on the posing tive width aspect of the wave form shape have a bearing at all?

The overriding factors with clocks are believe it or not, not just the actual clock module itself, it the whole circuit everything including the cables you attach to the said device, we in EE circles call this the transmission line, the total pathway from a/c entering the box to the point of entrance onto the connecting device everything in this path-way has an effect, large or small and by effecting many different aspects of the clock generation process and changing many of the parameters that really mater in delivering a quality, ultra-stable, ultra-low drift timing signal that will produce positive differences in your system sound quality.

So, let’s look at why you would wish to use an externally generated audio clock signal in the first place, after of the vast majority of audio dac’s (consumer) do not have the ability to accept a direct master clock feed frequency, I can only thing of seven brands off the top of my head right now. Compare that to the dozens of other makes which don’t.

To do this we need to look at how audio data is transmitted between boxes and internally within an all in one CD/Dac unit. The vast majority use a form of serial audio data transfer which is called I2S (Inter IC sound) or pronounced (Eye Squared ‘S’) it is loosely based on another serial data format I2C although that only has two lines of data as opposed to I2s which has three layers which are Audio serial data / Bit clock / Word select clock.

The word select (Clock) is for determining either left or right channel (1 or 0) low side for left AND HIGH SIDE LEADING EDGE FOR RIGHT. However it’s important is the overall timing of the serial data itself and we can use those, much touted word ‘Jitter’

So, the theory and model goes, the more faithful to the ideal square wave the better the result will be.

The starter for 10 is this function generated 10 Mhz Square wave form as base line comparison for you, the FG used is decent 16bit model as you can see its not as most people would expect, (It doesn’t look perfectly square) Oddly enough SW don’t exist in nature per say, they are made up of multi different frequency sine waves and ‘bolted together’
All the measurements at this stage are taken with identical settings: real time sample rate is 10GS/s with a real time memory rate of 5Ks with a HTB (horizontal time base) of 50ns per division.


This is the base line image for 10Mhz SWF generated by the function generator.

If we look at a current Influencer touted special 10Mhz clock we can see that it’s figure of merit for the same measurement article is 879Hz again a very good figure of merit this gives an accuracy percentage of 0.00879% again an excellent number.

For a control sample a reference 10Mhz clock I use here in the lab to keep all of the test equipment on superior time base.

Again the same settings used across the experiment, with at least 250K samples taken on each sample time.

Forgot to mention the scope bandwidth was set to 200Mhz, with digitals scopes, like audio Nyquist theory is relevant here so at least twice the BW is required to view the frequency you are viewing. However, the higher the actual frequency you are viewing in general the more BW you need.

So our 10Mhz square wave form personally I would use a 50Mhz BW setting on the scope, however for a more complete picture I am using 20 times the sampled frequerncy to ensure sufficient capability ofthe scope also the wave form has been adjusted to use the 12 bit ADC potential to the fullest as in the wave form is taking up at least 3/4 of the screen height.

Last section of the investigations this is a full serial data analysis with eye diagrams, X/talk (digital noise measurement), PLL (phase lock loop track) which looks at the secondary reclocking scheme that many clock employ to improve performance is shown as a similar visual as the jitter track in the pervious post. Full jitter suite which is more in depth than the previous (but still cracking Gromit jitter kit) goes more in depth, can also extrapolate various other insertable (explanation later) clock models and programmable matlab (third party software you can programme your own formula’s and models and import into then scope) modelling / TIE (time interval error) plots, bath tub jitter indicators (yes really) eye opening and measurements plus we can ask the scope to calculate the BER (Bit error ratio) to sit inside the eye diagram to check if the 'PC modelled' eye pattern matches the real world result.
It would take me a long while to fully explain all of the indicated window functions, which in all honest would bore you senseless, however these images are for the purpose of comparison just to show a real in depth breath of measurement abilities. they are also FFT spectrum plots of various different parameters etc. The eye patterns should be clean and open and cross evenly and look like an eye shape.

If anyone wishes to ask questions then no problem.