Especially that.
While chasing nanovolts, I think that Zero Signal Plane is worth to consider.
Interesting read.
As we all know, exp(jwt) is cos(wt)+j sin(wt) which is a complex number running in a circle.
A dac can play complex signals if you put real on left channel and imaginary on right channel. It's just a sin and a cos.
Acquire both channels with ADC (soundcard), and you get the original complex number back.
Multiply with exp(-jwt) and extract modulus and angle
angle gives phase shift between dac clock and adc clock, so you get a time domain view of jitter. No FFT!
As we all know, exp(jwt) is cos(wt)+j sin(wt) which is a complex number running in a circle.
A dac can play complex signals if you put real on left channel and imaginary on right channel. It's just a sin and a cos.
Acquire both channels with ADC (soundcard), and you get the original complex number back.
Multiply with exp(-jwt) and extract modulus and angle
angle gives phase shift between dac clock and adc clock, so you get a time domain view of jitter. No FFT!
I like where this is going.
This post and the whole thread are relevant. With four Channels you can make much more precise or deep measurements when it comes to close-in Phase Noise. The processing isn't totally complete, but I know what the Big Boys do with their multi-thousand $$ gear, which I'll probably never buy. The post and the graphs show the calculations with simulated data but using four Channels for the processing.
The hardware needs are quite specific. I have two such Quad-Channel ADC cards. I can provide one to someone who'd like to do a DIY build. This can be paired either with an FPGA or the computer for processing.
By the way, referring to earlier posts here, I see the microphony of caps. This is well known.
However, has anyone thought of checking the effects of vibration on the crystal or clock itself? That would be an interesting area of exploration because I believe the crystal itself can behave differently under vibration - there should be an incidence on CIPN and hence on SQ.
However, before worrying about all this, there's a lot of phenomena whose effects can totally mask those of CIPN.
However, has anyone thought of checking the effects of vibration on the crystal or clock itself? That would be an interesting area of exploration because I believe the crystal itself can behave differently under vibration - there should be an incidence on CIPN and hence on SQ.
However, before worrying about all this, there's a lot of phenomena whose effects can totally mask those of CIPN.
Not quite clear to me. Playback and recording from the same clock tends to be jitter canceling. You send out a sequence of samples. Say, they get played back and re-recorded at the same wrong time. You get back a sequence of samples. The assumption is that the sequence is evenly spaced in time. Any errors in sample values are presumed to be amplitude errors, maybe added noise and or nonlinearity in the converters. Am I missing something?
True!
Unless you send the sound card spdif output to a wm8805 based dac and you want to know how much jitter the chip adds. In this case having the spdif transmitter and the adc use the same clock is an advantage...
If you haven't got two channels at your disposal, you can also use one channel and filter off the sum product, like radio engineers have done for more than a century:
cos(omega t) = 1/2 exp(j omega t) + 1/2 exp(-j omega t)
Multiply by exp(-j omega t) and you get 1/2 + 1/2 exp(-2 j omega t).
Now all you have to do is filter and multiply by 2.
With either method, when the clocks are independent, you are bound to get a slope in the phase due to frequency differences.
Well then, that's the 'in the computer for processing' case I mentioned, isn't it?
It's been a while since I touched the processing code, but that would be the method used to derive the first graph.
There's another thing (state-of-the-art) I did which the best R & S and some others do and that's in the second graph and uses 4 Channels.
John, who wrote R.E.W., could easily adapt it to do the same thing IMO.
Yup. "software detector". That's how the "output stage shootout" works. Note if the frequency is such the period is an integer number of samples, you dont need to filter, only average over a period... because then exp(2jwt) averages to zero over the period... If it's not an integer number of samples it's more annoying.
Yes that's where it all starts
BTW I still regret not having used the (already bought) canned XO's for the WM8804. It was the intention to implement those but I could not find space for it including a separate regulator. As a kind of curse they tend to show up always when I don't need them.
BTW I still regret not having used the (already bought) canned XO's for the WM8804. It was the intention to implement those but I could not find space for it including a separate regulator. As a kind of curse they tend to show up always when I don't need them.
Last edited:
Of course they are basically the same but then again they aren't the same anymore.
The feeling that someone bakes you a very nice cake but then you have to go.