advertisement


Phase Noise

If I remember correctly, John also was speaking about that jitter different frequencies affecting differently and one single number does not reflect reality correctly.
 
sq225917 said:
The only thing that matters is at what level does it become audible?
Click to expand...
For a given sample rate, this depends on the DAC architecture. The classic parallel 16 bit DAC is far more tolerant than single bit Delta Sigma or DSD. Multi bit Delta Sigma (BB, Wolfson) are a compromise
 
gints said:
If I remember correctly, John also was speaking about that jitter different frequencies affecting differently and one single number does not reflect reality correctly.
Click to expand...
That is correct, it is related to the shape of the sideband skirts of the phase noise spectrum.
What you really don't want is some sort of periodic modulation to the jitter, this can lead to more audible beating effects. This is something ESS have been accused of
 
There was a couple of articles in WW some time back about a new concept which the authors called an Anti Jitter Circuit, IIRC, and which was claimed to have all sorts of advantages over PLL's etc. Example circuits were given and interested parties were invited to contact them for a licence to use the new tech (which was pretty simple actually).
I heard nothing more about it there after....
 
I have not seen this before and will have to think about it.
My first thought is that making a monostable that doesn't have a lot of jitter of its own is not trivial, so this might be one of those circuits that simulates well but is hard to do in practice
 
davidsrsb said:
I have started this thread for discussion about phase noise and its relationship with jitter.
I want to keep out of the Westlake thread as I am not a sponsor.

These are useful links
https://www.everythingrf.com/community/what-is-phase-noise

https://blog.bliley.com/ultimate-guide-to-phase-noise

Phase noise is an analogue RF thing. When you take an RF signal and square it up to be a clock, phase noise contributes to timing jitter.
Click to expand...
I think I got some ideas as to what this might be now. Basically you cannot produce a very tight frequency pulse/spike, you're always gonna get spill over into nearby frequencies and also the main frequency cannot be kept absolutely fixed due to the various inherent and theoretical limitations. This frequency 'spill-over' will cause time domain jitter.

The best way to reduce all this is to use a better crystal oscillator in the first place rather than phase locked loops and other tricks to fix/control the frequency drift/jitter.

Temperature and other environmental factors need to be controlled as far as possible first before trying to compensate for these effects with clever electronics.
 
Real world effects like locking make it very hard to get good isolation between a "clean" output clock and a very close input source clock. I have even seen ovened oscillator clocks jumping in frequency when the thermostat cycles
 
sq225917 said:
The only thing that matters is at what level does it become audible?
Click to expand...
My own rule of thumb from "a finger in the air" is that any jitter less than 200 ps from a CD player or DAC, as measured by Julian Dunn's "J-test", is likely to be inaudible. And this is regularly bettered in even modest products.

[TL;DR]There are various more complete answers in the literature depending on how you define "audible". But it's quite a complex subject since you have multiple variables to handle. I had to understand a little about the subject many years ago when professionally I took on a digital audio project. But I am not formally educated in the subject.

There are quite a few useful published papers by Julian Dunn on the topic. His most compact summary is in the graph in section 3.3 and figure 9 of "Jitter: Specification and Assessment in Digital Audio Equipment" where, to take just one point, sinusoidal jitter of 100 ps at 3 kHz is a "maximum inaudible jitter amplitude" and at 20 kHz the limit is down to about 20 ps. But IIRC audibility here is when converting full-scale digital signals to 120 dB SPL gives rise to unwanted output above 0 dB SPL. Very much worst case - especially at high signal frequencies in real music. But it does give room for the jitter to be signal-correlated.

There's a Stereophile article "Bits is Bits?" (a version of a published paper) in which Malcolm Hawksford and Chris Dunn go reasonably deeply into S/PDIF interfacing, e.g. between CD transport and separate DAC. Starting on page 7 there's "Audibility of jitter errors". The analysis is close in principle to Julian Dunn's and it gives similar numbers. Again it's for the worst case as you go up in signal frequency so I take low to mid-frequency (up to circa 3 kHz) numbers as most useful. [/TL;DR]
 
My only experience of this was a sound in sync system I worked on for the BPO in the early 80s, we were sampling at half TV line frequency with a clock phase locked to line sync. Jitter was gross and with the benefit of hindsight, it did spoil the SNR on a first generation telephony delta sigma ADC
 
davidsrsb said:
Real world effects like locking make it very hard to get good isolation between a "clean" output clock and a very close input source clock. I have even seen ovened oscillator clocks jumping in frequency when the thermostat cycles
Click to expand...

That certainly happens - better quality oscillators use proportional control on the oven heater but there are numerous pitfalls along the way.
 
nostromo said:
I think I got some ideas as to what this might be now. Basically you cannot produce a very tight frequency pulse/spike, you're always gonna get spill over into nearby frequencies and also the main frequency cannot be kept absolutely fixed due to the various inherent and theoretical limitations. This frequency 'spill-over' will cause time domain jitter.

The best way to reduce all this is to use a better crystal oscillator in the first place rather than phase locked loops and other tricks to fix/control the frequency drift/jitter.

Temperature and other environmental factors need to be controlled as far as possible first before trying to compensate for these effects with clever electronics.
Click to expand...

A PLL does not improve the phase noise of a crystal oscillator. It improves the close-in phase noise of an otherwise free-running oscillator by locking its phase to a crystal oscillator, typically running at a lower frequency. Crystal oscillators have excellent phase noise, but choice of frequency is limited, hence PLL-based synthesisers.
 
John Phillips said:
My own rule of thumb from "a finger in the air" is that any jitter less than 200 ps from a CD player or DAC, as measured by Julian Dunn's "J-test", is likely to be inaudible. And this is regularly bettered in even modest products.

[TL;DR]There are various more complete answers in the literature depending on how you define "audible". But it's quite a complex subject since you have multiple variables to handle. I had to understand a little about the subject many years ago when professionally I took on a digital audio project. But I am not formally educated in the subject.

There are quite a few useful published papers by Julian Dunn on the topic. His most compact summary is in the graph in section 3.3 and figure 9 of "Jitter: Specification and Assessment in Digital Audio Equipment" where, to take just one point, sinusoidal jitter of 100 ps at 3 kHz is a "maximum inaudible jitter amplitude" and at 20 kHz the limit is down to about 20 ps. But IIRC audibility here is when converting full-scale digital signals to 120 dB SPL gives rise to unwanted output above 0 dB SPL. Very much worst case - especially at high signal frequencies in real music. But it does give room for the jitter to be signal-correlated.

There's a Stereophile article "Bits is Bits?" (a version of a published paper) in which Malcolm Hawksford and Chris Dunn go reasonably deeply into S/PDIF interfacing, e.g. between CD transport and separate DAC. Starting on page 7 there's "Audibility of jitter errors". The analysis is close in principle to Julian Dunn's and it gives similar numbers. Again it's for the worst case as you go up in signal frequency so I take low to mid-frequency (up to circa 3 kHz) numbers as most useful. [/TL;DR]
Click to expand...

I tried my best to get something from this, but it's way beyond me. I'm just glad there are people (such as yourself) who do understand this stuff. It seems to me that nowadays for DACs, 'bits ARE bits' and the obvious jitter issue is now practically solved for audio with better clocks and synchronisation circuits.
 
You must log in or register to reply here.


advertisement


Back
Top