|
Post by svart on Sept 21, 2020 8:08:15 GMT -6
I think what RB proves is that like distortion we like certain types of jitter. There is nothing wrong with this concept, we just have to accept it. This is kind of where my head's at on the subject currently. After sifting through endless info on the subject and doing several shootouts, I'm totally open to the possibility that I might prefer a little jitter. Obviously, the only way to find out is to get the $7000 Antelope. There's different kinds of jitter. Deterministic jitter is non-random and has a correlated source, such as from a second clock/oscillator that bleeds onto the primary clock lines (periodic jitter). The harmonic output from this jitter can cause a rise in even or odd harmonic distortion depending on the frequency being modulated onto the primary clock. Deterministic jitter can have a correlated but aperiodic source if the ingress is from digital data lines. This type of jitter seems random but is not. Since the data is usually bounded by datarate, it can also have specific harmonic effects on the primary clock. Random (non-deterministic) jitter is always uncorrelated and has more of a spread-spectrum/dithering effect on the primary clock and the harmonic results are more evenly distributed between even and odd harmonics. Generally this is very low probability as noise ingress from other sources or reflections from external connections are much more likely. There's also manifestations of jitter through the hardware/software pieces as well. Additive jitter is simply how much of any kind of jitter a device adds to the signal. Both active and passive components can add jitter and they can be from deterministic and random sources and can be classified into a few different symptoms below. Cycle-to-cycle jitter is where the period (frequency) of the clock is changing. This could be from either random or deterministic jitter external sources or poorly designed clock sources like PLLs. Period jitter is where the duty cycle of a pulse is different from the duty cycle of the periods around it. Could be caused by a lot of things but generally from random noise sources. Phase jitter is an accumulation of all jitter over a period of time and is used as a measure of quality. Generally measured in RMS values of time, over a frequency range. An oscillator I'm looking at using for a design has a phase jitter measurement of 80fsec RMS over 12KHz-20MHz integration bandwidth. So, the type of jitter can have a profound effect on how it's perceived. Non-random jitter might cause specific harmonic issues whereas random jitter might alleviate harmonic issues by adding something similar to a dithering effect to signals.
|
|
|
Post by svart on Sept 21, 2020 8:22:23 GMT -6
I’ll wager 99% of the actual “clock issues” out there are actually termination issues on large systems, or clock cable issues. I once had to visit a customer who had an issue with some devices they bought from my company. These devices used a serial (LV-RS-232) cable connection and were daisy-chained together. We specified a maximum of 200ft of contiguous connections. The customer was adamant that they were only about 180ft apart and the last few units would work randomly and then stop. Turns out the devices were 180ft apart but the cables they used were run up through the cable trays and added another 100ft to the length through all the twists and turns they made. These were custom cables with unknown specs and they were also adamant that they were not going to exchange the cables nor add new ones. The customer was still unhappy and threatened to send all this gear back even though they were the ones who violated our cabling specifications.. But such is life when dealing with big corporate customers. Anyway, a few minutes with an oscilloscope showed a horrendous reflection between the last units and the ones 300ft away. The reflection was causing data bits to reflect back and forth causing a condition known to digital folks as "double triggering" where the reflected bits are interacting with the data from a subsequent transmission effectively causing the unit to see two corrupted commands. The units already had terminations, but I added a secondary external termination to continue to load down the data lines so that the reflection was loaded down so much that it was below the triggering threshold of the receivers. Quite the on-the-spot feat with the director of engineering for this large media conglomerate looking over my shoulder the whole time but it saved a few million dollars worth of sales (even though it was never our fault). Terminations matter!
|
|
|
Post by tkaitkai on Sept 21, 2020 8:55:08 GMT -6
svart That was a very concise explanation of something that has eluded me for a good while. You said more in a few paragraphs than people do in 20-page digital clocking flame wars on GS. I appreciate you sharing your insights. Excellent post.
|
|
|
Post by svart on Sept 21, 2020 8:59:19 GMT -6
svart That was a very concise explanation of something that has eluded me for a good while. You said more in a few paragraphs than people do in 20-page digital clocking flame wars on GS. I appreciate you sharing your insights. Excellent post. No problem. I'm afraid I overly simplified it a bit so that it's not too confusing, but it should suffice for most who are interested in the basic ideas.
|
|