The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Design >> High-Speed I/O Design >> Eye-Diagram
https://designers-guide.org/forum/YaBB.pl?num=1366215145

Message started by raja.cedt on Apr 17^th, 2013, 9:12am

Title: Eye-Diagram
Post by raja.cedt on Apr 17^th, 2013, 9:12am

Dear all,
i was referring some jitter basics from the following link. In slide 32 there are two eye diagram, it looks top eye diagram looks very clean and i would expect less jitter compared to the bottom, their Quality looks very far that's why i could comment on that. But measurements shows exactly opp conclusions in the slide 33&34. can any one point me some points regarding this??

http://www.home.agilent.com/upload/cmc_upload/All/ADMF2009_HowToMeasureJitterEffectively.pdf

Thanks,
Raj.

Title: Re: Eye-Diagram
Post by smlogan on Apr 17^th, 2013, 2:20pm

Hi Raj,

... it looks top eye diagram looks very clean and i would expect less jitter compared to the bottom, their Quality looks very far that's why i could comment on that. But measurements shows exactly opp conclusions in the slide 33&34. can any one point me some points regarding this??

A good question indeed! However, the conclusion drawn by the article is likely correct. The reason that the eye diagram on the bottom of slide 32, which appears to have a larger peak-to-peak jitter than the eye diagram on the top, is said to be superior lies in the measurement methodology and, ultimately, the type of jitter. If you examine slide 33 - which decomposes the jitter of the lower eye diagram on slide 32 - the random jitter component is measured as 2 ps rms and the deterministic components account for the remainder. Slide 34 - which decomposes the jitter of the upper waveform of side 32 - shows a random jitter component of 11 ps rms. In many applications where bit error rate is a concern, the peak-to-peak random jitter component is estimated for a BER of 1e-12 as 14*sigma. Hence, with a 2 ps rms random component, the peak-to-peak random jitter in a 1e-12 BER application is 14*2 ps rms = 28 ps. However, in the upper waveform on slide 32, the random jitter contribution in a 1e-12 BER application is 14*11 ps rms = 154 ps. Hence, the actual eye closure when observed over a significant period of time of the upper waveform in slide 32 is much greater than that shown in slide 32.

If the measurements shown in slide 32 were taken for a greater period of time, the eye closure of the top waveform would indeed show greater eye closure than that of the lower waveform. The fact that the measured data was taken over a relatively short period of time leads to the deceptive eye diagrams. Moreover, the short time period over which the measurements were taken probably contribute to a poor estimate of the random jitter standard deviation. Hence, a better overall assessment, in my opinion, is to capture both waveforms for a longer time period.

I hope this makes sense. Let me know if I have confused the issue.

Shawn

Title: Re: Eye-Diagram
Post by raja.cedt on Apr 18^th, 2013, 9:37am

Hello smlogan,
Thanks for detailed reply. Have you noticed, they both were measured for same time interval then how come the first one has more random jitter. I understood that if run for lesser time you would see less jitter random accumulation.

Bottom line is you can't comment eye diagram Quality with-our eye, isn't it?? or do you have any rough way of identifying good eye.

Looks like you have good understanding on eye-diagram so i would like ask some more basic's without posting new one.

1. is it good to depends on transient noise simulation to check RJ? if not how would you go with TX or RX design??
2.In slide 35 there is a bathtub curve, from the optimal sampling point if you move either side why random jitter start impact rather DJ initially and DJ at the end of cycle?

Sorry for many Questions...
Raj.

Title: Re: Eye-Diagram
Post by ywguo on Jun 8^th, 2013, 3:22am

Hi Raj.

Quote:

1. is it good to depends on transient noise simulation to check RJ? if not how would you go with TX or RX design??
2.In slide 35 there is a bathtub curve, from the optimal sampling point if you move either side why random jitter start impact rather DJ initially and DJ at the end of cycle?

1. Theoretically spectre simulator has the ability to simulate RJ with transient noise. But it is not often practical because Serdes is very high speed circuit. And you must have use devices of very high Ft. So you need to set very big value for Fmax to get accurate result. Since the Fmax of transient noise is inversely proportional to Fmax, the time step becomes very very small and the simulation becomes very very slow, especially for bigger circuit like whole TX, RX.

2. The bathtub is a probability of bit error. When the sampling point is near the begin and the end one UI, it is determined by the DJ, and near 50%. In the other part, the probability is determined by RJ as the equation shown in slide 35.

Best Regards,
Yawei