• State Resolved
  • Locked Locked
  • Replies 4 replies
  • Subscribers 20 subscribers
  • Views 717 views
  • Users 0 members are here
Support feedback
Options
Options
Related

This thread has been locked.

If you have a related question, please click the "Ask a related question" button in the top right corner. The newly created question will be automatically linked to this question.

Marginal jitter calculation

Julien Appaix
Prodigy 145 points

Hello there,

In the context of writing a specification of a LVDS systems interconnection I must precise what jitter requirements are expected (jitter at the LVDS transmitter ouput, jitter at the receiver input, maximum clock and cable jitter, and also eventually a margin jitter).

I have a serie of question which concern additional and margin jitters.

 

First of all, we can read and calculate output transmitter jitter (given a BER) thanks a datasheet. My question is : should we consider an additional jitter margin in order to secure ? In a first approach I do not think so due to the jitter is calculated given a BER... However, maybe telecommunication and high speed data link experts do usually take such an additional margin. If it is done so, does it exists rule of thumb values given for example in percent of the UI or in ps ??

In another hand, I am wondering the same question for jitter at the input of a LVDS receiver. Datasheets specify either the minimal eye opening width or both side maximal jitter a signal arriving to the receiver must have. I wonder if it should be considered a jitter margin?

Finally, I often read about marginal jitter in literature as it is more conservative jitter value since it is taken at non-zero threshold voltage (it is thus bigger than peak-to-peak jitter). Regarding this, I read a lot about measuring this however I never read anything about how to asses it. Is this a kind of value that we can only measure and that could not be assessed? Does it exists a way to calculate it in order to write a requirement?

 

Thank you for responding,

Best regards,

Julien A.

 

 

  • 0
    Prodigy 145 points

    Hello experts,


    No one wants to respond to my post ?

    Best regards,

    Julien

  • 0 in reply to Julien Appaix
    TI__Mastermind 28001 points

    Julien,

    This question sounds like something more specific to the application than our TI products, but my opinion is a s follows:

    Because random (RMS) jitter and BER are related and both are statistical values, it would seem that margin would not be necessary.  Nevertheless, I think that there is a tendency for customers to drive harder requirements than necessary.  For instance, I have had customers push the lower SONET 12 kHz integration limit for jitter for 10Gig Ethernet "because it's hard".  The reasoning is if we can meet the harder requirement, the we can also meet hte easier requirement.   Also, if there is any uncertainty then adding margin is one way to feel more confdent.  

    I think in terms of random jitter, but I have heard of datasheets where vendors requre that the input clock have some maximum peak to peak jitter.  But I struggle with this.  For instance, if my RMS jitter is 100 fs, can I meet a peak to peak jitter of 3000 fs?   Theoretically, if this is really a gaussian distribution, one can find the jitter arbitrarly large given a long enough observation time, although that time might be unrealitically long.

    I wish I had better answers to your questions, but I do think that they might be answered in a different way, depending on application.  

    Regards,

    Dean

     

  • 0 in reply to Dean Banerjee
    Prodigy 145 points

    Hello Dean,

    Thank for responding. What I have understood is "the more margin i (could) have, the more confident I am", and it sounds good to me.

    However I don't have understood what is quoted below:

    Dean Banerjee said:
    I think in terms of random jitter, but I have heard of datasheets where vendors requre that the input clock have some maximum peak to peak jitter.  But I struggle with this.  For instance, if my RMS jitter is 100 fs, can I meet a peak to peak jitter of 3000 fs?   Theoretically, if this is really a gaussian distribution, one can find the jitter arbitrarly large given a long enough observation time, although that time might be unrealitically long.

    Could you reformulate this please ?

    Thanks,

    Julien A.

  • 0 in reply to Julien Appaix
    TI__Mastermind 28001 points

    Julien,

    "the more margin i (could) have, the more confident I am"  sounds like a good statement for RMS jitter as this is a random event.  For instance, supposed my BER requirment was 1E-8 and I knew that a jitter of 1 ps corresponded exacthis corresponded theoretically to exactly a BER of 0.9999E-8.   Theoretically, one would say that this meets the requirement.   If one was to observe this infinitely long, you might see th BER of 0.9999E-8, but if you were to observe for a shorter time, it is possible that you would see a BER of slightly higher than 1E-8.  Just a shot in the dark,but but maybe this might have something to do with the marginal jitter, but not sure.

    Reformulating what I said before, take this example.  Suppose I have 1 ps of rms jitter.  This means that if was to compare the rising edge of the actual clock to ideally what it should have been, this error would theoretically be a gaussian distribution with standard deviation of 1 ps an mean of 0 ps.   68% of the time I would see an error betwen -0.68 and +0.68 ps, 95% of the time I would see an error between -2 and +2 ps and 99% of the time I would see an error between -3 and +3 ps.   Although the occurrence of an event might be astronomically small, it would be theoretically possible for one cycle of the clock to be off by 100 ps.  So if someone was to require this signal to never be off by more than 10 ps, I would claim that it would not meet this requirement.  However, if one was to say that the rising edge of this signal was to be off by no more than 3 ps for 90% of the time, I would say that this signal would meet the requirement.     The assumption that the random jitter has a Gaussian distribution might not be perfectly true, but I think it is pretty close to reality.  Perhaps if one was to take a scope and measure the jitter for 10000 clock cycles and do a histogram, this might give more confidence, but for the kinds of jitter than my group deals with (<< 1 ps), I have never seen this measured accurately with an oscilloscope.   Nevertheless, if it turned out that this distribution was not Gaussian,  then if the RMS jitter was 1 ps, then it could be more (or less) probably than 1 in 1000 that the difference beween the ideal rising edge of the clock and the measured edge of the clock would be more than 3 ps.

    I am not familiar with "marginal jitter", but maybe it has something to do with challenging the assumptions that jitter is a perfectly gaussian distributed random variable.

    Regards,

    Dean