[802.3ae_Serial] More about stressed eye calibration

Thread Links	Date Links
Thread Prev	Thread Next	Thread Index	Date Prev	Date Next	Date Index

All -

Since returning from Santa Rosa, I have had a few more thoughts on calibrating a stressed eye for Rx testing.

1. Achieving vertical closure (ISI) with only amplitude interference will add more pulse shrinkage than we probably want. To counter this, I referenced a 5 GHz filter in my presentation Wednesday. I would rather not use 5 GHz filter, but it looks like something is required to emulate closure from channel dispersion. 7.5 GHz provides less than might be expected in a link, but may be acceptable. It is certainly more readily available.

2. We can calibrate the 2 sine terms (amplitude interference and phase jitter) while transmitting square wave patterns. If on a scope, the interferer amplitude can be observed on the top or base line, or its jitter can be observed at the waist of the crossing. Obviously the phase modulation term can only be observed as jitter at the waist of the crossing.

The peak to peak values of these sine terms can be determined by placing cursors on the two modes or peaks of their histograms.

Either sine term can also be calibrated using RF spectral analysis while transmitting square wave patterns. I have the appropriate equations somewhere, and this method is usually quite accurate at least for phase modulation. I believe spectral analysis is what Agilent currently uses for calibrating the phase modulation term.

For simplicity, jitter from the sine terms are both sufficiently high probability that their peak to peak values can be added to predict their combined peak to peak total. Convolution analysis is not necessary.

3. As a concern, I am becoming convinced that calibrating final eye height to sufficient accuracy with short patterns may not be possible. I expect there will always be residual variation in amplitude and phase response across the very wide frequency range, and long patterns will inevitably show measurably worse baseline wander and jitter. So what do we do? Measure an rms value and extrapolate? As shown by Agilent's bathtub curves, the BLW and jitter distributions are expectedly truncated Gaussians.

Of course, some such low probability jitter is expected in real systems, and so perhaps we should welcome it. The E/O and other sources will add more. Given this, what calibration levels for jitter should be target? Steve Joiner proposed that jitter from the combined sine terms could be set to W+3sigma in the absence of low probability jitter.

However, since some low probability jitter will inevitably exist, I propose...

Thanks,