# What separates a "good" eye diagram from a "bad" one?

I'm running some USB verification testing at work, and the Agilent oscilloscope I'm working with returns a nice summary of pass/fail statistics along with a pretty eye diagram. Since the pass/fail is indicated within the scope, I don't need to do a ton of analysis on these diagrams.

I've looked at quite a few of these in the past few days and it's made me curious:In general, what separates a"good" eye diagrams from a"bad" one? In a lot of the tests I've run, the device failed, but the eye diagram looked very similar to one that passed.

I can understand a diagram where there are blatant crossings through the eye, but what other factors are taken into account when looking at these diagrams?

• Is that a diagram for a link that (sometimes?) fails? Jun 17, 2013 at 22:56
• Good question. I believe this was a passing diagram. I'll try to dig up a "fail". Jun 17, 2013 at 22:57
• LOL, I just read this in the ON Semi document bout interpreting Eye Diagrams: "Using the persistence mode of the oscilloscope, the superstition of millions of time−domain waveforms can be displayed." Break that scope and it's 7 years bad luck!
– Kaz
Jun 18, 2013 at 1:11
• This diagram will tell you very little other than data is 12Mb per second. Try triggering the scope where you have it but delaying the waveform a 100 (or more) bits equivalent - then take a look at it. It will give you an idea how much the edge syncs are moving about. Jun 18, 2013 at 7:24

The eye you show is firmly in the category of "good" eyes. There's nothing there that should cause an error any more often than once every 100,000 years or so.

Some things to look for:

• Are you sending real data while measuring the eye? An idle pattern or similar behavior could create an eye that looks much cleaner than the eye would be with real data.

• Is there any external interference that could be causing intermittent degredation in the eye (that wasn't captured in this case)?

The eye on the right here shows ISI with a shape that's characteristic of $\sqrt{f}$ attenuation (which typically comes from skin effect loss).