Equivalent time sampling oscilloscopes have greater bandwidth and lower noise than real-time oscilloscopes that cost more than twice as much. For eye diagrams, mask testing, jitter analysis, time domain reflectometry (TDR), and Q-factor, equivalent time (ET) is the way to go because of the lower noise floor. Real-time (RT) is the way to go when you don’t know much about the signal, or when the signal isn’t a repeating test pattern, or you don’t have access to a clock for triggering, or you want to implement a software processing model on the waveform, or... when you want to exercise your genius.
Where jitter analysis was the Great Problem that assured the job security of enterprising signal integrity (SI) acolytes a decade ago, and as equalization, pre/de-emphasis, and de-embedding have been the SI hot topics for the last five years, I think that crosstalk analysis in high-speed differential systems is the next “killer problem” in high-speed design.
Picture ten differential channels in parallel on PCB, each at 10Gb/s -- one of the prescriptions for 100 gigabit Ethernet. Let's assume that, other than crosstalk, the signal is perfect: no jitter, no inter-symbol interference, and no other electromagnetic interference (EMI). Don’t worry about skew (the variation of channel trace lengths) because it’s shuffled higher in the protocol stack -- worry about crosstalk.
Now picture the signal of one channel, the victim, on an oscilloscope.
When an aggressor makes a logic transition, the changing signal blasts electromagnetic radiation through the dielectric medium. With nine aggressors, it will be difficult to distinguish individual jolts of crosstalk noise in the scope trace. Even if there were just one aggressor, it would still be hard to discriminate crosstalk from other noise unless it was really bad.
With this image in mind, let’s resolve how crosstalk at these rates, with screaming rise/fall times, looks on a real-time scope. We’ll do equivalent time scopes in another post.
Trigger the RT scope on one of the ten channels, call it the victim. Accumulate and store a record. Now set the time base to display two or three full bit periods and scan through the waveform. We’re looking for jolts of noise that repeat in integral multiples of the bit period. If the rise/fall times of the signals are 30ps, then the jolts of noise will last 30ps. We won’t see identical jolts every period because the aggressors don’t make transitions every period. Most high-speed serial signals have 50 percent transition density, so identical jolts occur an average of every other bit for every aggressor in the system. A garden variety victim bit period will exhibit crosstalk from 4 to 5 disturbers.
The thing I like most about real-time scopes, as compared to equivalent time scopes, is that you can extract the data, take it offline, and analyze it to your heart’s content.
How would you identify and evaluate each of the nine aggressors?
Assume that neither the victim nor the aggressors experience any other signal degradation, but that the scope does have a noise floor and exhibits a bit of ringing. Do not assume that the ten signals are frequency locked; that is, they operate at the same nominal frequency, but each has an independent clock, so they don't have fixed phase relationships.
You can use two channels on your hypothetical RT scope, or use whatever ideas come to mind. Knock yourself out.
Tweet This
