It’s Not Jitter, It’s Noise

-December 28, 2015

An oscilloscope draws a two-dimensional view of a signal -- voltage versus time. In this post, let’s ponder noise in these two dimensions.

When we look at a digital signal on a oscilloscope, the only noise we care about is noise that causes errors. In high-data-rate signals, the vast majority of errors occur when the timing of a logic transition fluctuates across the sampling point. That sampling point should be close to the center of the eye. For a differential signal, it’s right around (t, V) = (n+½T, 0), where n is an integer, and T is the bit period.

If transitions had zero rise/fall time, jitter caused by phase noise would be the sole culprit. But voltage noise shifts edge timing, too.

Consider the rising edge of a 0-to-1 transition. If, for example, voltage noise pushes the edge down, then the time position at V=0 shifts to the right. If the rise time and voltage noise are large enough, the edge shifts past the sampling point, and what should have been a 1 is mistaken for a 0.

In high-speed serial systems, the channel is the dominant cause of eye closure. The channel acts like an attenuating low-pass filter. Rise/fall times increase as higher-frequency components are damped out. The effect is called intersymbol interference (ISI), because the shape of a particular bit in the waveform is affected by the frequency content of bits that surround it. For example, a sequence of alternating ones and zeros has higher-frequency content then a long string of identical bits.

The combination of more gentle edges and reduced peak-to-peak voltage causes errors in two ways. First, the gentle-sloping edge of a transition might pass right by the sampling point. Second, a transition may never be higher or lower than the sampling-point voltage. The amplitude of a 1-to-0 transition that follows a long string of identical bits may not drop below the sampling point through the entire bit period.

A more interesting point (and, as 100-gigabit Ethernet emerges, an increasingly relevant one) is crosstalk-induced jitter. Crosstalk is electromagnetic interference (EMI) generated by logic transitions of one signal, the aggressor, and picked up on the trace of another, the victim. The magnitude of the pickup depends on the victim-aggressor mutual inductance.

Both ISI and crosstalk cause jitter, but they are voltage noise, not timing noise.

Of course, equalization was invented to defeat ISI. It’s easy to amplify the high-frequency signal components that ISI damps out. You can do it at the transmitter with de-emphasis or at the receiver with a continuous time linear equalizer (CTLE), in either analog or digital form, or with a linear feed forward equalizer (FFE).

It’s worse for crosstalk, though. Simple equalization techniques can open eyes closed by ISI, but to a receiver, crosstalk looks like randomly occurring blasts of noise. No simple, linear equalization technique can reduce the effects of crosstalk. Depending on the relationship between the underlying clocks of the victim and the aggressor, decision feedback equalizers (DFEs) can help.

In both cases, to understand how both voltage and timing noise affect your system, don’t overlook noise analysis in your headlong rush to jitter analysis. Most advanced oscilloscope jitter analysis software includes noise analysis. Another way is to expand the two-dimensional V(t) analysis of an oscilloscope to three dimensions with a bit error rate (BER) tester or a BERTscope and look at BER(t, V), also known as the BER contour.

Loading comments...

Write a Comment

To comment please Log In