Signal Integrity

Power-aware Signal Integrity and EMI/EMC On High-speed Digital Chip-to-Chip Links

Signal Integrity header image 2

For Decision Feedback Equalizers, Beauty is in the Eye

Posted March 28th, 2012 · 2 Comments · Video


By Colin Warwick

“Beauty is in the eye of the beholder” is a phrase from the 1878 novel Molly Bawn by Margaret Wolfe Hungerford. It means that no matter what the intrinsic properties of a thing, it’s how you look at it that matters.

In our case, the “eye” we’re taking about isn’t the beholder’s eye but the eye pattern diagram of a chip-to-chip SerDes link that uses a decision feedback equalizer or DFE. People often wonder about it because has ugly-looking discontinuities in it. They even think something’s wrong with it. In fact this is how an eye pattern diagram should look with a DFE and in this video we’ll explore how and why. After watching the video, you’ll see the discontinuities are not ugly but are in fact beautiful.

decision feedback equalizer tutorial video

If you’re interested in the mathematical detail, please see our white paper Equalization: The Correction and Analysis of Degraded Signals

Tags: ··

2 Comments so far ↓

  • Joe S

    Real DFE shifts the threshold, not the waveform; and samples at (or near) the mid point of the UI. The discontinuities arise because to show an eye with the effect of DFE it is necessary to keep a constant vertical screen position (ie, constant threshold) and shift the waveform, instead.
    We faced this same problem, and got the same complaints. We decided that no real electronics responds discontinuously, and low pass filtered those discontinuities to match the bandwidth with which the signal was acquired. They are still very visible, but they don’t get the “that can’t be real” reaction. In fact, instantaneous jumps can’t be real. In a real DFE implementation, that samples only once per UI, it doesn’t it doesn’t have to be instantaneous

  • Colin Warwick

    Hi Joe,

    Thanks for your comments.

    The jumps in my simulation weren’t instantaneous (20ps rise/fall), but I agree the bandwidth I used in this illustration is bigger than on most electronics implementations and on most ‘scopes (unless you use an 90000X series with the highest true analog bandwidth 🙂 ). 20 ps time steps corresponds to f_Nyquist of 25 GHz. If we were to add a simulation of the finite bandwidth of the scope or the electronics, the discontinuities would indeed be smoother.

    I suppose one could envision two-part eye pattern diagrams and waveforms that display the absolute waveform and absolute threshold value separately, but I think they would be harder to interpret than the composite ones here (which are voltage relative to present threshold). They shows how much the eye/waveform is opening up as a result of the DFE. You can think of them as the difference voltage going into a differential slicer/comparator (receiver input minus threshold).

    — Colin

Leave a Comment