Performing eye diagram measurements on closed eyes

In a previous blog post, I mentioned the need for de-embedding the channel from your measurements. In this post, I will discuss silicon techniques used to compensate for channel degradation and ways you can characterize these systems.

Transmitter equalization

As a single data bit makes its way across the channel, two main effects occur: rise/fall times increase and amplitude gets smaller (bandwidth) and the pulse width gets bigger (dispersion). To compensate for the bandwidth degradation, the transmitter can “help” the signal along by pre-emphasizing or boosting its amplitude if there is an upcoming transition bit. This is also known as de-emphasis if the transition bit is at normal amplitude and the rest are attenuated. Signal characterization can be as simple as:

• characterizing the transition and non-transition bits individually since they have different SI characteristics and masks
• having the scope software add pre/de-emphasis if you are not able to program the Tx directly.

Receiver equalization

Dispersion in a 101 pattern can cause the “0” bit to look like a “1” or cause the last bit in a 10001 sequence to look like a “0.” This effect is referred to as ISI, or Inter Symbol Interference. In the same way that Transmitter Equalization is used to “help” the signal transition, Receiver Equalization like DFE (Decision Feedback Equalization) uses information from prior bits and known ISI to boost the signal before the Rx comparator. SI characterization of what happens in the Rx after equalization consists of the following:

• If the silicon vendor provides its “secret sauce” equalization information, you should plug this information (equalization type, number of taps, tap weights, etc.) into the scope’s equalization software.
• If the equalization information is not available, use the equalization coefficients that are called out in the standards document. Note that the standards documents mention minimum requirements, and the silicon implementation should at least meet these to be compliant.
• Finally, you can use the scope software auto-adapt function to determine the tap weights. Most scopes employ an LMS (Least Mean Square) adaptive algorithm that searches for the minimum eye closure in the middle of the eye opening.

Tx and Rx equalization has always been around, even if we are sometimes unaware that it is being implemented in the silicon. The reason many of us in SI are forced to deal with it now is that the Tx signals exhibit higher dB levels of pre/de-emphasis transitions and the inputs to the Rx show closed eyes that cannot be characterized unless equalization in the scope is used.