EQUALIZATION - Advanced Signal Integrity for High-Speed Digital Design - page 517

data eye for the equalized 200-bit random pattern is shown in Figure 12-23.
As expected, the equalizer gives an open data eye with approximately twice the
signal amplitude and eye height than those of the data eye from the passive
equalizer in Figure 12-18.
A potential source of confusion that merits discussion is the use of power gain
versus that of voltage gain. In our transfer function plots, we use power gain.
Since power is a function of the square of the voltage (or current) signal,

10 log P out
P in

gain power ( dB ) = (12-19)

Conversely, amplifier manufacturer datasheets often specify gain in terms of the
voltage:

20 log V out
V in

gain voltage ( dB ) = (12-20)

The moral here is that when working with gain figures, be sure to know whether
you are dealing with voltage or power gains, as mixing the two can lead to
erroneous results.
Active CTLEs provide gain, so they must dissipate some power, although
careful design can keep the power to less than 10 mW. In addition, active CTLEs
can be designed using higher-order high-pass filters, but the performance gain
typically may not justify the additional power consumed depending on the channel
response of the specific application. The active CTLE has some fundamental
limitations, notably the limited bandwidth of the amplifier and phase mismatch
between the two amplifier paths. For example, Kudoh et al., [2003] show 3-dB

bandwidths of 3 GHz for a conventional feedback amplifier and 10 GHz for an
improved design. Finally, tailoring the frequency response and the gain of the
filter requires the inclusion of additional control circuitry, just as it did with the
passive equalizer.

−

12.3 DISCRETE LINEAR EQUALIZERS

In Section 12.2 we dealt with equalizers that were completely analog in nature.
Contemporary high-speed components such as microprocessors, graphics proces-
sors, or memory controllers are manufactured using processes that are highly
optimized for digital circuit applications. Whereas the equalization problem is
very analog in nature, the discrete linear equalizer makes use of a combination
of digital and analog techniques. This allows us to better utilize the economies
of scale provided by Moore's law, thus satisfying the performance demand at
minimum cost. As such, they find wider use in computing equipment than do
their analog counterparts.
The linear time-invariant nature of our signaling system has another aspect that
we have not yet covered that will give us flexibility in implementing equalizer
solutions. In particular, the response of an LTI system does not depend on the