EQUALIZATION - Advanced Signal Integrity for High-Speed Digital Design - page 528

36
230
97
37
18

=

x in , adj mV

We then construct the input sample matrix and calculate the equalizer coeffi-
cients:

230 36 0 0 0
97 230 36 0 0
37 97 230 36 0
18 37 97 230 36
08770

x = mV

− 0 . 77

4 . 93

x − 1 y target

−

c = = 1 . 98

0 . 14

− 0 . 13

We apply the maximum signal swing constraint by using equation (12-22) to
adjust the equalizer coefficients:

− 0 . 097

0 . 624

c i
|

c ZFS i =

c i | =

− 0 . 250

0 . 017

− 0 . 016

Figure 12-31 shows the resulting equalized pulse response, which demonstrates
the effect of the zero forcing equalizer. Since the width of the nonequalized
pulse in Figure 12-30 is much wider than a single bit (which is 100 ps wide),
significant ISI is expected. The equalized pulse response is much more narrow and
the ISI has been removed (i.e., forced to be zero) at the equalizer sample points.
However, the figure also reveals the fact that the ZFS approach eliminates the ISI
only at the sampling points that correspond to the equalizer taps. The equalized
pulse shows ISI in the intervals between the sample points and at sample points
outside the equalizer.
The simulated output from the ZFS equalizer in Figure 12-32 shows an eye
opening of approximately 110 mV and 85 ps for a 300-bit pseudorandom bit
sequence. In addition, the worst-case data eye calculated from peak distortion
analysis on the equalized pulse response is 107 mV high and 80 ps wide. As we
can see, the zero forcing equalizer provides significant improvement in this case
over that of the nonequalized channel.