What is wrong with my simulation?

[ruwan2]

Hi,
I run a simulation about equalization on Serdes link with Simulink. It is from a demo file. Although it is effective, I cannot make up the equalization parameters with the result.
It is about 2 G Bit Per Second Serdes. Below picture shows that the distorted signal (up) and equalized signal (down):



I get the distorted channel response picture as:


The above curve looks like match to the general FR4 board. Please note that it is a linear horizontal axis as the usual case for serdes channel.


The adopted equalizer is a lead-lag filter:
nominator: Tau1=25e-9
denominator: Tau2=10e-9
Its frequency response curve is:



I use bodeplot, whose default horizontal axis is in log. My question is that the high frequency compensation between 1G to 3G is rather small with the channel attenuation. This is far beyond of my expectation on equalizer high frequency boost. From 1G to 10G, the equalizer has at most 5.8 dB, comparing the -15 dB loss at 3G for the channel. The equalization looks like not enough, but the eye diagram above seems OK on the other hand.
Could you agree these figures? What is wrong with me?

Thanks,

 

[zorro]

Hi ruwan2,

The first point is that it seems that there are some missing decinal points in yopur post, i.e. in order to be consistent with the Bode plot (and the rest) it should be
Tau1=.25e-9
Tau2=.10e-9

In these conditions, the equalizer has a zeto at about f=0.64 GHz (nor Grad/s) and a pole at f=1.6 GHz.

From the eye pattern of the distorted signal it can be estimated a rise time (from 10% to 90% of about .6 ns).
Assuming that the channel could be modeled roughly as a single-pole transfer function, its time constant would be .6ns/2.2 = 0.27e-9 ns .
This means that the zero of your erqualizer cancels the pole of the channel, moving the dominant pole of the response to Tau2.

The spectrum of the NRZ signal has a sin(x)/x shape with the first pass by zero at 2 GHz.
That means that the equalizer acts in the frequency band when the spectral content of the signal is important. Above 1.6 GHz or so the attenuation of the channel can be very high but that doesn't affect significantly the signal.

Conclusion: it seems that all is OK.
Maybe the wrong is that you confused rad/s with Hz, or neglected the fact that attenuation at very high frequency doesn't care.

Regards

Z