In a convolution operation, why the initial output bits shift down?

My speculations:

1. The reason is the negative-going dip of the ir, which is produced by the compensation. Using uncompensated ir, there will be no down-shifting on the output bits stream (see the fig on the bottom below).

 

2. The mechanism of producing such shifting.

It's basically a DC level shift. When a channel is hit by the first few bits, it is actually a process of charging the parasitic cap of the channel. Note there are two caps on a channel:

a. The shunt parasitic cap.

When this cap dominates, the output waveform will show up-shifting stream.

b. The SERIES cap produced by EQ.

This might be hard to understand since there is no coupling cap in the extracted channel. Well, the forming of this cap is due to the equalization, whether it is generated by FIR or CTLE. Aka, it is an effectual cap rather than a real one.

PS: From above two plots we can also see how the shunt cap produces ISI, and how the ISI suppresses the amplitude, meanwhile how the EQ mitigates the ISI effect.