The general symbol-level representation for a frequency-selective channel is:
$$ y[k] = \sum_{l=0}^{L-1} s[k-l]h_{l} + z[k] $$
where \$s[k-l]\$ is transmitted symbols, \$z[k]\$ is noise, \$L\$ is the channel length. Here \$h_l\$ only have values when \$l\$ is nonnegative, it's reasonable since the channel is causal (\$y[k]\$ depends only on \$a[k]\$, \$a[k-1]\$,...), but when I read Tse's book, it shows the sample-level channel representation:
$$ h_m = \sum_{i} a_i sinc[m-\tau_i W] $$
where \$a_i, \tau_i\$ means the strength and delay at i th path, \$W\$ means the bandwidth. It's clear that \$h_m\$ can have values in negative parts, so I wonder what's the relationship between the \$h_l\$ in the first equation and \$h_m\$ in the second equation? (i.e. how can we deal with the \$h_m\$ when \$m < 0\$ in the second equation?)
