On Aug 28, 5:02*am,
[email protected] (Steve Pope) wrote:
> julius *<
[email protected]> wrote:
> >On Aug 27, 4:54*pm, Eric Jacobsen <
[email protected]> wrote:
> >> Even if it were frequency selective ICI isn't a problem. * That's the
> >> benefit of "orthogonality" between subcarriers. * Frequency
> >> selectivity by itself doesn't disturb the orthogonality.
> >I think this is not quite correct. *The orthogonalization comes from
> >the FFT operator, which assumes periodic convolution. *So if the
> >channel length is larger than the cyclic prefix length, aka frequency
> >selectivity is too severe, then ICI will happen, too, since the effect
> >of convolving with the channel can no longer be modeled by periodic
> >convolution.
>
> Well, here's my thinking, which is orthogonal to either of the above.
>
> If the channel is frequency-selective, then you have the possible
> scenario of energy from stronger tones spilling over onto the weaker
> tones. *This is going to be more of a noticeable impairment than
> in the flat-fading case. *You still will get ICI in the latter
> case (from non-linearities or from different FFT timebases if you fail
> to correct for this), but you won't get disproportionate
> ICI on weaker tones.
>
> Steve
That still wraps back to the adequate CP-length argument. If there's
so large a variation in the channel from one tone to the other as to
cause disproportionate increase in interference on a neighbouring
tone, the sudden channel variation in frequency would result in the
impulse response spreading out over a large time duration exceeding
the designed CP. There's a limit to how much the CP can be increased,
and that depends on the symbol duration, which in turn depends on the
carrier spacing. A smaller carrier spacing produces an intermediate
carrier which is not as disproportionately affected by a sudden
channel fade, and allows for a larger CP.