Hi everyone,
while trying to understand what exactly the above frequency selective model does (which I now mostly do, after reading the relevant papers and some recent threads here), I started getting some doubts about the Rice channel implementation.
In short: the K factor is defined as the power ratio of the LOS component (i.e. first tap) to the NLOS components (rest of the taps). This means that in the case of a very large K factor, the NLOS paths become negligible and the channel converges towards a flat channel. This should hold regardless of the doppler frequency. Please correct me if I'm wrong here or anywhere.
However, the current implementation does not show this behaviour -- setting high K values seems to only have very minor effect on the channel's frequency response. I'm aware of the changes made in the 'next' branch - they do not fix the problem (I backported the changes locally into maint-3.7 since next is pretty broken for me at the moment and I didn't feel like troubleshooting that...).
Test flowgraph: Fast noise source -> Freq. Sel. Fading Model -> Qt freq. or fosphor sink.
The current implementation uses the K factor to compute the first tap but ignores it for the rest of the taps. A quick, ugly and probably wrong solution is to multiply the taps returned by flat_fader_impl::next_samples with the respective LOS/NLOS scaling factors. This actually works (in the sense of giving a flat channel for large values of K) but I didn't try to validate this any further. I suspect it might be the wrong approach since it uses the K factor twice (once for computing first tap, then for weighting).
PS: by 'taps' I mean the raw taps returned from all the flat faders, before interpolation.
Any thoughts on this? Am I completely off? :)
Cheers,
Tal