WARP Project Forums - Wireless Open-Access Research Platform

Hello all:

Just a quick doubt (hopefully). In "wl_example_siso_ofdm_txrx.m" the impulse response of a certain halfband 2x interpolation filter is given:

%Define a halfband 2x interp filter response
interp_filt2 = zeros(1,43);
interp_filt2([1 3 5 7 9 11 13 15 17 19 21]) = [12 -32 72 -140 252 -422 682 -1086 1778 -3284 10364];
interp_filt2([23 25 27 29 31 33 35 37 39 41 43]) = interp_filt2(fliplr([1 3 5 7 9 11 13 15 17 19 21]));
interp_filt2(22) = 16384;
interp_filt2 = interp_filt2./max(abs(interp_filt2));

Our main problem is that the interpolation filter, not being ideal, gives rise to interpolation errors (linear distortions due to the attenuation of the upper frequencies in the original band? Plain aliasing, as in non-linear distortions due to insuffcient rejection of the image components outside the original band?)

Without interpolation, the (noiseless channel) simulated received constellations are mathematically perfect (all points converge perfectly to a symbol and overlap). With interpolation, the constellations are somewhat a noisy cloud (very close to the ideal symbol location). Unfortunately, this interpolation error affects very differently to either OFDM or polyphase implementations of other waveforms.

Should a longest filter response alleviate such problems? Could you provide additional details on this type of filters, please or alternatively, can anyone kindly help me get started by pointing me to an online resource with information on how to make such filters longer?

Any information you can provide me would be greatly appreciated.  :)