mirror
/
dji_droneid
kopia lustrzana https://github.com/proto17/dji_droneid
1
0
Forkuj 0
Kod Zgłoszenia Projekty Wydania Wiki Aktywność

Update README.md

main
proto17 2022-04-30 12:27:46 -04:00 zatwierdzone przez GitHub
rodzic f312a1cc14
commit ca5fada2d0
Nie znaleziono w bazie danych klucza dla tego podpisu
ID klucza GPG: 4AEE18F83AFDEB23
1 zmienionych plików z 2 dodań i 2 usunięć
Pokaż statystyki Ściągnij plik aktualizacji Ściągnij plik porównania Expand all files Collapse all files

4
README.md
Wyświetl plik

@ -100,7 +100,7 @@ Since there's no change to the order of the symbols, and the 9 OFDM symbol case'
# Current Questions
## Cross Correlation
MATLAB can do a normalized cross correlation using `xcorr(X, Y, 0, 'normalized')` but it's *crazy* slow. So I wrote my own function ![normalized_xcorr.m](matlab/updated_scripts/normalized_xcorr_fast.m) that is ~ 8x faster than `xcorr` but over 100x slower than the `filter` function that was being used. If anyone has a better idea on how to do a truly normalized (0.0 - 1.0) cross correlation please let me know.
MATLAB can do a normalized cross correlation using `xcorr(X, Y, 0, 'normalized')` but it's *crazy* slow. So I wrote my own function [normalized_xcorr.m](matlab/updated_scripts/normalized_xcorr_fast.m) that is ~ 8x faster than `xcorr` but over 100x slower than the `filter` function that was being used. If anyone has a better idea on how to do a truly normalized (0.0 - 1.0) cross correlation please let me know.
## Burst Extraction
Right now the normalized cross correlation mentioned above is used to find the bursts using ![process_file.m](matlab/updated_scripts/process_file.m). Since the cross correlation is so slow, a file that contains tens of millions of samples takes a long time in MATLAB and an even longer time in Octave. The old method of using the `filter` function was blazing fast, but there was no way to know what the correlation thresholds needed to be without making multiple passes through the file. Energy detection would probably work, but that falls apart in low SNR conditions. A normalized autocorrelation would likely be as slow as the cross correlation, which probably rules out autocorrelating for the ZC sequence. I'd love to hear some ideas to help speed this process up.
Right now the normalized cross correlation mentioned above is used to find the bursts using [process_file.m](matlab/updated_scripts/process_file.m). Since the cross correlation is so slow, a file that contains tens of millions of samples takes a long time in MATLAB and an even longer time in Octave. The old method of using the `filter` function was blazing fast, but there was no way to know what the correlation thresholds needed to be without making multiple passes through the file. Energy detection would probably work, but that falls apart in low SNR conditions. A normalized autocorrelation would likely be as slow as the cross correlation, which probably rules out autocorrelating for the ZC sequence. I'd love to hear some ideas to help speed this process up.