[srslte-users] Performance difference in ./cell_search release1.4 vs. release 2.0

[David Rupprecht]

Dear all,

I noticed that the cell_search of release 1.4 performs much faster than
the cell_search of release 2.0. But I can not spot much differences in
the code.

cell_search v1.4 needs with the time command about *2 minutes* to scan
band 3:

60.61user 9.16system 2:15.38elapsed 51%CPU (0avgtext+0avgdata
27232maxresident)k
0inputs+152outputs (0major+14717minor)pagefaults 0swaps

cell_search v2.0 needs with the time command about *13 minutes* to scan
band 3:

83.70user 31.77system 13:12.29elapsed 14%CPU (0avgtext+0avgdata
26084maxresident)k
0inputs+152outputs (0major+1932minor)pagefaults

Furthermore, cell_search v2.0 always gets UHD overflows (O):

O[726/749]: EARFCN 1926 Freq. 1877.60 MHz looking for PSS.
O[727/749]: EARFCN 1927 Freq. 1877.70 MHz looking for PSS.
O[728/749]: EARFCN 1928 Freq. 1877.80 MHz looking for PSS.


The only difference I can spot is that cell_search v2.0 (to be exact the
libraries) are linked static with the binary. However, cell_search v1.4
uses shared linking. Thus I changed the linking behavior to shared in
v2.0 as well, but the results remain the same.

The constants to detect a cell remain also the same over the versions.

#define SRSLTE_DEFAULT_MAX_FRAMES_PBCH      500
#define SRSLTE_DEFAULT_MAX_FRAMES_PSS       10
#define SRSLTE_DEFAULT_NOF_VALID_PSS_FRAMES 10

Do you have any suggestions, what the problem might be? I was testing
this setup on a i7-6700HQ CPU with a standard ubuntu.

Best regards,
David

-- 
M.Sc. David Rupprecht

Ruhr-University Bochum
Research Group Information Security
Universitätsstraße 150
ID 2/130
44780 Bochum / Germany

Phone: +49 234 / 32 - 23508
Web: www.infsec.rub.de