forked from JiaoXianjun/multi-rtl-sdr-calibration
-
Notifications
You must be signed in to change notification settings - Fork 0
/
TODO
35 lines (26 loc) · 2.07 KB
/
TODO
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
TODO:
-1: Make/test it in Windows? Then we would have more potential developers?
0: (DONE. solution is using rtl_tcp directly. Not udp C program relay any more.)Test packets/signal loss. (Perhaps this can be caused by not-in-time read from rtl-sdr driver).
1: (DONE. rtl_tcp supports it natively.)Make the relay to not only localhost but also other connected computer via network?
2: (DONE. rtl_tcp supports it natively.)Would Async mode rtl-sdr read be better than Sync mode?
3: (DONE. No need now because using rtl_tcp now.) Make bi-directional UDP link to let C program accept Matlab's control?
4: In-fly multi-dongles calibration algorithm. (Which signal is good candidate for this? GSM? 1090MHz ADS-B?)
(Now actually synchronization of single dongle bas been completed. I am trying to demodulate information in SCH and BCCH. See gsm_sync_demod.m )
4.1:
continuous fft to locate time position of fcch, which has been done.
4.2:
Then optimize it for lower computation load by fft only in predicted position.
4.3:
Fft can estimate integer times of frequency offset. Then we compensate it firstly.
After that fractional frequency offset should be estimated by calculating phase step size between adjacent samples. Then compensate it also.
4.4:
Now we have frequency offset estimated and compensated. We should do sampling rate error estimation and compensation,
because in a communication receiver frequency offset mainly comes from tuner but sampling rate error mainly comes from a/d clock,
and they are not guaranteed heaving analytic relationship. We should estimate them separately and many times iteratively.
Because they are twisted for how signal "look like", and can be separated by iterative processing the signal.
Their is a Kalibrate-rtl project on github which did similar thing. But I think it only did with frequency error.
I should check if it also did with sampling rate error.
4.5:
Any other RF non-ideal factor should be compansated to improve accuracy? IQ imbalence? IQ skew?
4.6:
Optimization, optimization, optimization. C code? OpenCL? I have both experiences.