primozb
Newbie level 5
- Joined
- Feb 11, 2012
- Messages
- 10
- Helped
- 5
- Reputation
- 10
- Reaction score
- 4
- Trophy points
- 1,283
- Location
- EU
- Activity points
- 1,402
Hi
I am experimenting with capturing bursts from my mobile phone on uplink.
What I basically did is that I followed this paper where a concept of GSM passive receiver is described in this
document link.
In one sentence, the described concept is to locate start of normal burst by correlating normal burst training sequence data to a known training sequence (base station color number). Then using correlation peak to locate start of normal burst.
I did some experiments on captured IQ data from my mobile phone (a couple of meters from antenna).
The output of Viterbi detector looks like this:
Attached picture gsm_passive_receiver_file.png
As you can see the training sequence match perfectly (TRAIN_SEQ_NUM=1) for all captured bursts.
What bothers me is that starting 3-bits are never `000`.
I did experiments also with other captured data, the results were always the same: first starting 3-bits were never `000` (saw different combinations `111`, `001`, `110`, ..).
Pictures of captured normal burst IO, matched filtered and impulse response for detected bursts are here:
Attached pictures: burst_003.IQ.png, burst_003.mafi_filtered.png and burst_003.imp_response.png
My guess is that IQ data is too noisy to be detected without errors, but what confuses me here is that training sequence is always detected correctly. Only the start of burst is erroneous.
Any suggestions for what should I look at, are very welcome.
If you want to take a look at source code and captured data I used for experiments, it is located here source code link.
Any links to groups where such topics are discussed will be also very appreciated.
WBR
I am experimenting with capturing bursts from my mobile phone on uplink.
What I basically did is that I followed this paper where a concept of GSM passive receiver is described in this
document link.
In one sentence, the described concept is to locate start of normal burst by correlating normal burst training sequence data to a known training sequence (base station color number). Then using correlation peak to locate start of normal burst.
I did some experiments on captured IQ data from my mobile phone (a couple of meters from antenna).
The output of Viterbi detector looks like this:
Attached picture gsm_passive_receiver_file.png
As you can see the training sequence match perfectly (TRAIN_SEQ_NUM=1) for all captured bursts.
What bothers me is that starting 3-bits are never `000`.
I did experiments also with other captured data, the results were always the same: first starting 3-bits were never `000` (saw different combinations `111`, `001`, `110`, ..).
Pictures of captured normal burst IO, matched filtered and impulse response for detected bursts are here:
Attached pictures: burst_003.IQ.png, burst_003.mafi_filtered.png and burst_003.imp_response.png
My guess is that IQ data is too noisy to be detected without errors, but what confuses me here is that training sequence is always detected correctly. Only the start of burst is erroneous.
Any suggestions for what should I look at, are very welcome.
If you want to take a look at source code and captured data I used for experiments, it is located here source code link.
Any links to groups where such topics are discussed will be also very appreciated.
WBR
