Sometimes, the first mark bit does not always begin with at the start of the I/Q capture. There could be a number of I/Q samples recorded before the first time the mark bit is set to 1. In the example below, there are 5 frames before the first timestamp, which doesn't start until the sixth frame.
The first column of the IQ vector contains Q and the second column contains I.
N Frame Data (one I, one Q sample per frame)
0 [ Q-------0I-------T ]
1 [ Q-------0I-------T ]
2 [ Q-------0I-------T ]
3 [ Q-------0I-------T ]
4 [ Q-------0I-------T ]
5 [ Q-------0I-------T ] <- The first mark bit is here. This is where you start to build the first timestamp; this 'T' is the MSB of the timestamp
6 [ Q-------0I-------T ] <- 'T' is MSB - 1 bit of the timestamp
7 [ Q-------0I-------T ] <- 'T' is MSB - 2 bit of the timestamp
8 [ Q-------0I-------T ] <- etc.
...
To get the timestamp for frames N=0 though N=4, you must extrapolate the timestamp from frame 5 backwards. To get the timestamps for frames 6-68, you must extrapolate the timestamp forwards. The time between each frame is equal to (1/Output Data Rate)x(Number of I or Q samples per frame).
I/Q Bit Resolution
Time Between Each Frame
24
1/(Output Data Rate)
16
2/(Output Data Rate)
10
3/(Output Data Rate)
8
4/(Output Data Rate)
Once the 64 bits of timestamp is put together, you get a number that looks like:
[ S-------T-----0000]
Where 'S-------' is 32 bits specifying the timestamp in seconds since 1970 (time_t), 'T-----' is 28 bits specifying the offset from that second (in clock ticks at 114.375MHz), and '0000' are 4 unused bits.
Note
There could be some frames at the very end of the capture that have an incomplete timestamp because the capture stops before there is a complete group of 64 frames to make an extended frame . In that case you could extrapolate from the previous timestamp.