precise timing of the read in lag

[Enrico Segre]

Hi, I still have one question connected with my previous posts.
For the application I have in mind, I need to analyse online some data, and to provide trigger output depending on the result. It is not paramont that the online analysis is fast, but it is that the timing is known.
The principle of the ringbuffer and the variable data lag involved is clear to me. Having understood that the data seen by the computer is variably delayed with respect to reality, I try to measure on the fly this delay as precisely as possible. If this would be known, I could program the trigger to be sent in the [computer] future, after a suitable wait time, which is adjusted by subtracting the current lag. I have already seen that with some hackery (highest priority in the proper threads, clock timer checked frequently, wait VIs dinamically adjusted, etc.) I can achieve timings of the trigger with errors down to just a few msec. However, I still fail by several tens of msec in estimating the time lag. The formula I'm using is:

lag=[(buffer_pointer - 4* display_index) modulo (4*buffer_size_in_I32)]/2^21 + t2-t1,

where 2^21 stands for 2Msamples/sec, t1 is the timer value read by a clock located in a sequence box which encloses READ_POINTER (big orange while loop in state 4), and t2 is the timer value at the very end of my calculation, just before computing the final delay before the trigger. Is this correct? Is there some variable contribution to the lag (e.g., USB data transfer) which I'm overlooking?

Thanks a lot in advance, Enrico

[Coen]

There are indeed further time lags in the USB link. First, the transfer between Receiver and PC ringbuffer occurs in bursts of 64 kByte. This leads to a timing jitter of approx. 30 msec. Second, the receiver contains an additional FIFO buffer to handle interrupt latencies. The FIFO handles hick-ups in the USB transfer up to tens of msec that occur occasionally (caused by the Windows OS).

Best regards, Coen (BioSemi)

[Enrico Segre]

hmmm, interrupt latencies. This explains it. The filling of the receiver FIFO is deterministic, the lag between last processed and most recent data in the ringbuffer also (it is variable, but directly dependent on the pointers), but the link between the two depends on an interrupt and breaks the determinism. Am I correct?
This suggests me a possible way out: if the interrupt routine would either report how much data is still in the receiver FIFO, or the exact timestamp of the transfer, it would be possible to compute this lag. I understand that Biosemi itself developed the receiver driver, hence it should not difficult to insert this feature into the code. The information could be stored in a location read by READ_POINTER and passed further to Labview. Does this sound realistic? As it is turning out that this is very important for our research program, I am very interested in any progress in this direction.
Enrico