Fragment issues with multiple buffers

[sjkelly]

With the highlevel_loopback.jl example, some devices with a packet interface such as the USRP fail to produce the expected result when using multiple buffers. This seems to be addressed by using a single large buffer.

cc @mbaz who reported.

[sjkelly]

Note that there is a SOAPY_SDR_MORE_FRAGMENTS flag that is set in USRP: https://github.com/pothosware/SoapyUHD/blob/26b48cc11d3072b03b21793d980887a2e00091c8/SoapyUHDDevice.cpp#L292

It may be good to check this as well and modify our read loop accordingly.

[mbaz]

This is the result of a test where three blocks are looped back, using highlevel_loopback.jl. Each block is 10 ms and 10,000 samples long; sampling frequency is 1 MHz. The first and last blocks are all zeros, while the middle one is a sine wave. The plot is zoomed into the sine wave block: Note that the sine wave lasts less than expected: it starts at sample 11,400 and ends at sample 19,200. Moreover, there is a glitch at sample 21,400, where the block should have ended.

If, instead of writeing three blocks, first one large block is built in Julia and then it is sent once to SoapySDR, then the sine wave has the expected duration, and there are no glitches.

Test code here: https://gist.github.com/mbaz/3f2edfcc5d33412f02dd145cd8b571a1

[sjkelly]

Note: the USRP drivers have been updated in Julia. This issue was specifically mentioned in the UHD patch notes. Hopefully this is now fixed

[mbaz]

Since the LFSR RF loopback is working, I think we can close this and re-open if the problem resurfaces.