uhd rx_time tag handling broken

[mattes-bru]

I have following use case: I want to detect a signal and start the recoding of it dynamically. For that I use a USRP source with two channels (on for the detector and another for the recording path) The recording is done by sigmf sink which is controlled by it's message input. As the detector channel of the USRP is changing I get tags with the rx_time on both channels.

With that use case I experienced two issues:

1. The sigmf sink adds a new capture with every tag
2. the values for core:datetime and core:sample_start do not match

I checked the USRP source with a custom python block: the offset between two timestamps and their positions does match. second_ts == first_ts + (second_pos - first_pos)/samp_rate

[storborg]

Hi @mattes-bru , thanks for the report. We've tried to replicate what you're describing but we're a little unclear on exactly what is happening. I have a couple of questions:

• Are you experiencing any underruns in the sample stream? If so, that would cause exactly the behavior that you are describing, and it is expected to, because the USRP Source emits a sample stream which has gaps in time.
• When you say "one for the detector and another for the recording path", do you have a detector that is implemented in RFNoC? Or is the detector producing an analog signal externally to the USRP? If so, there could be a bug related to RFNoC-specific streamers which causes incorrect time handling.
• Lastly, if you could describe your specific hardware and software stack setup (USRP type, UHD version, transport type, and host platform), that would be great.

[mattes-bru]

Hi @storborg, thanks for looking at this.

To answer your questions:

• I am using a Ettus X310 with a TwinRX, a USRP Source with two streams. The source block is controlled by the message input, where I change the center frequency of the first channel. This results in tags on both streams. The SigMF sink, which is connected to the second channel of the USRP source. For completeness: Recent releases of GURadio and UHD (not master) installed through pybombs on Ubuntu 18.10
• I did not get any underruns in the stream, although there are tags in it due to the command messages. But the timestamps are at the expected positions for a gapless stream.

I also had a deeper look at the issue and I found, that:

1. the sigMF sink does not differentiate between a uhd time tag which is gapless and one with a gap. Currently I try to find a way to extend your code to handle that.
2. There is a offset or error produced by the double to string conversion of the sub second value. I could avoid that by using a fixed nanosecond precision. But there was also a bigger error at the very beginning of the recording, which I'll need to track down.

I hope this makes it clearer for you, what my issues are. I wanted to set up a test flow graph but it seems I'll need some more time for that.