|
From: | George Nychis |
Subject: | Re: [Discuss-gnuradio] inband timestamp issues |
Date: | Tue, 19 Aug 2008 09:49:55 -0700 |
User-agent: | Thunderbird 2.0.0.16 (X11/20080724) |
George, The timestamp semantics that we are using on the USRP2 is that for Rx, the timestamp refers to the time that we clock the sample out of the DSP Rx pipeline (n.b., not the Rx FIFO). For transmit, it's the time we clock the sample into the TX pipeline (not the Tx FIFO). In both cases, the reference point is at the edge of the DSP pipeline. This seems to be the only place that makes sense, since the group delay through the pipeline varies depending on the configuration. In all cases, the timestamp is in ticks of the master clock.
This sounds correct. I think the major issue is just that the current implementation is taking the time the packet is built as the timestamp. This is far from correct. Because of this, we don't have a timestamp that represents the first or last sample, and if the packet builder is halted then we get these timestamp "jumps."
- George
[Prev in Thread] | Current Thread | [Next in Thread] |