[USRP-users] Description of Timestamp & I/Q samples out of the radio & CHDR

Martin Braun martin.braun at ettus.com
Wed Dec 10 07:41:39 EST 2014


Hey Sebastian,

thanks for checking out RFNoC. I have answers inline.

On 12/10/2014 11:22 AM, seb fenn via USRP-users wrote:
> Hi
> 
> Can someone please refer me to some accurate documentation, or
> describe, the structure of (CHDR) data coming out of the radio and
> into the RFNoC network. In particular:

CHDR is described in the manual. You'll need to use your local build of
the UHD manual when working with RFNoC.

> a) What is the format of the timestamp

It counts ticks.

> b) Where is the timestamp is the packet

See the manual. Also, I recommend the CHDR dissectors for wireshark.

> c) How the I and Q samples are arranged in the 64  bit words d) Size
> and coding of the I/Q samples (e.g. 16 bit signed?)

CHDR does by itself not define the data type; the assumption is that the
the block knows what to do with the data.
For our typical operations, we use 16-bit signed (as indicated by the
"sc16" data description).

> e) What is a trailer (as indicated by bit 62 in the header)

CHDR does not have a trailer. Bit 62 is part of the packet type (see the
manual).

> I am coding up blocks to include in the FPGA fabric, not writing
> software.

Until we've finished all of our RFNoC goals, you'll probably need to do
both -- although for starters, adding an XML stub in
uhd/usrp/rfnoc/blocks/ will probably do.

> And is there more information on CHDR? I have gleaned some from
> simulations and the paper "Simplifying FPGA Design with a Novel
> Network on Chip Architecture". But could do with some more complete
> descriptions. For example I don't understand why the Packet Size
> field in Context Packets in various testbenches (e.g. noc_shell_tb.v)
> is set to 16. And there seems to be a three word acknowledgment of a
> write to the NoC Shell address space. What is the meaning of the data
> in these words, I have guessed some of it , but would like
> confirmation.

'context package' is actually a VITA-49 term. It still means the same
here, although in CHDR, we can know name all the packet types specifically.

> I have seen
> https://github.com/EttusResearch/uhd/wiki/RFNoC:--Specification , but
> more information would be welcomed. For example is there any guidance
> on the usage of the "Upstream Flow Control: Cycles per ACK" and
> "Downstream Flow Control: Buffer size" registers? Some of this
> information seems inaccurate, for example the " Upstream Flow
> Control: Packets per ACK" register seems to be tied low.
> 
> Thanks in advance. The RFNoC approach seems a great innovation but
> working out how to add my own logic is proving something of a slog at
> the moment and I'd welcome more documentation.

At this time, all we can offer is the entirety of our sources. Also, the
C++ API documentation is fairly complete, and also answers some of your
questions on nomenclature.

I guess what you really want is a step-by-step tutorial, and I agree it
would be nice to have all that. But doing that is would be a waste of
everyone's time until all APIs etc. have been finalized -- also, it
would slow down the active development, which is still going on at full
pace, be assured!

Cheers,
Martin






More information about the USRP-users mailing list