Martin- Just an update. It works now. On a whim based on your comment below I decided to rebuild with a larger input buffer on the input port of the noc shell. This fixed it. Originally it was set to 11 which is what the noc shell template is. I looked at the Ettus duc and it had 12 set for this and it probably has more fifo depth internally than I have in my component. I tried 14 since that was the threshold before invoking a cascaded fifo arrangement. That did the trick and now it keeps up. Seems to be always the fix for flow control weirdness, just throw more fifos in....
Thanks Jeff -----Original Message----- From: USRP-users [mailto:[email protected]] On Behalf Of Martin Braun via USRP-users Sent: Tuesday, January 09, 2018 12:52 PM To: [email protected] Subject: Re: [USRP-users] RFNOC design question Hey Jeff, at first glance, it looks like you're doing it all right. What's the size of your input buffer on your custom DUC? Also, can you post a full log of the streamer setup? -- M On 01/09/2018 08:41 AM, Long, Jeffrey P. via USRP-users wrote: > Looking for some advice and help on RFNOC.. > > > > I am creating a custom RFNOC block that will roughly do what the > noc_block_duc does but instead uses a fractional polyphase interpolator. > I believe I have everything working correctly in terms of the tlast and > EOB and for single timed bursts it seems to work fine which I can verify > on the scope and in chipscope. However when I switch to a streaming data > I start getting UUs. If I look at it in chipscope it appears that the > incoming data is just not being supplied fast enough. My block is not > pushing back. In terms of rates I am using a 1 gig Ethernet connection > and am running 32 times interpolation in my block with the radio block > being set to 200 MHz. This would imply that I have a 6.25Msps required > feed rate to my block. I think this should be well within the 1 gig > capability. Again I verified that my block is not pushing back so it > really appears the data is not coming down fast enough. My rfnoc graph > consists of the radio core connected to my block and then to the host. > Nothing else. > > > > I am wondering are there flow control things that I am not doing > properly in my block. Perhaps the host UHD code is not getting the right > feedback for the flow control? Perhaps sample line rate settings that I > should be doing? In my UHD code I am only setting the sample rate in the > radio core and it only accepts 200 MHz but does that somehow screw up > the flow control? Is there special sauce in the duc that makes it > necessary to use that with the radio core to get the rates correct? > > > > Thanks for any help. > > > > Jeff > > > > _______________________________________________ > USRP-users mailing list > [email protected] > http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com > _______________________________________________ USRP-users mailing list [email protected] http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com _______________________________________________ USRP-users mailing list [email protected] http://lists.ettus.com/mailman/listinfo/usrp-users_lists.ettus.com
