USRP N210 Latency and underrun

[Fabien PELLET]

Hello,

I'm trying to make work the attached flowgraph : a simple Audio source, 
some filtering and feeding an USRP sink (N210).

I have two problems with it :

- I measure the latency between audio input and RF output of the N210 
and it is too long for me as I have about 30-40ms at the minimum.

- I have some underruns <U> at the startup of the flowgraph and then 
sometimes others <U> appear. The problem is that at each new <U>, the 
latency is increasing.

I do not use that over hours but the latency seems to be out of control.

I'm working on a laptop DELL Precision M4800 with a I7-4800M, 8Go RAM 
and a NVidia Quadro K2100M.

GNURadio 3.8.2 is installed from source on Lubuntu 20.04 with UHD4.0. 
UHD and GNURadio are installed from source.

I try a lot of things without any beginning of a small optimization or 
correction of the problem :

- Working on a Lubuntu with a LXQT and lowlatency kernel,

- Try to play with the MTU and send/recv buffers,

- Install Cuda and OpenCL to play with GR-CLENABLED and GR-LFAST for the 
filters,

- Reduce the taps of my FFT bandpass filter,

- Test UHD bandwidth using benchmark provided : with 25Msps in TX and 
25Msps in RX at the same time, no underrun.

- Decimate the signal from the Audio Source around 10Ksps to get small 
number of taps for the FFT bandpass filter.

The only way to remove the underrun are :

- Feed the USRP sink with an little higher sampling than it expects (but 
signal at the TX output is wrong, I know).

- Remove completly the FFT bandpass filter and the Hilbert.

My flowgraph seems so simple that I hope I only miss a stupid thing. I 
hope someone here could help me find where is my mistake and how I can, 
at the minimum, warranty that my latency are stable even if it is at 
40-50ms.

Thanks for your help,

Best regards,

Fabien, F4CTZ.

test_N210_tx.grc
Description: Text document