[Discuss-gnuradio] OFDM transmit/response recording problems

[Dev Ramudit]

Hello all,

	For an experiment I'm working on I need to transmit an OFDM signal from 
one USRP and 'respond' from another USRP with an OFDM message with 
(hopefully) millisecond timing. I've used elements of 
benchmark_ofdm_tx/rx and the tunnel examples to construct the following:

USRP A                 |    USRP B
begin carrier sensing  |
                       |
carrier sensed = True  |    send 50 pkts
...                    |    ...
carrier sensed = False |    done sending
send 50 pkts           |    receive and wait ~.5 seconds
repeat                 |    repeat

Both USRPs have a TX graph and an RX graph running. The TX path is the 
same as the one in the benchmark_ofdm_tx files. The RX path goes to a 
custom OFDM module which doesnt demod packets, but only provides carrier 
sense. The rx_graph looks something like this:

---
self.rxpath = receive_path(self, callback, options)
self.connect(self.u, self.rxpath)

self.logfile = gr.file_sink(gr.sizeof_short, "/mnt/rd/logfile" +
str(self._usrpnum) + "dat")
self.ctolshort = gr.complex_to_interleaved_short()syn
self.connect(self.u, self.ctolshort, self.logfile)
---

The receive_path goes up to the ofdm_recv.chan_filt, so that the 
gr.probe_avg_mag_sqrd_c can provide carrier sense. I had to cut out the 
demod, as it seemed to slow down each side so much that I'd have an 
abundance of underruns. Right now I just constantly record the data from 
each side to a file, then later run it through a script to demod and 
correlate the timings.

The script above runs OK at a decimation of 12, interpolation of 24 and 
200 occupied tones. I only get one uO every 10000 pkts or so, which is 
acceptably low. However, if I run it with -d 10 and -i 20, I get a uO 
every 50 pkts and no usable data.

I've done everything I can to speed up the hard disk side of the 
equation by creating a RAID 0, running on both XFS and EXT2 with the 
fastest options I can find. However, it doesnt seem as if the drive 
writes are the issue. If I remove the file_sink or close the file socket 
while running, I still receive the uOs. I've kept track of both my RAM 
and CPU, but neither are maxing out during the runs, so I dont believe 
that's the problem either.

Basically, I want to run a tx and rx path at the same time on two USRPs, 
with the lowest decimation possible. I need to send from one and respond 
from the other within milliseconds. Any suggestions on how to reduce the 
number of uOs? Or a different approach?

Thanks,
Dev