Re: [Discuss-gnuradio] Repeatability of delivery ratios using benchmark_tx/rx

[Aditya Dhananjay]

On Wed, Oct 16, 2013 at 10:29 PM, Aditya Dhananjay <address@hidden> wrote:

On Wed, Oct 16, 2013 at 10:22 PM, Aditya Dhananjay <address@hidden> wrote:

Hi All,

I have two USRP N210 devices connected by an attenuator cable. I set up the following experiments.

-- BEGIN EXPERIMENT 1

Step 1) Use benchmark_tx on one of the USRPs and send 100 packets. At the receiver, I simply record the incoming samples and save them into a file called samples1.dat. This file is static and does not change from now on.

Step 2) Run benchmark_rx with the --from-file=samples1.dat option. The other parameters (modulation, fft-length, occupied-tones, etc) are identical to those I used at the transmitter in Step 1. Measure the packet delivery rate.

Step 3) Repeat Step 2 a bunch of times. Sometimes, I notice differences in the packet delivery ratios, even though all runs of the experiment use the identical 'samples1.dat' file and an identical set of parameters These differences are small (1-2 packet difference), but I don't understand why they exist.

Is this normal behavior? If so, could someone please let me know what the probable causes are?

-- END EXPERIMENT 1

The next experiment does not use either of the USRP devices.

-- BEGIN EXPERIMENT 2

1) Use benchmark_tx to generate 100 packets, and write the samples into a file (samples2.dat) using the --to-file option.

2) Use benchmark_rx with the --from-file=samples2.dat option and check the packet delivery rate. The parameters used (fft-length, etc) are obviously identical to those used to generate the samples in Step 1. I observe that the delivery rate is never even close to 100%. 

3) Change to a different parameter set and repeat steps 1 and 2. Across different runs, I observe that the delivery rate is anywhere between 30% and 85%.

This cannot be the expected behavior. Any pointers on what I'm

doing wrong would be much appreciated!

-- END EXPERIMENT 2

Thanks!

Aditya

I should have made a brief mention of my environment:

GNU Radio 3.7.1 installed from source

Ubuntu 12.04 LTS (64-bit)

Thanks!

Aditya

Update:

In experiment 1, I began to notice large variations in packet delivery ratios across different runs of step 3. If I modified benchmark_rx to just sleep for a few seconds between tb.start() and tb.wait(), these large variations disappeared. It seems like once the EOF is reached, it simply kills all threads that run inside the flow graph.

However, even with the sleep, there are small variations (~1-2 packets) in an experiment with a total of 1000 packets being generated. Could anyone please give me a little hint on why this might be the case?

Thanks.

Aditya