Send Image data over noisy 5G LTE network

[Rahul Ganesh]

Hello Team,

I am relatively new to the Powder platform and am planning to leverage this platform for my Capstone project which involves sending a bunch of images over a noisy mobile network, and see how the packet loss impacts Machine Learning models. For transmitting images from a noisy network using UDP, I plan to leverage the Powder Platform. Earlier, from a Call with Powder team, I got high level idea about the platform and how I could go about it. I have completed few of the related tutorial sessions such as Lab1 and Lab1 extension from oai_FALL_21 workshop. From this I understood, how to simulate a basic 5G E2E network, generate some traffic through ping, using telnet to modify channel parameters. However, I am new to the world of networking and have the following queries to proceed with my project:

i.  Whenever I try to change the parameter noise_power_dB to a high positive value such as 10 or 15 the network collapses with disconnection. (not sure such positive values are even valid). I would like to know what would be a viable value to set for parameters noise_power_dB and ploss to expect some distortions in the image because of packet loss.

ii. how can I extend my learnings from this lab, to send images over a noisy mobile network. I am planning to create a UDP socket and want the images to be sent to it through the lossy network (planning to write a Python script for the same). But as I am new to Computer Networks, a little bit details/tips/pointers on how should I go about would be greatly appreciated. 

Thanks for your time and help in advance.

Regards,

Rahul

[Dustin Maas]

Hi Rahul,

Sounds like you’ve made some good progress!

i. Whenever I try to change the parameter noise_power_dB to a high
positive value such as 10 or 15 the network collapses with disconnection.
(not sure such positive values are even valid). I would like to know what

would be a viable value to set for parameters *noise_power_dB* and *ploss *to

expect some distortions in the image because of packet loss.

I think this is expected. I’m not sure precisely how OAI implements noise_power_dB, but I imagine it is something like the ratio of noise power to gNB signal power. So, positive values may represent the noise signal being stronger than the gNB signal. (You can have a look at the source to be sure what they are doing.) ploss and noise_power_dB will have a combined affect on packet loss. Maybe it would be good to start with no path loss and effectively no noise (-100 or disabled if that is possible), start up iperf3 sending UDP traffic across the network, then start to add noise and path loss and see what levels start to cause lost packets.

ii. how can I extend my learnings from this lab, to send images over a
noisy mobile network. I am planning to create a UDP socket and want the
images to be sent to it through the lossy network (planning to write a
Python script for the same). But as I am new to Computer Networks, a little
bit details/tips/pointers on how should I go about would be greatly
appreciated.

I think a good next step would be to transition to the Indoor OTA Lab and repeat your experiments there. You could inject noise using the radios in that space, play with transceiver gains, etc.

Best,
Dustin

--
You received this message because you are subscribed to the Google Groups "Powder Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to powder-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/powder-users/f89e1bc6-b08d-4677-94e3-bc046c170564n%40googlegroups.com.