Clarification on Experiment Settings for Reproducing DRL Control CDF Plot in "Intelligence and Learning in O-RAN for Data-driven NextG Cellular Networks"

[erictseng610]

Hello,

I am currently working on reproducing the results shown in Figure 4 of your paper, "Intelligence and Learning in O-RAN for Data-driven NextG Cellular Networks," published in IEEE Communications Magazine (Vol. 59, No. 10, October 2021). Specifically, I'm interested in the Downlink Spectral Efficiency of the eMBB slice for different scheduling policies and with DRL control.

Despite following the experiment setup as detailed in the paper, my results significantly diverge from those presented in Figure 4. I understand from the paper that the tests were conducted in a challenging setup, including the random mobility of UEs. Consequently, I used the "rome_slow_close" dataset (tr0 to tr17) for all actions and calculated the rewards to plot the CDF. However, my results are markedly different from those shown in Figure 4.

I kindly request additional details regarding the experimental setup used in your study, specifically:

Which dataset(s) were used to obtain the experimental results presented in the paper?

Thank you for your time and assistance.

[salvododo]

Hi,

it's been quite some time since I processed the data and generated the plots to represent the results. Give me a some time to look into the repos and find the scripts we used to generate the figures. Will get back to you ASAP

Salvo

[hanghoo]

Hi @erictseng610, did you use dataset tr0-tr17 to train DRL?