Page 130 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

Fig. 1 – Block diagram of CAVIAR simulation with AI/ML in the simulation loop (INLOOP). In OUTLOOP simulations, the simulator can write iles that
will be later used for designing and assessing AI/ML models.

modeling mobility and virtual worlds discussed in this paper The rest of the paper is organized as follows. Methods
can alleviate the problem. The generated datasets are es‑ and software for CAVIAR simulation of 6G are presented
pecially useful when spatial consistency and time evolu‑ in Section 2. Section 3 explains some improvements in the
tion are important to assess an AI/ML technique applied RT simulation methodology. Section 4 presents numer‑
to the physical layer. ical results and their discussion. Finally, Section 5 con‑
cludes the paper.
The contributions of this paper are:
2. 6G SIMULATION IN VIRTUAL WORLDS
• A discussion of strategies and software for simu‑
lating Communication networks and icial intel‑
ligence immersed in VIrtual or Augmented Reality Gaming and other industries are driving the develop‑
ment of sophisticated tools to create virtual worlds, com‑
(CAVIAR).
posed of 3D models, physics engines and other compo‑
• A preview of a CAVIAR simulator that will be used in nents. The virtual world 3D scenery can be created from
the UFPA Problem Statement for the 2021 ITU AI/ML scratch by 3D design modelers, or from data imported
in 5G Challenge, which consists of a Reinforcement from the real world. For instance, the new Cesium plug‑in
Learning (RL) problem with the decisions taken by for Epic Game’s Unreal Engine integrates photogrametric
3
the RL agent changing the virtual world on‑the‑ ly (as information obtained from drones into 3D models avail‑
4
the simulation evolves). able via Cadmapper and other sites. This complements
5
• Discuss a new methodology using photogrametric tools such as Twinmotion, which facilitate the construc‑
tion of 3D virtual worlds. This paper promotes the vi‑
data available from the Internet to improve the re‑
sion that 6G and beyond will bene it from the availabil‑
alism of ray‑tracing simulations by automatically as‑
ity of virtual worlds to leverage ML/AI applied to com‑
signing electromagnetic properties to the materials
munication networks. Current investigations of AI ap‑
composing a scene, via semantic segmentation with
plied to 5G aim at inding how raw data from sensors such
deep neural networks.
as LIDAR and cameras can optimize the communication
• Results exposing trade‑offs between speed and accu‑ performance [12, 13, 14, 15]. But the possibility of hav‑
racy when generating channels via ray tracing. ing realistic 3D models, physics engines and other virtual
• Results of a reinforcement learning experiment in reality assets for simulations of communication systems,
beam selection realized in the CAVIAR environment. opens new horizons in terms of AI/ML applied to 6G and
beyond.
• Source code and datasets to reproduce the baseline
of 2021 ITU AI/ML in 5G Challenge. 2 3 https://cesium.com/blog/2021/03/30/
cesium‑for‑unreal‑now‑available/.
2 https://ai5gchallenge.ufpa.br/ 4 https://cadmapper.com.
5 https://www.unrealengine.com/en‑US/twinmotion.

114 © International Telecommunication Union, 2021