FAQ from GTC 2022

[faycalaa]

Many very interesting questions were asked during GTC 2022. We're posting them here, along with our answers, as we believe they might be of interest to many users.

• Are there any plans to support “real-world” datasets (as commonly done in machine learning)?

Sionna already provides a function that allows one to use datasets of channel measurements in-lieu of an actual channel model. Looking into the future, this might be extended to ingesting datasets from other types of measurements, e.g., recorded IQ samples.

• You mention that external parties can contribute. What kind of contributions are you looking for?

There are three ways in which external parties can contribute: GitHub Issues, pull requests, or simply suggestions for new features in the GitHub Discussion Forum. We are very interested in contributions for new channel models for THz communications, reconfigurable intelligent surfaces, optical communications, and any type of hardware impairments.

• I am working on semantic communications. Does Sionna support this?

Semantic communications is about extracting relevant information from data for a given task, prior to communicating it. Since all of Sionna’s components are differentiable, it is a perfect fit for such applications as everything can be trained from end-to-end.

• You mentioned RIS at the beginning of your talk. How are they currently supported?

Sionna has no channel model for RIS, yet. This is one of the areas in which we hope for contributions from the community.

• Are there any plans regarding hardware interfaces (using SDRs etc.)?

This is definitely on our agenda for future releases. We do not have any concrete timeline.

• Do you see Sionna as a purely academic tool?

Sionna is a tool written by researchers for researchers. It can be used for academic as well as industrial research. One of Sionna’s goals is to bring both worlds closer to each other by offering the flexibility needed for blue-sky research while simultaneously enabling realistic industry-grade simulations. However, Sionna does not have the ambition to be used for 3GPP-compliant system-level simulations.

• I do research on optical communications. Can Sionna be used for this?

You could use Sionna to simulate optical communication systems. However, it currently does not have any channel model for this. This is one area in which we hope for contributions from the community.

• Can Sionna be used for system-level simulations?

No. Sionna is a link-level simulation tool. Looking into the future, it might be very interesting to expand Sionna towards this direction to leverage GPU acceleration and differentiability.

• Can you explain what is meant by a differentiable link-level simulator?

Take your favorite physical layer component, e.g., a demapper: soft-symbols go in, LLRs come out. Differentiable means that Sionna can compute automatically the gradients of the LLRs with respect to the soft-symbols. Thanks to the chain rule of differentiation, this extends to an entire system from end-to-end. This allows you to optimize any component with respect to a performance metric computed at the output of any other component.

• Do I need to use Jupyter to work with Sionna or can I use a Python development environment like PyCharm?

You can use any tool you like. We work internally with VSC and the Jupyter notebook plugin.

The Sionna Development Team