Beamforming

[krzysztofrusek]

This pr introduces beam forming and better walls handling in the simulator.

TODO

• Refactor walls argument
• Add isotropic antena and free space loss
• Add monte-carlo walls
• Improve documentation. In particular we must decide whether to use classical antena gain or relative to the maximum.
• Add walls visualization

[m-wojnar]

Looks good 👍 For the further steps, I propose to add some class or function wrapper for a whole process. This will allow the user to call just one function instead of preparation of the wall matrix, antena gain matrix, and finally simulator.

Do you possibly know why unit tests are failing? It looks like a tfp or jax bug?

[krzysztofrusek]

Looks good 👍 For the further steps, I propose to add some class or function wrapper for a whole process. This will allow the user to call just one function instead of preparation of the wall matrix, antena gain matrix, and finally simulator.

I agree.

My idea is to have simulator as a callable with differentiable parameters.

This callable can be either :

• a pytree with __call__
• or a factory function returning differentiable simulator (optax style).

I am not sure right now what is better at this point, as we need to modularize the sim function.

Do you possibly know why unit tests are failing? It looks like a tfp or jax bug?

no, but indeed this looks like tfp issue, maybe we need to bump version.