[FEATURE] `compute_coverage_map`

[rwydaegh]

Terms

• Checked the existing issues to see if my suggestion has not already been suggested;

Description

There's no coverage map function. with_receivers_grid exists, compute_paths exists, but there's nothing to tie them together into a 2D power map. Two milestone issues already assume this capability exists:

• chore(tests): compare coverage map with Sionna #206: "Compare coverage map with Sionna"
• chore(docs): add notebook tutorial for radio coverage map using generative sampling-based RT #202: "Add notebook tutorial for radio coverage map"

smoothing.ipynb builds something like a coverage map manually (~40 lines of orchestration), using 0.5 ** bounces instead of actual EM.

def compute_coverage_map(
    self,
    frequency: Float[ArrayLike, ""],
    *,
    max_order: int = 2,
    grid_size: tuple[int, int] = (50, 50),
    height: Float[ArrayLike, ""] = 1.5,
    method: Literal["exhaustive", "hybrid"] = "hybrid",
    dB: bool = True,
    coherent: bool = False,
    bvh: TriangleBvh | None = None,
    **kwargs,
) -> tuple[Float[Array, "m n"], Float[Array, " m"], Float[Array, " n"]]:
    """2D received power map over a receiver grid.
    Returns (power_map, x_coords, y_coords).
    """

Internally it would place receivers on a grid, loop over orders 0 to max_order, call compute_paths then compute_received_power per order, and accumulate. Plus a draw_coverage_map in differt.plotting.

This depends on #427 (compute_received_power). Without actual EM there's not much point.

For testing: open scene with one TX should match FSPL falloff, scene with a reflector should show the interference pattern, and ideally a comparison against Sionna (which directly addresses #206).

Main practical concern is memory for larger grids. 50x50 receivers with many path candidates can get heavy, but batch_size and chunk_size should handle it if documented properly.

This needs compute_received_power from #427 first. Once that's in, this becomes straightforward, and it directly addresses #206 and #202.

[jeertmans]

Hi @rwydaegh! Yes, you're right that the current way I compute the received power is incorrect. And probably that the example in the Coherent vs. Non-coherent RT also contains some errors.

I believe computing coverage maps using batched-over-rx-positions methods is not the best, especially because it requires an enormous amount of memory. When #427 is merged, I think it would be also nice to see if we can compare the batched approach with a SBR method.