normal

sionna.phy.utils.normal(size: Sequence[int], *, mean: float = 0.0, std: float = 1.0, dtype: torch.dtype | None = None, device: str | torch.device | None = None, generator: torch._C.Generator | None = None) → torch.Tensor

Generate random normal tensor, compile-aware.

In eager mode, uses the provided generator for reproducibility. In compiled mode, uses torch.randn (which uses the Graph RNG) to ensure proper synchronization with other random operations like randint.

Note: torch.normal uses a different RNG stream than torch.randn under torch.compile, which can cause training issues. Using torch.randn with scaling ensures consistent RNG behavior.

Parameters:

• size (Sequence[int]) – Shape of the output tensor.

• mean (float) – Mean of the distribution. Defaults to 0.0.

• std (float) – Standard deviation of the distribution. Defaults to 1.0.

• dtype (torch.dtype | None) – Data type of the output tensor.

• device (str | torch.device | None) – Device for the output tensor.

• generator (torch._C.Generator | None) – Random number generator (used only in eager mode).

Outputs:

samples – Tensor with random normal values.