meddlr.ops.fftnc#

meddlr.ops.fftnc(input: Tensor, dim=None, norm='ortho', is_real: Optional[bool] = None) → Tensor[source]#

Apply nD centered fast fourier transform.

This function is a backwards-compatible wrapper for centered torch.fft.fftn(). It supports backwards compatibility with ``torch.fft``as implemented in torch<1.7.

Parameters

  • input (torch.Tensor) – A tensor (typically complex).

  • dim (Tuple[int]) – Dimensions to be transformed.

  • norm (str | bool, optional) – The normalization method. Defaults to 'ortho'. For torch<1.7, only 'ortho' is supported.

  • is_real (bool, optional) – If True, input is a real-valued tensor. If None or False and input.shape[-1] == 2, input is a real-view of a complex tensor.

Returns

torch.Tensor

Note

Real-valued tensors are not supported with torch<1.7.