LMMSEInterpolator initialization is very slow with GPU enabled

[vtyibin]

The __init__ function of LMMSEInterpolator runs very slow when GPU is enabled. Specifically, the _build_pilot_mask() function takes a very long time to finish.

The issue is that in _build_pilot_mask(), the variables mask and pilots are not converted to NumPy arrays, which force the following loops to run on the GPU. In my case, the _build_pilot_mask() function takes about 1 minutes instead of 0.2 second.

So the two assign statements from

sionna/sionna/ofdm/channel_estimation.py

Line 1734 in 0a4a22e

mask = pilot_pattern.mask

should be modified as:

 mask = np.array(pilot_pattern.mask)
 pilots = np.array(pilot_pattern.pilots)

[jhoydis]

Hi @vtyibin,

Sorry for the late reply to this issue. I cannot reproduce your observations. Could you provide a fully functional code snippet? I also do not think that any loop in the initialization is executed on GPU.

[vtyibin]

@jhoydis Please use the attached Jupyter Notebook file.
test_lmmse_interpolator.ipynb.zip