Open jkosciessa opened 1 month ago
Debugging in the k-Wave function kSpaceFirstOrder
suggests that this may arise from unreasonable tissue values somewhere. A major pressure line in this instance is estimated around the neck area after only a few executed time steps (far away from the transducer plane, suggesting a computational problem). I attached some plots of different time steps, that ultimately converge in NaNs. Note that this is from the pseudoCT implementation that uses continuous tissue property values.
This seems fixed in commit 361c869e0bee9c688d96626abfef36a77f68847b after regularizing minimum pseudoCT-based bone density to the water estimate. As also discussed here this suggests that the conversion of pseudo-HU into tissue values may not be very sensible yet and should be double-checked.
I additionally increased the time step to 10% of the limit (from 1%) in commit d83e970ae456580f4a68052584014b13eae07e9f, but the former should be the relevant change.
With the updated tissue attenuation, I run into sporadic acoustic simulations that are filled with NaNs. Those will eventually crash and are characterized by max ISPPA positions of 1 1 1:
The crash will happen at the attempt to plot, and will be of the format: