k-Wave

1. carljiri
   

   Member
   Joined: Jun '20
   Posts: 3

   Hey,
   thanks for the great tool.

   i simulate 3D a bowl shaped focused transducer which consists out of 8 ring elements.
   i got a problem for phase shift/focus shift simulation.

   I use Matlab 2019a
   Kwave Cuda 1.3
   gtx 3090

   I create 8 different signals for the 8 elements with time delayed signals.
   i is the number of the element: source.p(i, :) = filterTimeSeries(kgrid, medium, sig);

   If i use kspaceFirstOrder3D Simulation, it works perfect but needs 25 minutes with gpuarray-single for one simulation.

   If i use 3DG for cuda i got this error: Dataset p_source_input has wrong dimension size. iostream error Matlab also Puts out errors: h5readc (not found or unsupported characters)

   If i dont use phase shift, and only give 1 signal to the source for all elements. 3DG works perfect and fast and calculates everything in 90 seconds.

   so 90seconds vs 25 min is a huge difference for me.

   Maybe someone knows how to fix this porblem or did i miss, that this is not possible.

   Posted 3 years ago #

2. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Hi carljiri,

   Does source.p have the same number of signals as the sum(source.p_mask(:))?

   Brad

   Posted 3 years ago #

3. carljiri
   

   Member
   Joined: Jun '20
   Posts: 3

   Hey :)

   all elemements are present in the 3D array. where the element exists index == number of the element. so inner circle ==1 to most outer circle == 8.

   so i got everything from 1 to 8 in its....(p_mask)

   and in the source_p i got 8 rows of shifted signals.

   Posted 3 years ago #

4. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Just to clarify, does p_mask contain numbers 1 to 8, or just the numbers 1 and 0?

   Posted 3 years ago #

5. carljiri
   

   Member
   Joined: Jun '20
   Posts: 3

   Hey,
   it contains 0 for no mask
   and 1,2,3,4,5,6,7,8

   it gets really strange now.

   on the nvidia 2080 i can calculate it using cuda binaries. on the 3090, it crashes, but every other calculation where i got only 1 signal for all elements i can run on the 3090.

   i will try do delete the kwave toolbox and install them again on the 3090 server on monday. maybe it fixes the problem.

   Posted 3 years ago #

6. Bradley Treeby
   

   Developer
   Joined: Mar '10
   Posts: 1,643

   Hi carljiri,

   Ok, that makes sense now. The use of a labelled source mask is an experimental feature and not fully supported (I thought it wasn't supported at all in the C++/CUDA versions, but I'd need to go back and check). In a future version this will be more widely supported, but in a different format that also allows scaling.

   Brad.

   Posted 3 years ago #

7. Drainville1
   

   Member
   Joined: Jun '17
   Posts: 1

   Any update on the labelled source mask? Is there any way to reduce the size/dimension of source.p when using a large number of source points, that could otherwise be represented by a single signal for each unique element?

   Posted 6 months ago #

8. chatillon
   

   Member
   Joined: Sep '20
   Posts: 1

   Hello everyone,
   I encounter a problem that seems similar: I define a non-symmetrical 3D source and I do the same calculation in CPU (kspaceFirstOrder3D) and GPU (kspaceFirstOrder3DC) mode. The first gives satisfactory results while for the second I have the impression that there is an inversion of the X and Y axes of the source grid and/or of the sensor grid. Could you confirm the problem for me?
   thanks in advance

   Posted 6 months ago #

RSS feed for this topic

Reply

You must log in to post.