k-Wave User Forum » User Favorites: Farnazkhj

k-Wave User Forum » User Favorites: Farnazkhj farnazkhj Support for the k-Wave MATLAB toolbox en-US Tue, 12 May 2026 23:11:39 +0000 http://bbpress.org/?v=1.0.2 <![CDATA[Search]]> q http://www.k-wave.org/forum/search.php Bradley Treeby on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6931 Wed, 26 Jun 2019 10:46:35 +0000 Bradley Treeby 6931@http://www.k-wave.org/forum/ If you get a stability warning and reducing dt fixes the peak, then it does sound like a stability problem (with an easy fix). The stability is related to the frequency of the source, but the properties of the grid and medium. Farnaz on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6923 Tue, 25 Jun 2019 13:06:11 +0000 Farnaz 6923@http://www.k-wave.org/forum/ Dear Brad, Thank you for the quick response. Yes, I used filterTimeSeries but the peak is still there. Maybe this plot would demonstrate my question better: <img src="https://imagizer.imageshack.com/img922/6394/iEIIry.jpg" /> <img src="https://imagizer.imageshack.com/img923/4159/ypcT6l.jpg" /> The one with high peak in the frequency spectrum has the center frequency of 5e6hz whereas the other one with with center frequency of 1.5e6hz looks as I expect. The only parameter that changes is the frequency of input signal. note that I get the warning that the simulation might not be stable because of time steps but I am wondering why this happens for 5Mhz whereas the grid can support up to 17Mhz? (decreasing dt results in disappearing this peak) % ========================================================================= % DEFINE THE K-WAVE GRID % ========================================================================= % set the size of the perfectly matched layer (PML) pml_x_size = 66; % [grid points] pml_y_size = 65; % [grid points] pml_z_size = 10; % [grid points] % set total number of grid points not including the PML Nx = 1152 - 2 * pml_x_size; % [grid points] Ny = 1152 - 2 * pml_y_size; % [grid points] Nz = 32 - 2 * pml_z_size; % [grid points] % set desired grid size in the x-direction not including the PML x = 42.4e-3; % [m] % calculate the spacing between the grid points dx = 3.6765e-5; % [m] dy = dx; % [m] dz = dx; % [m] % create the k-space grid kgrid = kWaveGrid(Nx, dx, Ny, dy, Nz, dz); % ========================================================================= % DEFINE THE MEDIUM PARAMETERS % ========================================================================= % define the properties of the propagation medium medium.sound_speed = 1540 ; medium.density = 900 ; c0 = 1540; % [m/s] rho0 = 900; % [kg/m^3] medium.alpha_coeff = 0.5; % [dB/(MHz^y cm)] medium.alpha_power = 1.05; % create the time array t_end = (Nx * dx) * 2.2 / c0; % [s] kgrid.makeTime(c0, [], t_end); t_min_stable = checkStability(kgrid,medium) ; kgrid.t_array = (0:t_min_stable:t_end) ; % ========================================================================= % DEFINE THE INPUT SIGNAL % ========================================================================= % define properties of the input signal source_strength = 1e6; % [Pa] tone_burst_freq = 1.5e6; % [Hz] tone_burst_cycles = 2; % create the input signal using toneBurst input_signal = toneBurst(1/kgrid.dt, tone_burst_freq, tone_burst_cycles); input_signal = (source_strength ./ (c0 * rho0)) .* input_signal; input_signal = filterTimeSeries(kgrid,medium,input_signal) ; Bradley Treeby on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6921 Tue, 25 Jun 2019 11:49:33 +0000 Bradley Treeby 6921@http://www.k-wave.org/forum/ If the peaks do not increase as the simulation runs, it doesn't sound like a stability issue. Have you tried filtering the input signal, e.g., using <code>filterTimeSeries</code>? Farnaz on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6919 Tue, 25 Jun 2019 10:02:12 +0000 Farnaz 6919@http://www.k-wave.org/forum/ hi Brad, Thank you very much for the response. As far as I understood the stability depends mostly on dt. However, by changing input signal (increasing transducer center frequency (still far less than the frequency grid supports) or changing the number of cycles) I see peaks in the frequency spectrum where I do not expect, e.g. a relatively high peak at the highest frequency that grid supports or a very high peak centered around 0. Is this also stability issue? if not, how these peaks appear? Thanks in advance, Farnaz Bradley Treeby on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6915 Tue, 25 Jun 2019 08:48:47 +0000 Bradley Treeby 6915@http://www.k-wave.org/forum/ Hi Farnazkhj, Unfortunately there is no simple way to speed up the calculation without using more computational resources. We will release an MPI code that can scale across multiple nodes on a cluster later in the year. Unfortunately for now, the performance of a single recent GPU is likely to be better than the performance of a single node, even using multiple cores. The CFL depends on both dt and dx. So what you propose is only possible if you also change the grid spacing. As it stands, you can't sample with a lower frequency (i.e., decimate the output) in either the MATLAB or C++ code. For reference, the <code>-s</code> command line argument for the C++ code is identical to <code>sensor.record_start_index</code> in MATLAB. Hope that helps, Brad. Farnaz on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6904 Mon, 24 Jun 2019 11:22:58 +0000 Farnaz 6904@http://www.k-wave.org/forum/ Hello again, I realized by adding random scatterers the cross-talk artifact will be eliminated from the raw channel data. Although I have still questions regarding this simulation: -I am wondering how it would be possible to speed up this simulation?(it takes 10-11GB memory on GPU and about 15-20 minutes to run) -As it is not possible to run GPU code on parallel GPUs, how many CPU cores on a cluster do I need to run with same performance? - Is it possible to increase dt and compensate for it with decreasing cfl? - How can I sample in a lower frequency using MATLAB interface (I know it is possible to use -s argument using command line but I get error passing this argument in the MATLAB). Would this decrease the computation time? (I assume it won't and dt is an independent parameter for solving the equations, right?) Bradley Treeby on "Plane wave simulation, commercial probe" http://www.k-wave.org/forum/topic/plane-wave-simulation-commercial-probe#post-6893 Sat, 22 Jun 2019 20:04:02 +0000 Bradley Treeby 6893@http://www.k-wave.org/forum/ Hi Farnazkhj, 1. Yes, if you set the focus point to a large distance, you will get an unfocused transmit. 2. As long as the transducer is not within the PML, you can set Nz to be a small number. The only downside is that you might not capture the complete out-of-plane behaviour, although the effect on the image is likely to be very small. 3. Keep in mind a simulation in 2D is different to 3D (e.g., a point source in 2D corresponds to an infinite line source in 3D, so will spherical spreading will be less than a point source in 3D). If you can live with that, 2D will certainly be faster. Brad Farnaz on "Plane wave artifact" http://www.k-wave.org/forum/topic/plane-wave-artifact#post-6869 Tue, 04 Jun 2019 09:24:05 +0000 Farnaz 6869@http://www.k-wave.org/forum/ Hi all, I am trying to simulate a real transducer, I came across an artifact that I am not sure where it is coming from and whether it could be solved or it is just a physical limitation. I could not find my answer in the forum or the manual. The artifact seems to be some sort of cross talk. It would be a great help if you could share your insights. I am trying to simulate a 128 element transducer (64 elements of the transducer are activated during TX and RX), transducer width is 37.5 mm therefore only the 64 middle elements are active(18.75mm). I use the transducer class and CUDA code. My intention is to transmit a plane wave trough a medium and receive the echoes with same transducer. My grid size is 1024*1024*64 with pml size 64*64*16(outside). After extracting the scan-lines using transducer.combine_sensor_data(sensor_data) and ploting it, the area near transducer receives what appears to me as a horizontal wave propagated from the side elements. When I remove the kerf the artifact to some extend disappears (not completely though and I need kerf) -What can possibly cause such artifact? (e.g. can this be a result of small number of grid points in z directions?(I have restriction in computational resources right now)) -Can I avoid it in any way (increasing pml size, pml alpha, placing the transducer inside a highly absorbing material, increase or decrease dt, etc.) -What are the acoustic properties of the transducer and the kerf? -As the grid is too large, I would like to increase time step but the artifact seems to be even worse when I do so (I use t_min_stable = checkStability(kgrid,medium) and set t_end to this value, changing cfl value doesn't help too). How can I increase time step (or decreasing frequency or in general any solution that results in smaller data size) my code is as follows: % simulation settings DATA_CAST = 'gpuArray-single'; RUN_SIMULATION = true; f_max = 5e6 ; % set the size of the perfectly matched layer (PML) pml_x_size = 64; % [grid points] pml_y_size = 64; % [grid points] pml_z_size = 16 ; pml_x_alpha = 3; pml_y_alpha = 3 ; pml_z_alpha = 3 ; %total size x_size = 4.24e-2 ; %[m] y_size = 4.24e-2 ; dy = 0.0375/(128*4+127*4 ); %[m] dx = dy ; % dz = dy ; % Ny = y_size/dy grid points Ny = 1024 ; %ceil(y_size/dy)-2*pml_x_size ; Nx = 1024 ; %ceil(x_size/dx)-2*pml_y_size ; Nz = 32; % create the k-space grid kgrid = kWaveGrid(Nx, dx, Ny, dy, Nz, dz); % ========================================================================= % DEFINE THE MEDIUM PARAMETERS % ========================================================================= % define the properties of the propagation medium medium.sound_speed = 1540 ; rho0 = 900; % [kg/m^3] medium.alpha_coeff = 0.5 ; medium.alpha_power = 1.05 ; % create the time array c0_min = 1300 ; %m/s t_end = (Nx * dx) * 2.2 / c0_min; % [s] kgrid.makeTime(c0_min,0.3, t_end); % ========================================================================= % DEFINE THE INPUT SIGNAL % ========================================================================= % define properties of the input signal source_strength = 1e6; % [Pa] sampling_freq = 20e6 ; % [Hz] tone_burst_freq = 5e6; % [Hz] tone_burst_cycles = 2; input_signal = toneBurst(1/kgrid.dt, tone_burst_freq, tone_burst_cycles); input_signal = (source_strength ./ (c0_min * rho0)) .* input_signal; % ========================================================================= % DEFINE THE ULTRASOUND TRANSDUCER % ========================================================================= % physical properties of the transducer transducer.number_elements = 128; % total number of transducer elements transducer.element_width = 4; % width of each element [grid points] transducer.element_length =28; % length of each element [grid points] transducer.element_spacing = 4; % spacing (kerf width) between the elements [grid points] transducer.radius = inf; % radius of curvature of the transducer [m] % calculate the width of the transducer in grid points transducer_width = transducer.number_elements * transducer.element_width ... + (transducer.number_elements - 1) * transducer.element_spacing; % use this to position the transducer in the middle of the computational grid transducer.position = round([2, 2, 1]); % properties used to derive the beamforming delays transducer.sound_speed = c0_min; % sound speed [m/s] transducer.focus_distance = inf; % focus distance [m] transducer.elevation_focus_distance = inf; % focus distance in the elevation plane [m] transducer.steering_angle = 0; % steering angle [degrees] % apodization transducer.transmit_apodization = 'Hanning'; transducer.receive_apodization = 'Hanning'; % define the transducer elements that are currently active transducer.active_elements = zeros(transducer.number_elements, 1); transducer.active_elements(33:96) = 1; % append input signal used to drive the transducer transducer.input_signal = input_signal; % create the transducer using the defined settings transducer = kWaveTransducer(kgrid, transducer); % print out transducer properties transducer.properties; % ========================================================================= % DEFINE THE MEDIUM PROPERTIES % ========================================================================= % define a large image size to move across Nx_tot = Nx; Ny_tot = Ny ; Nz_tot = Nz; % define a random distribution of scatterers for the medium background_map_mean = 1; background_map = background_map_mean ; scattering_c0 = c0_min ; scattering_rho0 = scattering_c0 / 0.9 ; % define properties sound_speed_map = c0_min * ones(Nx_tot, Ny_tot, Nz_tot) .* background_map; density_map = rho0 * ones(Nx_tot, Ny_tot, Nz_tot) .* background_map; medium.sound_speed = sound_speed_map(:,:, :); medium.density = density_map(:, :, :); % % ========================================================================= % % RUN THE SIMULATION % % ========================================================================= % set the input settings input_args = {... 'PMLInside', false, 'PMLSize', [pml_x_size, pml_y_size, pml_z_size], ... 'PMLAlpha', [pml_x_alpha, pml_y_alpha, pml_z_alpha], 'DataCast', DATA_CAST, 'DataRecast', true, 'PlotSim', false}; % save input files to disk % run the simulation [sensor_data] = kspaceFirstOrder3DG(kgrid, medium, transducer, transducer, input_args{:}); scan_lines = transducer.combine_sensor_data(sensor_data) ; % save sensor_data % It would be really helpful if you can share your insights. Thanks in advance, Farnaz Farnaz on "Plane wave simulation, commercial probe" http://www.k-wave.org/forum/topic/plane-wave-simulation-commercial-probe#post-6850 Tue, 14 May 2019 12:41:39 +0000 Farnaz 6850@http://www.k-wave.org/forum/ Hi, I am new to k-wave and have some questions, I'm trying to simulate a commercial ultrasound transducer with 128 elements which 64 elements of it are activated in a plane wave scenario. 1- If I set the focus point to inf (or very large values) when I am defining transducer parameters will I have plane wave? If not, how I can simulate plane wave for a commercial probe then? 2- Based on the pitch and kerf parameters I need at least 1024 grid points in x and y direction (only for the transducer probe face)(3.75 cm probe 4 grid point for piezo and extra 4 grid points for kerf). The whole medium is then 4.24*4.24 cm and the probe plane is located just outside the PML. The same transducer will be used as sensor to record the data. As the transducer class only works in 3D mode, these many grid points are computationally expensive, what happens if I set Nz to a very small number while maintaining dz=dx=dy (say Nz = 16,32, etc...)? 3- If this is not possible to do what are your suggestions for my simulation? Should I use 2D simulation and set everything manually instead of using transducer class? (BTW i'm using 3D c++ version for CUDA but I run it in MATLAB (I have a GeForce GTX 1080 Ti))