Skip to content

Ultrasound Beamforming with MATLAB GPU Coder

Authors: Holoscan Team (NVIDIA), MathWorks Team (MathWorks)
Supported platforms: x86_64, aarch64
Last modified: March 18, 2025
Language: C++
Latest version: 1.0.0
Minimum Holoscan SDK version: 1.0.3
Tested Holoscan SDK versions: 1.0.3
Contribution metric: Level 4 - Experimental

This application does real-time ultrasound beamforming of simulated data. The beamforming algorithm is implemented in MATLAB and converted to CUDA using MATLAB GPU Coder. When the application is run, Holoviz will display the beamformed data in real time.

Folder Structure

matlab_beamform
├── data  # Data is generated with generate_data.mlx
│   └── ultrasound_beamforming.bin  # Simulated ultrasound data
├── matlab  # MATLAB files
│   ├── generate_beamform_jetson.m  # MATLAB script to generate CUDA DLLs on Jetson
│   ├── generate_beamform_x86.m  # MATLAB script to generate CUDA DLLs on x86
│   ├── generate_data.mlx  # MATLAB script to generate simulated data
│   └── matlab_beamform.m  # MATLAB function that CUDA code is generated from
├── CMakeLists.txt  # CMake build file
├── main.cpp  # Ultrasound beamforming app
└── matlab_beamform.yaml  # Ultrasound beamforming config

Generate Simulated Data

The required MATLAB Toolboxes are: * Phased Array System Toolbox * Communications Toolbox

Simply run the script matlab/generate_data.mlx from MATLAB and a binary file ultrasound_beamforming.bin will be written to a top-level data folder. The binary file contains the simulated ultrasound data, prior to beamforming.

Generate CUDA Code with MATLAB GPU Coder

x86: Ubuntu

In order to generate the CUDA Code, start MATLAB and cd to the matlab folder and open the generate_beamform_x86.m script. Run the script and a folder codegen/dll/matlab_beamform will be generated in the matlab_beamform folder.

arm64: Jetson

On an x86 computer with MATLAB installed, cd to the matlab folder and open the generate_beamform_jetson.m script. Having an ssh connection to the Jetson device you want to build the CUDA DLLs on, specify the parameters of that connection in the hwobj on line 7, also replace <ABSOLUTE_PATH> of cfg.Hardware.BuildDir on line 39, as the absolute path (on the Jetson device) to holohub folder. Run the script and a folder MATLAB_ws will be created in the matlab_beamform folder.

Configure Holoscan for MATLAB

If you have not already, start by building HoloHub: 

./dev_container build

x86: Ubuntu

Define the environment variable: 

export MATLAB_ROOT="/usr/local/MATLAB"
export MATLAB_VERSION="R2023b"
where you, if need be, replace MATLAB_ROOT with the location of your MATLAB install and MATLAB_VERSION with the correct version.

Next, run the HoloHub Docker container: 

./dev_container launch \
    --add-volume ${MATLAB_ROOT}/${MATLAB_VERSION} \
    --docker_opts "-e MATLAB_ROOT=/workspace/volumes/${MATLAB_VERSION}"

arm64: Jetson

The folder MATLAB_ws, created by MATLAB, mirrors the folder structure of the host machine and is therefore different from one user to another; hence, we need to specify the path to the codegen folder in the CMakeLists.txt, in order for the build to find the required libraries. Set the variable REL_PTH_MATLAB_CODEGEN to the relative path where the codegen folder is located in the MATLAB_ws folder. For example, if GPU Coder created the following folder structure on the Jetson device: 

matlab_beamform
└── MATLAB_ws
    └── R2023b
        └── C
            └── Users
                └── Jensen
                    └── holohub
                        └── applications
                            └── matlab_gpu_coder
                                └── matlab_beamform
                                    └── matlab
                                        └── codegen
the variable should be set as: 
REL_PTH_MATLAB_CODEGEN=MATLAB_ws/R2023b/C/Users/Jensen/holohub/applications/matlab_gpu_coder/matlab_beamform/matlab/codegen

Next, run the HoloHub Docker container: 

./dev_container launch