Background
The demands on the signal processing in high-end sensor systems are rising rapidly. The systems are based on digital antenna arrays with thousands of antenna elements, and they operate using advanced adaptive signal processing algorithms. Since the processing typically is embedded, with the constraints that follow, high processing efficiency is important. The competition on the future markets will require the ability to quickly innovate and develop new sensor functions. There is therefore also a focus put on efficient application development, i.e., engineering efficiency.
GPUs are in many respects a good fit for this class of data and performance-intensive class of applications. The devices often have impressive peak performance figures. The question is, however, how much usable performance we can achieve when executing our applications. At one end of the scale we have very performance demanding applications that shall be executed on powerful GPUs. At the other end we have a less performance demanding applications that shall be executed on small low-power GPUs. Common for all cases is that we want to maximize the utilization of the devices.
Description of the master thesis
The main task of this thesis work is to investigate how to maximize the utilization of GPUs for sensor signal processing. This requires a very good interplay between the algorithms and the computational architecture. To achieve this, how much effort and detailed GPU architecture knowledge is required from the application developer? How much support do we get from development software?
For the investigation we have a multi-channel radar signal processing benchmark. It is a scalable streaming application with a sequence of filters that process (multi-dimensional) matrices of data. The processing in the filters may be carried out along different directions of the matrices. The benchmark will be implemented in CUDA.
The expected results include answers on:
- The level of architecture detail that must be understood in order to truly utilize the GPU
- Required additional programming effort for going from “high performance” to “very high performance”
- How scaling of the applications affects the utilization
- Level of support that we have from the software stack in the mapping/programming
- Achieved GPU utilization for different types of filters/algorithms
- Performance results: Processing throughput and latency
Your profile
We are looking for 1-2 master degree students with an interest in:
- C/C++ and CUDA
- Parallel computing
- Signal processing
You are at the end of your master’s degree in Computer engineering, Electrical engineering, or equivalent, and is eligble for your 30 HP degree project.
This position requires that you will be approved in a security screening in accordance with the Swedish Protective Security Act.
What you will be a part of
You will collaborate with experienced engineers and professionals in an environment that fosters career development and personal growth. You will be part of a unit working with Software design for radar systems.
Surveillance, a Business Area within Saab, is a world-leading supplier of systems for detection of threats and self-protection. Business Unit Radar Solutions is responsible for Radar in airborne-, surface- and naval systems.
Last application day
2023-12-16
Contact information
Christer Åkerblom, Recruiting Manager
christer.akerblom@saabgroup.com
Anders Åhlander, Technical Specialist
073-437 9811
Saab is a company with a strong people-orientation. We offer a friendly work environment where we support and help each other to be at our best. Continuous learning, career & talent development and employee well-being are examples of areas where we always put the strongest effort to offer great opportunities.