Modeling and Software Synthesis for Multiprocessor Implementation of Wireless Communication Systems

You are here

Inside Signal Processing Newsletter Home Page

Top Reasons to Join SPS Today!

1. IEEE Signal Processing Magazine
2. Signal Processing Digital Library*
3. Inside Signal Processing Newsletter
4. SPS Resource Center
5. Career advancement & recognition
6. Discounts on conferences and publications
7. Professional networking
8. Communities for students, young professionals, and women
9. Volunteer opportunities
10. Coming soon! PDH/CEU credits
Click here to learn more.

News and Resources for Members of the IEEE Signal Processing Society

Modeling and Software Synthesis for Multiprocessor Implementation of Wireless Communication Systems

By: 
Lin, Shuoxin. University of Maryland, College Park

Advisor: Bhattacharyya, Shuvra S.

In recent years, the complexity of designing embedded signal processing systems for wireless communications has increased significantly based on the need to support increasing levels of operational flexibility and adaptivity, while also supporting increasing data rates and bandwidths. These trends pose important design and implementation challenges to meet the required demands on communication system performance, real-time operation, energy efficiency, and reconfigurability. Dataflow models of computation provide a useful framework that can be built upon to address these challenges.

Dataflow models provide high-level abstractions for specifying, analyzing and implementing a wide range of embedded signal processing applications. They allow designers to specify an application using high-level, platform-independent representations, and synthesize optimized embedded software that is targeted to specific types of hardware resources and design constraints.

The growing complexity of wireless communication systems, as motivated above, along with the complexity of system-on-chip platforms for embedded signal processing result in new problems that must be addressed in developing effective dataflow-based design methodologies. First, significant improvements to dataflow-based models and methods are needed to effectively utilize heterogeneous computing platforms and multiple forms of parallelism under stringent constraints on real-time performance and energy consumption. Second, effective modeling and analysis methods for handling dynamic parameters within dataflow graph components are needed for reliable and efficient management of system-level adaptivity and reconfiguration.

In this thesis, the authors address these problems by developing an integrated framework that exploits pipeline, data and task-level parallelism in dataflow models under memory constraints, and proposing novel dataflow modeling concepts and performance optimization techniques for design and implementation of dynamically parameterized communication systems. The main contributions of the thesis are summarized as follows:

(1) Software synthesis framework for heterogeneous signal processing platforms. The authors have developed an integrated dataflow-based design framework called DIF-GPU, which provides a toolset for specification, optimization and software synthesis of embedded software targeted to heterogeneous CPU-GPU platforms. DIF-GPU incorporates novel models and methods in the dataflow interchange format (DIF) that are geared toward design optimization of signal processing systems on heterogeneous architectures composed of multicore CPUs and GPUs. DIF-GPU helps to free developers from low-level, platform-specific fine-tuning, and allows them to focus on higher-level aspects of communication system design.

(2) Vectorization in DIF-GPU. In the context of dataflow models for embedded signal processing, vectorization is an important transformation for exploiting data parallelism. The authors have developed new techniques for integrated dataflow graph vectorization and scheduling on heterogeneous platforms. These techniques are developed in the DIF-GPU framework to provide optimized vectorization and scheduling capabilities for hybrid CPU-GPU platforms under memory constraints. For the targeted class of platforms, these techniques are shown to provide significantly better processing throughput compared to previous methods for a given memory constraint. The authors demonstrate their integrated vectorization and scheduling techniques by applying them to an Orthogonal Frequency Division Multiplexing (OFDM) receiver system.

(3) Modeling parameterized, dynamic dataflow behavior. The authors introduce a novel modeling method, called parameterized sets of modes (PSMs), that enables efficient representation and analysis of adaptive and dynamically reconfigurable signal processing functionality. PSMs can be viewed as high-level abstractions that model parameterized functionality involving groups of related regimes of operation ("modes") for dynamic dataflow models. The authors develop formal foundations for PSM-based modeling, and demonstrate the utility of this form of modeling by using it to develop efficient methods for scheduling dynamically parameterized dataflow graphs on different types of relevant platforms.

SPS on Twitter

  • DEADLINE EXTENDED: The 2023 IEEE International Workshop on Machine Learning for Signal Processing is now accepting… https://t.co/NLH2u19a3y
  • ONE MONTH OUT! We are celebrating the inaugural SPS Day on 2 June, honoring the date the Society was established in… https://t.co/V6Z3wKGK1O
  • The new SPS Scholarship Program welcomes applications from students interested in pursuing signal processing educat… https://t.co/0aYPMDSWDj
  • CALL FOR PAPERS: The IEEE Journal of Selected Topics in Signal Processing is now seeking submissions for a Special… https://t.co/NPCGrSjQbh
  • Test your knowledge of signal processing history with our April trivia! Our 75th anniversary celebration continues:… https://t.co/4xal7voFER

IEEE SPS Educational Resources

IEEE SPS Resource Center

IEEE SPS YouTube Channel