Embedded System Design Optimization and Adaptation using Compact System-Level Models

Maryland DSPCAD Research Group
Project Webpage


In this project, referred to as the Compact System-level Models (CSM) Project, we are developing new techniques to help advanced computing systems for signal processing better adapt to the environments in which they operate. This project is important because signal processing is everywhere (cell phones, computer networks, manufacturing systems, agriculture, etc.). Adapting to the environment helps these systems to operate more reliably by, for example, adapting to changing radio interference or the challenging radio environments presented by clusters of tall buildings. Many of these communication systems are also battery-operated or must run on limited energy; adapting to their operating environments helps to reduce their energy consumption and improve battery life. These techniques are particularly useful for cognitive radio, an emerging technology that allows devices for wireless communication (such as cell phones) to more efficiently use radio spectrum.

This project is developing new methods for creating software that can be reconfigured at run time. Typical software is created to operate in a particular mode; changing the software's operating conditions requires redesigning the software itself. New mathematical models and algorithms will allow system designers to create software that is designed to adapt itself dynamically to its environment. The project is addressing both models specifying the behavior of the software and for translating that specification into an efficient implementation.

The project is a collaboration among researchers at Georgia Institute of Technology, USA; Institut National des Sciences Appliquées (INSA) de Rennes, France; National Chiao Tung University, Taiwan; and University of Maryland at College Park, USA. The collaborators in Georgia, France, Taiwan, and Maryland provide complementary expertise in areas that include cyber-physical systems, cognitive radio algorithms, model-based design, and embedded signal processing. The collaboration also provides valuable international research experience for the Ph.D. student researchers involved in the project.




The project has produced a number of educational resources, including the following.

  • We published two educational videos on YouTube:
  • We delivered a Tutorial on Design for Low-Power Internet-of-Things (IoT) Systems at the 2018 International Symposium on Circuits and Systems in Florence, Italy. The tutorial was delivered jointly with Prof. Francesca Palumbo of Università degli Studi di Sassari, Italy, and Prof. Jarmo Takala of Tampere University of Technology, Finland. The tutorial consisted of four modules. The slides for these modules are available here:
  • We delivered a Tutorial on Machine Learning Methods for Performance/Power/Thermal Optimization of Signal Processing Systems at the 2018 IEEE Workshop on Signal Processing Systems in Cape Town, South Africa. The slides from the tutorial are available here: SIPS 2018 Tutorial Slides.

    This section summarizes software that has been developed in the project, and is available for download for experimentation and extension by other researchers. The documentation for this software is evolving. Questions on how to install, use or extend the software can be sent to: The DSPCAD-Help Mailing List.

    The CSM Package

    CSM is a software package for experimenting with Compact System Modeling in terms of Markov Decision Processes (MDPs). The CSM package includes an example of MDP-based design and implementation of an adaptive digital predistortion (DPD) system for wireless communications. More details on the MDP-based DPD techniques on which this MDP-based DPD system is based can be found in the following publication:

    [Li 2019] L. Li, P. Deaville, A. Sapio, L. Anttila, M. Valkama, M. Wolf, and S. S. Bhattacharyya. MADS: A framework for design and implementation of adaptive digital predistortion systems. IEEE Journal on Emerging and Selected Topics in Circuits and Systems, 9(4):712-722, December 2019.

    Another key component within the CSM package is GEMBench, which is a benchmarking tool for evaluating implementations of solvers for Markov Decision Processes (MDPs). GEMBench stands for the Gpu-accelerated Embedded Mdp testBench. GEMBench is targeted to a specific embedded GPU platform, the NVIDIA Jetson platform, and is designed for future retargetability to other platforms. GEMBench is a novel open source software package that is intended to run on the target platform. The package contains libraries of MDP solvers, parsers, datasets and reference solutions, which provide a comprehensive infrastructure for understanding trade-offs among existing embedded MDP techniques, and experimenting with novel techniques.

    More details about GEMBench can be found in the following publication:

    [Sapio 2019] A. Sapio, R. Tatiefo, S. Bhattacharyya, and M. Wolf. GEMBench: A platform for collaborative development of GPU accelerated embedded Markov decision systems. In Proceedings of the International Conference on Embedded Computer Systems: Architectures, Modeling, and Simulation, pages 294-308, Samos, Greece, July 2019.

    The CSM Package is available for download from the CSM Release Repository. Instructions on installing and getting started with the package are in the README.txt file in the top-level directory of the package.

    The DIF-DSG Sub-Package

    DIF-DSG is a sub-package of the Dataflow Interchange Format (DIF) Package that provides prototype software tools for design and synthesis using the dataflow schedule graph (DSG) model. DSGs provide a formal abstraction for representing schedules in dataflow-based design processes. The DSG abstraction allows designers to model a schedule as a separate dataflow graph, thereby providing a formal, abstract (platform- and language-independent) representation for the schedule.

    More details about the DSG model can be found in the following publications:

    [Lee 2021] K. Lee, Y. Lee, A. Raina, Y. Liu, J. Wu, C. Defrancisci, B. Riggan, and S. S. Bhattacharyya. Software synthesis from dataflow schedule graphs. SN Applied Sciences, 3(2):1-20, January 2021. Article No. 142.

    [Wu 2011] H. Wu, C. Shen, N. Sane, W. Plishker, and S. S. Bhattacharyya. A model-based schedule representation for heterogeneous mapping of dataflow graphs. In Proceedings of the International Heterogeneity in Computing Workshop, pages 66-77, Anchorage, Alaska, May 2011.

    The DIF-DSG Sub-Package is available as part of the DIF Package, which can be downloaded from the DIF Release Repository. Instructions on installing and getting started with the package are in the README.txt file in the top-level directory of the package.


    A list of publications from the CSM project can be found on the CSM Project Publications Page.


    This research has been supported in part by the Computer and Network Systems Program of the U.S. National Science Foundation under Grant No. CNS1514425 (University of Maryland at College Park), and CNS2002853 (University of Nebraska-Lincoln).


    This webpage was last updated on 10/28/2021.