Modeling of Block-Based DSP Systems.pdfVIP

MODELING OF BLOCK-BASED DSP SYSTEMS Dong-Ik Ko and Shuvra S. Bhattacharyya Department of Electrical and Computer Engineering, and Institute for Advanced Computer Studies University of Maryland, College Park, 20742, USA ABSTRACT Modeling semantics based on dataflow graphs are used widely in design tools for digital signal processing (DSP). This paper devel- ops efficient techniques for representing and manipulating block- based operations in dataflow-based DSP design tools. In this con- text, a block refers to a finite-length sequence of data items, such as a sequence of speech samples, an image, or a group of video frames, as part of an enclosing data stream. We develop in this paper a meta-modeling technique called blocked dataflow (BLDF) for augmenting DSP design tools with more effective blocked data support in an efficient and general manner. We compare BLDF against alternative modeling approaches through a detailed case study of an MPEG 2 video encoder system. 1. INTRODUCTION In the digital signal processing (DSP) domain, rapid prototyping tools based on coarse-grain dataflow semantics are widely used [2]. One important requirement in these tools is support for block- based processing, such as that involved in image and video appli- cations. We develop in this paper a blocked dataflow (BLDF) modeling approach for efficient handling of block-based data in dataflow-based DSP design tools. BLDF combines meta-model- ing, block-based processing, multidimensional representation, and dynamic parameter reconfiguration in a single, unified framework that leads to more efficient dataflow graphs for scheduling and software synthesis. In this paper, by a dataflow model of computation (dataflow MoC), we mean a programming model based on dataflow seman- tics. Programs in a dataflow MoC are thus represented as directed graphs in which vertices, called dataflow actors, represent compu- tational tasks, and edges represent logical FIFO communication channels between tasks. A d