Integrated Modeling Using Finite State Machines and Dataflow Graphs - Signal Processing Systems - page 958

dataflow graphs seem to be the best choice to establish a design methodology
upon it. One of the key problems is due to the fact that dataflow actors often
show control dominant behavior which cannot adequately be expressed in many
dataflow models. Consequently, different modeling approaches integrating finite
state machines ( FSMs ) with dataflow graphs have been proposed in the past.
However, this often comes with the drawback of decreased analysis capabilities.
Nevertheless, several successful academic system-level design methodologies have
emerged recently based on integrated modeling approaches.
In this chapter, we will present different integrated modeling approaches (Sect. 2 ) .
Section 3 is devoted to the problem of identifying static dataflow actors within a
general dataflow graph. Different design methodologies, namely Ptolemy II [ 28 ] ,
OpenDF [ 4 ] , and SystemCoDesigner [ 27 ] , will be presented in Sect. 4 . We will
focus particularly on the analyzability of the application model and the exploitation
of these analysis capabilities in the corresponding design methodology. In Sect. 5 ,
clustering approaches for static dataflow graphs embedded in more general dataflow
graphs will be presented. Finally, in Sect. 6 , we recap the important points of this
chapter.

2 Modeling Approaches

In this section, we will discuss modeling approaches combining dataflow graphs
with Finite State Machines ( FSMs ). Starting with the recapitulation of some
fundamental dataflow models like Synchronous Dataflow ( SDF ) [ 29 ] and Dynamic
Dataflow ( DDF ), the *charts (pronounced “star charts”) [ 18 ] formalism is intro-
duced. The *charts formalism was one of the first integrated modeling approaches
implemented in the Ptolemy project [ 12 ] . Furthermore, we will compare enable-
invoke [ 34 ] and core functional dataflow [ 34 ] with the *charts approach. Later on,
the CAL Actor Language ( CAL )[ 11 ] , Extended CoDesign Finite State Machines
( ECFSMs ) [ 35 ] , and S YSTE M O C [ 13 ] will be discussed.

2.1 Dataflow Graphs

In synchronous dataflow ( SDF ) [ 29 ] graphs, the firing of an actor is an atomic
computation that consumes a fixed number of tokens from each incoming edge and
produces a fixed number of tokens on each outgoing edge. The most restrictive
subclass of SDF graphs are Homogeneous SDF ( HSDF ) graphs, where each actor
exactly consumes and produces a single token from each incoming and on each
outgoing edge, respectively. Edges are conceptionally unbounded FIFO queues
representing a possibly infinite stream of data. The consumption and production
rates can be used to unambiguously define an iteration that returns the queues to