An Intro

  • Category: Theater
  • Words: 632
  • Grade: 100

Signal processing systems typically consist of both control and data flow parts. The control flow often consists of reactive control that implements higher protocol levels and flow control that governs the operation of the data flow parts. The data flow parts are used to model the computationally intensive parts that are most efficiently described as operations on infinite streams of data samples.

Data flow systems [2] are described as networks of processes performing the signal processing and signals connecting the ports of those processes. Static data flow requires the processes to consume and produce fixed numbers of samples (called rate) at their ports upon each activation. Dynamic data flow denotes the case when data rates can be computed dynamically during runtime.

In a variety of cases control and data flow parts can neither be specified nor analyzed separately since they are interacting in a tight loop. Packet-based transmission systems are an example for this class of applications. We refer to these as mixed control/data flow systems.

Abbildung 1: toplevel FSM

We will use the example of a wireless modem for low rate data communications [3] to exemplify this situation. On a high level of abstraction the modem functionality can be described as a hierarchical FSM. Its toplevel state diagram is depicted in figure 1. After the modem has been activated by an external event it enters an initial synchronization mode ( INIT) where the frequency hopping pattern and further transmission and protocol parameters are determined. Depending on these the communication continues in the BFSK or 4FSK mode.

Abbildung 2: modem operation in BFSK mode

Figure 2 shows the state diagram for the BFSK mode. After an initial delay, a timer is set and frequency estimation is performed. Next, the timing offset is estimated and packet synchronization is started. The packet synchronization can be cancelled by a timer in which case the modem continues with the timing preamble detection (after changing the frequency hopping cell and restarting the timer).

For the purpose of validation and system performance analysis (e.g. to determine the packet error rate) a sufficiently large number of packets - resulting in an even longer sequence of signal processing states - has to be simulated. Functional aspects to be modeled are

the existence of multiple (processing-) states
event driven state transition
generation of events by control and data-flow subsystems
while in state X: perform F(X)
upon state transition X Y: execute the procedure P(X,Y)
generate data in state X, consume it in state Y
Thus, one can neither replace the data flow part with a simplified (statistical) model nor can one assume the state of the control flow portions to be static.
The modeling technique used to specify and simulate such systems should fulfill a set of additional requirements related to efficiency considerations - efficiency of modeling (ease of use, expressive power), of analysis (compile time analysis, simulation speed), and implementation (quality of the link to implementation). These are

The modeling technique should be sufficiently abstract to model the functionality independent of a specific implementation.
It has to allow a modular approach to system design to support partioning a given problem into pieces of manageable size.
The respective modules should model the functionality indepent of the context they are used in to ensure good reusability.
A necessary precondition to be able to model and analyze functionality independent of a special implementation is that the specifications need to have determinate behaviour.
In the following section we will first review existing approaches to modeling and simulation of heterogeneous systems and describe their limitations. Next, we will introduce a new computational model named process coordination calculus (PCC). Then, we will treat the concept of scheduling constraints in greater depth and present an example of the expressiveness of PCC specifications.

ad 4
Copyright 2011 All Rights Reserved