Next: 4 Top Level command Program Architecture Up: Appnotes Index Previous:2 Introduction

RASSP Autocoding for DSP Control Application Note

3.0 What is a Command Program

A typical large-scale embedded signal processing architecture consists of a signal processing system communicating with a command and control system. The signal processing system performs the high-bandwidth numerical processing and typically executes on Digital Signal Processors (DSPs). It is controlled by a command program (CP) that is intimately related to the signal processing system. The command program serves as an interface between the signal processing program and the rest of the command and control system. This is done by translating system-derived or user provided inputs into commands understood by the signal processing system. The results processed by the signal processing system are then fed back to the command and control system. The command program serves as an interface between the signal processing program and the rest of the command and control system. This is done by translating system-derived or user provided inputs into commands understood by the signal processing system. The results processed by the signal processing system are then fed back to the command and control system. Figure 3 - 1, A Typical Large Scale Aerospace Architecture, shows the command program ambiguously residing in both the Signal Processing System and the Command and Control System. The generic controls over the signal processing program all reflect the graphical structure of that program, and thus the CP is frequently conceptualized as a part of the Signal Processing System. But this is only half of the story. The CP receives commands and forwards results to the Command and Control System, which is unaware of the graph structure of the Signal Processing System. This communication between the CP and the rest of the Command and Control System (CCS) occurs in the "language" of the CCS which usually is state oriented. For example the CCS may want to place an Airborne Radar Signal

Figure 3 - 1: A Typical Large Scale Aerospace Architecutre

Processing System into the Air-to-Ground mapping state. Since the CP also speaks this state oriented language it may be viewed as part of the CCS. Control a Signal Processing Program that has been graphically specified via a dataflow model of computation and autocoded, a generic set of controls over the signal processing graphs are made available to the CP. They define the complete set of commands understood by the signal processing program. (See the Data Flow Graph application note for a complete discussion of these capabilities.) The abilities provided fall into several broad categories:

• Graph instantiation - comprises loading signal processing program executables onto their associated suite of processors as well as obtaining handles for parameter and queue objects.

• Runtime program to processor mapping - maps SPP executable programs to DSP processor.

• Execution Control - commands that allow the CP to control the execution state of the signal processing programs. The SPP execute asynchronously to the CP.

• Data flow control (limited) - includes control of the standard process states such as run/stop/reset, as well as the ability to control the flow of input data into the signal processing graph. Because of the data-flow paradigm of the graph, this determines whether a "running" graph is stalled or executing,

• Write access to graph parameters - The CP is responsible for writing graph parameter data to the signal processor. Since graph execution is asynchronous to the CP, the CP must carefully control the graph execution state before writing data via graph parameters.

• Read/Write access to graph queues connected to the CP - allows CP to asynchronously read data from graph or write data to a graph.

Depending on real-time conditions, some data-flow may need to be rerouted either within a graph or among distinct graphs and is therefore also under the control of the CP. It is important to emphasize that most data-flow is not under control of the CP. That which is under CP control is used to perform infrequent, significant graph reconfigurations. The CP may manage data-flow within a graph by simply setting parameters controlling "switches" in the graph. Data flowing between distinct graph instances may be managed by connecting and disconnecting dynamic queues serving as pipes between the graphs. The CP can also be connected via dynamic queues to the signal processing program. Typically the data read/written on these queues is low rate.

Next: 4 Application Example Up: Appnotes Index Previous:2 Introduction