Foresight beats out hindsight ... but only in the case of model-based design: Interview with Dr. Jon Friedman, The MathWorks
EDITOR'S NOTE: It doesn’t matter which type of ware is being viewed. The purpose of modeling is always the same: The visionary gets to see the possibilities – before any investment is made. And the same applies to military embedded technology, according to The MathWorks’ Dr. Jon Friedman: Rather than use the same old Line Replaceable Unit (LRU) approach where snafus are found near the end of the design road, model based design enables engineers to “virtually” see the end from the beginning – and eliminate those costly late-game redesigns. We also found out some of the company’s hot new focal points and why they are – or will be – important to the defense industry.
MIL EMBEDDED: Your products run the gamut across use in systems engineering, algorithm development, code analysis, and test and measure-ment. Can you briefly summarize your focus areas for defense and aerospace markets?
FRIEDMAN: The MathWorks supports engineers in the aerospace and defense markets across a wide range of applications including aircraft control systems, land systems, and spacecraft as well as advanced communication systems and new sensor and signal processing applications. These focal points are facilitated primarily with two platform products: MATLAB and Simulink.
MIL EMBEDDED: OK, can you tell us a little about MATLAB first, then Simulink?
FRIEDMAN: Certainly. MATLAB is a platform for technical computing. To put the term “technical computing” into context, think about engineering activities like data analysis and visualization, test data and signal creation/synthesis, algorithm development/synthesis, and algorithm testing/analysis. Engineers developing sensors to capture images need to characterize and calibrate the sensor before they can create the algorithms that use the images for intelligence, surveillance, or reconnaissance work. To accomplish this task, they collect and analyze data from the sensor. They will then build a mathematical model of the sensor’s dynamics. Using this model, the engineers can then determine the optimal calibration for the sensor.
The other product, Simulink, is the platform for model-based design. In the past, using a traditional Line Replaceable Unit (LRU) design process, engineers gather requirements from several sources, which are combined to create a paper specification and often take up numerous binders on an engineer’s desk. An acceptable design is eventually achieved, then handed off to another team that performs verification and validation testing. Because testing occurs at the end of the design process, errors that are introduced throughout the design process are often found late, making them expensive to fix. In contrast, model-based design such as that offered by Simulink, begins with the creation of an executable specification that can be linked to the original requirements, providing two-way traceability between the design and the requirements.
MIL EMBEDDED: Though many systems engineers recognize the benefits offered by MATLAB and Simulink, you’re also encouraging more software programmers to use the tools and underlying data. What are the use cases and benefits?
FRIEDMAN: In the development of complex systems, engineers often focus on one specific area, such as controls algorithms, communications designs, or software programming. The outputs of these groups must ultimately meet the same requirements. However, when these groups are operating in completely different environments, there are many opportunities for designs to diverge. Model-based design such as that offered by MATLAB and Simulink enables engineers to collaborate early in the development cycle and to reuse models in different stages of design. Not only is the end design better optimized for functionality and performance, early-stage design iterations are also significantly less expensive than those made after code has been completed.
MIL EMBEDDED: How can your modeling (MATLAB and Simulink) and code analysis/verification (PolySpace) products work together for development teams?
FRIEDMAN: Model-based design bridges multiple disciplines, from controls to communications and from system engineering to testing. This enables teams to easily share requirements, models, data, and test cases. With model-based design, development teams use the same tool platform to incrementally design and test, building from requirements definition all the way to integration testing. For example, the team can model requirements and functional design in MATLAB and Simulink, produce code using Real-Time Workshop Embedded Coder, and test the code using PolySpace and Simulink verification and validation tools.
MIL EMBEDDED: Can you share an example of a current defense program where this more integrated development approach is making a big difference, and highlight the before and after?
FRIEDMAN: In 2004, BAE Systems was tasked with developing a military standard (MIL-STD-188-165A) satellite communications waveform for implementation in a Command, Control, Communications, Computers, Intelligence, Surveillance, and Reconnaissance (C4ISR) radio. At the same time, BAE Systems sought to evaluate a new design flow for reducing development time. Working together, BAE and Xilinx ran two projects in parallel, one using BAE Systems’ traditional design flow, which relied on hand-coding VHDL. The other project utilized a model-based design flow involving automatic code generation using The MathWorks and Xilinx tools. To ensure a fair comparison, each effort used an equivalent set of IP cores.
MIL EMBEDDED: So what were the results of the two differing design flow methodologies?
FRIEDMAN: Dr. David Haessig, manager of Waveform Products at BAE Systems, summed it up best when he said, “With the model-based approach, we developed a common model of the waveform, which was used for performance simulation, operational debugging, and code generation. As a result, we demonstrated more than a 10-to-1 reduction in the time to develop the signal processing chain of a Software-Defined Radio. This really illustrates the potential for improving productivity in SDR applications.”
The most impactful result was a reduction of more than 4-to-1 in overall project time including hardware integration and lab testing. Also, with model-based design, the Simulink model was directly connected to the resulting code, which forced the developer to capture all of the required waveform details in the model leading to earlier discovery. This meant bugs were discovered and removed early in the design flow – at the modeling stage, not later at the VHDL behavioral testing stage where they can be difficult and time-consuming to fix. Another big advantage with Simulink and System Generator was that the necessary clocking signals were generated automatically, and components were connected easily, reducing the need to study the data sheet for details concerning control, timing, and other options.
MIL EMBEDDED: What are some other hot areas for The MathWorks right now, and why are they important to the defense industry’s future?
FRIEDMAN: I’d have to say avionics safety standards and also sensor systems.
Regarding avionics safety standards, all commercial aircraft software and electronics must be certified as compliant with the DO-178B and DO-254 safety specifications to have authority to fly in commercial airspace. Many defense suppliers are also required to certify military systems or follow processes compliant to these standards for applications such as Unmanned Aerial Vehicles (UAVs). Estimates predict that conformance to DO-178B adds 50 to 200 percent to software development costs. However, model-based design is one way to reduce costs by integrating verification activities earlier in the design process to ensure conformance to standards. These tools can aid designers in four key areas: traceability, requirements validation, verification, and conformance.
As far as sensor systems are concerned, as you know, they are at the heart of many new defense programs and upgrades to existing military equipment. Unmanned systems rely on video and infrared sensors for navigation and tracking. Anti-IED technologies often employ a suite of sensors to help detect threats. These sensor systems can vary in terms of the types of signals being analyzed (RF, optical, infrared, or acoustic), the types of analysis being performed (spectral or object detection), and the platforms upon which they run (DSP, FPGA, or ASIC). However, most development activities fall into three categories: understanding and characterizing the sensor itself, processing data read from the sensor unit, and performing higher-level analysis. So we design our tools to help engineers throughout this process.
MIL EMBEDDED: Where is the next frontier for The MathWorks in developing tools for defense software development teams?
FRIEDMAN: One area of focus for the company is continuing to build capabilities to support verification and validation activities. Using MATLAB and Simulink to design and verify early in the process is well established. The MathWorks is focusing on developing tools to help designers use models to help them in the other verification and validation tasks with which they are faced. Examples include leveraging parallel computing capabilities for executing large numbers of simulations, complementing simulation with property proving on models and code, and using formal analysis to help in automatic test case generation.
The MathWorks 508-647-7000 www.mathworks.com