SWAP Beat: Toolkits for FPGA development

15This is the first of a planned series of columns by Robert Hoyecki, Director of Advanced Multi-Computing at Curtiss-Wright Controls Embedded Computing. Rob will discuss a range of signal processing and FPGA applications to help mil-aero developers introduce products more quickly while trimming recurring costs.

Vital for defense and aerospace, signal processing gives the cutting edge to sensor performance and target classification, voice and data communications, intelligence gathering, and security across the entire battle space electro-magnetic spectrum. This is a market that in many instances has severe environmental constraints and is continually seeking to make Space, Weight, and Power (SWAP) savings while increasing the complexity and performance of its embedded systems. A simple sensor such as a GPS receiver might be wholly integrated into an embedded system as an I/O module. Or the sensor could require a complex, real-time, heterogeneous multicomputing system if it is being used in an airborne SAR radar or signals intelligence application. No matter the system architecture, using FPGAs achieves a host of undeniable benefits, but realizing these benefits is not without (sometimes painful) trade-offs.

Unless the sensor is truly stand-alone, it will form part of a larger system that uses the incoming sensor data within the context of a weapons platform, such as an armored vehicle or combat aircraft. A signal processing application then becomes one element of a much larger development problem with two major parts. The DSP algorithms make up one part of the challenge. Engineers often focus their attention on these algorithms when selecting the overall sensor architecture and FPGA device. It is relatively simple to visualize, simulate, and scale the problem to the proposed sensor application using high-level tools such as The MathWorks' Simulink and Xilinx's System Generator. The second, and perhaps more important part of the problem, is how the FPGA design will deal with the following functions:

n Data acquisition

n Buffering

n Internal/external memory control

n Bandwidth of data paths

n Dissemination of results (DMA engines)

n Timing and synchronization

n Error recovery

n Fault reporting

n Instrumentation for test and verification

n Proof of operation over the platform's temperature extremes

Selecting the best architecture, whether based on an FPGA, general-purpose processors, or a mix of both types, involves a series of commercial and technology benefit trade-offs. Much more complexity comes into play with this choice than with deciding whether the DSP algorithm will fit the preferred FPGA device. Developers must consider time-to-market, the availability of the necessary in-house skills, the cost of training, the level of risk, and the potential return on investment. ROI might be measured in space, weight, or power saved, reduced recurring cost, selling in greater quantities, or enhanced competitive differentiation. However, in the defense and aerospace market, small production volumes highlight the need to keep development cost and risk to a minimum. As a result, developers are increasingly turning to off-the-shelf vendors to get a head start with hardware modules, libraries, and tools to save time and reduce risk.

COTS vendors typically offer a range of FPGA products for embedded computing systems, from a basic FPGA mezzanine module to complement a motherboard or single board computer to multiple FPGAs connected by high-performance switched fabrics to multi-computing engines. The tools available from these COTS vendors reflect the architecture and FPGA types that the vendors support, but all aim to kick-start development projects. They provide the key infrastructure to support the developer's application with IP cores for memory controllers, and DMA engines plus Serial RapidIO and PCIe endpoints. System-on-Chip architectures hook the various parts together. In addition, these IP cores are supported by software driver libraries or complete board support packages for selected real-time operating systems.

Larger fabric-based systems can use middleware and analysis tools for abstracted, consistent communications among FPGA or processing nodes. For example, Curtiss-Wright Controls, Embedded Computing (CWCEC) complements its signal processing FPGAs with its Continuum tools. Figure 1 shows the company’s VPX3-450 signal processing module incorporating a Xilinx Virtex-5 FPGA plus a Freescale Semiconductor dual-core 8640D general-purpose processor.

Figure 1

COTS vendor tools allow developers to more tightly focus on the DSP algorithms and the overall application and so add value for their customers. Toolsets from the FPGA vendors and other third-party suppliers, such as Synopsys and Mentor, are ideally suited to the development of these algorithms, producing code from abstract hardware description languages such as VHDL or Verilog that offer synthesis, placement, and routing with various intermediate stages of simulation. And, because COTS vendors also use the same tools for the development of their IP cores and libraries, easier integration and verification of the final FPGA bit stream is assured.

Significant space, weight, power, and recurring cost reductions through FPGA use have continued to prove elusive for defense and aerospace sensor developers seeking predictable, reasonable initial development costs. Yet considerable gains are possible by using readily available hardware modules, IP cores, libraries, and drivers. Many contractors have demonstrated that this is an effective strategy. In addition to continued intensive research into FPGA reconfigurability, development is ongoing to improve design tools and gain much more insight into system behavior by using real-time monitoring, data collection, and analysis tools. FPGAs will continue to be the solution of choice for the many and varied sensor systems deployed by the armed forces and security services.

Robert Hoyecki is Director of Advanced Multi-Computing at Curtiss-Wright Controls Embedded Computing. Rob has 15 years of experience in embedded computing with a focus on signal process products. He has held numerous leadership positions such as application engineering manager and product marketing manager. Rob earned a Bachelor of Science degree in Electrical Engineering Technology from Rochester Institute of Technology.

Rob can be reached at  info@cwcembedded.com.