Download this article in .PDF format
This file type includes high resolution graphics and schematics when applicable.

Fig. 1

Signal processing is often the difference between missing and identifying a radar target, delivering an effective electronic-warfare (EW) response, or maintaining secure communications. A number of digital components perform signal processing in military electronic systems, with digital signal processors (DSPs) and field-programmable gate arrays (FPGAs) often used separately or together. Finding the right component match for an application can be a challenge, although understanding key performance parameters and how they relate to different military electronic applications can help ease the process.

DSPs are essentially types of microprocessors aimed at performing math-intensive operations. The processing power of a DSP can be determined by its clock rate and the number of operations it can perform per clock cycle, such as how many additions or multiplications per second. Its processing capabilities are typically characterized in terms of millions of instructions per second (MIPS).

Real-World Solutions

As examples of the processing capabilities available from high-speed DSPs, models C6654 and CC6652 are (respectively) fixed- and floating-point DSPs recently introduced by Texas Instruments. Based on the firm’s KeyStone multiprocessor core architecture, and incorporating the new C66x DSP core, these DSP devices run at clock speeds to 850 MHz (C6654) and 600 MHz (C6652). With an extended operating temperature range of –40 to +100°C, the devices are suitable for avionics and other defense applications as well as for ruggedized commercial applications.

Each DSP features two on-chip phase-locked loops (PLLs), two channels of 8- or 16-b processing, 8 GB of addressable memory space, and I2C and SPI interfaces. The model
TMS320C6652 DSP provides a cycle time of 1.167 ns at 600 MHz, delivering 19.2 GMACS and 9.6 GFLOPS at 600 MHz. The model TMS320C6654 has a cycle time of 1.175 ns at 850 MHz, providing 27.2 GMACS and 13.6 GFLOPS at 800 MHz.

Additional suppliers of DSPs include Analog Devices, Infineon Technologies, Intersil, Microchip Technology, NXP Semiconductors, ON Semiconductor, STMicroelectronics, and Zilog.

As an example of a military-grade FPGA, the Virtex-7 X980T FPGA from Xilinx, has been used in a 24-channel portable radar beamformer and demonstrated as much as 987 GFLOPS capability with lower power consumption. Fabricated on a 28-nm silicon semiconductor process, the FPGA achieves as much as 2.8-Tb/s total serial bandwidth with as many as two million logic cells.

Atmel has been a longtime supplier of both radiation-tolerant and radiation-hardened (rad-hard) FPGAs for space-based applications (Fig. 1). As an example, model AT40KEL040 is a rad-hard FPGA dedicated for space use. It is based on a static-random-access-memory (SRAM) architecture with 46,000 application-specific-integrated-circuit (ASIC) gates and fabricated in a 0.35-µm silicon CMOS semiconductor process. The FPGA, which operates at internal clock rate of 60 MHz, provides 18-ns memory access speed over an operating temperature range of –55 to +125°C.

Altera, now part of Intel Co., offers high-performance Stratix 10 FPGAs that support advanced beamforming functions in many stationary and portable military radar systems (Fig. 2). Capable of operating at clock rates to 1 GHz, these FPGAs can perform at rates to 10 TFLOPS with power efficiency to 80 GFLOPS/W.

In addition, Microsemi Corp. supplies high-performance, low-power and radiation-tolerance FPGAs and systems-on-a-chip (SoCs) for commercial, industrial, and military applications. Earlier this year, the company acknowledged its support of defense contractor Lockheed Martin during the U.S. Navy’s successful 158th, 159th, and 160th test launches of the Trident II D5 Fleet Ballistic Missile (FBM). Lockheed Martin has been the U.S. Navy's strategic missile prime contractor, and Microsemi has played a strong part in achieving the signal-processing goals of these missile guidance systems with its FPGAs and SoCs.

In many defense systems, signal processing is not accomplished solely by DSPs or FPGAs, but through a combination of the two, along with multiple-core microprocessors and various other digital components. Suppliers of DSPs and FPGAs typically offer sample boards for different applications, such as secure communications systems and portable radar systems, to help users through their own system-design process. System designs employing both DSPs and FPGAs offer a great deal of flexibility, since different types of functions, such as fixed-point and floating-point operations, can be directed to the different types of processors.

In addition to sample boards, many component suppliers also support their devices with model libraries that can speed the design process, such as models for FIR filters, FFTs, and IIR filters. Design support should be a consideration in addition to component performance when selecting and specifying a DSP or FPGA.

A DSP must be preconditioned or programmed to perform a particular function, usually by "C" programming code or some other assembly language. DSPs are designed to work with external memory on a printed circuit board (PCB) or within a system, and can work with large amounts of available data for handling large processing operations. As a tradeoff, accounting is necessary for transferring data between a DSP and the memory locations, plus any required transfer times.

An FPGA, on the other hand, is a collection of transistor gates that are able to change functions in the field. The array of gates can be connected together to achieve different functions, such as addition or multiplication, without the prior programming required for a DSP. In some ways, an FPGA can be thought of as multiple DSPs or processors, each with its own dedicated memory, so that multiple functions can be performed.

Fig. 2

In addition to performing very-high-level operations—like finite-impulse-response (FIR) filters, infinite-impulse-response filters (IIR), or fast-Fourier-transform (FFT) functions—FPGAs can perform DSP functions (e.g., the aforementioned addition and multiplication) at very high clock rates. Again, with an FPGA, comparing performance is a matter of determining how many useful functions can be performed per clock cycle, or how many total operations can be performed per second.

In contrast to DSPs, FPGAs are designed and fabricated with internal memory. This enables functioning without any memory-transfer delays, with fast data input/output (I/O) rates. As a tradeoff, the amount of memory and the size of the data sets processed are limited compared to a DSP. Of course, when an FPGA must handle large, complex data sets, it can be designed into a PCB or module with dedicated external memory to increase its processing capabilities with minimum data I/O delays.

One way to think of the difference between an FPGA and DSP for a military system is to imagine how a block diagram of part of the system (a receiver, for instance) might be handled by each component. In an FPGA, equivalent receiver components (e.g., frequency mixers, filters, and amplifiers) are defined as specific functions within the FPGA. Once those functions are defined, the FPGA provides fairly efficient operation. However, the functions cannot be easily changed, since the gates within the device have been set for the desired signal processing.

In a DSP, though, the block diagram is defined by programming code in terms of its functionality. If necessary, a DSP programmed as a receiver could be reprogrammed as a transmitter, with the signal branching and complex decision making determined as responses to the programming code.

In many military electronic systems (portable systems, in particular), power efficiency is important in maintaining military design goals for size, weight, and power (SWaP). The power efficiency of signal processing as performed by multicore processors, DSPs, or FPGAs is usually compared in terms of millions of floating-point operations per watt (MFLOPS/W) and billions of floating-point operations per watt (GFLOPS/W). FPGAs typically excel in this area compared to DSPs.

Download this article in .PDF format
This file type includes high resolution graphics and schematics when applicable.