26-10-2010, 01:55 PM
[attachment=6821]
Guide to Using Field Programmable Gate Arrays (FPGAs) for
Application-Specific Digital Signal Processing Performance
Gregory Ray Goslin
Digital Signal Processing Program Manager
Xilinx, Inc.
2100 Logic Dr.
San Jose, CA 95124
Abstract:
FPGAs have become a competitive alternative for high performance DSP applications, previously dominated by general purpose DSP and ASIC devices. This paper describes the benefits of using an FPGA as a DSP Co-processor, as well as, a stand-alone DSP Engine. Two Case Studies, a Viterbi Decoder and a 16-Tap FIR Filter are used to illustrate how the FPGA can radically accelerate system performance and reduce component count in a DSP application. Finally, different implementation techniques for reducing hardware requirements and increasing performance are described in detail. Introduction: Traditionally, digital signal processing (DSP) algorithms are implemented using general-purpose (programmable) DSP chips for low-rate applications, or special-purpose (fixed function) DSP chip-sets and application-specific integrated circuits (ASICs) for higher rates. Advancements in Field Programmable Gate Arrays (FPGAs) provide new options for DSP design engineers. The FPGA maintains the advantages of custom functionality like an ASIC while avoiding the high development costs and the inability to make design modifications after production. The FPGA also adds design flexibility and adaptability with optimal device utilization while conserving both board space and system power, which is often not the case with DSP chips. When a design demands the use of a DSP, or time-to-market is critical, or design adaptability is crucial, then the FPGA may offer a better solution. The SRAM-based FPGA is well suited for arithmetic, including Multiply & Accumulate (MAC) intensive DSP functions. A wide range of arithmetic functions (such as Fast Fourier Transform’s (FFT’s), convolutions, and other filtering algorithms) can be integrated with surrounding peripheral circuitry. The FPGA can also be reconfigured on-the-fly to perform one of many systemlevel functions. When building a DSP system in an FPGA, the design can take advantage of parallel structures and arithmetic algorithms to minimize resources and exceed the performance of single or multiple general-purpose DSP devices. Distributed Arithmetic[1] for array multiplication in an FPGA is one way to increase data bandwidth and throughput by several order ofmagnitudes over off-the-shelf DSP solutions. One example is a 16-Tap, 8-Bit Fixed Point, Finite Impulse Response (FIR) filter[2]. The FIR design supports more than 8 million samples per second. This example can also be implemented using multiple bits, until a “Fully Parallel Distributed Arithmetic” algorithm is obtained for higher sample rates (i.e., 55.89 million samples per second). Figure 1 compares the 16-Tap FIR filter implemented in a state-of-the-art fixed-point DSP with that of the Xilinx FPGA. As published by Forward Concepts[5], “For 1995, the state-of-the-art fixed-point DSP is rated at 50 MIPS.” Such a device requires 20 nsec per Tap to implement a 16-Tap FIR filter, which translates to a theoretical maximum (with zero wait-states) sample rate of 3.125 million samples per second. An In-System Programmable (ISP) FPGA can also be reconfigured on the board during system operation. Taking advantage of the reconfigurability feature means a minimal chip solution can be transformed to perform multiple functions. For example, an FPGA could be the basis for a system that performs one of several DSP functions. Suppose, for instance, one function is to compress a data stream in transmit mode and another function is to decompress the data in receive mode. The FPGA can be reconfigured on-the-fly to switch, or toggle, from one function to another. This capability of the FPGA adds functionality and processing power to a minimum-chip DSP system controlled with an internal or an external controller. This “Reconfigurable Computing” technique is beginning to impact design methodologies.