This paper discusses the unique capabilities platform FPGA-based designs offer when used to help meet the challenges of today's demanding high-performance real-time DSP applications, both as stand-alone solutions and as complementary solutions for traditional DSPs. A detailed investigation of the use of model-based design methodologies reveals the ease with which platform FPGAs can be employed to accommodate these highly complex DSP implementations. Finally, an example in the wireless arena – Multiple Input, Multiple Output (MIMO) technology – is used to illustrate the concepts.

Changing standards
For the past two decades, the networking, imaging and defense industries have searched for new and more efficient ways to process and deliver content-rich data across disparate networks of commercial and consumer-level systems and devices. In the context of DSP, the wired and wireless communications markets are experiencing this "standards revolution" with WiFi, WiMAX, 3G, 4G, and WCDMA system standards, application standards such as OFDM and MIMO, as well as technology standards such as Software Defined Radio (SDR). Additionally, a continuous stream of changes is also erupting from the audio, video, and broadcast markets as well as the defense electronics industry.

Processing complexity
Every standard represents a substantial opportunity and challenge for companies in the DSP design community. However, the DSP processing requirements of many of these standards has produced a performance gap that traditional DSPs and general-purpose processors (GPPs) cannot cross on their own (Fig 1).

1. Traditional processor architectures are ill equipped to provide the massive parallelism needed to support new-generation standards in today's high-end real-time applications.
(Click above image to see a larger, more detailed version)

These new applications have driven processing complexity and computational workload from tens of mega-samples per second (MSPS) to 100s of MSPS. The digital convergence phenomenon (video, voice, data) and adaptation to user's individual needs has created the simultaneous demand for more sophisticated real-time processing and higher bandwidth to support on-demand content-rich multimedia data.

System Integration
Treating DSP system design as solely a signal processing problem addresses only part of the issue. For many applications, DSP function is not a separate entity, but rather one of several system-on-a-chip (SoC) functions, raising the need to address a host of system integration issues as well. These issues include:

• Connectivity using standards such as RapidIO, PCIe ATCA, 1G/10G Ethernet, and DDR/DDR2 memories.
• System control issues such as statistics monitoring, QoS/CoS, and traffic priorities.
• The need to meet reliability requirements by keeping power consumption within the standards set by Networking Equipment Building Systems (NEBS) and European Technical Standards Institute (ETSI), and signal integrity requirements set by bit error rates (BERs).

How DSP solutions compare
Amid the confusion of challenges, issues and concerns, it's easy to understand why making the correct selection of the most appropriate DSP solution is so critical. Fig 2 maps five of the most critical considerations for the leading DSP technologies.

2. FPGAs offer the best combination of performance, design flexibility and scalability; combining programmable DSPs and FPGAs is the best all-around solution.
(Click above image to see a larger, more detailed version)

As performance and complexity requirements increase, the massive parallelism available in FPGA-based DSP designs becomes the critical selection criterion. Add to this the FPGA's design flexibility to help adapt to changing standards and its ability to scale power and cost based on the particular algorithm being implemented, and it is easy to see why FPGAs have gained such prominence as a leading DSP technology.

Making the right selection is dictated in large part by the particular application. In some cases, the best choice is a combination of programmable DSP and FPGA, where the FPGA is used either as a pre-processor to the DSP, or as a co-processor.

DSP design with platform FPGAs
Over the past two decades FPGAs have evolved from a collection of gates for programmable logic to platform FPGAs integrating system-level capabilities:

• Clock management.
• Memories.
• Parallel and serial I/Os.
• Ethernet MACs.
• Microcontroller(s) and microprocessor(s).

Platform FPGAs provide numerous multiply-accumulate (MAC) functions – the fundamental DSP building block (Fig 3). In contrast, a traditional Digital Signal Processor provides only one to four such MACs.

3. The Xilinx Virtex-4 DSP block offers up to 512 18x18 MACs.
(Click above image to see a larger, more detailed version)

With the Virtex-4 family, Xilinx introduced a new concept called the multi-platform FPGA. Built upon an architectural innovation called the ASMBL architecture, the multi-platform FPGA enables tuning the ratio of key features on the FPGA to match the requirements of an application domain; e.g., the SX family of Virtex-4 FPGAs is optimized to deliver more MACs per dollar and lower power for DSP applications.