BDTI's DSP Insider Archives |
||
| HOME << |
||
This month:
PXA250-based PDA Performance Problems ProbedIntel’s PXA2xx processor family is making significant headway in the high-end PDA market. Last month, for example, Sony began shipping the first Palm OS-based PDA powered by a PXA2xx. Although the PXA2xx has gained wide acceptance in high-end PDAs, some reviewers (for example, at CNET.com) have complained that PXA2xx-based PDAs are not appreciably faster than PDAs based on its predecessor, StrongARM. Some reviewers have been particularly critical of the lack of improvement on multimedia applications. BDTI recently analyzed the PXA2xx’s signal-processing performance using the industry-standard BDTI Benchmarks™ . Although BDTI has not yet benchmarked StrongARM, it has benchmarked the closely related ARM9 core as well as the ARM7 and ARM9E cores. (Benchmark results for the PXA2xx, ARM7, ARM9 and ARM9E are available at http://www.BDTI.com/bdtimark/BDTImark2000.htm) These analyses show that the PXA2xx itself is probably not to blame for any lackluster multimedia performance of PXA2xx-based PDAs. Instead, BDTI’s analyses suggest that a 400 MHz PXA250 is roughly three times faster on signal-processing tasks than the fastest StrongARM processor, the 206 MHz SA1110. Clock speed is the most obvious contributor to this large gain in DSP performance. The PXA2xx achieves its higher clock rate partly by extending the five-stage StrongARM pipeline to seven stages. Although longer pipelines allow higher clock rates, they can also lead to an increase in branch-related stalls. However, the PXA2xx contains a branch prediction unit that mitigates this problem.
The PXA2xx also holds an advantage in architectural efficiency. The
PXA2xx instruction set includes a number of DSP-oriented operations
not supported by StrongARM. For example, the PXA2xx supports
instructions that perform two 16-bit fixed-point multiply accumulate
(MAC) operations per cycle. In contrast, StrongARM supports no more
than one 16-bit MAC per cycle. The PXA2xx is also able to complete
certain load/store operations in the background while it continues to
execute other instructions. In contrast, StrongARM is obliged to wait
for data transfers to complete before continuing execution. These and
other architectural efficiencies give the PXA2xx a significant
advantage: BDTI’s analyses suggest that even a 200 MHz PXA210 is
significantly faster than a
So why aren’t users consistently experiencing better multimedia
application performance with PXA2xx-based PDAs? The answer likely
involves a number of factors including memory system bottlenecks and
operating system inefficiencies. One key problem may be a shortage of
PXA2xx-optimized multimedia applications. Software developed for the
SA1110 won’t take advantage of some key performance-boosting features
on the PXA2xx; therefore, PXA2xx-based PDAs will only realize their
full performance potential when multimedia software has been optimized
for the PXA2xx.
Software Defined Radio: Ready for Prime Time?Last month BDTI participated in the Software Defined Radio Forum’s first conference. Judging from the presentations given at this conference, software defined radio (SDR) is quickly evolving from an interesting but somewhat academic concept to a mainstream technology. The main driver of this transformation may be the U.S. Department of Defense’s commitment to SDR. Today’s military operations require communications among different branches of the armed forces, but different armed forces often use incompatible radio systems. To address this problem, the DoD initiated the Joint Tactical Radio System (JTRS) program, which seeks to migrate all U.S. armed forces to SDR systems. The goal is to create radio systems that can communicate not only with each other, but also with legacy systems and with the systems of U.S. allies. The JTRS program has the potential to create a significant new market for communications equipment. The value of the first JTRS contract, which was awarded to a Boeing-led team of contractors this June, could exceed two billion dollars if the government exercises all of the contract options. This initial contract is expected to lead to the installation of SDR systems in over 10,000 military vehicles and aircraft. Eventually, vehicular and airborne SDR installations are expected to exceed 100,000 units. As the DoD expands JTRS to other applications, such as naval and hand-held radios, the program may encompass over 750,000 radios. If the JTRS program reaches this volume, it will reduce the development and per-unit costs associated with SDR and make this technology more attractive for commercial applications.
The types of vendors attending the conference reflected the growing
gap between the computational needs of leading-edge communications
applications and the capabilities of mainstream processors: FPGA and
specialized DSP architecture vendors were more visible than leading
DSP or general-purpose processor vendors. The presence of
specialized, high-performance DSP core licensors like Morpho,
QuickSilver, and picoChip was particularly notable. Historically,
high-performance DSP core licensors have had difficulty establishing a
foothold in commercial communications markets. It will be interesting
to see if these vendors find a warmer welcome in the government-driven
SDR market.
BDTI Case Study
This Month: Effective Error ManagementNearly all signal-processing applications use computational techniques that introduce small numeric errors. While the error for a single computation is typically very small, the cumulative effects of error over many computations can cause problems like noisy audio, grainy video, or unreliable communications. As a result, managing computation errors is a key challenge of DSP application development. The most common source of error in DSP applications is quantization, a type of error that occurs because processors represent numbers with limited precision. For example, a processor that uses 16-bit fixed-point data approximates the value of 1/3 as the fraction 10923 / 32768, which is 0.33334... rather than the desired value 0.33333... One way to reduce quantization error is to use a larger data type. For example, a 32-bit data type can approximate 1/3 as 0.3333333334... Inexact formulas are another source of error. Most processors lack the resources to compute functions like sin(x) directly. Instead, a function like sin(x) must be computed indirectly with an algorithm that produces the same result as the desired function. For example, the Taylor power series can be used to compute sin(x) with any desired precision. Unfortunately, this type of indirect approach often involves a large number of operations, making it necessary to use a shortened version of the algorithm. Because these shortened algorithms omit some operations, they produce approximate results. One way to reduce this type of error is to use a longer algorithm that more closely approximates the desired function. Of course, using larger data types or longer formulas increases memory use and/or computation time—commodities that must be carefully conserved in most DSP applications. Hence, managing error requires careful consideration of these trade-offs. Indeed, it is often possible to speed up an application by replacing an unnecessarily precise algorithm with one that is less precise. To manage error efficiently, a DSP engineer must understand the error tolerance of the application and how error arises and propagates through the constituent algorithms. Then the engineer can focus error-reduction efforts on the algorithms or operations that create the most significant errors. Over the last few years, BDTI has developed a number of proprietary tools and techniques (such as the fixed-point emulation library described in the July 2001 DSP Insider) that enable it to efficiently analyze and manage computational errors. In one recent example, an audio processing start-up engaged BDTI’s services to optimize an audio processing algorithm that included logarithmic functions. The original implementations of these functions required too much computation time. Using its proprietary tools and techniques, BDTI created new implementations that were much faster but still maintained the desired accuracy.
To learn how BDTI can help you manage error efficiently, contact
Jeremy Giddings (giddings@BDTI.com).
Impulse Response, by Jeff Bier
Easy MoneyIn October, ARM Ltd. announced two unhappy firsts: its first-ever quarter-to-quarter sales decline, and its first layoffs. Prior to these developments, ARM seemed to have discovered a bulletproof business model that was easy to mimic: just whip up a core, license it, and let the royalty checks roll in. Indeed, ARM’s previously uninterrupted ascent, coupled with the ready availability of venture capital, gave rise to a flood of new processor core licensors in the late 90’s. In truth, core licensing is anything but foolproof, and success requires far more than slapping together a core. Chip development is a difficult, expensive, and time-consuming process; forcing licensees to start with a poorly supported, bare-bones core just makes things worse. Stand-alone cores become even less attractive as increasing chip complexity and shrinking process geometries make chip development harder. Core vendors’ very survival—never mind successhinges on the availability of complementary offerings such as world-class development tools, application software components, and supporting services. Unfortunately, too many would-be ARMs vastly underestimated the importance of these complementary offerings. An all-too-common strategy among upstart core vendors was to offer a core and some rudimentary software development tools. In the past year, many of these companies have realized the error of their ways and have begun to develop comprehensive solutions for applications such as voice-over-IP and 802.11. But with so much ground to make up, these efforts may be too little, too late. In contrast, leaders like ARM have made complementary offerings a cornerstone of their strategies. For example, ARM immediately followed the announcement of its ARM11 core with the announcement of an ARM11-based version of its wireless platform. And some newcomers have emphasized their complementary offerings: since its launch last year, Adelante Technologies has promoted its cores as part of a package that includes complementary IP and tools, for example.
Although the flood of new core licensors has slowed to a trickle, a
few new core vendors continue to emerge. Can today’s market support
these new entrants? Maybe—but only if they deliver compelling,
comprehensive offerings—offerings that solve more problems than they
create.
BDTI Welcomes GlobalDSP MagazineEven in this gloomy economic climate, bright new ventures continue to appear. GlobalDSP is one such venture. GlobalDSP is a new vendor-independent electronic publication targeted at the DSP engineering community. GlobalDSP includes a variety of content, such as articles, product review, and upcoming events, and covers a wide range of applications for signal processing.
To subscribe, go to http://www.globaldsp.com.
Plan to Attend GSPx/ISPC This SpringThe International Signal Processing Conference (ISPC) and Global Signal Processing Expo (GSPx) will provide a forum for peer-to-peer interaction among engineers and managers who create and use signal processing technology. This new trade show and conference is organized by Amnon Aliphas, the founder of ICSPAT/DSP World, and will be held at the Hotel Intercontinental in Dallas, Texas, from March 31 to April 3, 2003. The technical conference will combine workshops and conference presentations with panel sessions led by industry leaders on compelling topics that impact the direction of DSP technology. The expo will provide a venue for companies to introduce and demonstrate new technologies and products.
More information on GSPx and ISPC is available on the GSPx Web site at
http://www.gspx.com/ISPC/index.html.
BDTI Communications Benchmark™ (OFDM)The first of BDTI’s DSP Application Benchmarks™ is available for licensing. The BDTI Communications Benchmark (OFDM) models an OFDM (orthogonal frequency division multiplexing) receiver, which is used in a variety of communications applications, such as fixed wireless systems. This new benchmark can be applied to a range of processor technologies, including DSPs and FPGAs—results for Altera’s Stratix FPGAs and Motorola’s SC140-based MSC8101 are published in BDTI’s new report FPGAs for DSP. To ensure fair comparisons, BDTI’s specification details key benchmark parameters such as sample rates, filter lengths, and channel-code constraint lengths. BDTI certifies benchmark implementations before publication of results.
http://www.bdti.com/products/services_benchmarking.html#app_benchmarks
Latest Technology Analysis Reports from BDTI
Inside the LSI Logic ZSP500A new addition to the Inside series, this report on the LSI Logic ZSP500 provides a comprehensive qualitative analysis of the processor architecture and features, along with a complete quantitative performance analysis based on the BDTI Benchmarks™ . For a free summary, see http://www.BDTI.com/articles/zsp500_summary_report.pdf. For info on the full report, go to http://www.BDTI.com/products/reports_zsp500.html.
FPGAs for DSPFPGAs for DSP provides a detailed assessment of the DSP capabilities of DSP-capable FPGAs and explains why FPGAs are a compelling solution for some DSP applications. The report discusses new offerings from Altera and Xilinx and includes benchmark results of FPGA DSP performance.
http://www.BDTI.com/products/reports_focus.html
Sounding Board Service for Marketing PresentationsIn these times, it is more important than ever to make sure your marketing presentation is technically correct and accurately targeted. BDTI’s Sounding Board service helps vendors of DSP-related products—chips, cores, tools, and softwaredevelop accurate and compelling marketing presentations. BDTI’s expert analysts will review your presentation, host a Q&A session, and give you specific, detailed suggestions for improvement. BDTI will help you achieve technical accuracy and ensure your message has the right focus for the target audience.
For more information on this new service, go to
http://www.BDTI.com/products/services_sounding.html
About BDTIBDTI is an independent source for DSP technology analysis and optimized DSP software. From rigorous technical analyses of processors for DSP, such as the Inside series of processor analyses, to highly regarded technology training classes, BDTI is the trusted independent source for reliable information on DSP technology.
For more information, visit our Web site at http://www.BDTI.com.
The next issue of BDTI’s DSP Insider is coming in January. Previous issues of BDTI’s DSP Insider are archived on BDTI’s Web site. Follow the link from http://www.BDTI.com/dspinsider.htm. If you have comments, suggestions, or other feedback about the DSP Insider, please send email to dspinsider@BDTI.com. BDTI’s DSP Insider is a free monthly electronic newsletter published by Berkeley Design Technology, Inc. If our newsletter was forwarded to you and you would like to receive it regularly, please register at http://www.BDTI.com/dspinsider.htm.
If you no longer wish to receive the DSP Insider send an email
message to dspinsider@BDTI.com with the words Remove me in the
subject line.
BDTI’s DSP Insider © 2002 Berkeley Design Technology, Inc. |