EDN recently interviewed Denis Labrecque, marketing-programs manager at Analog Devices. In the interview, Labrecque sounds off on DSPs’ past, present, and future in audio, graphics, and imaging. He provides his perspectives on competition within his company’s product line; against other DSP architectures; and versus other silicon-based processing architectures, such as CPUs and FPGAs.

You’ve navigated a unique professional path to get to your current position at Analog Devices. To begin, can you give a brief rundown on your educational and employment background for EDN’s readers?

To quote one of your favorite bands: “What a long, strange trip it’s been.” After hearing strange new electronic sounds coming out of my transistor radio in the early ’70s (in particular, the synthesizer sounds in Lucky Man by Emerson, Lake & Palmer and Baba O’Riley by The Who, not to mention Switched-On Bach by Wendy Carlos—but I digress), I knew I needed to be in the music biz. I started to pursue an EE to get into it from the electronics side, but quickly decided that a music degree would bring a more valuable perspective to the industry (not to mention requiring far less calculus).

I started in hardware production (acquired soldering skills that stay with me today) with Star Instruments, the electronic instrument company responsible for the SYNARE electronic drum and one of the first music companies to use a microprocessor in an instrument. I then spent many years as VP of Passport Designs, the first MIDI music software company (anyone else out there remember the Commodore 64?), developing MasterTracksPro MIDI sequencing and Encore notation software. Then a stint at Emu Systems as product marketing manager for the award-winning Emu APS AudioProductionStudio PC card, then back to software with the VP of marketing position at Staccato Systems. Staccato had developed a PC-based software synthesizer that was licensed by Analog Devices for their SoundMAX codec product and subsequently acquired by ADI in 2001.

I then moved to the incredible marketing team at the general-purpose DSP division of ADI, immersing myself in all that is embedded processing—industrial, medical, automotive. Recently, as I’m still known as the audio guy, I’ve taken on additional responsibilities with customers in the pro audio space as the Pro-Audio marketing manager, coming (somewhat) full circle back to the music biz.

Continued innovation in the technology sector at the system level requires continued demand for ever-more-capable silicon building blocks from system designers. Analog Devices has long been a tier 1 chip supplier to the audio industry with its SHARC 32-bit and floating-point-capable DSPs, so let’s begin there. Can you provide a historical background on how your audio customers’ needs—performance and otherwise—have evolved over time, and how ADI has responded to them?

There has been a constant demand for higher-performance floating-point audio processors. Demand started in the professional market, where programmable parts enabled new features and flexibility. Next came consumer applications driven by digital content and enabled by lower-cost parts.

That’s why Analog Devices created processors with a specific mix of peripherals, memory, and ROM aimed at particular market categories. For example, consider the 2126x and 2136x families. These parts satisfy the cost and performance requirements of audio/video receivers.

I find it quite interesting to look at the generic demands put on a processor. The desired attributes of high I/O channel count; complex, simultaneous data processing; large dynamic range; input analysis; and transformation are the same whether you’re designing a pro audio recording console or a patient-monitoring device for the medical industry.

With that said, certain peripherals—a large number of serial ports, flexible signal-routing units, and onboard sample-rate converters—have been enthusiastically embraced by the audio community. We constantly engage our customers across a variety of markets and, with their help, try to anticipate the processing needs for our future products. For example, the recently announced 21469 processor provides hardware accelerators for FIR and IIR filters as well as FFTs, reflecting the desire for custom room-equalization functions. We have seen the advantages of providing the most up-to-date decoders in our ever-evolving DSPs that enable our audio customers to bring new capabilities to their AVRs and other audio systems.

We’ve enabled the integration of high-performance decoders in our parts that have enabled such decoders to be available in mid-end and lower-end AVRs due to the price/performance advantages of our processors.

 ADI used to support a multicore product family called TigerSHARC. However, my understanding (please confirm or deny) is that the company has decided to develop no more single-die multicore SHARC devices, instead building high-speed DSP-to-DSP buses into its SHARC DSPs and directing customers to implement multichip configurations. Was TigerSHARC used in any high-end audio applications, and what are the strengths and shortcomings of the alternative multi-SHARC approach?

 While TigerSHARC is indeed currently used in several high-end applications and we will, of course, continue to support the existing TigerSHARC products, we are focusing future development on our SHARC and Blackfin processors.

With respect to the question of products where multiple cores are in a single die, or multiple DSPs continue to link to each other through high-speed buses, our real focus is on providing customers with the right system-level solutions that balance core performance as well as I/O performance—solutions that allow large amounts of data transfer as well as speeding up the execution of independent software modules. It’s about providing the right solution for the right price, and the goal is to provide the appropriate technical solution to match that intent.

SHARC isn’t the only DSP family in ADI’s portfolio, of course. The company’s DSP core development effort with Intel resulted in the Blackfin processor, which combines 16-bit digital signal processing and primary system processor capabilities in one device. For the moment, at least, Blackfin doesn’t support native floating point operations, although floating point can be emulated in software. Focusing first on audio, to what degree has Blackfin expanded the market serviceable by Analog Devices’ products? What criteria does a customer use when choosing between Blackfin and SHARC? And assuming that Blackfin will eventually add 32-bit MAC and native floating-point support (please confirm), how going forward will the Blackfin-versus-SHARC differentiation evolve over time?

SHARC and Blackfin fill complementary market positions. SHARC has always been and will continue to be ADI’s flagship high-end processor for not only the audio market but also medical and industrial. Blackfin, though a native 16-bit processor, is a very capable audio processor and is optimized to perform equally well for signal processing, control processing, and media processing, making it attractive for a broad range of multimedia applications. Its high clock speed, up to 2400 MIPS, and dual 16-bit MAC unit enable it to do a serious amount of audio processing. Even with the overhead associated with 32-bit, double-precision processing, a 600-MHz Blackfin can potentially do as much processing as a 150-MHz SHARC.

To help customers decide what to use, we need to know what specific audio application and functions they need. Is the primary function of the product audio processing for a home-theater system, automotive amplifier, mixing console, broadcast processor? Then use SHARC. However, if the product is primarily about communication, the user interface, or streaming but needs some audio features, then use Blackfin. Products falling into this category are portable media players, streaming network nodes, or automotive head units.

I can’t comment specifically on future features, but in general we see SHARC continuing to take the applications that require multichannel massive computational apps while Blackfin focuses on applications where communication and microprocessor features are desired. We view these differences as complementary as opposed to competitive. In fact, I’ve recently seen a new synthesizer product where both a Blackfin and multiple SHARCs are used.

Even lower-end audio digital signal processing needs are supported by ADI’s SigmaDSP family. How do you (and therefore your customers) differentiate between SigmaDSP, Blackfin, and SHARC, both now and in the future? And more generally, how do you as a company balances the desire to provide a robust portfolio of products to your customers against the likelihood that a too-diverse portfolio might result in customer confusion and purchase paralysis?

The SigmaDSP family was introduced to meet the needs of low-cost products with basic audio processing needs. The combined codec and audio processor provides a high level of integration. The capabilities of the SigmaDSP family continue to evolve: Initial parts debuted at 25 MHz, and the latest generation is running at 170 MHz. And the performance lines between SigmaDSP, Blackfin, and SHARC continue to blur as SigmaDSP matures.

However, relative performance is just one way of comparing the parts. The larger difference deals with the underlying programming model. SigmaStudio, the graphical audio-design environment for SigmaDSP, has revolutionized audio-system development. Audio-system development is now possible for engineers with basic audio and embedded-software development skills. This satisfies a large untapped segment of the market. In particular, we’re finding great success in the automotive audio segment with SigmaDSP.

But you’ve raised a very important point. Analog Devices must carefully educate customers regarding the best processor for their application. We must articulate how to select the ideal processor to avoid the confusion and paralysis you mention.

Back to an earlier statement, we need to know what specific audio application and functions they need. All three families have particular attributes that make them ideal for a specific application. SHARC and Blackfin appeal to engineers needing complete control and customization, while SigmaDSP is better for fixed-function applications.

Multimedia means more than just audio, of course. I’m aware, for example, that ADI has long promoted Blackfin for digital still cameras, and at least one manufacturer—Sigma—now uses Blackfin in some of its products. What are the key factors necessary for ADI to expand its beachhead in digital still, video, and hybrid still-plus-video cameras, both in an absolute sense and relative to competitive offerings? And in this latter case, both versus other independent semiconductor suppliers and versus internal IC development teams at companies like Canon?

While Blackfin has seen some success in the still-camera market, our current focus is imaging for industrial use. In particular, we have been very active in development of processing for industrial applications such as production-line monitoring and security cameras. Blackfin has many aspects that make it good at the image processing and communication required for industrial video. From an architectural and instruction-set standpoint, in addition to its dual 40-bit ALUs, the Blackfin core contains four 8-bit video ALUs that allow for numerous flexible configurations of quad 8-bit or dual 16-bit operations per cycle.

Examples of video ALU functions are Sum of Absolute Differences (used in motion-estimation routines) and a Huffman Coding-related instruction (used to increase the efficiency of entropy encoding found in both multimedia and communications applications). A specific algorithm example for the video ALUs is Discrete Cosine Transform (DCT, used in MPEG motion estimation), which can be accomplished in approximately 300 cycles. Additional instruction-set elements such as complex math, Viterbi dual add-compare-select operations, and CRC are further instruction-set architecture features that align to the multimedia and communications target segments.

In the area of device peripherals, on-board Ethernet MAC and USB interfaces also aid in adoption in the multimedia and communications segments. Specifically for video applications, the Blackfin Parallel Peripheral Interface (PPI) provides a very flexible front-end interface that can natively connect to both CMOS sensors directly and Analog Front End (AFE) components used to convert data captured by a Charged Coupled Device (CCD) from the analog to the digital domain. In some Blackfin products, we have also added fixed hardware functionality to cover such tasks as the color conversion used when outputting processed data to a display.

Speaking of fixed-function ICs, a recently published survey of audio-design engineers revealed robust interest in two non-DSP platforms for sound processing: x86 CPUs and FPGAs. The x86 CPU could even be the one already in a Mac or a PC, shifting expense from the audio peripheral to the more function-versatile host computer. How do you combat this attractive system-partitioning approach?

There are separate forces at work in the x86 and FPGA trends. I’ll start with x86 and further divide this category into native processing, applications that are executing on a PC or Mac, and embedded x86 typified by the new Atom processor. Native processing has long been a viable solution for audio-production and -postproduction work. These applications are dominated by their user interface and storage needs. However, it is unlikely that the PC environment will ever achieve the low latency and real-time reliability for true professional-audio applications. Do you really want the integrity of your audio system dependent upon Windows? Do you know how threads are active on your PC? A large number of professional-digital-audio workstations in fact move native processing to internal/external audio processors—Universal Audio, SonicCore, SSL, and several others all produce accelerator cards just for this purpose. This improves throughput, simultaneous channel count, multiple plug-in effects count, and more. It also has the added benefit of reducing the rampant piracy that permeates the audio-software industry.

Embedded x86 processors are taking hold in pro-audio applications needing connectivity or a raft of standard peripherals, such as USB or flash memory support. The main attraction is a set of ready-made device drivers and support libraries. Embedded x86 processors compare more with Blackfin than SHARC. In order to maintain our place in the market, we continue to expand our set of software-support libraries, called system services, and to augment the Blackfin core with a rich set of audio and connectivity peripherals. At the end of the day, Blackfin’s place in the market is secure.

Much has been written lately about the increasing capabilities of FPGAs. In terms of pure processing performance, they can beat out programmable DSPs. FPGAs are good at algorithmically simple operations, such as filters and FFTs. These can be easily described in gates and implemented efficiently. We’re familiar with these advantages and embedded fixed-function hardware accelerators in the latest SHARC 21469 processor. These accelerators offload computationally intensive operations, allowing the core to focus on other tasks. The combination more than doubles the overall numeric performance of SHARC.

And what about FPGAs? You and other DSP suppliers such as TI have a seemingly odd relationship with companies like Actel, Altera, and Xilinx. Sometimes you partner together, when either a standalone DSP or FPGA doesn’t sufficiently address the particular market’s performance or other requirements. But in parallel, you’re slugging it out in the market against FPGAs for more moderate-performance applications. Friend, foe, or both, where do you currently stand, and how will this tug-of-war play out over time?

Expanding on my answer to the previous question, in many of our customer applications, we have found that we coexist nicely with FPGAs, and, in most of these cases, each product brings its special advantages to the design. ADI expects to continue supplying floating- and fixed-point devices to run most algorithms. We have seen FPGAs used to implement highly specialized and customized functions that are not available in a general solution, such as handling large amounts of data movement specific to an application. The requirements are not always related to speed but instead are related to the functional needs of the system design.

Is there anything else we haven’t covered in this interview that you’d like to offer readers in closing concerning the topic of DSPs (in general, and ADI in particular) in multimedia?

Lately, I’ve been touting my version of Moore’s law. I call it the More law: No matter what you develop, someone will want more. First, there’s a push for more I/O and control. The number of channels that need to be processed is growing exponentially along with the support needed for device drivers, system services, high-level language, and uClinux.

Performance is another area. Higher speed and precision are required to process multiple data streams in real time, and increasingly complex algorithms need a high degree of integration and functionality. There is also a call to get better and better at standby and active low-power consumption. Then there’s connectivity, because handling multiformat media and advanced media types requires the support of multiple standards: Ethernet, USB, CAN, ATAPI, removable storage interfaces.

And the more you add to digital signal processing, the more system integration you need to get to lower BOM cost, reduce the device size, and drive down development time and risk. I sincerely believe that by listening and responding to these requests, we will be very successful in helping our customers create new products.