Designers are demanding end-to-end sensor processing, right out of the box: An interview with Dipak Roy, Chairman of D-TA SystemsStory
May 25, 2010
Editor's Note: Sure, there are lots of signal processing, I/O, and FPGA co-processor boards on the market. And many of them are amazingly fast, sporting leading-edge performance and software. The trouble is, designers still need to design a system around these boards, which often come from disparate suppliers. I caught up with Dipak Roy, who has resurfaced with his new company, D-TA Systems. Edited excerpts follow.
Editor's Note: Sure, there are lots of signal processing, I/O, and FPGA co-processor boards on the market. And many of them are amazingly fast, sporting leading-edge performance and software. The trouble is, designers still need to design a system around these boards, which often come from disparate suppliers. I caught up with Dipak Roy, an old friend from my COTS sonar days, who has resurfaced with his new company, D-TA Systems. Dipak is a bona fide sonar expert, and his former company, ICS (aquired by Radstone, now a part of GE Intelligent Platforms), invented the FPDP interface that became a VITA standard. The fact that Dipak is now relying on 10 GbE for data transfer came as no surprise. Edited excerpts follow.
MIL EMBEDDED: According to your literature, D-TA’s value proposition is box-level, preconfigured systems. What does that mean?
ROY: The conventional norm for building embedded systems for military applications is to start from COTS boards from various vendors. One has to select a computer platform, OS, enclosure, and power supply, then procure software device drivers and application routines. Next, the processes of system integration and debugging commence, followed by laboratory tests and field trials. The whole exercise is time consuming, expensive, and often leads to significant cost and time overruns.
However, D-TA Systems’ approach changes the whole paradigm for demanding sensor-processing applications by offering box-level, user-reconfigurable systems based on 10 Gigabit network technology. Since our products come fully tested and require minimum customer reconfiguration, the deployment cost and time savings are dramatic.
MIL EMBEDDED: Describe the types of integration challenges customers face with board-level point solutions.
ROY: Starting from board-level products (COTS boards) and building a complex multichannel system is a time consuming and expensive proposition. For example, a user trying to implement a 16-channel HF radar system with four-channel COTS PMC ADC cards (the maximum ADC count available on current PMC cards) needs to integrate four such cards, acquire a clock source, and develop a clock distribution strategy to maintain synchronization before any application development can commence.
To help solve these issues, a fully integrated and tested functional module that can be quickly configured for a particular application is a real time-saver. Following a successful trial, customers often require customization services for repackaging, ruggedization, functional modifications, and so on, for volume deployment.
MIL EMBEDDED: What are your target markets and applications?
ROY: Defense, aerospace, wireless, and test and measurement. Our specialty is sensor-array processing, with “plug-and-play” products that cover an application space from sonar/acoustic to radio/radar.
We are targeting a very wide range, from sonar to wireless. In the sonar/acoustic area, we are looking at high-frequency and high-channel-count applications such as mine sonar, shallow-water sonar, obstacle avoidance, vibration analysis, ultrasound, and real-time acoustic simulation. In the radio and radar area, the products are particularly suited to multiantenna applications such as phased array radar, RF direction finding, smart antenna base station, signal intelligence and target location, sonobuoy, RF test, and channel simulation/emulation. Our specialty is to offer precise synchronization, from RF to baseband, across a large number of input and output channels.
Interestingly, we have also recently established an advanced sensor processing laboratory at a local university, and several faculty members and graduate students are now engaged in developing new and exciting applications.
MIL EMBEDDED: Where does 10 GbE fit into the need for deployment speed?
ROY: 10 Gigabit network technology offers data throughput rates that are faster than any computer bus. And, more importantly, it allows better system partitioning that is particularly advantageous in mixed-signal applications. Analog functions can be isolated from the “noisy” computer bus for better analog performance. Moreover, it is part of the IEEE standards process and is continually upgraded for speed and bandwidth.
MIL EMBEDDED: You are building systems using the same COTS hardware and software as everyone else. Why would a customer choose your solution?
ROY: All our products are designed from the ground up. Unlike the typical COTS board vendor, we are not constrained by board size, predefined connector pin-outs, power supply ratings, enclosure size, and so on. So we can offer an uncompromising solution that is designed to solve the problems encountered in demanding sensor-processing applications: dynamic range, analog performance, processing speed, and instantaneous bandwidth, to name a few.
Also, as mentioned, virtually all of our products are 10 Gigabit network attached, and most of them are housed in a 1U high, standard 19" rack-mountable enclosure. Our mandate is to offer end-to-end solutions for demanding sensor-processing applications. For radio and radar applications, examples include tunable RF, multichannel IF (software radio), and 10 Gigabit record and playback systems. For high-frequency sonar and acoustic applications, there are high-precision signal conditioning, 24-bit digitization, and a high channel count. These products can be seamlessly connected to build any complex sensor processing systems in a matter of days, not years. Other factors in making a designer’s life easier include libraries of pretested FPGA cores and real-time multithreaded software application routines. So our value proposition is very compelling: The customer can focus on their applications and not on data acquisition system development.
MIL EMBEDDED: What are the software considerations?
ROY: The 10 Gigabit network connection is OS agnostic, so we typically use the 10 Gigabit links for data transfer and 1 Gigabit network for control. The base SDK supplied with the products includes a Control API and Data API. The Control API allows the control of our boxes, while the Data API allows the user to build applications using example codes provided. This structure shields users from socket calls and simplifies data access. We have also developed many DSP functions for multithreaded, multicore platforms that seamlessly attach to the base SDK.
Looking toward the future, I believe that the current trend in multithreaded software processing in platforms with multiple GPP cores will revolutionize sensor processing; this will allow more real-time signal processing functions to be housed in software and GPPs, rather than in FPGAs. It will also save development time and cost and enable rapid reconfiguration.
MIL EMBEDDED: Will the sensor-processing world adopt emerging 40 or even 100 GbE technology?
ROY: D-TA Systems is the first company to introduce 10 Gigabit sensor processing. It is new and just getting started. As opposed to datacom applications, sensor-processing applications require sustained processing at a high data rate. Most applications cannot tolerate missing samples or data packets. There is no “resend” in sensor processing applications. The processor has to keep up with the data rate. That is why the sensor-processing world trails behind the datacom world in terms of sheer network speed.
As the processor gets faster and faster, we will see introduction of 40 and 100 GbE networks for sensor processing applications for the simple reason that more signal channels with higher bandwidths can be handled by each network. Moore’s law dictates that the processing power will increase to allow signal processing at these rates.
MIL EMBEDDED: How has sensor-processing technology evolved in the past 20 years?
ROY: One has to remember that not long ago, all sensor processing was done in the analog domain. Over the years, it has evolved significantly and in many ways. Today, the processing has become mostly digital, primarily because of increased data converter speed and accuracy. This has not only allowed detection and processing of low-level signals, it has also allowed data converters to get closer and closer to the actual sensor, thereby simplifying front-end analog design. For example, the advent of Sigma-Delta technology has revolutionized acoustic signal processing. The advent of FPGAs has allowed many DSP functions to be implemented in real time. The availability of high-speed networks and computer buses has allowed raw or preprocessed sensor data to be accessed by a general-purpose computer for further processing or display.
MIL EMBEDDED: What are the biggest challenges facing your customers?
ROY: The biggest challenge facing our customers is time to market. In the current tight-budget scenario, cost overruns cannot be tolerated. Further, most customers these days want to see a working demonstration before funding any development. D-TA products are designed to precisely address these problems: Virtually any sensor-processing systems can be built quickly by selecting products from our catalog.
MIL EMBEDDED: Which new technology do you think we’ll see in the next two years that’s not evident today?
ROY: Going forward, we will see more evolution in data converter technology in terms of higher speed and accuracy. The other focus will be on lower-power devices that will also allow increased channel counts. During the next several years, it might be possible to connect the data converters (ADCs and DACs) directly to the antenna. This will require not only very high-speed converters but also faster FPGAs and processors to handle the very high data rate. In the short-term, we will also witness significant evolution in RFIC [Radio Frequency Integrated Circuit] technology, which will significantly reduce the size and cost of RF subsystems.
Dr. Dipak Roy is Chairman of the Board at D-TA Systems Inc. His embedded industry experience began briefly in the form of Research Engineer. Then he took the entrepreneur route: His first company, ICS (now part of GE Intelligent Platforms), won defense contracts including the U.S. Navy’s SQQ-89 sonar upgrade program. Additionally, Dipak has authored a patent and published articles in 30 technical publications. He has received many awards, notable among them is the American National Standards Institute (ANSI) award for the invention and standardization of FPDP, a high-speed data flow concept that simplifies system integration. In 2007, he was appointed by the government of Canada to the board of Sustainable Development Technology Canada (SDTC), a $1 billion foundation developing “Clean Technology” companies in Canada. He received his PhD in Electronics Engineering from Carleton University, Ottawa. Dipak can be contacted at [email protected]
D-TA Systems Inc. 613-745-8713 www.d-ta.com