ASIC Technology

Caution: Wireless IC Design Ahead

The experiences of a company developingtheir first wireless products.

by Grant Hulse

Have you been thinking of putting a wireless link in your digital product lately? If so, you're not alone. The allure of being connected, yet unconnected, to the wired infrastructure of corporate file systems and communication channels has created an enormous market interest in wireless radio data and voice connections. But if your background is CMOS design, plunging into the RF world, even at the baseband level, will be a new experience. We know, because we've made that jump ourselves. This article is written to share our experiences in our approach to wireless design.

Background Over a year ago, American Microsystems Inc. (AMI) and Seattle Silicon became interested in the local-area private radio systems: radio links that go down the hall, around campus, or around the community. For these private systems, spread-spectrum technology, a frequency-coding method adopted from the military, has become the preferred approach. The FCC and foreign governments promote the use of spread spectrum in the unlicensed spectrums of usable airspace by allowing higher antenna power (longer ranges) for radio links that implement it properly. This allows even the small wireless system provider a venue to offer very powerful radio systems for local-area environments.

Unfortunately, many of these "niche" opportunities, often championed by smaller companies, have been hampered with market-limiting high costs because the companies do not have the product volume to cover the expense of full integration. We decided to enable these markets with a family of general-purpose, programmable, spread-spectrum transceiver ICs that can be used in a variety of wireless niche applications. Our products are CMOS baseband devices with a heavy dose of register-programmable digital functions and some analog interface circuitry.

Breadboard early and often Like all good CMOS designers, we began with a "system-on-a-chip" mentality. We toyed with putting the IF modulators and demodulators in silicon, adding more of the IF receiver circuitry, and supporting different packet modes and preamble codes. Hindsight clearly reveals that getting the design ideas produced very early in the lab was our most important decision. Although we funded a radio prototype development primarily to verify our receiver synchronization and tracking capabilities, we soon discovered the lab radio provided answers for every marketing or specification decision that had to be made. Modeling our device with an FPGA for the digital portions and using discrete components for our analog portions, allowed us to verify the radio's performance with any configuration of programmable options we wanted to check. In all cases, the breadboard testing guided our decisions.

The lab radio work made us very cognizant of all aspects of the system design during the early design phase. As we worked through various design ideas, we costed out the radio right down to the bill of materials. RF component lead times, prices, performance, and supplier reputations quickly became part of our design equation. At first, we planned to offer a variety of popular modulation schemes (BPSK, QPSK, DBPSK, DQPSK, QAM, MPSK), and to allow the user to customize packet styles and preamble coding. Our Verilog simulations showed no problems with the variety, but, once implemented in the radio, several combinations were discovered that could not be supported due to poor tracking, phase inversions, and other "radio" problems. The radio prototype work added man-months of cost to the design phase of the project, but probably saved us several turns of silicon.

Figure 1. The relative timing relationships of the data with the actual digital modulation showing the increased bits required for encoding the data signal.

Listen to experience The U.S. engineering community has broad capability in radio design and development. Based on our experience, we would advise you to use it. We brought our radio consultant in at the beginning of the specification phase and continued to leverage the expertise throughout the design. Since our receiver synchronization and tracking loop interfaces to analog pieces both on and off the IC, modeling the loop properly both in the simulation deck and in the lab radio was paramount to success. Long discussions about what silicon engineers could do with CMOS and what radio engineers had to have for performance resulted in some proprietary breakthroughs on several fronts in the receiver loop. Because we involved our consultants so deeply in the process, it is hard to say how tough the road would have been had we used them more sparingly, but intuition suggests it would have been a much harder challenge.

That said, it's important to select an RF consultant who can deliver the goods, since radio design charges alone can quickly exceed $50,000.

Watch the design boundaries Our cost analysis revealed that leaving much of the IF functions in discrete analog outside our IC provided the best system cost and design ver satility for the radio designer. We found a variety of monolithic IF and RF ICs in production or in development built in BiCMOS, GaAs, or bipolar technologies that offered better specification performance than we felt we could reliably achieve, so we concentrated our radio-side effort on supporting good interfaces to this group of products. We added on-chip programmability for a variety of modulation formats and various control signals for off-chip IF and RF circuits. Overdoing the integration of the baseband CMOS portion with today's capabilities is something that should be guarded against.

The system-level power budget was also key. Probably the most critical decision in our design was the selection of a synchronization, or correlation, technique. Parallel correlators have been developed that provide synchronization in only one or two symbol times. However, these require digital-processing techniques that, at high chipping code rates (we wanted a 64 MHz maximum baseband rate), have to use very expensive and power-hungry A/D converters to get the signal data in digital form for manipulation. We chose instead to implement a serial, or sliding, correlator that can accept longer signals and higher-frequency codes at much more reasonable system power and price points. The downside is longer initial synchronization time. This forced us to target applications that could tolerate longer setup times in order to keep their power usage down.

We were also careful to add separate sleep modes and power down modes for the transmitter and receiver. We also specified a 3V power supply requirement, though many systems still need 5V levels in the RF sections.

Decisions about power and integration levels provide function, performance, and cost tradeoffs that reach far beyond the boundaries of the IC. They must be well understood at the system application level very early in the design process.

Figure 2. The normalized frequency spectral density of the basic and encoded data depicting the processing gain.

Design the radio not the IC Our product goal is to offer, via programming on one IC, all possible combinations of symbol rates, PN (psuedo-noise) code lengths, and PN chipping rates up to the practical limitations of the unlicensed spectrum bands at 915 MHz, 2.4 GHz, and 5.7 GHz. We allow PN code lengths from 3 to 2047 chips; any symbol rate below 1 Mbps; and all the resulting PN chipping rates below 64 MHz (at baseband). We also offer a variety of symbol modulation formats such as BPSK, QPSK, and QAM. And we give users the ability to look at noise measurements in any of their target channels and use that information to make changes on the fly in PN codes, code lengths, data rates, and antenna power levels.

Taken to its extreme, this level of programmability gives us hundreds of possible radio configurations. While this sounds great-even possible-from our digital CMOS perspective, the reality is that each configuration is indeed a different radio with different channel-center frequencies, different RF filtering requirements, and different spectrum signatures. To simulate and prove digitally that each of our combinations are properly in place is a straightforward IC test function. But to test or guarantee a specific programmable radio configuration can only practically be done in our customer's actual radio design.

Incorporating several different code lengths or symbol rates into a single radio adds different filtering paths and selection circuitry. Component costs can vary sharply depending on performance, range, and the chosen spectrum. We have received cost quotes for radios that use our device that vary from $70 to several thousand dollars per unit based on the performance needed.

Gear up for support High-frequency equipment for radio evaluation and support is relatively expensive and should be factored into the business plan. A spectrum analyzer, a high-frequency signal generator, modulation equipment, and a new assortment of cables and discretes topped our growing shopping list. The learning curve required to operate and interpret the information from the equipment can also push out lead times.

Because we have continuing wireless IC plans, we have made those purchase decisions, but we also contracted most of our early -RF lab work to an outside source because of our lack of know-how and equipment. Both approaches, however, will add time and money to your schedule, so it's important to review your equipment needs early.

Another dimension of wireless support is conformance to FCC rules and guidelines. Since we provide evaluation radios that showcase the features of our devices, we needed a functional understanding of at least the "Part 15" regulations. This has not been as complicated as we first expected. We have found FCC officials to be very helpful and informed, but the certification process does take time and must be planned for in the time-to-market calculations.

Here's another reminder: Since the radio needs to meet FCC broadcast regulations in each of its possible settings, it must be certified at each of those settings. We suggest that you contact your FCC office early and often to learn the requirements and then make sure your schedule includes enough time.

Putting together a wireless business plan and design approach is an arduous task. It's important to plan well, but to be prepared for surprises. We jumped in with a little less knowledge than we now have, but the market potential and value we feel we are adding to this burgeoning opportunity clearly made it a good decision.

In the course of bringing our spread-spectrum products to market, we have talked to hundreds of potential designers. Many are very aware of the wireless challenges and are prepared to meet them, many are very aware of these challenges and sagely prefer to buy completed radio modules from someone else, and many more still believe that getting into wireless won't be too tough once they dive in. With the right upfront information, all three of these perspectives can work. So my advice is "Come on in­the water's fine­(once you get used to it)!"

Grant Hulse, director of the Standard Products Division of American Microsystems Inc., joined AMI in 1993. He has an engineering degree from Bringham Young University and a MSEE from Stanford University. Grant worked at AMI prior to his graduate studies and worked at NEC Electronics before returning to AMI.

To voice an opinion on this or any Integrated System Design article, please e-mail your message to: michael@asic.com.

integrated system design  September 1995