Wireless USB in a dongle
Click to enlarge
The Cypress USB controller attaches directly to Wisair's 531 baseband/MAC (media-access-controller) chip. This device, which Wisair designed in 130-nm CMOS, creates everything but the RF portion of a WiMedia UWB (ultrawideband) radio platform. The chip permits the wireless-USB link to operate at 53 to 480 Mbps and still coexist with nearby Bluetooth and 802.11a/b/g transceivers without external filters.
Click to enlarge
A wireless-USB hub starts with a connection to the bus. Seeing no differential advantage in designing its own USB interface until it could integrate it into a single chip, Wisair purchased the wired-USB transceiver, protocol engine, RAM, and general-purpose microcontroller for its dongle in one package—in what appears to be a Cypress Semiconductor 7C68013A USB high-speed peripheral chip. Doing so gives it a fast 8051 and 16 kbytes of memory to work with for system control, interfacing the USB port to the wireless-baseband device and driving the activity-monitor LEDs.
The Wisair 531 teams with a 180-nm SiGe (silicon-germanium) BiCMOS RF device, the Wisair 502 wireless-USB PHY (physical layer). The 502 is a full OFDM (orthogonal-frequency-division-multiplexing) RF section, deploying three 528-MHz UWB sub-bands within the 3.1- to 4.8-GHz band. To aid integration, the chip requires less than 10 external components, including only one crystal, and a integrates both a voltage-controlled oscillator and a bandpass filter. The chip supports two-way antenna diversity, and the company claims that the device achieves effective bit rates to 480 Mbps at ranges of greater than 10m.
Taking advantage of the RF chip's ability to handle two antennas, the reference design provides two loop antennas embedded in the pc board. Note that they are both in the same plane, keeping everything compact but limiting the ability of antenna diversity to capture difficult signals. The silver connectors at the bases of the antennas are test points.
Circumstances are conspiring to make the use of reference designs almost mandatory. Consider, for example, the case of wireless USB. The basic idea is irresistible: Use a short-range, multiple-access wireless network to replace the tangle of cables that USB promised to eliminate but in fact reinforced. UWB (ultrawideband) radios operating at very high frequencies make possible data rates comparable with the fastest USB interconnects. Atmospheric and structural attenuation will keep the signals localized, minimizing interference problems with nearby systems.
But then come the problems. Except for Bluetooth radios, which operate at such low data rates and over such small distances as to be inappropriate even for USB replacement, UWB radios are not generally compact, low-powered or inexpensive—all attributes that would be absolutely necessary for a USB cable-replacement technology. We are, after all, talking about replacing two connectors and a chunk of cable, not hauling away a rack of equipment.
The solution, as in the case of Bluetooth before, is integration. Ideally, you could cost-effectively replace the connectors and cable with two radios, if the radios are on little tiny dice in very-high-volume production.
But Wireless USB isn't a standard yet, so it would be very hard for any vendor to get into high-volume production with an existing chip set. And since the necessary operating frequencies are very nearly beyond the reach of RF CMOS, the IC part of the radio will have to be either a pretty substantial die built in some exotic process, or at least two dice in different processes, only one of which could be commodity digital CMOS.
So the wireless-USB adapter will have to comprise at least two chips, and maybe more. And, given the very small amount of real estate available—inside a mouse, for example—the entire device, including antennas, will have to be quite compact.
And that's where the reference design comes in. Very few, if any, peripheral-equipment vendors have USB-controller expertise and UWB RF and baseband expertise and antenna design skills for multiantenna systems and the ability to lay out a highly compact circuit board with an RF section operating in the neighborhood of 5 GHz. In short, most of the users of this technology will depend on reference designs.
Hence a company like Wisair, a five-year-old Israeli company, can't expect to get a quick start in the wireless-USB market simply by selling a working radio. The company has to package up its silicon solution in a reference design—preferably, a production-ready reference design already laid out, tested, and supplied with the necessary software.
That's exactly what the company has done with its USB Host Dongle reference design (block diagram). The project began with an architectural specification for a short-haul UWB data radio several years ago, well in advance of the FCC approval of the spectrum allocation and well in advance of the standards work on Certified Wireless USB (WUSB). This work was realized in a pizza-box sized prototype built from FPGAs and discrete RF components. Intel and the FCC used that prototype for examining the WUSB concept.
Success of the big prototype led to the development of an initial chipset in October 2004, covering eight 528-MHz bands in the spectrum between 3.1 and 7.5 GHz. And that, in turn, led to the development of the two-chip configuration in the current product. Wisair director of marketing and business development Serdar Yurdakul says that the two-chip solution is itself an interim step on the way to a single-chip radio due to sample in December 2006.
The dongle reference design gives us a nice look into the architectural and the make-versus-buy decisions that a company faces on the way to single-chip integration. To start with, Wisair decided that spending engineering resources to develop a wired USB interface made no sense, with companies like Cypress selling single-chip USB microcontrollers at commodity prices. So into the design went a Cypress USB/MCU.
Then there was the matter of the radio. Architecturally, Wisair—a company whose roots are deep in Israel's government-sponsored digital-signal-processing infrastructure—decided to take a frequency-domain approach to UWB. "The OFDM [orthogonal frequency division multiplexing] receiver is really just a 128-channel spectrum analyzer," Yurdakul explains. "We chose to implement it in the digital domain with Fast Fourier-transform [FFT] techniques, allowing us to have very sharp edges in the frequency domain without a lot of time-domain filtering. Similarly, the transmitter is driven from an inverse FFT."
That strategy led to the selection of dense, 130-nm CMOS for a baseband chip that would include a custom datapath for bit-level operations, and what Yurdakul calls "a fairly busy ARM-9" for symbol-level operations and system control. At the bit level, the chip's transmit path is conducting scrambling, interleaving, convolution codes, inverse FFTs, and digital-to-analog conversion, including one of the first uses of a Viterbi algorithm at these kinds of frequencies. On the ARM side, there is the 128-bit AES encryption and easy ID necessary to the WiMedia standard, plus a detect-and-avoid algorithm that allows the radio a certain amount of agility in coexisting with 802.11 or Bluetooth neighbors. This is handy for reducing the need for external RF filtration, but it may also prove key in getting European and Japanese regulators to approve the frequency allocation for the WiMedia standard.
Interestingly, the baseband design provides for at least two-way antenna diversity. This is not, Yurdakul explains, intended to cleverly subtract two signals to eliminate multipath interference. "In OFDM, multiple paths just allow us to gather up more energy," he says. "The idea of multiple antennas is to get as much energy as possible." One might add that two small antennas gather more energy than one, but don't require the RF filters for WiFi and Bluetooth suppression that a single, higher-gain antenna demands.
Unfortunately, the commodity 130-nm mixed-signal CMOS process was not suitable for producing either the 3.1- to 4.9-GHz wideband low-noise amplifier or the similarly wideband power amplifier the design required. That responsibility fell to a separate chip designed in IBM's 180-nm SiGe process.
So what is the next step? Clearly the company can't rest on its laurels with three chips in the dongle. In fact, integration is underway, using a foundry off-the-shelf RF CMOS process. "Given the bandwidth and noise budget involved, we are moving very carefully from 180-nm SiGe into an all-CMOS RF design," Yurdakul admits.
Once the single-chip solution is unleashed, Wisair is hoping for wider adoption of the WiMedia standard—and of course of its chip. That will lead to more hands-on experience for OEMs, and to the design of more application-specific, single-chip devices, in which the WUSB radio simply becomes another block in the device's SOC (system on chip). And so the three-chip reference design will become a single-chip customer design. But it will still probably include a good deal of reference-design input from Wisair. Layout at 5 GHz on FR-4 is still not trivial.
The above is an extended version of an article that appeared in shorter form in the print edition of EDN. This PDF file shows the printed version.