'Whisky and soda' solution to software-defined radio architecture
A group of old hands from the Cambridge wireless technology business is licensing a software-defined modem for cellular air interfaces.
Cognovo was founded last year by Tony Milbourn and several of his cofounders at TTCom, the Cambridge wireless technology company bought by Motorola in 2006. The idea was to develop an air interface modem that could be programmed to work with cellular standards ranging from GSM to LTE-A, in order to build low-cost multimode handsets.
"We set up Cognovo knowing that the time was coming for a software-defined approach to handsets," said Richard Fry, business development director of the new company, who was responsible for intellectual-property (IP) licensing deals at TTPCom.
The Cognovo team went looking for a processor architecture upon which it could build the high-performance yet low-power signal-processing engine necessary for a modern air-interface modem. Fry says an LTE modem may need to handle up to 60bn operations a second.
ARM Holdings, the Cambridge-based IP licensing company, had been developing a vector signal-processing core for a while, using tools it got hold of when it acquired Belgian design house Frontier Design, and the results of work with the University of Michigan on a research programme called SODA.
The SODA project had focused on defining the ideal instruction set for a processor handling wireless protocols. ARM used the Optimode tools developed by Frontier Design to turn that instruction set into the first iteration of what became known as the Ardbeg processor, named after an Islay malt whisky.
"The first generation of Ardbeg was just exactly that," said Fry, "but just having the processor is not enough."
ARM came to a deal with Cognovo, spinning off the Ardbeg techology to the new company in return for a stake.
"We then had the complementary parts of the jigsaw," said Fry, "and the TTPCom legacy also gave ARM the confidence that we would be able to do the business development side of it as well."
According to Mark Collins, chief operating officer of Cognovo and former commercial director of TTPCom, the TTPCom business model was to licence hardware IP to chip makers to include in chips they sold to handset makers, to whom TTCom would then license software. Since the major OEMs wanted to work with well-established chip suppliers, who were not necessarily enthusiastic about licensing anyone else's IP, this left TTPCom selling to Tier Two OEMs who were prepared to take a chance on less-established silicon suppliers.
At Cognovo the idea is to license the hardware and software IP directly to Tier One handset makers, some of whom (think Apple with the A4 chip in the just-released iPhone 4) are keen to take greater control of the technologies that are key to their businesses.
"They're the guys who can see software-defined modems are going to happen," Fry said. "The problem with selling to the big OEMs is that they only want to deal with big suppliers. But in an IP licensing model, so long as they trust the IP then they can decide who makes [the resultant chips]."
The Cognovo Software Defined Modem platform includes the Ardbeg-based Modem Compute Engine, an operating system and a development suite. It is meant for use in chips for handsets and portable devices capable of LTE Category 4, 150Mbit/s datarate connections, but can be scaled to support multi-mode operation with other standards. Cognovo claims that a single-engine design supporting WCDMA, HSPA, HSPA+, LTE and WiMAX enables a multimode baseband IC to be realised in a die area of less than 6sq mm on a 32nm process.
The development tools have been set up so that designers can choose different numbers of Ardbeg cores, depending on the complexity of the protocol being implemented: according to Collins, whereas supporting HSPA may take one core, supporting LTE-A may take six.
According to Fry, once designers have written the code to implement the most complex protocols they want to support, "the others become almost trivially easy". Power management features in the platform enable designers to switch off unused capacity when a large chip is running a simple protocol. Fry also claims that the die size of the programmable solution is "very competitive" with single-mode ASICs, and that an ASIC running a complex protocol would have to dedicate further die are to the hardware necessary to run legacy protocols.
The development flow also includes tools to take C-language models from the MatLab simulation environment and compile them directly onto the Cognovo processor.