Software and digital logic are encroaching further on analogue territory in mobile radio.
The US Federal Communications Commission recently raised a record $44.9bn in bids from companies wanting mobile spectrum around 1.7 and 2.1GHz - double the last big US spectrum sale in 2008.
It’s a sign of how regimented and crowded the airwaves are, forcing individual systems into narrow chunks of the overall spectrum. More agile radios could change the way we manage radio spectrum, opportunistically using spare capacity where it arises. The ability to conjure up different radios in fractions of a microsecond using the same digital fabric that can run applications to process sensor data is a utopian dream for wireless.
Mainstream consumer devices need no longer be limited to communicating on fixed frequency channels, instead they could hop promiscuously around the airwaves exploiting the estimated 90 per cent of spectrum that isn’t in use at any one time.
Moreover, being made from code, such radios could even be ‘deleted’ when stolen, or upgraded to a higher performance for premium users. Most crucially, the technology would benefit from the kinds of size and power reductions digital circuitry often sees. But that calls for a change in the architecture of RF systems from the top to the bottom to make them digital. The front-end, predominantly analogue sections of today’s radios are highly resistant to change. A British team has taken a big step towards changing that.
Late last year engineers at the product design and technology consultancy Cambridge Consultants built what appears to be the first true software radio. Developed from off-the-shelf programmable logic and very little else, the device is a transmitter that can send data at rates of gigabits per second at carrier frequencies up to 14GHz, switching between different carrier frequencies almost instantaneously.
“If you analysed the output your best guess is that we had somehow achieved the impossible and were running a band pass sigma delta converter at many GHz,” said Monty Barlow, Cambridge Consultants’ director of wireless technology.
True software radio has, however, remained elusive since it was proposed in the early 1990s. The closest we’ve got is software-defined radio (SDR), which is based on a combination of digital and analogue circuitry.
In SDR, low-frequency signals are generated and analysed by software but the conversion of these signals to and from radio frequency uses analogue hardware. The conversion typically only operates on a certain range of frequencies (at best a 10:1 range). Changing frequency takes at best several microseconds since the analogue takes a while to settle after each change.
Analogue methods exist for faster switching of radio carrier frequency, signal bandwidth and sample rate, but they tend to be expensive, power-hungry or both because they depend on the duplication of parts.
A true software radio transmitter would feed a stream of numbers into a super fast digital-to-analogue converter (DAC) connected to a radio antenna. A fully software receiver would be able to take a high frequency signal from an antenna directly into an analogue-to-digital converter (ADC), with a digital signal processor transforming the resultant bitstream into useful data.
The hurdle the Cambridge Consultants team has overcome, on the transmitter side at least, has been converting quickly and accurately enough between the digital and the analogue domains, without relying on exotic and expensive chip manufacturing technologies such as gallium arsenide.
They began the development last year in expectation of the imminent arrival of digital power amplifiers – a technology the company is also working on. They reached their goal in just three months, making the transmitter from a field programmable digital chip called the Virtex-5 FPGA, made by Xilinx, connected to an antenna.
The input to the FPGA is low-frequency digital data bits, while the output is a high-frequency digital bitstream that comes out of one of four serial channels. In between, the algorithm takes advantage of parallel processing. “Our trick is to run several buffered streams in parallel. The more parallel we go, the more memory we use but the slower we can run the processing in each stream. Multiple streams are then combined to form one digital output signal,” Barlow explains.
In this way, the majority of the fabric of the device, whether it is an FPGA as now or in the future a custom digital chip, can run at a speed of around 100MHz. FPGA-implemented logic needs to run comparatively slowly because the reprogrammability of the devices tends to lead to longer connections between gates than in full-custom chips, so a key element of these parts is the serialiser-deserialiser, which takes parallel data and packages it up to be streamed in high speed serially.
The use of serial channels for off-chip communications generally results in lower noise and higher effective speeds, as it is easier to condition a few serial channels for multi-gigahertz operation than try to make tens of parallel I/O lines run well at hundreds of megahertz.
Breaking the task into many parallel chunks that can run at low speed inside an FPGA is not a new idea. What has eluded other design groups is finding a way to avoid discontinuities when joining the pieces back together. The Cambridge Consultants team considers its breakthrough to be ‘re-stitching’ with maximum fidelity.
“It’s a way of breaking it into block processing, with each chain working on, say, 1000 bits in a row,” says Barlow. The company has applied for patents around the algorithms used to compute the software radio signal in parallel blocks and then tie them together coherently.
The highest carrier frequency achievable is around half the bitrate of the serial ports, which can be up to 28Gbit/s a Virtex-5. “In practice we’d need to take a lot of care with the design of the main digital clock to get a usable signal at 14GHz carrier frequency, but lower frequencies such as 10GHz are comfortable,” adds Barlow. The team is now scaling the technology to ensure that it still functions well using the higher-speed serial modes.
The firm’s technology demonstrator board runs at a 3Gbit/s data rate, uses a carrier frequency of 900MHz suitable for GSM and EDGE protocols, and supports 100MHz of bandwidth. All modulation schemes are apparently possible, because the fidelity of the signal is very high within the passband – the range between the upper and lower cut-off frequencies. “We chose GSM/EDGE for our example because we could put out loads of different signals across 80MHz or so. Specifically, 14 narrowband signals around 200kHz wide, each equivalent to a single carrier on a base station. But LTE, UMTS or any other signal would have been equally possible,” says Barlow.
One could dismiss Cambridge Consultants’ ‘all digital’ claim because it still has an analogue filter on the demonstrator board. Analogue filtering is needed to meet any legal requirement for transmitting without interference with adjacent radio bands. “If you want to operate around 900MHz and not destroy 1GHz and 800MHz communications while you’re about it, you need to roll it off,” Barlow explains.
“With a digital transmitter, every time we go faster and we make our passband wider, we can reduce the performance of the analogue filters needed, allowing them to have a much gentler roll-off,” says Barlow. Continued silicon scaling will help reduce the amount of space the circuitry needs on future processes. “That is not the trajectory that analogue is on.”
Without a power amplifier, the transmitter and antenna combo can only transmit to a receiver positioned on the other side of a small room. An analogue PA would work perfectly well to increase the range but a highly efficient, fast digital PA would be better, and naturally would need a digital stream. “You would filter it after you’d amplified it,” says Barlow. Digital power amplifiers may one day be integrated on-die on chips containing purely digital radios.
Receiver design is in progress but is an order of magnitude more difficult than the transmitter. “With a transmitter, you know what you’re going to send in advance so you can look 10 or 100 bits ahead and add only a tiny bit of latency, whereas in a receiver, that analogue input is there for an instant in time. You need to perform the conversion to digital very quickly,“ says Barlow.
Despite the challenges in receiver design, there are some promising avenues to explore. Conventional receivers use a mixer to ‘slide down’ high frequency radio signals to a more manageable low frequency. This process involves synthesising fast, clean signals to mix with. The team is looking at digital equivalents since it can now synthesise very fast analogue-like waveforms. Equally, the very fast DAC the team has built could form the core of a fast ADC.
William Webb, president of the IET and previous director of Ofcom, has been an advocate of dynamic spectrum access, that is moving from band-to-band as a way of using spectrum more efficiently. He is, nevertheless, cautious about the implications of Cambridge Consultants’ development. “It will add a lot of flexibility for prototyping. If one had a small radio chip flexible enough to be programmed so you could test the market easily that would be very useful for some applications,” he said. “Would the dynamics of market change or the way we allocate spectrum? Well, I guess eventually people would use the flexibility of such a technology but I think that would involve a long time frame.”
Cambridge Consultants is not necessarily looking to own or immediately license the transmitter technology but wants to show that it is possible while looking at the implications for clients who use radios in their products or develop radio chips. Part of this process is predicting the point at which a transition to pure software radio will make sense for the myriad of applications the firm works on using conventional radio designs.
Barlow and colleagues have already looked at the economics of implementing the transmitter circuitry on a 28nm CMOS process and calculated that it would occupy a fraction of a square millimetre and cost around 7 cents now and perhaps 4 cents next year, which would make it available to integrate for anyone making moderately complex digital chips.
Another line of enquiry is whether there is a fixed overhead with true software radio that means unless you need a certain level of sophistication you needn’t bother. “Our feeling is that it is going to be possible to make a limited but incredibly cheap digital radio as well as sophisticated ones with this approach,” says Barlow.
Today the software transmitter has a relatively high power consumption and complexity, but for niche applications it might provide a benefit not just in terms of product flexibility but service cost. Over their lifetime smart meters could be expected to send data on energy usage to a variety of service providers using different communications systems as consumers switch supplier. A fully reprogrammable RF transceiver would make the switch possible without having to send someone out to swap radio modules.
The technology as it stands also opens up some interesting options for building wireless transmitters, including sending radio streams out of more than one digital serial channel and doing beam-forming and other sophisticated radio techniques that reduce the transmit power required by focusing the energy in a specific direction.
Three or four years on (with help from Moore’s Law), the company thinks Wi-Fi access points could benefit from software radio. “Maybe the same piece of kit could be sold worldwide and handle other parts of the spectrum such as TV white space at around 700MHz or above 6GHz. Currently you buy a unit and it does whatever frequency it came with,” suggests Barlow.
Today’s smartphones do not need to be agile in their use of spectrum, and the analogue circuitry they use is very tailored to the application. Once software radio technologies have shrunk and matured to the point where they are almost cost-free, arguably it might make sense to change standards to make use of the flexibility of the technology, and fulfill the dream of cognitive radio.
For Cambridge Consultants, the development carries one clear message: in time our radios will be made from code. “However disbelieving someone is or how wedded to analogue they are today, at some point they will have to accept the inevitable,” says Barlow.