THE biblical account of the construction of the Tower of Babel tells about an ancient building so tall that its designers hoped it would reach the heavens. In the story, God thwarts the plan by making the workers speak different languages so that they cannot understand one another and the ambitious tower is never finished.
Today, this monument to confusion lives on in the world of telecommunications. Take the transmission languages of mobile phones. These differ from region to region and even from one phone system to another. Fly from Europe to America and the chances are that your mobile phone will not recognise the local telecoms lingo when you arrive.
Worse still, when mobile phone companies change their language, old phones are unable to learn it, rendering them obsolete overnight. The risk of losing customers this way stops telecoms companies from investing in new, more efficient telephone systems and represents a major obstacle to improving services. 鈥淚n the US, we鈥檙e still using a cellular technology based on an electronic format that was developed and formalised in 1978,鈥� says Vanu Bose, a researcher at the Massachusetts Institute of Technology in Cambridge. 鈥淭here have been tremendous advances in technology since 1978 but we don鈥檛 see them because of the huge cost of upgrading cellular base stations and handsets.鈥�
Advertisement
This is about to change. Researchers in the US are developing transmitters and receivers that can learn any transmission language, an ability that makes them almost infinitely upgradeable. Known as 鈥渟oftware radios鈥�, the new devices will be far more flexible and capable than any conventional radio. Inside them, powerful chips will analyse digital versions of radio signals. And since mobile phones are essentially radios they will be among the first to benefit.
The amazing power of digital analysis allows a receiver to simultaneously monitor every transmission made over a huge range of frequencies, scanning them for news broadcasts, weather forecasts or keywords and phrases鈥攁nd, of course, any mobile phone calls. Learning a new digital language is a simple matter of switching software.
Brain wave plug in
But the best is yet to come. Replace the antenna from a software radio with, say, a sensor that can pick up electrical signals from our bodies and you have the basic components of an electroencephalograph (EEG) for measuring brain waves or an electrocardiograph (ECG) for heart beats. Load the software that can make sense of these new signals and away you go. Suddenly the fantastic potential of software radios becomes clear: they are what will make possible a new generation of electronic superdevices capable of taking on almost any task.
To understand how software radios work, take a look at the most common example of incompatible transmission languages鈥擜M and FM radio broadcasts. In simple terms, amplitude modulation encodes a signal鈥攁 music recital, for example鈥攂y varying the amplitude of a radio wave. In a conventional radio receiver, this signal is picked up by an antenna, amplified and passed through a tuning circuit that filters out all but the desired frequency. This signal is then decoded, or demodulated, using a circuit that produces an outgoing signal proportional to the changes in amplitude of the incoming signal.
Frequency modulation, on the other hand, encodes the recital by changing the frequency of the broadcast. The signal can only be demodulated with a circuit that produces an output that depends on these changes.
Both FM and AM receivers are hard-wired for their separate purposes. So, since an AM demodulator is obviously unable to demodulate an FM signal and vice versa, a conventional radio can only receive both types of signal if it has both types of circuit.
Software radios take an entirely different approach. 鈥淒igital technology has progressed to the point where there鈥檚 very little about a communications device that requires dedicated hardware,鈥� says John Guttag, professor of computer science and engineering at the MIT. With his colleague David Tennenhouse, Guttag leads MIT鈥檚 research programme in software radios. 鈥淚f you have a piece of wire for an antenna, a pre-amplifier, a device known as an analogue-to-digital converter and a few computer chips, everything else can be done in the software,鈥� he explains.
The trick is to turn the analogue broadcast signal into a digital signal that a computer can handle. This is done by an analogue-to-digital converter (ADC) which carries out a task that is rather similar to plotting the measurements of an analogue signal on a piece of graph paper. Along the bottom of this graph are intervals of time indicating the sampling rate. This must be at least twice the frequency range of the signal that is being digitised to get a useful result鈥攁nything less would not give a good measurement of the signal.
Up the side of the graph, an ADC would plot the amplitude of the wave, divided up into a predetermined number of intervals known as bits. 鈥淭he more bits you have per unit of time, the more detail you have about the information encoded in a radio signal, and more detail means greater clarity,鈥� explains Guttag.
This clarity is crucial because software radios will not be dealing with simple sine waves but vastly complicated ones that span huge frequency ranges. The processors inside software radios must make sense of these waves. The better the resolution of the digital 鈥減icture鈥� of the wave, the easier this becomes.
Fast scanning
To make sense of these waves a chip must break them down into their component parts: the transmissions in which it is interested. This is possible with a technique known as Fourier analysis and one that computers are ideally suited to carrying out. This is the way in which software radios are able to monitor all the transmissions over vast ranges of frequency simultaneously. And once the transmissions are in digital form the chips can play around with them, selecting specific broadcasts and decoding them with a specific transmission language like FM or AM depending on what software is loaded.
Apart from an antenna, a software radio consists of an ADC, a processor for analysing the digital signals and a high-speed link for shunting data around. Each of these devices is pushing current technology to its limits.
ADCs are a good example. To digitise broadcasts at 2.5 gigahertz (2500 megahertz) which have a bandwidth of 100 megahertz, an ADC must have a sampling rate of twice this or 200 megahertz. But suitably powerful ADCs are not yet available, so the team at MIT has demonstrated the techniques with lower-frequency broadcasts which have much narrower bandwidths.
AM radio is confined to frequencies that range from 535 kilohertz to 1605 kilohertz鈥攁 bandwidth of about 1 meghertz. So an ADC with a sampling rate of twice this frequency, 2 megahertz or more, is enough to analyse the entire band. Like computer chips, ADCs have been growing in speed and power for more than a decade. 鈥淭here are ADCs that can grab 500 million samples a second,鈥� says Bose. 鈥淏ut speed isn鈥檛 a problem. Precision is.鈥�
The problem with these ADCs, which operate using only 8 bits, is that they produce a poor resolution picture of the wave鈥檚 amplitude. This is the same as saying that there are only eight squares on the vertical axis of the digital graph paper. 鈥淓ight bits isn鈥檛 enough to do the sophisticated signal processing that software radios need to do,鈥� says Bose. 鈥淔or that you need at least 12-bit ADCs.鈥�
Today, the best 12-bit ADCs have a sampling rate of about 40 megahertz. More than enough to sample the AM bands and the entire FM frequency band from 88 megahertz to 108 megahertz but not yet anywhere near enough to handle the 2.5 gigahertz broadcasts.
This digitising process generates some 480 million bits of information per second. Pushing this amount of data around is tough.
Help is at hand, however, from a technique known as direct memory access which is used to move data around inside ordinary PCs. Mike Ismert, another member of the MIT team, has modified the technique for the prototype software radio so that the data passes from the ADC to the processor at a billion bits per second, more than enough for these purposes.
The task becomes more difficult if the antenna and ADC are physically distant from the processor: for example, in wireless computer networks where the signals picked up by each receiver are sent to a central computer for processing.
Today, a typical office network carries up to 10 million bits per second, nowhere near enough capacity. To handle the increase in data, the MIT team has turned to a new technique for pumping data. Known as asynchronous transfer mode or ATM, it is capable of handling more than a billion bits per second on a single optical fibre. By comparison, the best telephone modems manage only 56 thousand bits per second. Guttag, Tennenhouse and their team use an ATM network to pump information between the ADC and the processors.
Even with ATM, a bottleneck occurs when this data is passed into a processing chip. The reason is that ATM is designed to pump information for many users at once around a network. The data is sent in packets which must be labelled and sorted, and this is done extremely quickly inside the network. However, the operating software in a chip is not so good at this task and the process of inspecting the packets is time consuming. 鈥淧rocessed data comes out of the device at maybe one-third of the speed data is coming in,鈥� notes Bose. 鈥淭hat鈥檚 not good enough.鈥� Only the help of a few tricks allows the data to be used effectively.
Processing the data is another huge task. Computer chips carry out simple operations, such as adding or integrating, on each packet of data that passes through them. Normally, a series of operations are needed to perform a specific task. For example, the analysis required to demodulate FM signals can be done in four simple steps. Unfortunately, each step must be carried out on every sample. With 40 million samples passing through each second this requires a processor that can perform 160 million operations per second. This is within the grasp of the chips in workstations but not yet possible with the ones in PCs.
Sealed in a chip
Carrying a workstation around instead of a Walkman radio would not be much of a trade-off. But there is another type of chip known as a digital signal processor that can do the job instead. DSPs are hard-wired to process the raw data generated by ADCs and are already used in some mobile phones. The latest models can easily outperform general-purpose computing chips such as Pentiums when it comes to real-time processing.
But it won鈥檛 be long before general-purpose chips can do this too. 鈥淭he speed and memory capacity of computer chips are doubling every eighteen months. The best solution to a lot of the problems with speed and flexibility is just to wait a year,鈥� says Guttag.
Perhaps the biggest problem will be devising a portable energy source to power software radios. Wideband ADCs and high-speed chips are power-hungry creatures. And while battery research is proceeding apace, few expect a handy-sized solution this century. So the first applications for software radios will be at fixed sites with access to mains power.
In the longer term, the first large-scale beneficiary will be the US military which uses more than 100 different types of radio. Infantry troops need low-frequency radios for short-distance communications that can pass through walls, foliage and around small hills. Air forces, on the other hand, need to send data faster and farther, a task demanding higher frequencies, different antennas and different signal formats. Today, these radio systems chatter away in the equivalent of foreign languages.
Enter the US Department of Defense and a project which it calls Speakeasy, designed to build software radios that will talk to each other. 鈥淪oftware radios would allow military units using different radio formats to communicate on more secure channels,鈥� says Joe Mitola, an engineer with the Mitre Corporation, a defence consultancy near Washington DC. These radios would even negotiate with each other to find the most secure encryption format that they can share. 鈥淭his is similar to the way fax machines from different manufacturers negotiate the highest speed at which they can communicate,鈥� explains Mitola.
Speakeasy radios are some time from hitting the battlefield but simpler applications are already available. This month, AirNet Communications, a telecommunications company based in Florida, begins selling a software-based cellphone base station. The station is designed to understand a transmission language known as GSM. But with the aid of software radio technology, the station can be modified to understand almost any other language that may be devised in future. 鈥淲e can even run two different formats off the same equipment just by adding more software,鈥� says John Chenoweth, an engineer and marketing manager for AirNet.
At MIT, their prototype radio can already monitor and demodulate the AM and FM bands, and the team is working on software that will understand the electronic languages used in mobile phone systems. Next may come software that can make sense of signals from the global positioning system. A mobile phone that uses GPS to know where it is would be able to dial into local traffic and weather reports or display maps of the area and local information.
Beyond that, researchers are working on software that encodes signals for different broadcasting conditions. So when the signal is susceptible to noise, in heavy rain or in an urban canyon, for example, the phone would automatically switch to an electronic language with stronger error correction built in. Of course, all this must happen seamlessly without disturbing the user. Just how the on-board computer would be able to detect the outside conditions and know how to switch isn鈥檛 yet clear. 鈥淭here are lots of systems and software issues that still need to be solved,鈥� admits Guttag. 鈥淲e鈥檙e prototyping and testing different software algorithms, but the technology still has a way to go.鈥�
Perhaps the most exciting application for software radios will be superdevices that can take on almost any measurement by changing the transducer that takes the reading. Ambulances might carry all-in-one medical devices that act as heart monitors, brain-wave monitors and thermometers with different attachments.
The cost savings could be enormous. 鈥淢edical instruments are tremendously expensive,鈥� says Bose. 鈥淏ut when you buy an ultrasound machine and a heart monitor, you鈥檙e buying similar signal processing systems. But with software radio technology, you would buy several machines for the price of one.鈥�
Guttag and his colleagues expect software radio technology to trickle into fixed installations within five years. 鈥淎nywhere there鈥檚 ample power,鈥� predicts Guttag. But when the battery problems are solved, expect to see software radios everywhere. 鈥淯ltimately, a single software radio device can replace any of your current devices,鈥� says Bose. 鈥淏ut, more importantly, it can replace all your devices with something better.鈥�