10 November 2011
The analogue barrier
The trend in RF design over the past 20 years has clearly been towards greater levels of digitisation in order to squeeze more capacity out of some of the most congested parts of the radio spectrum. At the same time, operators want to reduce power consumption in their basestations and hubs while users want better battery life out of their handsets. These factors are uniting to force big changes on the analogue front-end.
But as you get closer to the antenna, it becomes harder to simply transfer more work onto a digital signal processor (DSP). Moving towards more digitally focused protocols tends to reduce efficiency, rather than increase it, when you look at devices such as power amplifiers.
In the shift from 2G protocols such as GSM to the code-division multiple access (CDMA) schemes of 3G, the modulation techniques became more complex. GSM uses the Gaussian minimum shift keying technique, which shifts the phase of a constant-amplitude sine wave backwards and forwards to relay data. The developers of the standard adopted this technique because it allows a linear power amplifier to operate in its most efficient, saturated region, helping to maintain battery life.
The demand for high data rates that came with 3G pushed protocol designers into adopting more complex modulation schemes, such as quadrature amplitude modulation, in which phase and amplitude are altered. To improve the accuracy of the signal output, the power amplifier has to back off further into its linear region, which is less power-efficient than working close to the distinctly non-linear saturated region that would be used traditionally. Digital pre-distortion can improve the situation by making it possible to get closer to the saturated region, but it's not a complete solution to the problem. And the situation is not improving for the power-amplifier designer.
As modulation schemes become more complex, the peak to average ratio worsens. In LTE, peak-to-average values are close to 10dB, compared to 7dB for 3G UMTS and 3dB for GSM (see fig 1).
One answer was to look back in history at a power-amplifier design developed before the Second World War. In a concept demonstrated in May 1936 using vacuum tubes by Bell Labs researcher WillFhiam Doherty, two amplifiers are used in parallel: one supplies the bulk of the power at a constant level; the second provides additional energy for modulation peaks. Because the main amplifier can be run closer to saturation, overall efficiency increases, but the secondary amplifier provides control over the signal.
There is no need to stop at two amplifiers. Engineers from Pohang University of Science and Technology in Korea have suggested that, under ideal conditions, a five-way design could be useful, although their work has so far resulted in just a three-way implementation tuned for wideband-CDMA protocols. At the moment, peak-to-average ratios are not so large that adding more amplifiers makes sense. But future standards could make more complex Doherty-based designs viable (see fig 2).
However, there is a second issue that complicates the design of power amplifiers. Some 14 individual frequency bands are available for LTE, with different regions and countries picking and choosing from the list to fit in with existing cellular, TV and radio networks.
Concepts such as white-space radio are pushing radio designers to make their circuitry even more flexible: sniff the airwaves to find free space and then move into it, leaving that frequency band when too many other transmitters show up. For the most part, a more flexible radio means using software to control a set of standard hardware building blocks, rather than tuning a hardware circuit to a specific RF range.
Close to the baseband, the situation is relatively simple: use more DSP. But as the signal reaches the antenna, things get more complicated. Power amplifiers are optimised for operation in fairly narrow frequency bands and that includes Doherty architectures, although the multi-way version help deal with the high crest factors of the newer wideband protocols such as LTE. It is possible to design amplifiers that have a wider frequency range, but these suffer in terms of power efficiency.
Another 'blast from the past' offers a way to claw back some efficiency. First described a year after Doherty's proposal, envelope tracking continually adjusts the supply voltage to the power amplifier so that it gets just enough to deliver the required signal (see fig 3). This reduces the amount of energy dissipated as heat from the notional over-voltage in a conventional amplifier design. The efficiency improvement from envelope tracking tends to increase the further the amplifier operates from its frequency sweet-spot.
But the circuitry is not entirely straightforward to implement and building a power-supply modulator accurate enough for the job takes some effort. One approach, used by startup Nujira, is to predict the amplitude of the RF output from the modulated signal and then apply a function that can be used to derive what the drain voltage should be. However, the calculation imposes a delay so the trick is to delay the RF signal on its way to the amplifier, so voltage and output match up.
In general, envelope tracking is being implemented in handsets more quickly than with the somewhat larger power amplifiers of basestations, where the tracking is harder to implement efficiently.
For the basestations, changes to more exotic processes are showing greater promise. Gallium arsenide has gradually given way to silicon laterally diffused metal-on-semiconductor (LDMOS), largely because the silicon devices can handle very high power outputs. However, gallium nitride and silicon carbide may supplant LDMOS in the future, thanks to their ability to work efficiently at high temperature, reducing the need for active cooling.
DSP can help; power amplifiers are now being designed to handle a level of predistortion, intended to compensate for frequency-dependent losses in the amplifier subsystem. It increases the amount of compute horsepower needed in the RF front-end, but this generally needs a lot less energy than that wasted by the power amplifier itself if predistortion is not used.
There is a similar problem on the receive side. In principle, in a software radio, a single A/D converter feeds into a high-speed DSP that uses digital demodulation to pull the signal of interest out of the rest of the radio soup. In reality, neither the A/D converters nor the DSPs available today can offer the level of horsepower needed – unless you consider the arrays in battlefield military radios that sit in the back of jeeps and helicopters.
The problem for the A/D converter is not so much the sampling rate – there are designs that can work at rates of more than 1GHz – but the resolution. The demodulation process requires very high dynamic ranges to let the computer find the buried signal.
The increase in DSP power allows some use of wideband techniques, where an A/D converter uses subsampling to extract a range of frequency bands. However, some radio standards have stringent demands that make it hard to realise this architecture. GSM, for example, demands that a handset be able to pluck a weak signal from between channels that may contain some very noisy transmitters.
As a result, an array of fixed surface-acoustic wave (SAW) filters often works out to be more effective than using a more digitally-oriented approach.
The antenna is, potentially, where some of the biggest changes over the next few years will be seen. Handsets now need to support many different radio standards, from Bluetooth to a variety of LTE-compatible bands: can one antenna handle all of those? The answer is, surprisingly, a cautious yes. But, as with filters, the answer is not through using a wideband antenna, but through integrating many different antennas into one package.
There have been attempts to develop a true wideband antenna: fractal designs showed some promise about ten years ago. As the name suggests, these antennas exploit the self-similar nature of fractal geometry to allow one printed structure to behave as many different antennas. This design does work, but that is where the good news ends. In tests, researchers from the University of Brasilia found performance drops with frequency – they work best in RF bands below most of those used in digital communications. As a result, the fractal antenna may be useful where DSP horsepower and amplification technology can overcome inefficiencies and where the ability to cope with arbitrary frequency bands is vital. For everything else, you need a different approach.
With careful design, it is possible to create an antenna structure that can handle multiple frequencies simultaneously, but the effectiveness of the design depends on which radios will be used simultaneously. As a result, the multimode antenna has to be custom-designed for every use case.
Moving away from an entirely passive design can help improve performance. It is possible to put transistors or diodes into the antenna itself to switch between different radio modes. Another approach is to stick with a more generic antenna design and then tune electrically the matching network used to convey the signal to the receiver circuitry. Although it is trickier to design, tunable antennas generally perform better than tunable matching networks. If the tuning range is pushed too far, the matching network can begin to incur significant signal losses. The result is likely to be antennas that themselves resemble printed circuit boards, complete with embedded components. But, even with those changes, the fundamental design of the antenna, as with the power amplifier and filters, remains rooted in traditional techniques.