Practical Chip Design

Jun 25 2008 12:01PM | Permalink | Email this | Comments (0) |
Blog This! using:  Blogger.com | LiveJournal |
Digg This | Slashdot This | add to Del.icio.us

In a recent feature, EDN explored the integration of RF circuitry into SoCs for wireless mobile devices. There is a trend, certainly, toward the single-chip radio, excluding the power amplifier (PA) and antenna switch that are for most air interfaces just too taxing for today's digital CMOS processes. But despite the trend, there is still a booming market in small-signal RF components. One example comes from Analog Devices, which recently introduced a pair of power detector ICs: one for base stations and the other for handsets. Each illustrates a good argument for keeping this function outside of both the PA and the baseband SoC.

First, the function itself. In wireless applications, power detectors form the feedback loop between the amplifier output and the baseband processor. Feedback control of power is necessary for several reasons. First, process, voltage, and especially temperature variations make open-loop control of output power imprecise without a lot of software compensation. Second, in the handset, harmless-looking actions like changing the orientation of the phone near a large metal surface, or—RF desingers' favorite--putting your hand on the antenna can make a significant change in VSWR, changing the entire power picture independent of the gain setting of the PA.

Why is precise power control important? In most air interfaces except GSM, precise control over output power turns out to be very important, according to Analog Devices applications engineer Carlos Calvo. OFDM radios, and even to some extent HSDPA systems, exhibit high peak-to-RMS ratios, and hence operate their PAs in their linear region. But at lower gain, the PA is less efficient. So for best efficiency, and hence best battery life, you want the PA tuned as close to saturation as possible. Yet not too close: if you start clipping those very high wave crests, you will waste energy, impair signal quality and violate the regulations for spurious radiation. So getting the PA output power just right is a big deal.

Fine—so why not just detect the output power in the PA, or in the system SoC? The big reason, according to Analog Devices product marketing manager Dale Wilson, is that integrated power detectors are usually diode-based. That's a simple, easily integrated approach, but it suffers from limited dynamic range, is strongly temperature-dependent (inside a component that tends to be the warmest part of the radio) and has difficulty with large crest-factors. There is a lot of secret sauce in doing an accurate, stable power detector.

And there are some application-specific problems revolving around that issue of wide dynamic range. In base stations, the radiated power needs to change with the proximity of the handset, over a very wide range. To cover the full range of power settings for a base station transmitter, you need a logarithmic detector, not a linear one: hence ADI's 5513 logarithmic amplifier power detector.

In handsets, the problem is not so much dealing with a huge range of RMS output power as dealing with those high crest factors. For this application, ADI's 5502 offers a unique capability: a combination of sampling peak-detection and RMS power detection in the same package. This allows the baseband processor to continuously monitor the peak-to-average ratio during transmission, and to detect the onset of clipping by the tell-tale reduction in the ratio. Using this data, the baseband chip can turn up the PA gain just to clipping, back off a little bit, and be running at maximum efficiency for this particular waveform. And by continuously monitoring the power and the ratio, you will get an early warning of a sudden change in the VSWR.

Being able to monitor crest factor in real time is also appealing to base-station designers, by the way. Wilson says that this feature is very likely to appear in a leaded part for base stations in the near future.

Calvo admits that you can do about the same things with a diode-based detector by taking the raw data back to the baseband processor and using a look-up table to correct the signal for voltage, temperature, and crest-factor. But very likely the energy consumed in executing this software would be comparable to the energy consumed by the separate power-detector chip. And since power analysis has to be done during transmission, the task would have to run just at the point when the baseband processor is the busiest.

So there are a couple of arguments for a separate power-detector IC: proprietary circuit features and, compared to software, potential energy savings. As in many other instances, the integration that saves chip count and appears to save power may come at the expense of sophistication and expertise that may turn out to be important to user-visible advantages. Integration isn't automatically the answer.