As part of the Portage project I need to design a way to detect the RF energy emitted by operating cellular phones and other mobile devices. Here's my first attempt, using a LT5534 RF log detector:

Basically the exact same circuit as the evaluation circuit given in the datasheet. The layout is mine however, with 1206 SMD parts rather than 0603's. The layout is definitely "cargo cult" rf design, for instance the microstrip from the antenna isn't a calculated width. All the same the gEDA files and gerbers are obtainable from my monotone repository from branch io.pkt.client.portage.rf-log-detector.lt5534

Testing

Interestingly connecting an antenna to the SMA jack seemed to reduce performance rather than increase it. The noise floor of reported rf power would go up dramatically, almost to the point of saturating the LT5534's output and making actual signals undetectable. In any case testing was done with the board as pictured above, nothing conntected to the SMA input and the devices to be tested were placed within a few inches of the board. A Tektronix TDS 1001B with 200mhz probes was connected to TP103, the LT5534's output. Vcc was 3.1V. I did try simply soldering a 916mhz ceramic chip antenna on the back as well, the blue stick in the picture, but it seemed to have no effect on performance.

Bluetooth

Nokia N95 in bulk transfer from an Apple Macbook.

Signal strength seemed to be very directional, varying greatly in different orientations. The above plots show max strength, however the if the N95 is rotated to minimum strength the signal is completely undetectable.

Wifi

D-Link Airplus in 802.11g mode, ping flooding a remote target.

Those are very narrow spikes, just a couple of µS wide. Though the peak amplitude is high, there isn't enough energy in them to detect by simple averaging, say with a capacitor, and looking for an increase.

GSM - single phone

i-mate SP3i on Rogers network.

Very strong signal that varies linearly with distance. Strength varies a lot based on orientation and occlusion, for instance if the users hand is covering the phone. Still the phone is detectable from as much as 20ft away. Detecting the signal would be easy as well; there is a lot of energy present and even just simple averaging would work.

GSM - two phones

i-mate SP3i and Nokie N70, both on Rogers network.

This test consisted of two different calls, between the two phones. In both tests both phones were near the detector, allowing you to see the different timeslots used by the phones. In the first two screen captures, from the first call, the timeslots happen to be right next to each other, in the second two captures they are spaced apart. With work it may even be possible to not only detect the phones operating, but roughly how many are operating in an area.

CDMA

Samsung (unknown model) on Virgin Mobile network

No logs captured, but phone simply shows up as an increase in noise rather than a distinct time-based signal. Not detectable until phone is very close (<3") to detector.