Question: RMS detectors with a Linear-in-dB transfer function such as AD8362, AD8363, AD8364, ADL5902 and ADL5906 produces a peaking in detection voltage when driven by RF Bursts when there is no rms averaging capacitor present. Adding an external rms averaging capacitor seems to eliminate the peaks. Could you explain what could possibly produce this kind of peaking?
Think of the rms detector as a VGA/AGC loop (RF Input drives VGA, VGA drives low range rms detector, output of low range detector is compared to setpoint, difference drives an integrator which drives the gain control input of the VGA). When there is no input signal, the VGA gain is very high. When you apply a bursted signals, the VGA gain initially remains really high which produces a large signal at the VGA output and at the output of the squaring cell. This causes the integrator output to rise sharply (overshoot) if there is no rms filter cap present (the rms filter capacitor sets the response time of the integrator). The loop will eventually settle and the integrator output voltage (which is also the rms output) settles to a steady (but incorrect) final value. The attached step response plot is from the AD8362 datasheet.
Operating a linear-in-dB rms detector with no external rms averaging capacitor is not really meaningful. With no filter cap present, there is very little rms averaging going on and the device will give an incorrect output voltage. In order for the device to give the correct rms answer there must by R(oot), M(ean) and S(quare root) going on. If you don't do enough M(ean), that is, averaging, you rms computation is not valid and the dc voltage at the output will be incorrect.
So a minumum value of filter capacitance must be present. On the newer rms detector datasheets such as ADL5902 and ADL5906, we have experimentally determined the minimum required rms averaging capacitor size for various modulation schemes. The minimum filter cap size is generally in the 1nF to 10nF range. Once you add a reasonably sized rms filter cap, you also notice that the Vout spike will not be present (you have effectively reduced the loop bandwidth of the AGC loop, preventing the integrator it from over-reacting).
If you want a detector with a fast response time and no spike, then a log amp might be better (e.g. AD8318, AD8319, AD8317 or ADL5513). It does not give you a correct rms answer (but neither does the rms detector when operated with an rms averaging capacitor that is not present of set too small).