R board Unit #2

A second R board was constructed with a focus on the digital step attenuator.  Care was taken to ensure sufficient solder paste on the QFN ground pad.  Channel 2 was constructed as a direct AC bypass channel for reference.

A point to note, the mux datasheet comment on DC blocking capacitors on all ports is not just a good idea, its required.  During construction I forgot and skipped a step in adding the capacitors on channel 2 bypass.  This resulted in DC coupling of an input mux port and output mux port.  Even when these ports were not selected the net effect being to cause problems with the input mux pHEMT switching bias voltages.  The result was that the RF input signal was propagated to all ports irrespective of what ports were selected.

Unfortunately, little has changed relating to the 4.7MHz impulse noise on the digital step attenuator.  The following picture captures the RF output port using a 200MHz input at approximately -10dBm.  The attenuator settings vary from 0dB, -20dB, and -30dB.

It seems that the low frequency impulses are intrinsic to this type of digital step attenuator.  Based on studying various articles, I am guessing the impulse noise is from an internal negative voltage generator.  The part uses GaAs pHEMT transistors to switch the signal around or into resistive attenuator networks.  The pHEMTs required a negative bias voltage on the gate to turn on.  There are parts available where the negative gate voltage is supplied directly (external to the part), however, the device I used allows direct digital values.  I suspect the negative voltage is generated using something like an inductor on the high side of a transistor, into two opposite direction diodes followed by a capacitor on the output (a basic charge pump with positive current pumped into ground and negative current pumped away from a capacitor at the output of the negative voltage).  I am sure it is much more involved internally but this type of structure would produce impulses at the charge pump frequency. The exact level and distribution would depend on how many pHEMT switches are active in the attenuation/bypass network but as they all use the same negative bias voltage with ripple/impulses the frequency domain signature would be the same irrespective of attenuation value and input signal.

While I was not expecting this low level spurious signal, as a large signal attenuator and as a small scale delta attenuator for certain measurements and applications in narrow band configurations the R board should prove useful.  This is particularly true when a 0 dB attenuation channel without the impulse noise is provided as in the second unit build.