LO Feed-through improvement

Are you able to turn on a noise marker in your setup so that you can get a reading of the output noise in dBm/Hz? If not, can you let me know what the resolution bandwidth setting is when you measure -70 dBm of noise? For example, if the RBW was I MHz, then your noise floor in dBm/Hz would be -140 dBm/Hz (i.e. -70 dBm - 10log(1e6)). The datasheet includes various plots the  that show how the noise floor is affected by the output frequency, the output power and the VCTRL VGA control voltage. Not also that any noise on your baseband inputs or on the LO will degrade the output noise dB-for-dB. I don't think that what you are seeing is LO Leakage. The LO leakage will manifest itself as a tone at the output that will have the same frequency as the LO. 

As a next step, can we do a baseline measurement where the four baseband inputs are tied to a low noise 1.4V source (or ac-coupled to ground). Set the VCTRL voltage to 3.3V (AGCTRL bit set to 1). Set the LO to 1.5GHz (or your frequency of interest). Then measure the noise floor at 1.6 GHz with the resolution bandwidth set to 1Hz (if the machine will allow this). Under these conditions, the datasheet says you should see a noise level of about -155 dBm/1Hz (see plot at bottom left corner of page 10). This assumes that your LO source is very low noise. 

Please try this and let me know how it goes. - Eamon 

Hi Jan,

I'm better understanding now the time domain and frequency domain plots. I would have expected both to be equal. Shouldn't   the marker in the frequency plot (-63 dBm) be equal to the time domain level (-47dBm). But, leaving that aside, yes, if you have the inputs ac-coupled, your dominant output power should come from LO Leakage. 

I think that you should be able to get more out of the I and Q offset adjustment. The datasheet says that it can be reduced from -43 dBm to -64 dBm. There is a section in the datasheet that discusses the approach to tweaking the I and Q offset. I'm also attaching a link to an old ap-note on the topic. In this we suggest sweeping the I offset until you find an initial minimum. Then with I value held, you sweep Q and look for a deeper trough. Finally, you can do a third sweep of I to fine tune things. It can be done in any order, that is, I first or Q first. But when the temperature changes, much of this good work goes out the window (unless you are willing to re-cal over temperature). 

The noise floor does look a bit messy in the frequency domain plot. But it's well below the fundamental tone (the LO Leakage) at -63 dBm. It may be from your signal generator or it may be pick-up. 

Hi Jan,

I'm glad to hear that this worked. Here is the link to the application note.

Eamon

AN-1039: Correcting Imperfections in IQ Modulators to Improve RF Signal Fidelity | Analog Devices