AD9361 with FMCOMMS3 spurious tone when no input signal is feeding the board depending on fOL

Is your RX input terminated to 50ohm while you are observing this spectrum?

The spectrum looks like the sigma delta  ADC noise floor is dominating the RX path noise floor. Refer to the below posts for details on it:

https://ez.analog.com/wide-band-rf-transceivers/design-support/f/q-a/80160/ad9364-zynq-xilink-matlab-toolbox-problem-with-spectral-lobes/170625#170625

https://ez.analog.com/wide-band-rf-transceivers/design-support/f/q-a/78834/ad9361-passband-at-low-rx-gain-isn-t-flat 

Yes, it is terminated to 50 ohm, I attach a picture of the configuration. Actually, my concern is not about the sigma-delta noise due to the ADC. My concern is about the clear harmonic observed at 2 MHz in the picture when the fOL is set to 642 MHz or 638 MHz (it could be 102 MHz, 112 MHz etc and the harmonic appears in the same way). 

So for lower values of the gain, the harmonics are not observed, but they are observed for larger values of the gain, for instance from 30 dB onwards.

To give another example in order to clarify the problem, I attach two images where the ADC noise can be avoided and we just focus on the harmonic that appears for different fOL values when the input is noise:

Is it possible to measure the power of the tone powers seen,

If you give a tone at say -20 dBm , at what dBc levels are you observing these tones.

Yes, I've done some measurements and the spurious tones are at about -130 dBm, I just measured a calibrated tone at some frequency and compared with this value the values of the tones. I attach an image of the situation, I can see spurious at even other frequencies. I suspect that this is due to the synthesiser, there is some leakage but I am not sure about that.

In the image, the reference tone is at 2.003 MHz and its power is about -112 dBm, therefore the spurious are about -130 dBm.

The problem is that for applications where the power received is very low, like EMC for instance, these harmonics will introduce uncertainties.

 Couldn't fully understand your measurement here.

You have input -20 dBm CW signal at 2.003 MHz offset at  Rx port and you are measuring -112 dBm as power.?

spur at -130 dBm power should be well below sensitivity levels and should not impact performance.

My imput is not -20 dBm CW signal at 2.003 MHz, actually it is a CW signal at 2 GHz with a power of -112 dBm that is downconverted afterwards to 2 MHz. In order to observe the signal, I need to use the gain of the amplifier, it can be clearly observed since the noise is mainly attributed to the receiver chain and not the ADC. (In fact, this is the problem, the spurious effect when the gain of the amplifier is high enough, I said before I starts to see the spurious for G>30dB approx).

In the example, I feeded with a signal that was not in the same position of the spurious after downconverting to IF, otherwise you are not able to distinghis between them. Therefore, using a gain of 70 dB and feeding with -112 dBm I am able to observe the spurious about 20 dB below my tone, that means that my spurious are at about -130 dBm before the amplifier. 

I hope this is clear now, and with this issue, if you are measuring very low levels of signal, lets say for instance -110 dBm and your frequency of interest is at the same frequency of the spurious it will definitely impact the performance of your system, since it introduces uncertainties in your measurement in amplitude and phase.

-130 dBm levels of spurious will be the limitation of chip and you may not be able to further improve this.

If you need further reduction , you may have to do some amount of frequency planning and implementation of custom digital filters in baseband to remove these spurs.