I use the NI RFmx drivers to setup a spectrum analyzer, at 10 ms fetch data, and a CW appears at the center frequency when the span is >= ~500MHz. I reduce the span then the CW disappears. Can someone explain why?
The VST uses a homodyne or Zero-IF architecture, so the LO used for the RF downconverter stage is centered in the 1GHz of available instantaneous bandwidth and the spectrum is downconverted to DC. When the bandwidth is greater than IBW/2, that downconverter signal may have a DC offset, which translates to content at the Zero-IF, or center frequency. When you capture less than half the available 1GHz IBW, the driver can offset the LO outside of the frequency range of interest (Low IF mode), avoiding that artifact in your measurement.
The VST self calibration routine minimizes the effects of the in-band DC offset contribution within the specification limits of the instrument, but temperature changes after self-calibration or other dynamic effects can reduce the effectiveness of the correction technique.
Here are various resources that discuss the DC offset and LO leakage effect with Zero-IF architectures. The issue can appear on Transmit and Receive signal chains.