SDR Dongles: Cause of Anomalous Frequency Domain Response in the R802T2 Tuners?

[tggzzz]

I'm interested in the sensitivity of the cheap-as-chips SDR dongles containing an R802T2 tuner and RTL2832U demodulator. Since my principal interest is in wideband spectrum and network analysers, I don't care about the demodulator and am using the simple "rtl_power" application under Lnux

If I drive the input with a wideband white noise source, then in an ideal world, rtl_power would show a horizontal line across the 25-1500MHz band. Naturally it isn't like that, but there are some peculiar features I'd like to understand. The frequency domain sensitivity is shown in the attachment. The principal curious features are the "wiggle" between 1050 and 1150MHz, and the sharp 5dB transitions below 400MHz. Neither are level dependent, and they are visible in both rtl_power and gqrx applications.

I'm guessing the 1.1GHz "wiggle" is due to some anti-resonance effect associated with the R802T2 tuner or layout.

The sharp 5dB transitions, e.g. at 246.2MHz, are more troubling. They occur between adjacent FFT bins, even if the bins are only 1kHz wide. They appear to be related to the RF filter and/or mixer, since the voltage measured at the mixer's power detector output (pdet2, pin 5) also shows a sharp transition at the same frequency. (The 1.1GHz wiggle is also visible at the pdet2 output).

So, any comments on what might be the cause and/or cure?

Thanks

[LukeW]

How consistent is it?
Can it be normalised out accurately? Or does it drift?

Take the power spectrum from the noise source without the network under test, then connect the DUT, measure it and subtract it?

How consistent and reproducible is it between different RTLSDR dongles?

I wonder if you could take your signal source or noise source, split it in half, take one half through the DUT network into one RTLSDR, and take the other half into another, separate RTLSDR dongle. If these are the cheap $10 ones then this is affordable.

Then you might be able to set up software to take the difference between the two measurements in real time?

If you need clock sync between the two dongles you could connect them together on a common 28.8 MHz clock source, like is done when using these modules for passive radar applications.

[tggzzz]

Thank you for your thoughtful reply.

The effects are remarkably stable, and, as you suggest, can be completely removed by subtracting a reference/calibration sweep.

I've continued to think about the issues and to keep an eye out for other people's experience. I've documented them at https://entertaininghacks.wordpress.com/2015/06/24/rtl-sdr-dongles-anomalous-frequency-domain-response/ . I've come across another set of results that are very similar, so it looks like it is not a "rogue" dongle.

I've continued to expand my experiments, and they are documented at https://entertaininghacks.wordpress.com/category/frequency-domain-analysers/

I'm gradually getting around to publishing an improved TDR (i.e. part 2) and my utilities' source code.