The notch at 0 Hz

[Greg Troxel]

There is a theory/dogma that RTL-SDRs do not receive at 0 Hz relative to
configured frequency. I more or less understand the theory (but not
the edges).

I have been looking at data, and not really seeing this.

Does anyone have data where they see a sensor drop out as it crosses the
0 Hz point? A curve of signal level vs frequency? How wide is the
notch, if so? Anything else quantitative to say? Is 2 Hz away ok?
0.1? 10?

I am however seeing lots of interesting things:

diurnal variations in signal strength (but not two state e.g. as would
be expected from some interfering device)

bimodal behavior where the system is in one state with signal levels
and then suddenly changes (probably a restart, need better logs), to a
new mode where signal levels are consistently better.

that I will try to understand and write about. I am also paging in my
python code to convert json logs to xplot for visualizing this data, and
will be cleaning it up (personal list of devices in source needs to be
moved to config file with config file reader, or made unnecessary).

[Facility 406]

> Does anyone have data where they see a sensor drop out as it crosses the
> 0 Hz point? A curve of signal level vs frequency? How wide is the
> notch, if so? Anything else quantitative to say? Is 2 Hz away ok?
> 0.1? 10?

Hook up a DC battery to the antenna port through a resistor, and see
what the SDR indicates?

Kurt

[Greg Troxel]

I didn't mean 0 Hz, absolutely, but 0 Hz relative to the LO.

To measure, one probably has to set up a signal that's locked to a GPSDO
or otherwise very stable, and of a signal level so you are getting valid
SNR output (vs the "-0.5 dB") signal level), and then log while the
RTL-SDR warms up and drifts across.

I was trying to ask if anyone had actually measured, as it seems from
looking at very limited data (from real transmitters) that if there is a
notch it is much narrower than one would expect given the comments on
the list and the issue tracker, to the point where it's pretty much a
non-problem in practice. But, that's from a hazy, non-rigorous view,
and I don't yet believe it.

I looked at one emitter (EcoWitt) and it has fsk frequencies of
(eyeballing the contains-almost-all from a graph of -M signal):

914.963 to 914.975
915.017 to 915.031

so an LO of

-f 915.00M -s 250k

is actually ok, as there is 17 kHz of headroom to the closest signals.
I don't see any evidence of fewer decodes.

I could pick 914.900, but that would only give me to 915.025, if I'm
doing the math right. I don't really want to increase the bandwidth.

I then looked at all 4 ecowitt emitters at the same time, and I don't
see any signals really close to 915. (Yes, I realize if there were ASK
devices it would be different).

My graphs show x for freq1, + for freq2 and a synthetic box for the mean
(central frequency at which there are no actual signals), and then a
line for the average of the central values. The lines are hard/impossible to see
at this zoom but the values are:

914.993662
914.993914
914.994860
914.998039

which is remarkably tight. Overall 914.977 to 915.014 has ~no signals,
saying I should adjust to 914.995 if I were to be excessive.