Beginner question about rtlsdr

This is merely a historical archive of years 2008-2021, before the migration to mailman3.

A maintained and still updated list archive can be found at https://lists.osmocom.org/hyperkitty/list/osmocom-sdr@lists.osmocom.org/.

Scott Cutler scott at scottcutler.net
Sun Oct 21 23:04:43 UTC 2012


Happy to help.  I had to guess a bit on your level of background 
knowledge, but it sounds like I didn't go too over or under your head.

If you just want the energy over the entire sampled bandwidth, you are 
correct that you can just average the samples squared.  I believe this 
is equivalent to a "1 sample FFT", which as it happens is just the 
original value.  So averaging the squares over time will give you a nice 
smoothed energy value.  Note that by changing the sample rate, you can 
change the window size--the rtl-sdr chip supports from 250 kHz to 3.2 
MHz.  So that at least gives you coarse-grained control over your window.

That said, you will almost certainly want to use a frequency view once 
you start really digging into this.  So keep that in mind, even if in 
the short term you play with something simpler.

Your goals with interferometry are ambitious, but not absurdly so!  The 
rtl-sdr units have problems with frequency stability, and will be hard 
to synchronize--but these aren't necessarily insurmountable problems.  
It would be awesome to see progress here.

Unfortunately, I don't have any good references for texts.  I honestly 
use Wikipedia for most of my information.  Maybe some others here have 
recommendations.

-Scott


On 10/21/2012 3:24 PM, Michel Pelletier wrote:
> Scott,
>
> Thanks so much for your detailed reply.  As an optical amateur I am
> really excited to be immersed in a new world of very interesting
> concepts related to radio astronomy.  Thanks for patiently explaining
> to me how I was getting it wrong.  I've read the wikipedia pages on
> FFT and some of the history and I think I get, at least abstractly,
> the distinction now.
>
> On Sun, Oct 21, 2012 at 1:12 PM, Scott Cutler <scott at scottcutler.net> wrote:
>> To get from the time domain (the raw samples) to the frequency domain
>> (waterfall display, etc.), you need an FFT.  The FFT operates on some given
>> buffer size, and outputs a buffer of the same size filled with complex
>> frequency levels (where the magnitude squared is the signal energy).
> I don't think I want a waterfall "frequency domain" display however (I
> didn't know that, until I read the references you pointed me toward),
> I suspect that what I want is the time domain.  From a very simple
> perspective, and initially to recreate the Jansky experiment, I just
> want the signal strength over a wide bandwidth plotted over time.
> Jansky used a simple pen plotter of signal strength (again, according
> to wikipedia) to chart the passage of a strong radio signal across his
> local zenith.  After discovering the period was a sidereal day he
> consulted a sky chart and realized it was Sagittarius.
>
>> The FFT width determines your frequency resolution, not the bandwidth.  The
>> bandwidth is determined by the sample rate--in your case, 1.4 MHz.
>>
>> So you choose your FFT based on your resolution and performance
>> requirements.  In your case, the FFT will return frequencies from -0.7 to
>> 0.7 MHz around the center frequency--no matter what the FFT width.  The FFT
>> results store the positive frequencies from 1..(N/2-1) and the negative ones
>> from (N/2+1)..(N-1).  Sample 0 is 0 Hz and N/2 is 0.7 MHz (you can't really
>> use this last one since it aliases).  The other frequencies scale linearly
>> between these points (sample N/4 is 0.35 MHz, etc.).  For a waterfall
>> display, this generally means you want to swap the left and right halves of
>> a buffer.
>>
>> You can read as many samples as you want from the device. Conceptually, it's
>> similar to an audio device--if you need more samples, you just wait longer.
>> You could collect 64M samples if you wanted, and get some nice sub-Hz
>> resolution, but it would take 45 s to record at your settings.
> Ok, I think I'm getting the picture of this, and it's very
> enlightening.  I think that I just want the signal strength of the
> entire bandwidth over time.  Which I believe I can get by simply
> summing the magnitude squared of all my samples taken, and plot that
> over time.  Does that sound right to you?
>
> I eventually plan to try and do some basic interferometry, where
> multiple observations are taken and combined to synthesize an aperture
> capable of resolving point sources.  But I'm way far away from doing
> that or even really understanding it other than from a high level.
> There is an excellent python package called aipy that is developed by
> a group of radio astronomers that is used for a large scale 64 antenna
> array in southern Australia that I would eventually like to use to
> combine multiple sources.  The whole sky images they have developed so
> far from cheap (compared to multi-meter dishes) crossed dipole arrays
> are very impressive.  This may be utterly foolish thinking, but I
> really hope something like rtlsdr, or something like it terms of
> price, can bring large scale interferometry to the masses.
>
>> Incidentally, none of this is peculiar to rtl-sdr--even the fanciest and
>> most expensive units operate the same way, though they have higher sample
>> rates and such.  In fact, the stuff I said is fundamental to all signal
>> processing--if you're willing to get your hands dirty with math, you might
>> want to read the wiki articles on Fourier Transforms and other signal
>> processing subjects.
> Thanks Scott, I'm digging deep and consuming as much information on
> the subject as I can.  Can you recommend any standard works on signal
> processing I should dig into?
>
> -Michel





More information about the osmocom-sdr mailing list