I was playing around with the rtl_sdr dongles and came up with a trivial
hack to build a receiver with multiple coherent channels. I do this
basically by unsoldering the quartz clock on the slave units and cable
the
clock from the master rtl dongle to the slave units (I’ve attached some
pictures).
You still have to do sample alignment in software, but this is
relatively
easy. There are a lot of cool applications, such as a dual frequency
beacon
satellite receiver, interferometry, or passive radar that you can now do
with $16.
On Sep 24, 2013, at 11:41 AM, Marcus D. Leech [email protected] wrote:
Marcus L.
Marcus, (appreciate you may have done a lot more than your brief
description above, but just in case.)
The type of cheap 2 pin oscillator used with the Realtek chips will be
connected across an internal inverting buffer amplifier in the IC with
shunt capacitance and all the circuit goodness that makes such thinks
work. If you are going to replace that with a buffered clock source such
as a bench signal source or expensive TXCO you’re normally going to only
drive the crystal input pin and leave the other unconnected.now which
pin that is I can;t tell you because the data sheet/schematic isn’t
available to my knowledgebut hey, its $8 so trial and error!
Might also want to consider series termination for each cable to the
boards to minimize SI issues also.
Of course in Juha’s case he’s just using the original clock-osc and
getting lucky that it’s still oscillating cleanly with the two IC’s
driving the crystal.
Just tried a series termination on each dongle, consisting of a 1000pF
cap in series with a 200Ohm resistor on each arm. It still is “sane”
with a
+3.3dBm sinusoidal input, but there’s no difference in the relative
phase-noise between both channels.
Based on my very limited understanding on electronics, osclilators and
other such things, I would expect oscillator effects to cancel out when
looking at the relative phase of two channels. I would expect the phase
jitter on the relative phase to be dominated by what happens in the ADC
and
the digital down conversion. That is why I am running the two dongles
off
the same clock.
I have some measurements of the relative IQ signal (z_1/z_2) at 100 kHz,
10
Hz, and 1 Hz sample rates. I did a power spectrum estimate using a
Hanning
window to look at relative phase noise. I didn’t do any incoherent
averaging on any of these, so there is some statistical noise. I only
see
some “imperfections” in the 10 Hz and 1 Hz spectrum. However, I see
these
kinds of effects in 100 times more expensive receivers too. For my
purposes, the relative phase behaviour of the two channels is good
enough.
I have attached the spectra. I have also attached the 1 Hz IQ signal,
which
shows a small systematic wiggle, but very little phase difference
between
the two channels.
The samples also are aligned over two hours of sampling, which means
that
there cannot really be a very large amount of clock drift between the
two
systems.
Sure, the thing has it’s faults. But with $16, you really can’t beat the
price.
I modified my clock sharing so that I only insert a signal in the
Xtal_In
pin of other dongle. This way I won’t have two circuits driving the same
crystal, as Ian pointed out. The pin next to the edge of the dongle
turned
out to be the Xtal_In pin (the input of the opamp on the slave dongle).
The
dual coherent rtlsdr dongle still works the same. I guess I was lucky to
get it working the first time.
I am working towards setting up a fanout buffer, to do this properly.
The main thing to note is that square-wave clock input should be fed to
xtal_in for the elonics chip.
If you read that patent, it may give you some ideas about feeding a
clock to
xtal_in of other chips. Maybe other chips will also expect a sawtooth
in
the linear region of voltage swing, not rail-to-rail square wave. Note
the
comments in that patent on jitter at square wave, etc.,
Nobody on this thread has stated whether they are using e4000 or r820.
That
would probably be helpful.
Bah!!! The main thing to note is that the e4000 expects a clock input
on
XTAL_OUT. That’s right, the patent says that out can be an in, for the
e4000. Sorry I made a post in which I made a thought mistake and typed
the
opposite.
I wonder, since you are emulating the crystal output, if the R820T would
ordinarily drive the crystal to rail-to-rail, square wave output. It
might
be worthwhile using a high impedence probe to see what the signal you
are
replacing looks like.
I say this because the elonics patent suggests that staying in the
linear
region of the amplifiers in the feedback circuit (not rail to rail)
reduces
jitter.