Tag Archives: txco

R820T, RTL2832U SDR USB stick: some more findings about temperature sensitivity.

Another hack you can do with the SDR USB sticks – mount the clock crystal (28.8 MHz) remotely, feed a stable RF signal at any frequency you desire, and use it a as a thermometer. This will work great at room temperature and up to about 50 deg C, but beware, the ready will be ambiguous at higher temperatures. Here is a quick experiment:

Signal was supplied at 1000 MHz, from a virtually perfectly stable source. At some point, the crystal was touched with a finger (making sure not to touch any of the traces or components, as this could capacitively affect the oscillator). Surprisingly, the frequency first goes down (-0.6 ppm), and then up (+2.5 ppm).
temp effect (hot to cold to hot) ppm

Why is this so surprising – let’s have a quick look at typical crystal oscillators, there are several types, AT-type (most common), SC-type, and closely related IT-type, and others, less common. Nearly exclusively have I come across AT-type so far, for all kinds of cheap clock application. AT type means: inversion point at room temperature, then the frequency decreases a bit with temperature, to about 60 degrees C, then it increases again. See the black curve in this diagram (note that the absolute values are just typical, arbitrary, and change with cut angle tolerance):
r820t quartz ref osc temp dependence
Red line shows the typical characteristics of a SC (or IT)-cut crystal.

Rather than the expected AT-typical transition through a minimum frequency, when going from about 60-70 deg C, down to 35 deg C (body temperature), we observe just the opposite behavior (more like SC-type) – reference frequency goes through a maximum (which is a minimum of the shown signal frequency, if the signal is a constant 1000 MHz).

So it seems, the manufacturer did actually consider a relatively high operating temperature of this device when selecting the crystal, which is running at about 60 degrees – just the rough temperature of the board/crystal.
An AT-type would have been much counter-productive, because this is optimized for constant frequency over a broad range of temperatures, say, -10 to 60 deg C. However, for the SDR USB stick, the temperature related frequency change should be minimal at and around the hot operating condition of the board – I don’t think this is just coincidence, but somebody acutally put some thought into getting a device frequency stable, without any ovens or other compensation devices.

R820T, RTL2832U: SDR USB stick hack – clean and stable reference!

One of the shortcomings of these handy and cheap SDR USB sticks is the offset of the reference, and the drift. First, lets see what we have: in the meantime, I have two more or these little units to play around, and all have about 70-100 ppm offset, positive.
The reference is derived from a small 28.8 MHz crystal, and while such crystals are pretty much suitable for clock generation, they do drift with temperature, and the SDR USB stick is getting hot during use… Quick look inside (case can be easily opened, no damage):

r820t sdr usb stick opened
The small silvery metal can is the crystal. Any temperature change will cause this to slightly change frequency.

So, what about the crystal drift? Easy enough, just hooked up one of the sticks to a counter (with stable timebase; you need to use high impedance probe, otherwise, proble will pull the crystal frequency), and logged the frequency values for an hour, after “cold” startup (stick with no case).
The result:
r820t ref clock drift
1 to 2 ppm, that’s not all that bad! Still, in the world of precision oscillators, it’s ridiculously drifting. So for any precise characterization of the SDR USB device, we need to get this under control. Some have tried to replace the crystal by a TCXO, but even with this, it will be challenging to break the 1 ppm (1 kHz at 1 GHz!) mark.

A quick look at the datasheets reveals that pins 8 and 9 of the R820T are the input/output of the xtal drive circuit, and pin 10 forward the clock signal to the RTL2832U, to safe some parts, and cost.

That’s how the oscillator output looks like, probed at pin 9, with a 10 Meg probe (0.5 V/div, 10 ns/div)
r820t ref clock signal 0.5 v-divy 10 ns-divx
The signal, amplitude is about 1.5 Vpp, with a DC bias of 1.2 V.

Therefore, if we want to substitute the crystal, we have to feed a few dBms of power at 28.8 MHz into pin 8, and leave pin 9 unconnected. The feed line (50 Ohms) requires some adequate termination. We also need to provide a DC block, with a little coupling capacitor.
That’s the little hack:
r820t sdr usb ref clock hack
A SMC connector, terminated with 82 Ohm (which will be in parallel with the impedance of the R820T, hopefully giving about 50 Ohm, or close enough), and with 0805 10 nF capacitor, connected to pin 8 of the R820T.

As for the new reference source – nothing less than a HPAK 8662A, which is a real marvel of engineering and one of the best sources I can suggest for any tests that require low phase noise close to the carrier. It is stable to better than 0.0005 ppm, per day – compare this to the 1 ppm, per hour…..
Sure, not the mention – the 8662a carries 80 pounds of electronics, a big fan, and uses about 300 Watts of power to keep things clean.

The new clock source for the SDR USB stick:
r820t new clock source 8662a

The reference level needed to drive to R820T oscillator – just by trial and error, things start to work at about 400 mV, and up to 1 V of signal doesn’t seem to change anything. So I set the level to 500 mV into 50 Ohm, and this seems to work well.

Interestingly, the R820 T seems to work from about 28.75 to 28.85 MHz reference, with no change in performance (at least nothing obvious), except, of course, for the frequency shift. At frequencies below about 28.72, and above 28.89, the stick crashes-no more data comming.

Some tests: 28.800 MHz reference, 1 GHz signal (note the small frequency shift, which is related to SDRSharp software, not to any hardware offsets)
28800

28.75 MHz reference – still 1 GHz signal
28750

28.85 MHz reference – still 1 GHz signal
28850

Checking the math, this all makes sense – for 28.75 MHz, a reading of 1001.739 MHz would be expected, and 998.267, for the 28.85 MHz reference.