I know very little about RFID, but I can make these observations on
how I would probably perform a naive demodulation of this signal.
- Tune the USRP to your carrier frequency and bring your signal
down to baseband (centered around DC).
- Choose an appropriate number of samples per symbol or bit (I’d
arbitrarily choose 6 … or 8).
3. Filter the incoming signal to some appropriate spectral mask.
4. Find the magnitude (or even power since it saves a sqrt) of the
5. Develop some algorithm which accounts for frequency offset/tracking.
- Perform hard decisions from the soft decisions you make from the
algorithm developed in step (5).
Are you trying to decode the reader transmission, or the tag reply? You
be able to use a matched filter as a first step to clean up the signal
demodulating. I’m not really sure what 13.56 MHz RFID signals look like,
helps a lot at 915 MHz.
Regardless, I think step 3 is the key to get you started. You don’t want
throw out the I or Q channel. You probably want to use the
and then the signal will look like what you expect.
To try to answer Marco’s original question about I/Q directly: the phase
info available from I/Q samples is not directly useful for RFID ASK
However, the calculated magnitude provides slightly better SNR than
samples of magnitude alone.
I’ve done a lot of work with 13 MHz RFID. Most common form of modulation
Amplitude Shift Keying, a form of AM, with the data stream Manchester
A good example of channel filtering and AM demodulation using
is usrp_am_mw_rcv.py. That example is setup for AM broadcast, so you’ll
to widen the channel filter BW out to accomodate the baud rate you’re
and, you can eliminate the audio filtering and sink code. The
output is the ‘envelope’ of the signal, with an offset the depends on
percentage of modulation achieved. The RFID reader in Marco’s case
100% modulation, so the offset will be 50% of peak level, like this
font spacing req’d):
___ ___ ___
| | | | | |
0 | |___| |___| |
RFID tags typically achieve much lower modulation percentages, depending
coupling with the reader, like this:
___ ___ ___
| |___| |__| |
0 | |
In either case, you can threshold the data at some value less than the
average peak to do a crude conversion from envelope to bits. In turn,
can examine the bitstream to locate bit transitions. The bit stream can
be decoded. There are better ways, e.g., using matched filters, to
Manchester codes directly from the envelope data, but this should get