Re: [Discuss-gnuradio] Benchmark scripts

[Marcus D. Leech]

> Make a simple little FFT sink in GRC and use it on one of the USRPs to
> determine the received signal offset from the other USRP while it is
> transmitting. Or receive a signal from a signal generator of known
> frequency and note the offset for both USRPs. Or transmit a signal from
> each USRP and receive it using a different receiver and note the
> difference between the frequencies of the received signals. Or vary the
> frequency of either benchmark_rx or benchmark_tx via trial and error
> until you get proper transmission/reception. You will probably find less
> than 20kHz of offset at 2.4GHz.
>
> --n
This topic comes up repeatedly on this list.  When you have a radio 
"tuned" to a specific frequency, there is
  *nearly-always* a certain amount of residual frequency error. 
 Synthesized LOs (local oscillators) have a frequency
  offset that is proportional to the PPM tolerance of the reference 
oscillator that they're using.

Let's say that the oscillator has a 10PPM tolerance (which is typical, I 
don't, off the top of my head, know what the
  PPM spec is for the XO in the USRP1).  So, that's 10Hz for every MHz 
of crystal frequency.  That error "carries through"
  the PLL synthesizer.  In this case, we have an LO frequency of 
2.4GHz, so, that 2.4GHz/10PPM which gives us a potential
  frequency offset error of 24KHz--let's be charitable and assume that 
means anywhere +/- 12KHz.

For wideband modulations (higher data rates), a small frequency offset 
is generally mostly-harmless, since *most* of the
  modulation energy falls within your passband, even if the edges of 
your passband don't fall where you think they fall.
  But for narrowband modulation schemes (low data rates, or narrow OFDM 
buckets), frequency offset can be disasterous,
  since an offset in the band-edge causes most of the energy to fall 
*outside* of your passband.

In "real" data receivers, particularly narrowband ones, there's 
generally a feedback mechanism that tugs on the LO circuit
  a little bit to try and zero-in on the correct frequency.  In the 
analog world, FM stereo receivers have a so-called AFC
  circuit that uses noise estimates to steer the LO.  Television does 
the same thing (so-called AFT).

In the world of software-defined radio, it's easy to forget about these 
things, because, hey, it's all *digital*, and therefore
   *perfect*, right?  Wrong.

So, an SDR-based analog data communications system needs to be able to 
deal with this.  There are a couple of ways of doing this:

     1) lock your PLL Synthesizer to a high-quality reference clock, 
usually improving the PPM error by an order of magnitude or more
         On the USRP2/N2XX/E100 there are explicit inputs for an 
external 10MHz reference clock, and UHD makes it easy to
         enable this feature when you create the UHD source/sink.

     2) Have your demodulator provide feedback to the frequency-setting 
code to tweak the actual LO frequency (or DDC frequency,
          which is usually faster).  This is the most general approach, 
since it makes your code work well even on a platform that doesn't
          have a high-quality external reference.  Note this is on the 
*receive* side.  No sense in tweaking the transmitter when you have
          potentially-many receivers.  The conventional thing to do is 
to have the receiver track wherever the transmitter is right now.

This class of problem is in no way unique to SDR hardware.  Ham radio 
operators on the list will tell you about many adventures
   (particularly in the old days) of tweaking the LO performance on 
their VHF receivers to allow "full quieting" reception of the local
   FM repeater, and as I observed, FM radios and televisions have had 
some kind of automatic-fine-tuning for many many decades.



--
Marcus Leech
Principal Investigator
Shirleys Bay Radio Astronomy Consortium
http://www.sbrac.org