Question about LFTX and RFTX

GNU Radio
1

Hi,

I have a question about using the LFTX/RX boards to communicate at base
band frequencies. At the moment I’m using RFX2400 boards and frequency
synchronisation is always an important issue.

My question is, with the LFTX/RX boards, is the communication truly
baseband such that frequency synchronisation can be totally ignored?

We’re heading towards using the USRPs and LF boards as an ADSL test bed.

Looking forward to your replies,

  • Ismail
2

Ismail M. wrote:

My question is, with the LFTX/RX boards, is the communication truly
baseband such that frequency synchronisation can be totally ignored?

Not quite.

With the LFTX/RX boards, DC does indeed remain DC at the other end.
However, the crystals in each USRP still govern the sampling rate, and
each can be up to 50 PPM off (new USRP hardware has 20 PPM spec’d
crystals now.)

So the same issues with respect to symbol timing recovery exist.

In practice I’ve never encountered a USRP that had more than a few PPM
crystal offset–but you can only rely on the spec range, especially in
variable environments.

–
Johnathan C.
Corgan Enterprises LLC
http://corganenterprises.com

3

Hi again,

Recently there has been talk about the I and Q components of the
signal. I understand the analytic signal concept, also the
implementation as shown on:

http://gnuradio.org/trac/wiki/UsrpRfxDiagrams

where it says,

“Right idea, except that there is effectively a single complex
multipler. I.e., I2+jQ2 = (I1+jQ1) * exp(jwt).”

So the analytic signal is generated on the motherboard and mixed up to a
higher carrier on the daughterboard, i.e. to a new ‘w’ but same I and Q.

My question is that, with the LFTX and LFRX, if I tune the daughterboard
to DC, this would make w=0. So there would be no carrier. How are the
I and Q components separated? Or is it simply that each alternate
sample is I and Q, and in essence we are sending a purely real signal,
but saying each second sample represents the imaginary?

I hope the question makes sense.

Regards,

  • Ismail
4

Hi again,

What is the best way to do symbol timing recovery at baseband (i.e.
using the LFRX)?

I’ve tried a simple method based on power detection, and it’s not too
bad. I’m using a pwr_squelch block and the first non-zero sample that
comes in is considered as the first sample of my symbol. This method is
ok, but doesn’t solve my problem totally due to the large phase offset
when the sampling is totally out of phase.

The second option I had was to use a fractional_interpolator block to
make up for the slight sampling frequency offset between Tx and Rx.
I’ve measured the frequency offset at about 14ppm in my setup, so I
don’t think the fractional_interpolator block will help that much as
16bit float accuracy would not be enough.

I had another idea of oversampling on the receiver but I haven’t tried
it out fully yet.

Anyway, I would really be appreciative if anyone could point me in the
right direction to get over this problem.

Many thanks,
Ismail M.

5

On 8/30/07, Ismail M. [email protected] wrote:

“Right idea, except that there is effectively a single complex
multipler. I.e., I2+jQ2 = (I1+jQ1) * exp(jwt).”

So the analytic signal is generated on the motherboard and mixed up to a
higher carrier on the daughterboard, i.e. to a new ‘w’ but same I and Q.

My question is that, with the LFTX and LFRX, if I tune the daughterboard
to DC, this would make w=0. So there would be no carrier. How are the
I and Q components separated? Or is it simply that each alternate
sample is I and Q, and in essence we are sending a purely real signal,
but saying each second sample represents the imaginary?

From my limited DSP understanding, at baseband, I and Q are simply
phase relationships and amplitude information around and within the
unit circle.

With that being said, I don’t think I and Q can be combined into a
single real signal while preserving information simply because DC has
no phase component.

On the other hand, if the LF cards have both I and Q inputs/outputs,
then you should be able to see essentially two signals come out and
maintain their proper information. If it’s only real - you’ll
probably have to mix up to some IF to maintain all information.

I believe the real question here is if you can do complex sampling on
a real signal centered strictly around DC - which I don’t think you
can.

Anyone wish to comment? I am most likely wrong, but I’d definitely
like to know the real answer to this as well.

Brian

6

Ismail M. wrote:

Anyway, I would really be appreciative if anyone could point me in the
right direction to get over this problem.

Many thanks,
Ismail M.
There’s a Mueller and Muller timing recovery block that should handle
this for you. There’s both a complex and floating point version.
gr_clock_recovery_mm_cc
gr_clock_recovery_mm_ff

Tom