Re: [Discuss-gnuradio] USRP noise at higher frequencies

[Matt Ettus]

On 07/10/2010 03:44 AM, Joachim Roop wrote:
> Hi,
>
> I posted about this before (High frequency OFDM), but could not get a
> reply. I tried my best, but am completely stuck by now and would
> accept any wild guesses.
>
>
> I have two USRPs (TVRX and WBX-TX) connected via cable that transmit
> a OFDM signal (courtesy of gnuradio-examples/python/ofdm, default
> parameters) in between them. I have manually taken care that the
> carrier frequency is centered correctly on the receiver and that the
> amplitudes at both the transmitter and receiver are configured
> optimally.
>
> Now when I transmit said OFDM signal at around 100MHz, everything is
> looking perfectly fine. If I look at the transmitted and received
> signal, I see a perfectly sharp edged OFDM block in the spectrum and
> get 100% of the data from one USRP to the other. Nice. When I switch
> that same configuration into higher areas of the spectrum (400MHz+),
> again adjusting carrier-frequency and amplitudes, I also see the same
> sharp edged blocks. The SNR is a little worse, but still well beyond
> 30dB. This is where my problem lies. For some reason, most of the
> packets are broken or bursts are missing completely. If I switch to
> wider subcarrier distance, the transmission works again - but that
> isn't what I am looking for.
>
>
> 1) What happens at higher frequencies that "invisibly" destroys the
> OFDM block? Phase noise? Impulsive noise? Something to do with the
> sampling rate conversion? Any tips or pointers would be of great
> help.
>
> 2) How could I model a high frequency transmission (including phase
> noise) in GNU Radio, so that I can more easily see the problem?


This is due to frequency offset.  Oscillator error is in parts per 
million (roughly 5 to 10 ppm for a USRP1).  The higher the carrier 
frequency, the higher the absolute error.  At 100 MHz, 10 ppm error is 1 
kHz.  At 400 MHz, 10 ppm error is 4 kHz.

The code you are using only works at a limited frequency error.  The 
allowable frequency error is proportional to the tone spacing.  So when 
you go to higher tone spacing it is more tolerant of the higher 
frequency error, and that is why it works.

What you need to do is modify the code to search for the signal over a 
wider frequency band.  Or use the wider tone spacing.

Matt