Re: [Discuss-gnuradio] PSK Demodulator performance

[Marcus Müller]

Hi Mike!

Well, from a purely algorithmic point of view, there is no approximate magic happening inside the PSK demapper

(from constellation.cc):

    unsigned int
    constellation_psk::get_sector(const gr_complex *sample)
    {
      float phase = arg(*sample);
      float width = M_TWOPI / n_sectors;
      int sector = floor(phase/width + 0.5);
      if(sector < 0)
        sector += n_sectors;
      return sector;
    }

This is "perfect" up to the degree that arg() works well enough, which is mainly a matter of float accuracy, I guess, and should therefore be well below any significance.

The other thing is the finite-length root-raised cosine matched filtering; I think you can derive losses for that, but I think Martin's approach of measuring things might be faster and more accurate in the end.

Greetings,
Marcus

On Sat, May 3, 2014 at 10:59 AM, Mike Willis <address@hidden> wrote:

Marcus

 

Thanks for the answer – what goes into the link budget is a maximum error rate – 10^-5 or 10^-6 would by typical – so assuming a perfect transmission system if you have an ideal demodulator you can work out the bit error rate curve which might give say 12dB Eb/No. In practice the demodulator won’t be perfect, it might need 13dB or 12.5 dB to achieve the same error rate as an ideal demodulator. That’s what I meant. No soft decisions here, regrettably its a hard decision, 0 or 1.

 

I agree other parts of the chain add their own losses, we account for those also.

 

To explain, we have a DBPSK packet with up to 280 bytes in it. So a reasonable bit error rate is maybe 10^-5 to not lose too many.  

 

Mike

 

From: Marcus Müller [mailto:address@hidden]
Sent: 03 May 2014 09:50
To: Mike Willis
Subject: Re: [Discuss-gnuradio] PSK Demodulator performance

 

Hi Mike,

to me, a link budget is essentially the minimum SNR that my system has to work under.
That is basically: If every component of the transmission chain has its impact on overall SNR, the SNR is available at the point of symbol decision, which is your PSK demod (channel coding aside).

So what's the loss for a PSK demodulator?
If we look at the psk_demod available in the grc, all it does is take complex samples as input, use the root raised cosine pulse shape it assumes the signal to have to synchronize (and maximize SNR). From my point of view, that actually introduces processing gain, not loss. Maybe you were comparing it to a soft output decoder?

Greetings,

Marcus

 

On Sat, May 3, 2014 at 9:35 AM, Mike Willis <address@hidden> wrote:

Does anyone have any information on the performance – specifically the implementation loss of the Gnuradio PSK demodulator block? I am trying to work out a link budget. Typically this would be 1dB or so from ideal.

 

Mike

_______________________________________________
Discuss-gnuradio mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/discuss-gnuradio