Re: [Discuss-gnuradio] Softbits in gr-ieee 802.11 ?

[Bastian Bloessl]

I'm not sure if I understand you correctly, but you have to 
differentiate what performance you mean. IT++ might have better decoding 
performance (since it's using soft-bits), but the SIMD implementation 
has better computational performance (i.e., needs less CPU).

To be honest, I didn't do a lot of testing between IT++ and the current 
version. (I think I've seen a paper that compared soft-bits vs. 
hard-bits and there was a 2-3dB improvement, but you have to check). 
Since the decoder was a CPU bottleneck, I was more than happy to merge 
the SIMD implementation, which was developed by Gonzalo Arcos based on 
the gr-dtv decoder. See here

http://www.ccs-labs.org/bib/arcos2016accelerating/

Best,
Bastian

On 1/18/2018 13:41, sumit kumar wrote:
> Thanks for confirming. I believe that IT++ version will work with 
> corresponding GNU Radio version.
>
> Also what do you think (since you have developed the code for long time 
> :) ) : IT++ version will be as efficient as the current one? I mean the 
> current version uses SIMD for Viterbi.
> I have tested the current version for success rate. I can share the 
> results(attached). Success rate mentioned is nothing but ratio of 
> received to transmitted frame per second using a R&H VSG. The test was 
> done over the air.
>
>
>
> On 18 January 2018 at 13:06, Bastian Bloessl <address@hidden 
> <mailto:address@hidden>> wrote:
>
>     Hi,
>
>     you are right. The current implementation does not use soft-bits but
>     feeds hard bits into the decoder.
>
>     In 2013, I used the soft-input Viterbi decoder from the IT++ library.
>
>     http://itpp.sourceforge.net/4.3.1/convcode.html
>     <http://itpp.sourceforge.net/4.3.1/convcode.html>
>
>     Best,
>     Bastian
>
>     On 01/17/2018 11:36 AM, sumit kumar wrote:
>
>         Hi,
>
>         I am doing some proof of concept using gr-ieee 802.11.
>         The paper by author [1] says that it uses softbits, however in
>         the method
>
>         *constellation::decision_maker(const gr_complex *sample)*, I see
>         outputs as hardbits.
>         Am I missing something here i.e. any particular commit or branch
>         where the softbits implementation is there.
>
>            Its very important for my algorithm to use softbits :-/
>
>         [1]
>         https://conferences.sigcomm.org/sigcomm/2013/papers/srif/p9.pdf
>         <https://conferences.sigcomm.org/sigcomm/2013/papers/srif/p9.pdf>
>
>         Sumit
>
>
>
>
> --
> Sumit Kumar