[Ns-developers] wireless PHY project

Thierry Turletti Thierry.Turletti at sophia.inria.fr
Fri Oct 27 07:37:23 PDT 2006



Ruben Merz a écrit :
>
> Lee, Thierry,
>
>>> Actually, in some cases, it is interesting to get a "realistic" bit
>>> error mask by packet -- for example, when you want to evaluate the
>>> quality of  a "robust"  video transmission using UDPLite. This could be
>>> an interesting _option_...
>>
>> DCCP supports partial coverage CRC and is also a candidate for Video 
>> transmission. Of course, if lower layers drop the packet than it 
>> isn't very useful, so having the option for known incorrect/imperfect 
>> frames would be useful.
>
> Out of curiosity, do you need to have the full packet (understand all 
> the bit) transmitted to the upper layer(s) or (a) only a subset, say 
> only the set of bit that are erroneous or (b) only the statistics of 
> the error process on the bit?
>
> Because, if you only have to generate some statistics, it is in terms 
> of complexity much lighter than to generate all the bits of a packet 
> and then to have a look at all of them for potential errors. In other 
> words, should the simulator stop at the packet-level or be able go 
> down to the bit-level?
>
as an option only, yes, I believe it would be interesting to go down to 
the bit-level. For example, the bit-error masks can be applied to real 
video packets. These masks could be generated by some PHY model . You 
can also imagine to implement Viterbi decoding as you mention below and 
get soft state per bit, which would be much more cpu greedy -- and I 
already  know several experts in image/DSP that would love to get such 
an option ;-)
> I believe NS-3 should be a packet-level network simulator and should 
> avoid dealing with bit directly, mostly because the complexity is way 
> too high: you would not like to simulate a Viterbi decoder in every 
> node. On the other hand, computers are becoming more and more 
> powerful, and maybe this should be an option. Mathieu, are there any 
> guidelines about this? Also, what are thinking other people on this list.
>
> Ruben

Thierry


More information about the Ns-developers mailing list