Re: [Discuss-gnuradio] GMSK Viterbi Equalization

[Bob McGwier]

isaacgerg wrote:
> Is this scheme as simple as doing an LMS decision-directed equalization
> scheme where the Viterbi algorithm used to make the symbol decision?
>
>
>   
No there is no "equalizer" per se. It is a misnomer, but a popular one. 
What you have is a model of the filter the channel applies to the data 
symbols emitted by the source to produce the signal. You do dynamic 
programming forward in time over the putative symbols and the metric is 
the distance between the observed data and the filter applied to the 
paths in reverse. So the channel filter is applied to the "paths" as you 
try new symbols. Just like viterbi decoding, you have coalescence. When 
all paths backwards terminate in the same node, that is when you emit a 
decision in theory. Practically speaking, this is usually done after a 
fixed interval.

The channel sounding symbols are used to update the "channel filter".

This is much better than trying to "invert the channel" which is what an 
equalizer does. Those places where the channel transfer function has a 
zero in its spectrum (or a deep null) you have significant difficulties 
when inverting. You enhance the noise there because you are "nearly 
dividing by zero".

Bob

--
ARRL SDR Working Group Chair. Member: ARRL, AMSAT, AMSAT-DL,
TAPR, Packrats, NJQRP, QRP ARCI, QCWA, FRC.
“Trample the slow ....  Hurdle the dead"