Viterbi Soft Decision

I have implemented a Viterbi soft decision (1/2 k=7) for a BFSK
modulation.

I need to know the minimum level deeper into the trellis to get symbol. 

In a book, i read something about a good value is 4 or 5 times the
constraint lenght. 

Now, for example, after 32 levels (or states) i get the first symbol, but
then o have a delay of 32 bits.

Can i get a good measure of metric after 8 or 16 states, or the optimum is
to wait 32 states?

JAlbertoDJ wrote:

> I have implemented a Viterbi soft decision (1/2 k=7) for a BFSK
> modulation.

Great.

> I need to know the minimum level deeper into the trellis to get symbol. 

The deeper - the better.

> In a book, i read something about a good value is 4 or 5 times the
> constraint lenght. 

It depends. The higher is the code rate, the deeper you have to dig into 
the trellis. For r=1/2, somewhere 5...7 x constraint length is a good 
compromise value.

> Now, for example, after 32 levels (or states) i get the first symbol, but
> then o have a delay of 32 bits.

Yes, of course.

> Can i get a good measure of metric after 8 or 16 states, or the optimum is
> to wait 32 states?

The optimum is to wait for an infinite number of states. You are trading 
off the delay for the efficiency of the code.


Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Ok, then you advised me 32 bits of delay (4 characters aprox.)??

The software is for ionospheric communications.

And when you says, for r=1/2, somewhere 5...7 x constraint length is a
good compromise value, you refer to Data Input Sequence or Received
Sequence?

For example, I have this message to send:

10100101 (8 bits)   Data input

then, i code data input and get coded sequence for r=1/2:

1100110111100100  (16 bits)  Data coded


5...7 x constraint length is a good compromise value for data coded or for
input data???

thank you

JAlbertoDJ wrote:

> Ok, then you advised me 32 bits of delay (4 characters aprox.)??
> The software is for ionospheric communications.

I advise you to seek for professional advice.

Vladimir Vassilevsky

DSP and Mixed Signal Design Consultant

http://www.abvolt.com

Ok, but my software is for radioamateur communications, in others words,
for hobbie. And i passed in this forum for learning and share knowledge,
not for advertising free.

JAlbertoDJ wrote:
> Ok, but my software is for radioamateur communications, in others words,
> for hobbie. And i passed in this forum for learning and share knowledge,
> not for advertising free.
> 

The cheese is for free only in the two kinds of places:  mouse traps or 
trash bins.

VLV

On Feb 16, 6:54 am, "JAlbertoDJ" <nietoro...@yahoo.es> wrote:
> And when you says, for r=1/2, somewhere 5...7 x constraint length is a
> good compromise value, you refer to Data Input Sequence or Received
> Sequence?
>
> For example, I have this message to send:
>
> 10100101 (8 bits)   Data input
>
> then, i code data input and get coded sequence for r=1/2:
>
> 1100110111100100  (16 bits)  Data coded
>
> 5...7 x constraint length is a good compromise value for data coded or for
> input data???
>
> thank you

The traceback depth is counted in units of coded symbols. In your case
a coded symbol is a pair of bits exiting the encoder, produced by one
information bit entering the encoder. If your decoder has a delay of
5*K coded symbols, that corresponds to 5*K information bits.

John

Thank you John