Gr-trellis: convenc and viterbi with own mapper/de-mapper

GNU Radio
1

Hi all,

I am trying to use the convolutional encoder and Viterbi provided by the
gr-trellis class within another environment. I have my own mapper and
de-mapper blocks which I want to use. So I tried to use the feed the
viterbi_combined with this arguments:

va_combined =
trellis.viterbi_combined_fb(fo,nsymbols,0,-1,1,[-1,1],trellis.TRELLIS_EUCLIDEAN)

My de-mapper outputs soft bits between -1 and +1. Here is an example
output of my test script:

data: [0, 1, 1, 0, 0]
encoded: (0, 3, 2, 2, 0)
unpacked: (0, 0, 1, 1, 1, 0, 1, 0, 0, 0)
modulated: ((1+0j), (1+0j), (-1+0j), (-1+0j), (-1+0j), (1+0j), (-1+0j),
(1+0j), (1+0j), (1+0j))
demodulated: (-1.0, -1.0, 1.0, 1.0, 1.0, -1.0, 1.0, -1.0, -1.0, -1.0)
decoded: (0, 0, 1, 0, 0, 1, 0, 1, 0, 1)

Another thing I don’t understand is why the decoder outputs 10 values
instead of 5. I would be glad if someone told me what I am doing wrong.

Regards,
Jonas

2

On Mon, 16 Aug 2010, Jonas M. Börner wrote:

unpacked: (0, 0, 1, 1, 1, 0, 1, 0, 0, 0)
modulated: ((1+0j), (1+0j), (-1+0j), (-1+0j), (-1+0j), (1+0j), (-1+0j), (1+0j), (1+0j), (1+0j))
demodulated: (-1.0, -1.0, 1.0, 1.0, 1.0, -1.0, 1.0, -1.0, -1.0, -1.0)
decoded: (0, 0, 1, 0, 0, 1, 0, 1, 0, 1)

Another thing I don’t understand is why the decoder outputs 10 values instead of 5. I would be glad if someone told me what I am doing wrong.

Regards,
Jonas

If I remember correctly, when you start pushing symbols through the
trellis, it has some zero values before the input and some zero values
after your last symbol pushed into the trellis. I think this is
typically the
length of your trellis. So at some point you will have to truncate the
trellis output.

Any one else care to comment?

–Colby B.

3

Hi Colby,

thanks for your reply. My trellis is created like this:

t=trellis.fsm(1,2,[91,121])

The constraint length is 7 so it doesn’t look like it was connected to a
trellis-specific thing. As I remember from the gr-trellis examples they
didn’t do any truncating there before comparing the source with the
destination for error calculation…

Regards,
Jonas

Am 17.08.2010 um 08:10 schrieb [email protected]:

4

---------- Forwarded message ----------
From: Achilleas A. [email protected]
Date: Thu, Aug 19, 2010 at 4:19 AM
Subject: Re: [Discuss-gnuradio] gr-trellis: convenc and viterbi with
own mapper/de-mapper
To: [email protected]

Jonas,

there is a very extensive doc file in gr-trellis that hopefully
answers your questions,
as well as a lot of examples.
Please read it and let me know if you still have questions about it.
BTW: your issue has nothing to do with truncation as was suggested in a
reply.

In short: one way to implement your system is to mapp the encoded
symbols directly to
QPSK symbols (essentially this is what you are doing!) and thus
produce 5 demodulated symbols
that you will feed in the VA.

Achilleas

5

On Tue, 17 Aug 2010, “Jonas M. Börner” wrote:

My de-mapper outputs soft bits between -1 and +1. Here is an example output of my test script:
Regards,
Jonas

If I remember correctly, when you start pushing symbols through the trellis, it has some zero values before the input and some zero values after your last symbol pushed into the trellis. I think this is typically the length of your trellis. So at some point you will have to truncate the trellis output.

Any one else care to comment?

–Colby B.

I have not used the GNU Radio trellis before. In the past, I have only
used the the matlab trellis and I remember I had to deal with
truncating.

–Colby

6

Jonas,

one idea is to add a method “set_K(K)” to the viterbi class.
This method sets d_K AND calls the set_multiple() method with the new
value.
This method will be called from , say , the python code.

Achilleas

On Sat, Oct 23, 2010 at 1:59 PM, “Jonas M. Brner”