What is the proper way to use a Viterbi Decoder in continuous mode?

I'm trying to replicate the viterbi decoding operation in this example:

openExample('comm/DigitalVideoBroadcastingTerrestrialExample','supportingFile','commdvbt.slx')

It's doing okay, but I noticed that the Inner Decoder (Conv Encode) seems to be performing poorly compared to other blocks.

Example (comparing bits of input to output ):

OFDM Error rate: 1.000000

Mapping Error rate: 0.107143

Inner Int Error rate: 0.019761

Inner Dec Error rate: 0.029412

Iteration: 1

...

OFDM Error rate: 1.000000

Mapping Error rate: 0.074625

Inner Int Error rate: 0.003676

Inner Dec Error rate: 0.371936

Iteration: 28288

...

OFDM Error rate: 1.000000 (makes sense because no two input and output symbols will be exactly the same after noise)

Mapping Error rate: 0.058128

Inner Int Error rate: 0.012408

Inner Dec Error rate: 0.035539

Iteration: [last]

Resultant Error rate [all iterations]: 0.98776

If I use 'trunc' and 'hard' I get better results, but the viterbi decoder still sees consistent error rates between 0.01 - 0.50.

I believe the reason is because I'm somehow not achieving continuity between my iterations. What can I do?

trellis = poly2trellis(7,[171 133]);                %Trellis for 3/4 encoder
puncpat = [1 0 1 0 1 1]';                           %Puncturing Pattern

for i =  1: packets*ofdm_symbols + (2244/204)
    [ Some operations]    
    ....
    ....
    %% Inner Coder (Conv Encode)
        msg_encoded(startidx_out:endidx_out) = convenc(bin_interleaved(startidx_in:endidx_in), trellis,puncpat);
    ... 
    ...
    [ Some operations]
    ...
    ...
    %% Inner Decoder (Vit Dec)
        msg_decoded(startidx3_out:endidx3_out) = vitdec(Deinterleaved_packet(end-2175:end)', ...
        trellis,length(msg),'cont','unquant',puncpat);
    ...
    ...
end

[Dylan Ditta]

Oh yeah and my noise is based on:

ofdm_mod = awgn(ofdm_mod, 22,'measured')

22 dB of Signal/noise shouldn't screw up the signal this badly

Answer 1

[Sivsankar]

Hi @Dylan Ditta,

In the example model, the Viterbi decoder has a traceback depth of 136. So, make sure that you employ that in your code. Changing to the correct traceback depth may reduce the error rates. As a general estimate, the Traceback depth value is approximately two to three times (k – 1)/(1 – r), where k is the constraint length of the code and r is the code rate

In the model, the output from the Viterbi decoder is carried forward to a delay block and that is considered as the decoded output message in the model. So maybe you could try to implement that in your code before comparing the input and output bits.

You can leverage the following documentation on how you can implement the vitdec function to decode punctured signals:

https://www.mathworks.com/help/comm/ref/vitdec.html?s_tid=doc_ta#mw_7a459ebd-6025-41f9-8549-5077fc9525dd

Try to see if the above implementation decreases the error rates.

Thanks