Trellis in generic_mod_demod

I want to use convolution code to transmit packet over USRP. So I ended
modifying the generic_mod_demod in digital.

The original flow graph was

packed_to_unpacked(bits_per_symbol) —> gray_encoder —>
differential_encoder —> chunks_to_symbol(constellation)
—> rrc_filter —> USRP

USRP —> fll_band_edge —> pfb_clock_sync —>
0) —> demod_pkts

Now I have modified it to the following.

packed_to_unpacked(bits_per_trellis_input) —> trellis_encoder(fsm)
chunnks_to_symbol(constellatin) —> rrc_filter —> USRP

USRP —> fll_band_edge —> pfb_clock_sync —>
4)(symbol) —> viterbi_combined —> demod_pkts

I find this structure ends up in more errors than before. ( almost all
errors )

  1. Does the packet size at the output of has anything to do with
    block_length in viterbi_combined?

  2. Will it make any difference if I specify the starting and ending

There is no logic at the moment to make the trellis encoder and
decoder sync their blocks with the packets. I would have thought that
if you don’t specify the starting and ending states, and the blocks
are reasonably long, then it would still work adequately but I could
be wrong.

If you want to sync the blocks with the packets then you could add a
stream tag to the start of each packet, and detect that with the
encoder and decoder.