[Question] Is the tail of the sequence created by qdetector_*_create_gmsk correct?

[xerpi]

qdetector_*_create_gmsk calls gmskmod_modulate with the _sequence symbols passed (of length _sequence_len) plus the 2*_m extra symbols corresponding to the filter delay (_m).
Those extra symbol delay symbols are passed as 0s:

liquid-dsp/src/framing/src/qdetector.proto.c

Lines 211 to 212 in 9f32eb7

	for (i=0; i<_sequence_len + 2*_m; i++)
	gmskmod_modulate(mod, i < _sequence_len ? _sequence[i] : 0, &s[_k*i]);

Since we are making the extra symbols a "hard" 0, the generated samples' phase looks like:

This was generated by calling qdetector_cccf_create_gmsk with:

• samples per symbol = 10
• filter delay = 7
• beta = 0.4
• sequence: {0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1} (24 symbols)

Notice how the tail stays at a "low position".
Shouldn't the tail be all 0s instead?
Otherwise when correlating, those samples will have an influence.
I have the impression that something like this would be desirable:

[xerpi]

I've played a bit around it.
I have added a new function, gmskmod_modulate_raw which lets you pass the raw sample value (_x), then at qdetector_*_create_gmsk I have splitted the for (i=0; i<_sequence_len + 2*_m; i++) loop into two:

1. for (i=0; i<_sequence_len; i++) which just calls gmskmod_modulate
2. for (i=0; i<2*_m; i++) which calls the new gmskmod_modulate_raw passing 0.0f as the sample value

This generates the follow sample sequence:

Notice how the tail of the phase (green) goes to 0.