Linrad Notes
Source Issues

Coherent CW and Morse decoding

Linrad's CW processing is designed for weak, non-perfect signals with QSB, chirp and drift. It extracts the signal's carrier phase coherently and can decode Morse to ASCII. The algorithm is a state machine driven by cw_detect_flag; the author describes it in z_MORSE_DECODING.txt.

Coherent CW / Morse decoding

Three bandwidths

The baseband time function is computed at three bandwidths, all complex and with their phases rotated to match the carrier (so the signal is in I and any chirp/FM appears in Q):

In the source

The three bandwidths and the phase-alignment convention are described in z_MORSE_DECODING.txt. baseb_carrier comes from the narrow carrier filter in the second mixer (mix2.c:246); keying_spectrum is also computed in mix2.c. CWDETECT states are defined in sigdef.h:5.

Finding the keying speed

The power spectrum of the I component of baseb_wb, averaged over ~150 dots, has peaks at the Morse "clock" frequency and its harmonics/sub-harmonics (keying_spectrum). evaluate_keying_spectrum() picks the fundamental and derives the dot length in samples, cwbit_pts. make_ideal_waveform() then builds the ideal dash shape as seen through the selected baseband filter.

In the source

evaluate_keying_spectrum() — coherent.c:77; make_ideal_waveform() — coherent.c:212; collect_ramp() (power detector → baseb_ramp) — coherent.c:156; detect_cw_speed() in cwspeed.c. Driven from coherent_cw_detect() states CWDETECT_CLEARED / CWDETECT_SEARCH_SPEED (coherent.c:283).

Fitting dashes, then decoding

With a waveform template established (CWDETECT_WAVEFORM_ESTABLISHED), the decoder:

  1. collects the average dash shape from regions where the power detector (baseb_ramp) shows the signal high for ≈ 3·cwbit_pts, refining cwbit_pts;
  2. steps through the data fitting long regions to the dash shape, storing good fits in cw[];
  3. collects wideband dash/dot shapes and characterizes the intra-dash phase drift with two derivatives (frequency drift during a symbol), building noise-free reference functions;
  4. interpolates over short gaps and guesses the symbol pattern where one option is clearly best (short_region_guesses());
  5. guesses characters across undecoded gaps by least-squares fitting the surrounding decoded signal, keeping only guesses valid in the Morse alphabet and choosing the best S/N (character_guesses());
  6. reconstructs a more accurate carrier from the guessed keying and re-detects from scratch.

The result is drawn on the coherent-CW graph and, for meteor scatter and similar, output as ASCII.

In the source

State machine coherent_cw_detect() — coherent.c:283: first_find_parts() / second_find_parts() (CWDETECT_SOME_PARTS_FITTED, CWDETECT_LIMITS_FOUND), init_cw_decode_region() / cw_decode_region() (CWDETECT_REGION_*), init_cw_decode() / cw_decode() (CWDETECT_SOME_ASCII_FITTED); on failure it resets to CWDETECT_CLEARED. Waveform fitting is in cohsub.c (fit_dash, store_symmetry_adapted_dash, get_wb_average_dashes); auto-decode and speed estimation in cwdetect.c, cwspeed.c and morse.c. The 11 steps are enumerated in z_MORSE_DECODING.txt.

The narrow carrier used here is produced by the second mixer; the coherent phase extraction shares the amplitude/phase measurement used for adaptive polarization. The same recovered carrier phase drives the coherent output modes (Coh1–Coh4) for synchronous and binaural listening.