FM and wideband-FM
FM demodulation runs on the complex baseband. Narrowband FM produces audio directly from the instantaneous frequency; wideband FM (broadcast) decodes the full stereo composite, including the 19 kHz pilot, the 38 kHz stereo subcarrier and 57 kHz RDS.
FM detector#
The detector computes the instantaneous frequency as the time derivative of the baseband phase (fmfix()). How that phase is estimated is selected by the operator with the FM detection mode. With two channels the second channel is detected in parallel. If the requested audio bandwidth already equals the baseband bandwidth, the detector output goes straight to baseb_out.
Detection modes (FM0–FM3)#
The FM button on the baseband graph cycles bg.fm_mode through four states. Mode 0 differentiates the raw atan2 phase directly — lowest latency, best on strong signals. Modes 1–3 instead estimate the phase from the dominant bin of a short windowed FFT (finding the peak and interpolating across it), which rejects off-frequency noise and improves weak-signal FM; a larger FFT gives finer frequency resolution at the cost of time resolution and computation.
| Button | Phase estimate | FFT size |
|---|---|---|
| FM0 | direct atan2, per sample |
— |
| FM1 | windowed FFT, dominant bin + interpolation | 8 |
| FM2 | windowed FFT | 16 |
| FM3 | windowed FFT | 32 |
Internally the button maps to the FFT order fm_n = bg.fm_mode + 2, which is then forced to 0 (the atan2 path) when it would be below 3 — so FM0 is fm_n == 0 (fm_size == 1) and FM1–FM3 are fm_n 3/4/5. Changing the mode rebuilds the baseband stage.
In the source
detect_fm() — fm.c:93 (called from mix2.c:1838); phase/frequency estimator fmfix() — fm.c:45: atan2 when fm_n == 0 (fm.c:51), otherwise the windowed-FFT peak with 3-bin interpolation (fm.c:64). The mode → FFT-size mapping (fm_n = bg.fm_mode + 2, if(fm_n < 3) fm_n = 0, fm_win = make_window(4,…)) is at baseb_graph.c:4138; the FM button (BG_TOGGLE_FM_MODE) wraps at MAX_FM_FFTN-2 == 3 (baseb_graph.c:2798, label FM%d screen.c:2839; MAX_FM_FFTN / MAX_FM_FFTSIZE sigdef.h:2). The audio-bandwidth decision (use_audio_filter) is at fm.c:108. BG parameters fm_mode, fm_subtract, fm_audio_bw are in globdef.h:1032.
Wideband FM: composite, stereo and RDS#
For broadcast WFM the demodulated composite is split with low-pass and subcarrier filters into the mono sum (L+R), the stereo difference (L−R, recovered from the 38 kHz subcarrier regenerated from the 19 kHz pilot) and the 57 kHz RDS channel; the sum and difference are combined into L/R audio with de-emphasis. This path is taken only with one channel, a detected pilot, and fm_subtract != 2; otherwise Linrad uses the narrowband path (nbfm). An optional subtract mode removes the dominant carrier before further processing to cut intermodulation from adjacent channels.
In the source
WFM steps are numbered in the comments of detect_fm(): composite low-pass (fmfil55/fmfil70) fm.c:326; 19 kHz pilot detection fm.c:372; 38 kHz stereo and 57 kHz RDS fm.c:387; audio filter fm_audiofil fm.c:417; RDS low-pass fmfil_rds fm.c:451. The nbfm branch is selected at fm.c:189. Filter and channel arrays (fmfil*_fir, baseb_fm_sumchan/diffchan, fm_pilot_tone, baseb_fm_pil2/pil3) are declared in sigdef.h:226. Several WFM_* network modes stream the raw or partially processed baseband for analysis.
Related#
FM runs after the baseband stage; the second channel it uses (baseb_raw_orthog / baseb_carrier) comes from the second mixer.