Three fixes from Thetis-Enhanced (Yurij eu2av), measured on Orion MK2 /
Anvelina PRO3 hardware:
- drop overrange samples (env_TX*hw_scale > 1.0) before the cubic
xbuilder fit - they distort the fit and produce a wrong rx_scale
- optional median-ratio + MAD outlier rejection before the fit,
controlled by new SetPSOutlierSigma (0 = off)
- fallback rx_scale estimate from the top amplitude intervals when the
xbuilder fit fails or is rejected by rxscheck
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Model the pair on fmd.c / fmmod.c, but with the parts that a 75 kHz
deviation forces:
- the demodulator discriminates with arg(x[n] * conj(x[n-1])) rather
than a PLL; an omegaN in the tens of kHz cannot track 75 kHz.
- emphasis is a one-pole RC (tau 75 us) on both ends, not fmd's 1/f
fc_impulse FIR. A 1/f FIR from f_low = 20 Hz would sit ~+57 dB at
20 Hz, where broadcast FM specifies flat below the corner.
- TXA_WFM leaves the shared preemph block off; wfmmod carries its own.
- wfmmod clamps bp_fc = deviation + f_high to 0.45 * samplerate, since
+/-90 kHz exceeds Nyquist at the rates the narrowband modes use.
Scope is mono: no 19 kHz pilot, no 38 kHz stereo subcarrier, no RDS.
Both mode enums are appended to so the ABI stays stable for clients.
The JNI bindings are deliberately left alone.
Verified against a wfmmod -> wfmd loopback: the discriminator is exact,
the dc-removal one-pole tracks |H_lp(f)| * ain * sdelta to ratio 1.000,
and a 700 + 1900 Hz two-tone comes back with THD+N ~ 3e-5 % at +/-37.5
kHz deviation. Note the demodulator aliases unless samplerate exceeds
2 * deviation * |aud|max, so 192 kHz is the practical floor.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
ew8bakandClaude Fable 5