Also known as: signal fingerprints, symptom signatures, diagnostic fingerprints
Signal-quality signatures are measurement patterns that recur in GopherTrunk’s issue tracker often enough to be diagnostic on sight. Each row below pairs a specific numeric fingerprint with what it looks like at first glance and what it actually meant when it was finally run to ground. Recognizing a row saves the detour the original investigation took; the diagnostic playbook covers how to collect these measurements in the first place.
| Signature | What it looks like | What it actually means | Source issue |
|---|---|---|---|
| Phase imbalance ≈ +78°, image rejection ≈ 3 dB | A badly miscalibrated tuner | A real-sample stream misread as interleaved I/Q | #454 |
| Sample histogram rail-pinned (~25% at 0 and ~25% at 255), RMS near 0 dBFS, FFT looks clean | Weak or absent signal, since nothing decodes | Front-end overload; the fix is less gain, fixed | #881 |
| Wideband FFT carrier SNR higher, in-channel EVM much worse | The DSP breaks at the higher sample rate | Sampling-clock phase noise; the damage is in the captured samples | #764 |
| Dibit histogram 50/0/50/0 — only outer symbols | Corrupt demodulator output | Slicer collapse: symbol amplitude far above the slicer’s calibrated levels | #275 |
| Sync correlation hits at exactly one-frame cadence | Just another intermediate metric | Proof the sync layer works, independent of downstream CRC | #553 |
offset_hz ≈ ±12.5 kHz on a locked control channel |
A drifty crystal | Locked onto an adjacent channel’s carrier; the decoded identity is another site’s | #815 |
| Uncorrectable frames at exactly 100% | Hopelessly weak signal | A structural bit-alignment bug — signal degrades gradually, never to a deterministic 100% | #489 |
| EVM improving while CRC yield stays at 0 | An equalizer that is helping | The blind-equalizer trap: better modulus is not better bits | #1001 |
+78° phase, 3 dB image rejection: real samples read as complex
The Airspy R2/Mini stream bare real ADC samples at twice the IQ rate; converting to complex baseband is the host’s job. A driver that pairs adjacent real samples as I and Q produces this textbook fingerprint: neighbouring samples of an oversampled signal are highly correlated, so the apparent phase imbalance lands near 90° (measured +78.1°) with image rejection of only ~3.3 dB. After a proper Hilbert-pair converter the same metrics read −0.0007° and ~70.8 dB. See Airspy rate selection.
Rail-pinned histogram: overload that decodes anyway
In #881 a raw capture had 24.9% of its 8-bit samples at exactly 0 and 24.9% at exactly 255 — half the stream pinned to the rails, RMS at +1.3 dBFS — yet the FFT showed clean-looking carriers and an offline decode still recovered 3 of 4 control channels, because constant-envelope C4FM carries its information in phase and survives hard limiting. That combination (clean FFT, saturated histogram, dead live decode) misdirected the investigation toward a compelling but wrong DSP hypothesis. The counter-intuitive rule: on an overloading front end, more gain makes it worse — the fix was a fixed low gain, not AGC. See SDR gain & overload.
Carrier-clean but modulation-degraded: clock phase noise
In #764 the wideband FFT showed the carrier cleaner at 10 MS/s (SNR 36.2 dB vs 33.1 dB) while in-channel demod quality was ~10 dB worse (EVM 22.5% vs 7.4%). Neither capture clipped (peaks near −48 dBFS), ruling out overload. A high carrier-to-noise floor with smeared modulation is the signature of phase noise on the sampling clock (reciprocal mixing): the energy is all there, but its phase is jittered. The deficit was proven to be in the captured samples themselves via an independent-resampler cross-check.
50/0/50/0 dibit histogram: slicer collapse
When the four-level slicer’s calibrated thresholds sit far below the actual symbol amplitude, every symbol reads as an outer level — the dibit histogram collapses to two bins. The treacherous part: frame-sync correlation still works, because the P25 sync word uses only outer symbols, so the decoder looks half-alive. A symbol-domain AGC restored a healthy 28/22/27/23 spread in #275. Compare the eye diagram view of the same failure: all trajectories pass wide of the inner levels.
Sync hits at exact frame cadence: the sync layer is proven
The decisive positive signature from #553: TETRA training-sequence correlation went from 0 hits to 97 hits with a modal spacing of exactly 1020 dibits — one TETRA frame. Random correlation noise does not arrive on a frame grid. This proves sync acquisition works on real air independent of downstream CRC results, which cleanly splits “sync is broken” from “descramble/deinterleave/FEC is broken.”
offset_hz near ±12.5 kHz: you locked the neighbour
GopherTrunk locks whatever compatible carrier the matched filter finds in the channel passband and reports the configured frequency. A measured carrier offset sitting at the channel spacing (±12.5 kHz for narrowband P25) means the lock is an adjacent channel — and every decoded site identity belongs to the neighbour. Full treatment in carrier offset & adjacent-channel lock.
Exactly 100% uncorrectable: structural, not signal
Weak signal produces a distribution of frame quality. In
#489,
ldus=1622 uncorrectable_ldus=1622 — exactly 100%, deterministically — meant the
voice-frame bit offsets were structurally wrong (the decoder omitted the link
control blocks interleaved between voice subframes), so the FEC never saw aligned
codewords. When every frame fails, suspect alignment tables and bit ordering
before antennas. The round-trip tests passed because encoder and decoder shared
the same wrong offset table.
EVM improving while CRC stays 0: the blind-equalizer trap
A blind CMA equalizer minimizes deviation from a constant modulus — not decoding correctness — and its cost surface has spurious minima. During the TETRA equalizer work (#1001), a numerically unstable variant showed differential EVM collapsing from 34% to 8% while CRC-valid frame yield stayed at exactly zero: the equalizer had made the constellation rounder, not righter. The durable rule recorded in the project’s engineering notes: never conclude an equalizer helps from EVM — decode all the way to CRC yield, the only trustworthy metric.
Provenance
- #454 — Airspy real-sampling stream misread as I/Q; the +78°/3 dB fingerprint.
- #275 — first P25 lock; the 50/0/50/0 slicer-collapse histogram.
- #489 — 100% uncorrectable LDUs as a structural-misalignment tell.
- #881 — rail-pinned overload behind a clean-looking FFT.
- #764 — carrier-clean but modulation-degraded = sampling-clock phase noise.
- #553 — TETRA lock; correlation hits at exact frame cadence as sync proof.
- #815 — the ±12.5 kHz adjacent-channel lock fingerprint.
- #1001 — TETRA equalizer work where the EVM-vs-CRC trap was pinned down.