PhosphorRTTY — Decoder Test Results

Independent accuracy testing of the PhosphorRTTY RTTY decoder using the CLI test harness (RTTYDecode). The CLI uses identical DSP code to the iOS app via symlinks — same source files, same binary logic.

Methodology

• All tests run via the RTTYDecode CLI tool, which shares DSP source files with the iOS app via symlinks (single source of truth).
• Each file tested twice: once in AUTO mode (AFC + auto-baud + auto-shift) and once with fixed manual parameters (--mark, --shift, --baud, --no-afc).
• Error counts are framing errors reported by the DPLL decoder (failed stop-bit check).
• Total frame counts include all decoded characters plus framing errors.

Decoder Settings (defaults unless overridden)

• Sample rate: 8 kHz (resampled from source)
• FFT: 512-point, Hann window, ~15.6 Hz/bin
• Tone detection: I/Q quadrature demodulation, biquad LPF (75 Hz cutoff at 170 Hz shift)
• Envelope: asymmetric decay-average (attack = symbollen/4, decay = symbollen×16)
• Bit decision: W7AY ATC method
• Squelch threshold: 0.002
• DPLL: per-transition phase nudge, α = 0.05
• AFC: 3-frame pre-acquisition, 3-frame retune confirmation, 2-frame LOST confirmation

Test Files

* W6IWI labels this file as 60 Bd, but auto-baud detection and manual testing confirm it is 45.45 Bd. Fixed tuning at 60 Bd produces ~975 errors; at 45.45 Bd it produces 0.

Results

Decoded Output Samples

RTTY_Test_850.wav — Fixed Tuning (0 errors)

websdr_2022...14086kHz — Fixed Tuning, 20m OTA (0.7% errors)

websdr_2024...170Hz — AUTO mode, OTA (5.6% errors)

Notes on Specific Files

random850.wav — Known Broken Test File

The ~53% error rate is not a decoder fault. The W6IWI signal generator encodes stop bits as space (0) instead of mark (1) for approximately 53% of frames. The data bits themselves are correct — verified using --lenient mode which bypasses the stop-bit check. This affects random170.wav from the same generator identically.

websdr_2022...7086kHz — 40m Night with Deep Fading

The worst-performing file at 17–31% errors. This is a nighttime 40-metre recording with severe multipath fading typical of skip-zone propagation. The signal periodically drops below the noise floor entirely. In AUTO mode, each fade triggers AFC LOST re-acquisition (131 events in 26 minutes), burning several frames per cycle. Fixed tuning halves the error rate by avoiding re-acquisition overhead, but the remaining 17% represents genuine signal corruption where even a perfect decoder would fail.

RTTY_Test_850.wav — Mislabelled Baud Rate

W6IWI's website labels this file as 60 Bd. Testing confirms it is 45.45 Bd. At 60 Bd fixed tuning: 975/1,638 errors (59.5%). At 45.45 Bd fixed tuning: 0 errors. The auto-baud detector correctly identified 45.45 Bd without intervention.

AFC Impact

The difference between AUTO and fixed tuning quantifies the cost of AFC re-acquisition:

Each AFC LOST event costs approximately 3 frames during re-acquisition. On noisy signals, AFC re-acquisition is the dominant source of errors rather than bit-level decoding failures.

Summary

The PhosphorRTTY decoder demonstrates strong performance across a range of real-world and synthetic RTTY signals:

• Clean signals: 0% error rate with correct tuning. Perfect decode of QBF test sentences, RYRY sequences, and callsign identification.
• Moderate OTA signals (20m daytime): 0.4–4.2% error rate with fixed tuning; 4.4–6.5% in full AUTO mode. Callsigns, numbered sentences, and RYRY patterns decode reliably.
• Severe conditions (40m night, deep fading): 17% error rate fixed, 31% AUTO. The signal itself is intermittently below the noise floor — errors here are propagation-limited, not decoder-limited.
• Auto-detection: Auto-baud correctly identified a mislabelled 45.45 Bd file that the source claimed was 60 Bd. AFC successfully acquires and tracks signals across 170 Hz and 850 Hz shifts.
• The core decoder (DPLL + I/Q demodulation + W7AY ATC) is accurate. The majority of errors in AUTO mode stem from AFC re-acquisition cycles during noise bursts, not from bit-level decoding failures.