Results | Unique measurements (CSV)

Experimental Methods

Waveform and frame

GMSK, BT=0.5, h=0.5; 8 samples/symbol. Symbol rates: 4800, 9600, 19200, 38400, 57600 and 115200 symbol/s. C0 uses RS(255,223). C1, C2 and C3 add convolutional coding at rates 1/2, 2/3 and 3/4. Interleaver depths: 1, 2, 3, 4, 5 and 8.

Each independent burst contributes one randomly selected RS codeword to FER. Every recovered information byte is checked. Acquisition and decoding failures remain in the denominator. An interleaved burst is not counted as multiple independent trials.

Channels and energy

AWGN; AWGN plus orbital Doppler; AWGN plus Doppler, oscillator fluctuations and a weak echo. Doppler uses a synthetic equatorial overhead pass at 435 MHz and 550 km circular SSO, including Earth rotation and propagation-delay time scaling; elevation is at least 10 degrees. This is a reference geometry, not a reconstruction of BY70-4 telemetry.

Each transmitter and receiver oscillator has independent stationary OU frequency fluctuations (3 Hz RMS, 50 ms correlation time) and phase fluctuations (0.5 degree RMS, 0.5 ms correlation time). The echo has -15 dB relative power, 2 microseconds delay, random burst phase and +0.2 Hz differential Doppler. These are assumed parameters, not measured radio specifications. Carrier tracking estimates noisy observations; it is not forced to the true frequency.

Es/N0 refers to received symbol energy after one scalar burst-energy calibration. Intra-burst fluctuations and ISI remain. Complex noise variance is sps/10^(Es/N0/10), split equally between I and Q; no code-rate factor is applied. C/N0 [dB-Hz] = Es/N0 [dB] + 10 log10(Rs [symbol/s]). AWGN results are shared across rates without increasing their sample count.

Timing lock

The common uncoded probe has 32 alternating preamble symbols, a 32-symbol marker and random payload, with total lengths of 4096 and 32768 symbols. It does not depend on code or interleaver selection.

The initial mean timing error must be within half a symbol, with no subsequent run of 32 scored symbols outside half a symbol. Integer-symbol slips are retained. Scoring uses the final two-pass trajectory over the burst, excluding the initial 64 and final 8 symbols; it does not measure real-time acquisition delay. Timing lock alone does not certify carrier lock.

Sampling and intervals

11,778 unique points. The initial experiment used 65,536 trials per point. Registered continuations add 1,505,510,400 trials at 4,459 FER points, for a total of 2,277,393,408 independent trials. Final counts range from 65,536 to 1,668,352 per point. The 1,482 timing points are unchanged. FER: 0 to 8 dB; timing: -2 to 12 dB; both use 0.25 dB spacing. Earlier SCNet, benchmark and qualification observations are excluded.

Intervals invert the Beta(1/2,1/2) mixture likelihood ratio at alpha=0.04/512. They cover all registered points and sample sizes simultaneously with probability at least 96% for each mode (441 comparisons); this is not a 96% joint claim over all 144 modes.

Supplemental sample allocations were chosen using the initial results and then held fixed. The same time-uniform confidence sequences remain valid under this continuation; the confidence level is unchanged. Intervals are recomputed from total event and trial counts, not averaged from two estimates. Precision targets are checked against the final intervals; completing the allocated trials does not guarantee the target width.

The original and supplemental executions use the same binary and frozen receiver/channel sources. Raw task rows and checkpoint roll-ups were checked against their registered seed ranges. Repeated checkpoint records, original samples already present in supplemental CSVs, and AWGN display copies are not counted again.

Lines connect measured points only. Downward triangles mark upper bounds for zero observed frame errors. With 0/262,144 errors, the displayed upper bound is approximately 6.205515e-05 (0.00621%); zero observed errors do not mean zero true FER. All nonzero estimates are plotted. Lower confidence limits below 10^-6 continue beneath the graph and are given exactly in the tables. Tested thresholds use the confidence bound and do not certify interpolated values.

Adjacent-point checks use the simultaneous 95% full-table confidence sequences. Fixed-sample Fisher tests are not applied to sample counts selected using earlier outcomes. A first confirmed threshold refers only to a tested point; it does not certify all higher or interpolated values.

Precision summary | Measurement details | Receiver diagnostics (CSV) | Adjacent-point checks (CSV)

Typeface: Computer Modern. Font | Font licence.