dynpower: describe the adaptive SNR thresholds

[MUSTARDTIGERFPV]

The Dynamic Power and Signal Health pages both present the fixed per-rate SNR thresholds as the whole algorithm. Since 4.0 those values are only the starting point, so both pages describe behavior the firmware no longer has.

The algorithm

The TX keeps a 48 sample window (StdevAccumulator.h, WINDOW_SIZE = 48) of the SNR reported by the receiver and derives both thresholds from its mean and standard deviation.

Behavior	Source
Lower power at mean + 0.5 x stdev	dynpower.cpp, (snr_stat_mean + (snr_stat_stdev / 2)) / 16
Raise power at mean - (scale x stdev)	dynpower.cpp, (snr_stat_mean - (snr_stat_stdev * scale_factor_numerator) / 4) / 16
scale is 3.25 at 100% LQ, floors at 0.25 from 85% LQ down	numerator 13 to 1 across LQ 100 to 85, std::max(..., 1), over 4
Raise threshold held at least 1dB below the lower threshold	snr_thre_up_limit = snr_thre_dn_scaled - SNR1dB
Samples only collected while instantaneous LQ >= 99%	if (lq_current >= 99) dynpower_stat_snr.add(snrScaled)
Window cleared on packet rate change	dynpower_stat_snr.reset() on RF index change
Raising on SNR also requires a weak signal	rssi <= sens + 21 || lq_avg <= 95
More than one level can be added per update, ~2dB each	while loop with snrScaled += SNR_SCALE(2)

The 0.5 coefficient is 4.1 specific; 4.0 used 1.5. The page is worded to scope this.

Signal Health table corrections

Two cells did not match the rate tables in common.cpp:

• 2.4GHz 250Hz lower power 8.5 -> 9.5dB. Rate index 6 is SNR_SCALE(9.5). The Dynamic Power page already said 9.5dB, so the two pages contradicted each other.
• 900MHz 50Hz raise power 1.0 -> -1.0dB. Rate index 3 is SNR_SCALE(-1).

The other ten cells were checked against the rate tables and are correct. The table intro now says these are starting thresholds and links to the algorithm.

Starting Power

The existing claim is correct but incomplete. Dropping to minimum power is done by ResetPower(), which also runs after a committed configuration change, not only at power up. It deliberately does not lower the power while armed, and with Dynamic Power disabled it applies the configured Max Power instead.

Receiver overload protection

Not previously documented. An RSSI of -5dBm or higher lowers the power one level, and this check sits above the if (!config.GetDynamicPower()) return; guard, so it runs even with Dynamic Power disabled. This is a common source of confusion on the bench.

Smaller fixes

Three boundary conditions were written as strict inequalities but are inclusive in code: the 50% LQ hard limit, the 85% LQ top up, and the FLRC lower power RSSI limit.