weighted-average v0.1.0

Weighted mean of a batch.

Application scenarios

• Recent samples are more trustworthy than older ones.
• A sensor fusion step gives different confidence to each source.
• A calibration batch weights the center of the range more.

Contract

uint16_t weighted_mean_u16(const uint16_t *samples, const uint8_t *weights, uint8_t count);

• count must be 1..255.
• Weights are uint8_t; a zero weight sum returns 0.
• Result is rounded to the nearest integer.
• The product sum fits uint32_t for 255 samples with 8-bit weights.
• No allocation; O(N).

Parameters

• Weights are fixed for the batch; document their meaning.
• All-zero weights return 0; avoid that configuration.

Memory and timing

• No state; the caller owns both arrays.
• One pass; one division at the end.

Integration

const uint16_t samples[3] = {10, 20, 30};
const uint8_t weights[3] = {1, 2, 3};
uint16_t value = weighted_mean_u16(samples, weights, 3);

When not to use

• Weights change every sample; use a stateful filter with explicit coefficients.
• Impulses must be removed; use trimmed-mean or median.

Verification

make test covers a weighted result, zero weights, zero count, and a null input.

Behavior reference: Padauk Soft_Filter/AD_filter_08.