trimmed-mean v0.1.0

Average a batch after dropping one minimum and one maximum.

Application scenarios

• PIR acquisition uses a 6-sample batch; one bad sample must not move the result.
• An ADC burst contains a spike from a switching load.
• A sensor batch has one stuck or saturated reading.

Contract

uint16_t trimmed_mean(const uint16_t *samples, uint8_t count);

• count must be 3..255. Smaller batches return 0.
• One minimum and one maximum are removed; duplicates are handled as values.
• Result is rounded to the nearest integer.
• No allocation; O(N).

Parameters

• Batch size: 5..9 is typical. Larger batches reject more impulses but add latency.
• Duplicate min/max reduce the effective trim; document the expected distribution.

Memory and timing

• No state; the caller owns the batch.
• O(N) with one pass for sum/min/max.

Integration

uint16_t batch[6];
/* fill batch from the sensor, then: */
uint16_t value = trimmed_mean(batch, 6);

When not to use

• A continuous stream without batching; use median or moving-average.
• The batch may contain more than one bad sample; a larger trim or median is safer.

Verification

make test covers the 6-sample PIR batch, 3-sample minimum, duplicates, and invalid counts.

Behavior reference: a production trimmed-mean filter; this implementation is original.