Statistical Methods
Record Boundary Assumptions in Polyphase Resampling
Define the task
Changing a signal's sampling rate changes both its grid and its filtering history. A polyphase resampling procedure upsamples, applies a low-pass finite impulse response filter, then downsamples. Record the original rate, target ratio, sample axis and filter specification. This differs from selecting every few observations without filtering. The practical audit must include the beginning and end of the record, where assumed values outside the observed signal can affect results.
Carry out the check
Specify the integer up and down factors, filter design and padding rule. Default zero extension is an assumption, not evidence that the true signal was zero before and after recording. Compare boundary behavior under scientifically plausible alternatives and preserve timestamps for the output grid. A suitable antialias filter is required when reducing the effective rate. Avoid interpreting filter transients as events, and do not silently concatenate records across gaps as if they were one continuous signal.
Worked hypothetical example
For a hypothetical input rate of 250 Hz and up/down factors 2/5, the output rate is 100 Hz, with a nominal interval of 0.01 seconds instead of 0.004 seconds. This rate arithmetic does not specify actual filtered values or prove that event timing is preserved. Validate a synthetic signal with known timing and inspect edge samples before using the transformed record. Keep raw data and the full preprocessing record so the transformation can be revisited without repeated filtering.
Method checked 19 November 2026 against Primary method documentation: resample_poly.