downsampling

Downsampling is the process of reducing the sampling rate or data density of a signal or dataset. In digital signal processing and data analysis, it typically means forming a lower-resolution representation by using a subset of samples or by aggregating neighboring samples. It is the counterpart of upsampling.

In digital signals, straightforward subsampling can produce aliasing if the signal contains frequencies above the new Nyquist limit. Therefore, downsampling is usually preceded by a low-pass anti-aliasing filter that attenuates high-frequency components. The common approach is decimation by an integer factor M: filter, then keep every M-th sample. For non-integer factors, resampling with interpolation and filtering is used.

In image and video processing, downsampling reduces spatial resolution and frame rate, respectively. Techniques include area averaging, subsampling, or interpolation-based methods such as bilinear, bicubic, or lanczos resampling. Anti-aliasing considerations apply to still images to avoid moiré patterns.

Applications include data storage reduction, preprocessing for machine learning, multi-rate signal processing, and transmission bandwidth reduction. The main trade-off is information loss and potential degradation of temporal or spatial fidelity, balanced against reduced computational load and storage.

Example: downsampling an audio signal by a factor of two after applying a low-pass filter with a cutoff below half the original sampling rate yields a new sampling rate that is half the original.