undersampled

Undersampled is an adjective used to describe a signal, image, or dataset that has been captured or collected with fewer samples than would be required for accurate representation or processing. The meaning depends on context, but it generally implies potential loss of information or distortion due to insufficient sampling.

In signal processing, undersampling occurs when the sampling frequency is less than twice the highest frequency component of the signal, violating the Nyquist-Shannon sampling theorem. This leads to aliasing, where high-frequency content folds into lower frequencies and distorts the reconstructed signal. Remedies include applying an anti-aliasing low-pass filter before sampling, increasing the sampling rate, or using reconstruction methods that incorporate prior information.

In imaging and digital communications, undersampling can produce aliasing artifacts such as jagged edges or moiré patterns. Accelerated imaging modalities, such as some magnetic resonance imaging (MRI) techniques, intentionally undersample data to reduce acquisition time. Reconstruction methods, including compressed sensing or parallel imaging, attempt to recover plausible images from undersampled data.

In data science and statistics, undersampling can refer to reducing the number of samples, sometimes to balance class distributions (undersampling the majority class) or to decrease dataset size for efficiency. While it can mitigate bias toward overrepresented groups, it risks information loss and degraded model performance. In time series and sensor networks, undersampling can obscure temporal trends and frequency content.

Overall, undersampling highlights a deficit in sampling density. Whether addressed through filtering, alternative acquisition, or advanced reconstruction and modeling, the goal is to mitigate information loss while acknowledging the constraints that led to undersampling.