How a signal resembles itself after a delay; reveals periods and correlation times.The autocorrelation function measures how a signal correlates with itself at different time lags. For a time series, it quantifies the similarity between the signal and a delayed copy of itself. A high autocorrelation at a particular lag indicates that the signal repeats with that period, while a slow decay of autocorrelation with lag reveals the characteristic correlation time of the process. It is a fundamental tool for analyzing random processes, detecting periodicities, and estimating the memory of a system.
The concept was introduced by G. Udny Yule in the 1920s in the context of time-series analysis, and it was later formalized by Norbert Wiener and others in the 1940s.
For a discrete time series, the autocorrelation at lag k is computed by taking the product of the signal at time t and at time t+k, averaging over all t, and normalizing by the variance. The result is a function that starts at 1 for zero lag (perfect correlation) and typically decays to zero for large lags if the signal is random. Periodic signals produce peaks at lags equal to their period, while noise produces a sharp drop to zero. The autocorrelation function is also the Fourier transform of the power spectrum, linking time-domain and frequency-domain analyses.
It is used to estimate the correlation time of turbulent flows, to detect periodic signals in noisy data, and to validate the randomness of generated sequences in Monte Carlo simulations.
The autocorrelation function of a pure sine wave is a cosine wave of the same frequency, which is why it can reveal hidden periodicities even when the signal is buried in noise.
| \tau | variable | integral autocorrelation time — autocorrelation time second |
| C(t) | variable | autocorrelation function — autocorrelation function |
| t | variable | time lag — time second |
| \int_{0}^{\infty} | operator | integration from zero to infinity — definite integral |







