A periodogram is a useful graphical tool in the analysis of time series. Consider a time series consisting of n observations x1, x2,…, xn, corresponding to the times t=1, 2,…, n. The mean of the n observations is denoted by x¯. The series may be thought of as resulting from a mixture of cyclic variations of different periodicities so that, for n odd,where In the case where n is even the same expressions hold true, but now and there is one additional term in the equation for xt namely, .
The dominating contributions are indicated by large values for aj or bj. The periodogram is a plot of I(j) against j, for , whereandSome authors define the periodogram using multiples (e.g. 2/π) of some of these formulae.
The population counterpart of the time series is the stationary process X(t), and the counterpart of the periodogram is the spectral density function (or power spectrum or spectrum).
The periodogram is not a consistent estimator of the power spectrum and it is usual to give more weight to the first m terms of the summation in the periodogram, usingwhere m<n and the weight λk is the lag window. A typical value for m is about . Popular lag windows include:
Bartlett window | | k=1, 2,…, m, |
Daniell window | | k=1, 2,…, m, |
Hamming window | | k=1, 2,…, m, |
Parzen window | | 1≤k≤½ m, |
| | ½m≤k≤m, |
Tukey window | | k=1, 2,…, m, |
Direct analysis of the observed values {xj} is analysis in the time domain whereas analysis in terms of periodic functions is analysis in the frequency domain.