The power spectrum of a time series describes how the variance of the data is distributed over the frequency domain, into spectral components which the series may be decomposed. This distribution of the variance may be described either by a measure or by a statistical cumulative distribution function the power contributed by frequencies from 0 up to . Given a band of frequencies , the amount of variance contributed to by frequencies lying within the interval is given by . Then is called the spectral distribution function of . Provided is an absolutely continuous function, then there exists a spectral density function . In this case, the data or signal is said to possess an absolutely continuous spectrum. The spectral density at a frequency gives the rate of variance contributed by frequencies in the immediate neighbourhood of to the variance of per unit frequency.
The nature of the spectrum of a function gives useful information about the nature of , for example, whether it is periodic or not. The study of the power spectrum is a kind of generalisation of Fourier analysis and applies to functions which do not possess Fourier transforms.
An analogous definition applies to a stochastic process . Furthermore, time may be either continuous or discrete.
Intuitively, the spectrum decomposes the content of a signal or of a stochastic process into the different frequencies present in that process, and helps identify periodicities. More specific terms which are used are the power spectrum, spectral density, power spectral density, or energy spectral density.
The variance of has units which are the square of the units of . Therefore, these are also the units of or , and so the units of the spectral density are the square of the units of per unit frequency. In the case of the voltage of an electric signal, is proportional, except that it has the wrong units, to the power of the signal (implicitly assuming a constant resistance), and so even in statistical applications which use different units, the spectral distribution function and density function are often referred to as the power spectral distribution function and the power spectral density function, although the word power is often omitted for brevity in contexts where no misunderstanding will arise. The use of the power spectrum is most important in statistical signal processing and in the branch of statistics consisting of the analysis of time series. It is, however, useful in many other branches of physics and engineering, and may involve other units. Usually the data is a function of time but they may be a function of spatial variables instead.
- 1 Explanation
- 2 Preliminary conventions on notations for time series
- 3 Motivating example
- 4 Definition
- 5 Estimation
- 6 Properties
- 7 Related concepts
- 8 Applications
- 9 See also
- 10 Notes
- 11 References
- 12 External links
Any signal that can be represented as an amplitude that varies with time has a corresponding frequency spectrum. This includes familiar concepts such as visible light (color), musical notes, radio/TV channels, and even the regular rotation of the earth. When these physical phenomena are represented in the form of a frequency spectrum, certain physical descriptions of their internal processes become much simpler. Often, the frequency spectrum clearly shows harmonics, visible as distinct spikes or lines at particular frequencies, that provide insight into the mechanisms that generate the entire signal.
In physics, the signal is usually a wave, such as an electromagnetic wave, random vibration, or an acoustic wave. The power spectral density (PSD) of the signal, when multiplied by the appropriate factor, describes the power contributed to the wave, by a frequency, per unit frequency. Power spectral density is commonly expressed in watts per hertz (W/Hz).
For voltage signals, it is customary to use units of V2 Hz−1 for the PSD and V2 s Hz−1 for the ESD (energy spectral density). Often it is convenient to work with an amplitude spectral density (ASD), which is the square root of the PSD; the ASD of a voltage signal has units of V Hz−1/2. For random vibration analysis, units of g2 Hz−1 are frequently used for the PSD of acceleration. Here g denotes the g-force.
Although it is not necessary to assign physical dimensions to the signal or its argument, in the following discussion the terms used will assume that the signal varies in time.
Preliminary conventions on notations for time series
The phrase time series has been defined as "... a collection of observations made sequentially in time." But it is also used to refer to a stochastic process that functions as the underlying theoretical model for the process that generated the data (and thus includes consideration of all the other possible sequences of data that could have been observed, but were not). Furthermore, the 'time' can be either continuous or discrete. There are, therefore, four different but closely related definitions and formulas for the power spectrum of a time series.
If (discrete time) or (continuous time) is a stochastic process, we will refer to a possible time series of data coming from it as a sample or path or signal of the stochastic process. To avoid confusion, we will reserve the word process for a stochastic process, and use one of the words signal, or sample, to refer to a time series of data. However, the reader should be aware that in the engineering literature especially, this distinction is often conveyed instead by speaking of a deterministic signal (for the sample) but of a random signal for the process. Hence the value at a point in time of the deterministic signal is a number, denoted by , but the value at a point in time of a random signal is a random variable, denoted by the same symbol, . Nevertheless, this article will preserve the distinction and use capital letters for the random variables and processes.
Suppose , from to is a time series (discrete time) with zero mean. Suppose that it is a sum of a finite number of periodic components (all frequencies are positive):
The variance of is, for a zero-mean function as above, given by . If these data were samples taken from an electrical signal, this would be its average power (power is energy per unit time, so it is analogous to variance if energy is analogous to the amplitude squared).
Now, for simplicity, suppose the signal extends infinitely in time, so we pass to the limit as . If the average power is bounded, which is almost always the case in reality, then the following limit exists and is the variance of the data.
Again, for simplicity, we will pass to continuous time, and assume that the signal extends infinitely in time in both directions. Then these two formulas become
The root mean square of is , so the variance of is . Hence, the contribution to the average power of coming from the component with frequency is . All these contributions add up to the average power of .
Then the power as a function of frequency is , and its statistical cumulative distribution function will be
The variance is the covariance of the data with itself. If we now consider the same data but with a lag of , we can take the covariance of with , and define this to be the autocorrelation function of the signal (or data) :
If it exists, it is an even function of . If the average power is bounded, then exists everywhere, is finite, and is bounded by , which is the average power or variance of the data.
It can be shown that can be decomposed into periodic components with the same periods as :
This is in fact the spectral decomposition of over the different frequencies, and is related to the distribution of power of over the frequencies: the amplitude of a frequency component of is its contribution to the average power of the signal.
The power spectrum of this example is not continuous, and therefore does not have a derivative, and therefore this signal does not have a power spectral density function. In general, the power spectrum will usually be the sum of two parts: a line spectrum such as in this example, which is not continuous and does not have a density function, and a residue, which is absolutely continuous and does have a density function.
Energy spectral density
Energy spectral density describes how the energy of a signal or a time series is distributed with frequency. Here, the term energy is used in the generalized sense of signal processing; that is, the energy of a signal is
The energy spectral density is most suitable for transients—that is, pulse-like signals—having a finite total energy. In this case, Parseval's theorem  gives us an alternate expression for the energy of the signal in terms of its Fourier transform,
Here is the frequency in Hz, i.e., cycles per second. Often used is the angular frequency . Since the integral on the right-hand side is the energy of the signal, the integrand can be interpreted as a density function describing the energy per unit frequency contained in the signal at the frequency . In light of this, the energy spectral density of a signal is defined as
As a physical example of how one might measure the energy spectral density of a signal, suppose represents the potential (in volts) of an electrical pulse propagating along a transmission line of impedance , and suppose the line is terminated with a matched resistor (so that all of the pulse energy is delivered to the resistor and none is reflected back). By Ohm's law, the power delivered to the resistor at time is equal to , so the total energy is found by integrating with respect to time over the duration of the pulse. To find the value of the energy spectral density at frequency , one could insert between the transmission line and the resistor a bandpass filter which passes only a narrow range of frequencies (, say) near the frequency of interest and then measure the total energy dissipated across the resistor. The value of the energy spectral density at is then estimated to be . In this example, since the power has units of V2 Ω−1, the energy has units of V2 s Ω−1 = J, and hence the estimate of the energy spectral density has units of J Hz−1, as required. In many situations, it is common to forgo the step of dividing by so that the energy spectral density instead has units of V2 s Hz−1.
This definition generalizes in a straightforward manner to a discrete signal with an infinite number of values such as a signal sampled at discrete times :
where is the discrete Fourier transform of and is the complex conjugate of . The sampling interval is needed to keep the correct physical units and to ensure that we recover the continuous case in the limit ; however, in the mathematical sciences, the interval is often set to 1.
Power spectral density
The above definition of energy spectral density is most suitable for transients, i.e., pulse-like signals, for which the Fourier transforms of the signals exist. For continued signals that describe, for example, stationary physical processes, it makes more sense to define a power spectral density (PSD), which describes how the power of a signal or time series is distributed over the different frequencies, as in the simple example given previously. Here, power can be the actual physical power, or more often, for convenience with abstract signals, can be defined as the squared value of the signal. For example, statisticians study the variance of a set of data, but because of the analogy with electrical signals, it is customary to refer to it as the power spectrum even when it is not, physically speaking, power. The average power P of a signal is the following time average:
The power of a signal may be finite even if the energy is infinite. For example, a 10-volt power supply connected to a 1 kΩ resistor delivers (10 V)2 / (1 kΩ) = 0.1 W of power at any given time; however, if the supply is allowed to operate for an infinite amount of time, it will deliver an infinite amount of energy (0.1 J each second for an infinite number of seconds).
In analyzing the frequency content of the signal , one might like to compute the ordinary Fourier transform ; however, for many signals of interest this Fourier transform does not exist.[N 1] Because of this, it is advantageous to work with a truncated Fourier transform , where the signal is integrated only over a finite interval [0, T]:
Note that .
Using such formal reasoning, one may already guess that for a stationary random process, the power spectral density and the autocorrelation function of this signal should be a Fourier transform pair. Provided that is absolutely integrable, which is not always true, then
The Wiener–Khinchin theorem makes sense of this formula for any wide-sense stationary process under weaker hypotheses: does not need to be absolutely integrable, it only needs to exist. But the integral can no longer be interpreted as usual. The formula also makes sense if interpreted as involving distributions (in the sense of Laurent Schwartz, not in the sense of a statistical Cumulative distribution function) instead of functions. If is continuous, Bochner's theorem can be used to prove that its Fourier transform exists as a positive measure, whose distribution function is F (but not necessarily as a function and not necessarily possessing a probability density).
Many authors use this equality to actually define the power spectral density.
The power of the signal in a given frequency band can be calculated by integrating over positive and negative frequencies,
where is the integrated spectrum whose derivative is .
More generally, similar techniques may be used to estimate a time-varying spectral density.
The definition of the power spectral density generalizes in a straightforward manner to finite time series with , such as a signal sampled at discrete times for a total measurement period .
In a real-world application, one would typically average this single-measurement PSD over several repetitions of the measurement to obtain a more accurate estimate of the theoretical PSD of the physical process underlying the individual measurements. This computed PSD is sometimes called periodogram. One can prove that this periodogram converges to the true PSD when the averaging time interval T goes to infinity (Brown & Hwang) to approach the Power Spectral Density (PSD).
If two signals both possess power spectral densities, then a cross-spectral density can be calculated by using their cross-correlation function.
Properties of the power spectral density
Some properties of the PSD include:
- The spectrum of a real valued process is an even function of frequency: .
- If the process is continuous and purely indeterministic, the autocovariance function can be reconstructed by using the Inverse Fourier transform
- it describes the distribution of the variance over frequency. In particular,
- It is a linear function of the autocovariance function in the sense that if is decomposed into two functions , then
The integrated spectrum or power spectral distribution is defined as
Given two signals and , each of which possess power spectral densities and , it is possible to define a cross-spectral density (CSD) given by
The cross-spectral density (or 'cross power spectrum') is thus the Fourier transform of the cross-correlation function.
where is the cross-correlation of and .
By an extension of the Wiener–Khinchin theorem, the Fourier transform of the cross-spectral density is the cross-covariance function. In light of this, the PSD is seen to be a special case of the CSD for .
For discrete signals xn and yn, the relationship between the cross-spectral density and the cross-covariance is
The goal of spectral density estimation is to estimate the spectral density of a random signal from a sequence of time samples. Depending on what is known about the signal, estimation techniques can involve parametric or non-parametric approaches, and may be based on time-domain or frequency-domain analysis. For example, a common parametric technique involves fitting the observations to an autoregressive model. A common non-parametric technique is the periodogram.
- The spectral density of and the autocorrelation of form a Fourier transform pair (for PSD versus ESD, different definitions of autocorrelation function are used). This result is known as Wiener–Khinchin theorem.
- One of the results of Fourier analysis is Parseval's theorem which states that the area under the energy spectral density curve is equal to the area under the square of the magnitude of the signal, the total energy:
- The above theorem holds true in the discrete cases as well. A similar result holds for power: the area under the power spectral density curve is equal to the total signal power, which is , the autocorrelation function at zero lag. This is also (up to a constant which depends on the normalization factors chosen in the definitions employed) the variance of the data comprising the signal.
- Most spectrum graphs really display only the power spectral density. Sometimes (e.g., Bode plot, chirp) the complete frequency spectrum is graphed in two parts, amplitude versus frequency and phase versus frequency (which contains the rest of the information from the frequency spectrum). The original function cannot be recovered from the amplitude spectral density part alone — information about the phase is lost. See spectral phase and phase noise.
- The spectral centroid of a signal is the midpoint of its spectral density function, i.e. the frequency that divides the distribution into two equal parts.
- The spectral edge frequency of a signal is an extension of the previous concept to any proportion instead of two equal parts.
- The spectral density is a function of frequency, not a function of time. However, the spectral density of small windows of a longer signal may be calculated, and plotted versus time associated with the window. Such a graph is called a spectrogram. This is the basis of a number of spectral analysis techniques such as the short-time Fourier transform and wavelets.
The concept and use of the power spectrum of a signal is fundamental in electrical engineering, especially in electronic communication systems, including radio communications, radars, and related systems, plus passive [remote sensing] technology. Much effort has been expended and millions of dollars spent on developing and producing electronic instruments called "spectrum analyzers" for aiding electrical engineers and technicians in observing and measuring the power spectra of signals. The cost of a spectrum analyzer varies depending on its frequency range, its bandwidth, and its accuracy. The higher the frequency range (S-band, C-band, X-band, Ku-band, K-band, Ka-band, etc.), the more difficult the components are to make, assemble, and test and the more expensive the spectrum analyzer is. Also, the wider the bandwidth that a spectrum analyzer possesses, the more costly that it is, and the capability for more accurate measurements increases costs as well.
The spectrum analyzer measures the magnitude of the short-time Fourier transform (STFT) of an input signal. If the signal being analyzed can be considered a stationary process, the STFT is a good smoothed estimate of its power spectral density. These devices work in low frequencies and with small bandwidths.
- Coherence (signal processing), for use of the cross-spectral density.
- Noise spectral density
- Spectral density estimation
- Spectral efficiency
- Spectral power distribution
- Brightness temperature
- Colors of noise
- Spectral leakage
- Window function
- Some authors (e.g. Risken ) still use the non-normalized Fourier transform in a formal way to formulate a definition of the power spectral density
- or, equivalently, by the measure of that interval, .
- or, equivalently, is absolutely continuous with respect to Lebesgue measure.
- Gérard Maral (2003). VSAT Networks. John Wiley and Sons. ISBN 0-470-86684-5.
- Michael Peter Norton and Denis G. Karczub (2003). Fundamentals of Noise and Vibration Analysis for Engineers. Cambridge University Press. ISBN 0-521-49913-5.
- Michael Cerna and Audrey F. Harvey (2000). "The Fundamentals of FFT-Based Signal Analysis and Measurement" (PDF).
- Alessandro Birolini (2007). Reliability Engineering. Springer. p. 83. ISBN 978-3-540-49388-4.
- C. Chatfield (1989). The Analysis of Time Series—An Introduction (fourth ed.). Chapman and Hall, London. p. 1. ISBN 0-412-31820-2.
- Oppenheim; Verghese. Signals, Systems, and Inference. pp. 32–4.
- Stein, Jonathan Y. (2000). Digital Signal Processing: A Computer Science Perspective. Wiley. p. 115.
- Hannes Risken (1996). The Fokker–Planck Equation: Methods of Solution and Applications (2nd ed.). Springer. p. 30. ISBN 9783540615309.
- Fred Rieke, William Bialek, and David Warland (1999). Spikes: Exploring the Neural Code (Computational Neuroscience). MIT Press. ISBN 978-0262681087.
- Scott Millers and Donald Childers (2012). Probability and random processes. Academic Press. pp. 370–5.
- Dennis Ward Ricker (2003). Echo Signal Processing. Springer. ISBN 1-4020-7395-X.
- Robert Grover Brown & Patrick Y.C. Hwang (1997). Introduction to Random Signals and Applied Kalman Filtering. John Wiley & Sons. ISBN 0-471-12839-2.
- Storch, H. Von; F. W Zwiers (2001). Statistical analysis in climate research. Cambridge University Press. ISBN 0-521-01230-9.
- An Introduction to the Theory of Random Signals and Noise, Wilbur B. Davenport and Willian L. Root, IEEE Press, New York, 1987, ISBN 0-87942-235-1
- William D Penny (2009). "Signal Processing Course, chapter 7".