The Spectral Correlation Function

Spectral correlation in CSP means that distinct narrowband spectral components of a signal are correlated-they contain either identical information or some degree of redundant information.

Spectral correlation is perhaps the most widely used characterization of the cyclostationarity property. The main reason is that the computational efficiency of the FFT can be harnessed to characterize the cyclostationarity of a given signal or data set in an efficient manner. And not just efficient, but with a reasonable total computational cost, so that one doesn’t have to wait too long for the result.

Just as the normal power spectrum is actually the power spectral density, or more accurately, the spectral density of time-averaged power (or simply the variance when the mean is zero), the spectral correlation function is the spectral density of time-averaged correlation (covariance). What does this mean? Consider the following schematic showing two narrowband spectral components of an arbitrary signal:

scf_schematic
Figure 1. Illustration of the concept of spectral correlation. The time series represented by the narrowband spectral components centered at f-A/2 and f+A/2 are downconverted to zero frequency and their correlation is measured. When A=0, the result is the power spectral density function, otherwise it is referred to as the spectral correlation function. It is non-zero only for a countable set of numbers \{A\}, which are equal to the frequencies of sine waves that can be generated by quadratically transforming the data.

Let’s define narrowband spectral component to mean the output of a bandpass filter applied to a signal, where the bandwidth of the filter is much smaller than the bandwidth of the signal.

The sequence of shaded rectangles on the left are meant to imply a time series corresponding to the output of a bandpass filter centered at f-A/2 with bandwidth B. Similarly, the sequence of shaded rectangles on the right imply a time series corresponding to the output of a bandpass filter centered at f+A/2 with bandwidth B.

Let’s call the first time series Y_1(t, f-A/2) and the second one Y_2(t, f+A/2). Since these time series, or signals, are bandpass in general, if we attempt to measure their correlation we will get a small value if A \neq 0. However, if we downconvert each of them to baseband (zero frequency), we will obtain lowpass signals, and there is the possibility that these new signals are correlated to some degree.

As a limiting case, suppose each of the Y_j(t, \cdot) signals were noiseless sine waves. By the construction of the figure, their frequencies must be different if A \neq 0, and so their correlation will be zero. However, if each sine wave is perfectly downconverted to baseband, the resulting signals are simply complex-valued constants, and the correlation between the two constant time series is high.

In general, the narrowband time series Y_1(t, \cdot) and Y_2(t, \cdot) are not simple sine waves, but are sample paths of complicated random processes. Nevertheless, the correlation between separated spectral components–spectral correlation–is still a highly useful characterization of the signal for a large class of interesting signals.

The spectral components (the individual downconverted narrowband spectral components of the signal) are most often obtained through the use of the Fourier transform. As the transform of length T slides along the signal x(t), it produces a number of downconverted spectral components with approximate bandwidth 1/T.  The two involved time series of interest are then renamed as X_T(t, f-A/2) and X_T(t, f+A/2). A measure of the spectral correlation is given by the limiting average of the cyclic periodogram, which is defined by

\displaystyle I_T^A (t, f) = \frac{1}{T} X_T(t, f-A/2) X_T^*(t, f+A/2), \hfill (1)

as the amount of processed data increases without bound, and then and only then the spectral resolution (B = 1/T) is allowed to decrease to zero,

\displaystyle S_x^A (f) = \lim_{T\rightarrow\infty} \lim_{U\rightarrow\infty} \displaystyle\frac{1}{U} \int_{-U/2}^{U/2} I_T^A(t, f) \, dt. \hfill (2)

The limit spectral correlation function we just wrote down is a time-smoothed (time-averaged) cyclic periodogram. But the limit function can also be obtained by frequency smoothing the cyclic periodogram,

\displaystyle S_x^A(f) = \lim_{\Delta\rightarrow 0} \lim_{T\rightarrow\infty} g_\Delta(f) \otimes I_T^A(t, f), \hfill (3)

where g_\Delta(f) is a unit-area pulse-like smoothing kernel (such as a rectangle). In (3), the symbol \otimes denotes convolution.

The Significance of the Frequency A

The spectral correlation function (SCF) S_x^A(f) is typically zero for almost all real numbers A. That is, for almost all signal types, most pairs of narrowband frequency components are uncorrelated. Those few A for which the SCF is not identically zero are called cycle frequencies (CFs). The set of SCF CFs is exactly the same as the set of cycle frequencies for the cyclic autocorrelation function (CAF)! That is, the separation between correlated narrowband signal components of x(t) is the same as a frequency of a sine wave that can be generated by a quadratic nonlinearity (for example, a squarer or a delay-and-multiply device) operating on the original time-series data x(t).

The Cyclic Wiener Relationship

It can be shown that the Fourier transform of the CAF is equal to the SCF (The Literature [R1], My Papers [5,6]):

\displaystyle S_x^\alpha(f) = \int_{-\infty}^\infty R_x^\alpha(\tau) e^{-i 2 \pi f \tau}\, d\tau, \hfill (4)

which is called the cyclic Wiener relationship. The Wiener relationship (sometimes called the Wiener-Khintchine theorem)  is a name given to the more familiar Fourier transform relation between the conventional power spectral density and the autocorrelation

\displaystyle S_x^0 (f) = \int_{-\infty}^\infty R_x^0(\tau) e^{-i 2 \pi f \tau} \, d\tau, \hfill (5)

where S_x^0(f) is the conventional power spectrum and R_x^0(\tau) is the conventional autocorrelation function.

It follows that the cyclic autocorrelation function is the inverse Fourier transform of the spectral correlation function,

\displaystyle R_x^\alpha(\tau) = \int_{-\infty}^\infty S_x^\alpha(f) e^{i 2 \pi f \tau} \, df \hfill (4a)

and the conventional autocorrelation is the inverse transform of the power spectral density

\displaystyle R_x^0(\tau) = \int_{-\infty}^\infty S_x^0(f) e^{i 2 \pi f \tau} \, df .\hfill (5a)

The mean-square (power) of the time series x(t) (or variance if the time series has a mean value of zero) is simply the autocorrelation evaluated at \tau = 0. This implies that the power of the time series is the integral of the power spectral density

\displaystyle P_x = R_x^0(0) = \int_{-\infty}^\infty S_x^0(f) \, df. \hfill (5b)

Conjugate Spectral Correlation

This post has defined the non-conjugate spectral correlation function, which is the correlation between X_T(t, f-\alpha/2) and X_T(t, f+\alpha/2). (The correlation between random variables X and Y is defined as E[XY^*]; that is, the standard correlation includes a conjugation of one of the factors.)

The conjugate SCF is defined as the Fourier transform of the conjugate cyclic autocorrelation function,

\displaystyle S_{x^*}^\alpha (f) = \int_{-\infty}^\infty R_{x^*}^\alpha (\tau) e^{-i 2 \pi f \tau}\, d\tau . \hfill (6)

From this definition, it can be shown that the conjugate SCF is the density of time-averaged correlation between X_T(t, f+\alpha/2) and X_T^*(t, \alpha/2-f),

\displaystyle S_{x^*}^\alpha (f) = \lim_{T\rightarrow\infty} \lim_{U\rightarrow\infty} \displaystyle\frac{1}{U} \int_{-U/2}^{U/2} J_T^\alpha(t, f) \, dt, \hfill (7)

where J_T^\alpha(t, f) is the conjugate cyclic periodogram

J_T^\alpha(t,f) = \displaystyle \frac{1}{T} X_T(t, f+\alpha/2) X_T(t, \alpha/2-f). \hfill (8)

The detailed explanation for why we need two kinds of spectral correlation functions (and, correspondingly, two kinds of cyclic autocorrelation functions) can be found in the post on conjugation configurations.

Illustrations

The SCF in Figure 2 is estimated (more on that estimation in another post) from a simulated BPSK signal with a bit rate of 333.3 kHz and carrier frequency of 100 kHz. A small amount of noise is added to the signal prior to SCF estimation. The non-conjugate SCF shows the power spectrum for \alpha = 0 and the bit-rate SCF for \alpha = 333.3 kHz. The conjugate SCF plot shows the prominent feature for the doubled-carrier cycle frequency \alpha = 200.0 kHz, and features offset from the doubled-carrier feature by \pm 333.3 kHz. More on the spectral correlation of the BPSK signal can be found here and here. The spectral correlation surfaces for a variety of communication signals can be found in this gallery post.

A closely related function called the spectral coherence function is useful for blindly detecting cycle frequencies exhibited by arbitrary data sets.

ww_1
Figure 2. Non-conjugate and conjugate spectral correlation functions for a BPSK signal with bit rate of 333 kHz and carrier frequency 100 kHz.

Now consider a similar signal: QPSK with rectangular pulses. Let’s switch to normalized frequencies (physical frequencies are divided by the sampling rate) here for convenience. The signal has a symbol rate of 1/10, a carrier frequency of 0.05, unit power, and a small amount of additive white Gaussian noise. A power spectrum estimate is shown in Figure 3.

spec_corr_example_psds
Figure 3. The power spectrum of a rectangular-pulse QPSK signal and the spectra of four selected narrowband components of the signal. The symbol rate is (normalized) 0.1 and the carrier frequency is 0.05.

Consider also four distinct narrowband (NB) components of this QPSK signal as shown in the figure. The center frequencies are 0.0, 0.1, 0.08, and 0.18. We know that this signal has non-conjugate cycle frequencies that are equal to harmonics of the symbol rate, or k/10 for k = 0, \pm 1, \pm 2, \ldots. This means that the NB components with separations k/10 are correlated. So if we extract such NB components “by hand” and calculate their correlation coefficients as a function of relative delay, we should see large results for the pairs (0.0, 0.1) and (0.08, 0.18), and small results for all other pairs drawn from the four frequencies. The correlation coefficient for two random variables X and Y is

\displaystyle \rho_{XY} = E \left[ \frac{(X-\bar{X})(Y-\bar{Y})^*}{\sigma_X \sigma_Y} \right] \hfill (9)

where E[\cdot] is the expectation operator, \bar{X} is the mean value of X, \bar{Y} is the mean value of Y, and \sigma_X and \sigma_Y are the standard deviations of X and Y. Typically E[X] = \bar{X} = E[Y] = \bar{Y} = 0. In the measurement below, we use temporal averaging in place of the expectation.

So let’s do that. We apply a simple Fourier-based ideal filter (ideal meaning rectangular passband) with center frequency f_0 \in \{0.0, 0.1, 0.08, 0.18\}, frequency shift to complex baseband (zero center frequency), and decimate. The results are our narrowband signal components. Are they correlated when they should be and uncorrelated when they should be?

Here are the correlation-coefficient results:

spec_corr_example_cc
Figure 4. Correlation coefficients for various pairs of the narrowband components of the QPSK signal shown in Figure 3. Large correlation coefficients are obtained only when the difference between the narrowband frequency components’ center frequencies equals a harmonic of the symbol rate k/10.

Here the signals y_j(t) arise from the bandpass-filter center frequencies \{0.0, 0.1, 0.08, 0.18\}. So the spectral correlation concept is verified using this example: the only large correlation coefficients are those corresponding to a spectral-component center-frequency difference that is equal to a known cycle frequency for the QPSK signal (k/10). One can also simply plot the decimated shifted narrowband components and assess correlation visually:

spec_corr_example_time
Figure 5. Time series plots of the narrowband components of the QPSK signal shown in Figure 3 (after frequency shifting to zero frequency and decimating). Which pairs appear to be correlated?

Some readers question the need for the downconversion step before measuring correlation, so I’ve also performed our QPSK-based four-spectral-components measurement without doing the downconversion. That is, I just correlated the outputs of the four narrowband filters. The correlation coefficients and the outputs of the filters themselves (analogs of Figures 4 and 5) are shown in Figures 6 and 7. See also the discussion in the Comments section below.

Figure 6. Correlation coefficients for the four narrowband components in Figure 3, but without downconverting them to zero prior to performing the correlation-coefficient calculation.
Figure 7. Plots of the four narrowband components of the QPSK signal shown in Figure 3. No downconverting of these signals was performed, unlike in Figure 5.

Estimators

I’ve written several posts on estimators for the spectral correlation function; they are listed below. I think of them as falling into two categories: exhaustive and focused. For exhaustive spectral correlation estimators, the goal is to estimate the function over its entire (non-redundant) domain of definition as efficiently as possible. For focused estimators, the goal is to estimate the spectral correlation function for one or a small number of cycle frequencies with high accuracy and selectable frequency resolution.

Exhaustive Estimators

Strip Spectral Correlation Analyzer (SSCA)

FFT Accumulation Method (FAM)

Focused Estimators

The Frequency-Smoothing Method (FSM)

The Time-Smoothing Method (TSM)

Support the CSP Blog and Keep it Ad-Free

Please consider donating to the CSP Blog to keep it ad-free and to support the addition of major new features. The small box below is used to specify the number of $5 donations.

$5.00

Author: Chad Spooner

I'm a signal processing researcher specializing in cyclostationary signal processing (CSP) for communication signals. I hope to use this blog to help others with their cyclo-projects and to learn more about how CSP is being used and extended worldwide.

Leave a Reply to Claudio DiasCancel reply

Discover more from Cyclostationary Signal Processing

Subscribe now to keep reading and get access to the full archive.

Continue reading