Spectral Density: Facts and Examples

Total Page:16

File Type:pdf, Size:1020Kb

Spectral Density: Facts and Examples Introduction to Time Series Analysis. Lecture 15. Spectral Analysis 1. Spectral density: Facts and examples. 2. Spectral distribution function. 3. Wold’s decomposition. 1 Spectral Analysis Idea: decompose a stationary time series {Xt} into a combination of sinusoids, with random (and uncorrelated) coefficients. Just as in Fourier analysis, where we decompose (deterministic) functions into combinations of sinusoids. This is referred to as ‘spectral analysis’ or analysis in the ‘frequency domain,’ in contrast to the time domain approach we have considered so far. The frequency domain approach considers regression on sinusoids; the time domain approach considers regression on past values of the time series. 2 A periodic time series Consider Xt = A sin(2πνt)+ B cos(2πνt) = C sin(2πνt + φ), where A, B are uncorrelated, mean zero, variance σ2 = 1, and C2 = A2 + B2, tan φ = B/A. Then µt = E[Xt] = 0 γ(t, t + h) = cos(2πνh). So {Xt} is stationary. 3 An aside: Some trigonometric identities sin θ tan θ = , cos θ sin2 θ + cos2 θ = 1, sin(a + b) = sin a cos b + cos a sin b, cos(a + b) = cos a cos b − sin a sin b. 4 A periodic time series For Xt = A sin(2πνt)+ B cos(2πνt), with uncorrelated A, B (mean 0, variance σ2), γ(h)= σ2 cos(2πνh). The autocovariance of the sum of two uncorrelated time series is the sum of their autocovariances. Thus, the autocovariance of a sum of random sinusoids is a sum of sinusoids with the corresponding frequencies: k Xt = (Aj sin(2πνj t)+ Bj cos(2πνj t)) , j=1 X k 2 γ(h)= σj cos(2πνj h), j=1 X 2 where Aj,Bj are uncorrelated, mean zero, and Var(Aj )= Var(Bj)= σj . 5 A periodic time series k k 2 Xt = (Aj sin(2πνj t)+ Bj cos(2πνj t)) , γ(h)= σj cos(2πνj h). j=1 j=1 X X Thus, we can represent γ(h) using a Fourier series. The coefficients are the variances of the sinusoidal components. The spectral density is the continuous analog: the Fourier transform of γ. (The analogous spectral representation of a stationary process Xt involves a stochastic integral—a sum of discrete components at a finite number of frequencies is a special case. We won’t consider this representation in this course.) 6 Spectral density If a time series {Xt} has autocovariance γ satisfying ∞ h=−∞ |γ(h)| < ∞, then we define its spectral density as P ∞ f(ν)= γ(h)e−2πiνh h=X−∞ for −∞ <ν< ∞. 7 Spectral density: Some facts ∞ −2πiνh 1. We have h=−∞ γ(h)e < ∞. This is because |eiθ| = | cos θ + i sin θ| = (cos2 θ + sin2 θ)1/2 = 1, P and because of the absolute summability of γ. 2. f is periodic, with period 1. This is true since e−2πiνh is a periodic function of ν with period 1. Thus, we can restrict the domain of f to −1/2 ≤ ν ≤ 1/2. (The text does this.) 8 Spectral density: Some facts 3. f is even (that is, f(ν)= f(−ν)). To see this, write −1 ∞ f(ν)= γ(h)e−2πiνh + γ(0) + γ(h)e−2πiνh, h=X−∞ hX=1 −1 ∞ f(−ν)= γ(h)e−2πiν(−h) + γ(0) + γ(h)e−2πiν(−h), h=X−∞ hX=1 ∞ −1 = γ(−h)e−2πiνh + γ(0) + γ(−h)e−2πiνh hX=1 h=X−∞ = f(ν). 4. f(ν) ≥ 0. 9 Spectral density: Some facts 1/2 5. γ(h)= e2πiνhf(ν) dν. Z−1/2 1/2 1/2 ∞ e2πiνhf(ν) dν = e−2πiν(j−h)γ(j) dν −1/2 −1/2 j=−∞ Z Z X ∞ 1/2 = γ(j) e−2πiν(j−h) dν j=−∞ −1/2 X Z γ(j) = γ(h)+ eπi(j−h) − e−πi(j−h) 2πi(j − h) jX6=h γ(j) sin(π(j − h)) = γ(h)+ = γ(h). π(j − h) jX6=h 10 Example: White noise 2 For white noise {Wt}, we have seen that γ(0) = σw and γ(h) = 0 for h 6= 0. Thus, ∞ f(ν)= γ(h)e−2πiνh h=X−∞ 2 = γ(0) = σw. That is, the spectral density is constant across all frequencies: each frequency in the spectrum contributes equally to the variance. This is the origin of the name white noise: it is like white light, which is a uniform mixture of all frequencies in the visible spectrum. 11 Example: AR(1) 2 |h| 2 For Xt = φ1Xt−1 + Wt, we have seen that γ(h)= σwφ1 /(1 − φ1). Thus, ∞ 2 ∞ −2πiνh σw |h| −2πiνh f(ν)= γ(h)e = 2 φ1 e 1 − φ1 h=X−∞ h=X−∞ σ2 ∞ = w 1+ φh e−2πiνh + e2πiνh 1 − φ2 1 1 h=1 ! X σ2 φ e−2πiν φ e2πiν = w 1+ 1 + 1 1 − φ2 1 − φ e−2πiν 1 − φ e2πiν 1 1 1 2 −2πiν 2πiν σw 1 − φ1e φ1e = 2 −2πiν 2πiν (1 − φ1) (1 − φ1e )(1 − φ1e ) 2 σw = 2 . 1 − 2φ1 cos(2πν)+ φ1 12 Examples 2 White noise: {Wt}, γ(0) = σw and γ(h) = 0 for h 6= 0. 2 f(ν)= γ(0) = σw. 2 |h| 2 AR(1): Xt = φ1Xt−1 + Wt, γ(h)= σwφ1 /(1 − φ1). 2 σw f(ν)= 2 . 1−2φ1 cos(2πν)+φ1 If φ1 > 0 (positive autocorrelation), spectrum is dominated by low frequency components—smooth in the time domain. If φ1 < 0 (negative autocorrelation), spectrum is dominated by high frequency components—rough in the time domain. 13 Example: AR(1) Spectral density of AR(1): X = +0.9 X + W t t−1 t 100 90 80 70 60 ) ν 50 f( 40 30 20 10 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 ν 14 Example: AR(1) Spectral density of AR(1): X = −0.9 X + W t t−1 t 100 90 80 70 60 ) ν 50 f( 40 30 20 10 0 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 ν 15.
Recommended publications
  • Moving Average Filters
    CHAPTER 15 Moving Average Filters The moving average is the most common filter in DSP, mainly because it is the easiest digital filter to understand and use. In spite of its simplicity, the moving average filter is optimal for a common task: reducing random noise while retaining a sharp step response. This makes it the premier filter for time domain encoded signals. However, the moving average is the worst filter for frequency domain encoded signals, with little ability to separate one band of frequencies from another. Relatives of the moving average filter include the Gaussian, Blackman, and multiple- pass moving average. These have slightly better performance in the frequency domain, at the expense of increased computation time. Implementation by Convolution As the name implies, the moving average filter operates by averaging a number of points from the input signal to produce each point in the output signal. In equation form, this is written: EQUATION 15-1 Equation of the moving average filter. In M &1 this equation, x[ ] is the input signal, y[ ] is ' 1 % y[i] j x [i j ] the output signal, and M is the number of M j'0 points used in the moving average. This equation only uses points on one side of the output sample being calculated. Where x[ ] is the input signal, y[ ] is the output signal, and M is the number of points in the average. For example, in a 5 point moving average filter, point 80 in the output signal is given by: x [80] % x [81] % x [82] % x [83] % x [84] y [80] ' 5 277 278 The Scientist and Engineer's Guide to Digital Signal Processing As an alternative, the group of points from the input signal can be chosen symmetrically around the output point: x[78] % x[79] % x[80] % x[81] % x[82] y[80] ' 5 This corresponds to changing the summation in Eq.
    [Show full text]
  • Power Grid Topology Classification Based on Time Domain Analysis
    Power Grid Topology Classification Based on Time Domain Analysis Jia He, Maggie X. Cheng and Mariesa L. Crow, Fellow, IEEE Abstract—Power system monitoring is significantly im- all depend on the correct network topology information. proved with the use of Phasor Measurement Units (PMUs). Topology change, no matter what caused it, must be The availability of PMU data and recent advances in ma- immediately updated. An unattended topology error may chine learning together enable advanced data analysis that lead to cascading power outages and cause large scale is critical for real-time fault detection and diagnosis. This blackout ( [2], [3]). In this paper, we demonstrate that paper focuses on the problem of power line outage. We use a machine learning framework and consider the problem using a machine learning approach and real-time mea- of line outage identification as a classification problem in surement data we can timely and accurately identify the machine learning. The method is data-driven and does outage location. We leverage PMU data and consider not rely on the results of other analysis such as power the task of identifying the location of line outage as flow analysis and solving power system equations. Since it a classification problem. The methods can be applied does not involve using power system parameters to solve in real-time. Although training a classifier can be time- equations, it is applicable even when such parameters are consuming, the prediction of power line status can be unavailable. The proposed method uses only voltage phasor done in real-time. angles obtained from PMUs.
    [Show full text]
  • Logistic-Weighted Regression Improves Decoding of Finger Flexion from Electrocorticographic Signals
    Logistic-weighted Regression Improves Decoding of Finger Flexion from Electrocorticographic Signals Weixuan Chen*-IEEE Student Member, Xilin Liu-IEEE Student Member, and Brian Litt-IEEE Senior Member Abstract— One of the most interesting applications of brain linear Wiener filter [7–10]. To take into account the computer interfaces (BCIs) is movement prediction. With the physiological, physical, and mechanical constraints that development of invasive recording techniques and decoding affect the flexion of limbs, some studies applied switching algorithms in the past ten years, many single neuron-based and models [11] or Bayesian models [12,13] to the results of electrocorticography (ECoG)-based studies have been able to linear regressions above. Other studies have explored the decode trajectories of limb movements. As the output variables utility of non-linear methods, including neural networks [14– are continuous in these studies, a regression model is commonly 17], multilinear perceptrons [18], and support vector used. However, the decoding of limb movements is not a pure machines [18], but they tend to have difficulty with high regression problem, because the trajectories can be apparently dimensional features and limited training data [13]. classified into a motion state and a resting state, which result in a binary property overlooked by previous studies. In this Nevertheless, the studies of limb movement translation paper, we propose an algorithm called logistic-weighted are in fact not pure regression problems, because the limbs regression to make use of the property, and apply the algorithm are not always under the motion state. Whether it is during an to a BCI system decoding flexion of human fingers from ECoG experiment or in the daily life, the resting state of the limbs is signals.
    [Show full text]
  • Kalman and Particle Filtering
    Abstract: The Kalman and Particle filters are algorithms that recursively update an estimate of the state and find the innovations driving a stochastic process given a sequence of observations. The Kalman filter accomplishes this goal by linear projections, while the Particle filter does so by a sequential Monte Carlo method. With the state estimates, we can forecast and smooth the stochastic process. With the innovations, we can estimate the parameters of the model. The article discusses how to set a dynamic model in a state-space form, derives the Kalman and Particle filters, and explains how to use them for estimation. Kalman and Particle Filtering The Kalman and Particle filters are algorithms that recursively update an estimate of the state and find the innovations driving a stochastic process given a sequence of observations. The Kalman filter accomplishes this goal by linear projections, while the Particle filter does so by a sequential Monte Carlo method. Since both filters start with a state-space representation of the stochastic processes of interest, section 1 presents the state-space form of a dynamic model. Then, section 2 intro- duces the Kalman filter and section 3 develops the Particle filter. For extended expositions of this material, see Doucet, de Freitas, and Gordon (2001), Durbin and Koopman (2001), and Ljungqvist and Sargent (2004). 1. The state-space representation of a dynamic model A large class of dynamic models can be represented by a state-space form: Xt+1 = ϕ (Xt,Wt+1; γ) (1) Yt = g (Xt,Vt; γ) . (2) This representation handles a stochastic process by finding three objects: a vector that l describes the position of the system (a state, Xt X R ) and two functions, one mapping ∈ ⊂ 1 the state today into the state tomorrow (the transition equation, (1)) and one mapping the state into observables, Yt (the measurement equation, (2)).
    [Show full text]
  • Lecture 19: Wavelet Compression of Time Series and Images
    Lecture 19: Wavelet compression of time series and images c Christopher S. Bretherton Winter 2014 Ref: Matlab Wavelet Toolbox help. 19.1 Wavelet compression of a time series The last section of wavelet leleccum notoolbox.m demonstrates the use of wavelet compression on a time series. The idea is to keep the wavelet coefficients of largest amplitude and zero out the small ones. 19.2 Wavelet analysis/compression of an image Wavelet analysis is easily extended to two-dimensional images or datasets (data matrices), by first doing a wavelet transform of each column of the matrix, then transforming each row of the result (see wavelet image). The wavelet coeffi- cient matrix has the highest level (largest-scale) averages in the first rows/columns, then successively smaller detail scales further down the rows/columns. The ex- ample also shows fine results with 50-fold data compression. 19.3 Continuous Wavelet Transform (CWT) Given a continuous signal u(t) and an analyzing wavelet (x), the CWT has the form Z 1 s − t W (λ, t) = λ−1=2 ( )u(s)ds (19.3.1) −∞ λ Here λ, the scale, is a continuous variable. We insist that have mean zero and that its square integrates to 1. The continuous Haar wavelet is defined: 8 < 1 0 < t < 1=2 (t) = −1 1=2 < t < 1 (19.3.2) : 0 otherwise W (λ, t) is proportional to the difference of running means of u over successive intervals of length λ/2. 1 Amath 482/582 Lecture 19 Bretherton - Winter 2014 2 In practice, for a discrete time series, the integral is evaluated as a Riemann sum using the Matlab wavelet toolbox function cwt.
    [Show full text]
  • Measurement Techniques of Ultra-Wideband Transmissions
    Rec. ITU-R SM.1754-0 1 RECOMMENDATION ITU-R SM.1754-0* Measurement techniques of ultra-wideband transmissions (2006) Scope Taking into account that there are two general measurement approaches (time domain and frequency domain) this Recommendation gives the appropriate techniques to be applied when measuring UWB transmissions. Keywords Ultra-wideband( UWB), international transmissions, short-duration pulse The ITU Radiocommunication Assembly, considering a) that intentional transmissions from devices using ultra-wideband (UWB) technology may extend over a very large frequency range; b) that devices using UWB technology are being developed with transmissions that span numerous radiocommunication service allocations; c) that UWB technology may be integrated into many wireless applications such as short- range indoor and outdoor communications, radar imaging, medical imaging, asset tracking, surveillance, vehicular radar and intelligent transportation; d) that a UWB transmission may be a sequence of short-duration pulses; e) that UWB transmissions may appear as noise-like, which may add to the difficulty of their measurement; f) that the measurements of UWB transmissions are different from those of conventional radiocommunication systems; g) that proper measurements and assessments of power spectral density are key issues to be addressed for any radiation, noting a) that terms and definitions for UWB technology and devices are given in Recommendation ITU-R SM.1755; b) that there are two general measurement approaches, time domain and frequency domain, with each having a particular set of advantages and disadvantages, recommends 1 that techniques described in Annex 1 to this Recommendation should be considered when measuring UWB transmissions. * Radiocommunication Study Group 1 made editorial amendments to this Recommendation in the years 2018 and 2019 in accordance with Resolution ITU-R 1.
    [Show full text]
  • Alternative Tests for Time Series Dependence Based on Autocorrelation Coefficients
    Alternative Tests for Time Series Dependence Based on Autocorrelation Coefficients Richard M. Levich and Rosario C. Rizzo * Current Draft: December 1998 Abstract: When autocorrelation is small, existing statistical techniques may not be powerful enough to reject the hypothesis that a series is free of autocorrelation. We propose two new and simple statistical tests (RHO and PHI) based on the unweighted sum of autocorrelation and partial autocorrelation coefficients. We analyze a set of simulated data to show the higher power of RHO and PHI in comparison to conventional tests for autocorrelation, especially in the presence of small but persistent autocorrelation. We show an application of our tests to data on currency futures to demonstrate their practical use. Finally, we indicate how our methodology could be used for a new class of time series models (the Generalized Autoregressive, or GAR models) that take into account the presence of small but persistent autocorrelation. _______________________________________________________________ An earlier version of this paper was presented at the Symposium on Global Integration and Competition, sponsored by the Center for Japan-U.S. Business and Economic Studies, Stern School of Business, New York University, March 27-28, 1997. We thank Paul Samuelson and the other participants at that conference for useful comments. * Stern School of Business, New York University and Research Department, Bank of Italy, respectively. 1 1. Introduction Economic time series are often characterized by positive
    [Show full text]
  • Random Signals
    Chapter 8 RANDOM SIGNALS Signals can be divided into two main categories - deterministic and random. The term random signal is used primarily to denote signals, which have a random in its nature source. As an example we can mention the thermal noise, which is created by the random movement of electrons in an electric conductor. Apart from this, the term random signal is used also for signals falling into other categories, such as periodic signals, which have one or several parameters that have appropriate random behavior. An example is a periodic sinusoidal signal with a random phase or amplitude. Signals can be treated either as deterministic or random, depending on the application. Speech, for example, can be considered as a deterministic signal, if one specific speech waveform is considered. It can also be viewed as a random process if one considers the ensemble of all possible speech waveforms in order to design a system that will optimally process speech signals, in general. The behavior of stochastic signals can be described only in the mean. The description of such signals is as a rule based on terms and concepts borrowed from probability theory. Signals are, however, a function of time and such description becomes quickly difficult to manage and impractical. Only a fraction of the signals, known as ergodic, can be handled in a relatively simple way. Among those signals that are excluded are the class of the non-stationary signals, which otherwise play an essential part in practice. Working in frequency domain is a powerful technique in signal processing. While the spectrum is directly related to the deterministic signals, the spectrum of a ran- dom signal is defined through its correlation function.
    [Show full text]
  • Time Series: Co-Integration
    Time Series: Co-integration Series: Economic Forecasting; Time Series: General; Watson M W 1994 Vector autoregressions and co-integration. Time Series: Nonstationary Distributions and Unit In: Engle R F, McFadden D L (eds.) Handbook of Econo- Roots; Time Series: Seasonal Adjustment metrics Vol. IV. Elsevier, The Netherlands N. H. Chan Bibliography Banerjee A, Dolado J J, Galbraith J W, Hendry D F 1993 Co- Integration, Error Correction, and the Econometric Analysis of Non-stationary Data. Oxford University Press, Oxford, UK Time Series: Cycles Box G E P, Tiao G C 1977 A canonical analysis of multiple time series. Biometrika 64: 355–65 Time series data in economics and other fields of social Chan N H, Tsay R S 1996 On the use of canonical correlation science often exhibit cyclical behavior. For example, analysis in testing common trends. In: Lee J C, Johnson W O, aggregate retail sales are high in November and Zellner A (eds.) Modelling and Prediction: Honoring December and follow a seasonal cycle; voter regis- S. Geisser. Springer-Verlag, New York, pp. 364–77 trations are high before each presidential election and Chan N H, Wei C Z 1988 Limiting distributions of least squares follow an election cycle; and aggregate macro- estimates of unstable autoregressive processes. Annals of Statistics 16: 367–401 economic activity falls into recession every six to eight Engle R F, Granger C W J 1987 Cointegration and error years and follows a business cycle. In spite of this correction: Representation, estimation, and testing. Econo- cyclicality, these series are not perfectly predictable, metrica 55: 251–76 and the cycles are not strictly periodic.
    [Show full text]
  • STATISTICAL FOURIER ANALYSIS: CLARIFICATIONS and INTERPRETATIONS by DSG Pollock
    STATISTICAL FOURIER ANALYSIS: CLARIFICATIONS AND INTERPRETATIONS by D.S.G. Pollock (University of Leicester) Email: stephen [email protected] This paper expounds some of the results of Fourier theory that are es- sential to the statistical analysis of time series. It employs the algebra of circulant matrices to expose the structure of the discrete Fourier transform and to elucidate the filtering operations that may be applied to finite data sequences. An ideal filter with a gain of unity throughout the pass band and a gain of zero throughout the stop band is commonly regarded as incapable of being realised in finite samples. It is shown here that, to the contrary, such a filter can be realised both in the time domain and in the frequency domain. The algebra of circulant matrices is also helpful in revealing the nature of statistical processes that are band limited in the frequency domain. In order to apply the conventional techniques of autoregressive moving-average modelling, the data generated by such processes must be subjected to anti- aliasing filtering and sub sampling. These techniques are also described. It is argued that band-limited processes are more prevalent in statis- tical and econometric time series than is commonly recognised. 1 D.S.G. POLLOCK: Statistical Fourier Analysis 1. Introduction Statistical Fourier analysis is an important part of modern time-series analysis, yet it frequently poses an impediment that prevents a full understanding of temporal stochastic processes and of the manipulations to which their data are amenable. This paper provides a survey of the theory that is not overburdened by inessential complications, and it addresses some enduring misapprehensions.
    [Show full text]
  • Use of the Kurtosis Statistic in the Frequency Domain As an Aid In
    lEEE JOURNALlEEE OF OCEANICENGINEERING, VOL. OE-9, NO. 2, APRIL 1984 85 Use of the Kurtosis Statistic in the FrequencyDomain as an Aid in Detecting Random Signals Absmact-Power spectral density estimation is often employed as a couldbe utilized in signal processing. The objective ofthis method for signal ,detection. For signals which occur randomly, a paper is to compare the PSD technique for signal processing frequency domain kurtosis estimate supplements the power spectral witha new methodwhich computes the frequency domain density estimate and, in some cases, can be.employed to detect their presence. This has been verified from experiments vith real data of kurtosis (FDK) [2] forthe real and imaginary parts of the randomly occurring signals. In order to better understand the detec- complex frequency components. Kurtosis is defined as a ratio tion of randomlyoccurring signals, sinusoidal and narrow-band of a fourth-order central moment to the square of a second- Gaussian signals are considered, which when modeled to represent a order central moment. fading or multipath environment, are received as nowGaussian in Using theNeyman-Pearson theory in thetime domain, terms of a frequency domain kurtosis estimate. Several fading and multipath propagation probability density distributions of practical Ferguson [3] , has shown that kurtosis is a locally optimum interestare considered, including Rayleigh and log-normal. The detectionstatistic under certain conditions. The reader is model is generalized to handle transient and frequency modulated referred to Ferguson'swork for the details; however, it can signals by taking into account the probability of the signal being in a be simply said thatit is concernedwith detecting outliers specific frequency range over the total data interval.
    [Show full text]
  • 2D Fourier, Scale, and Cross-Correlation
    2D Fourier, Scale, and Cross-correlation CS 510 Lecture #12 February 26th, 2014 Where are we? • We can detect objects, but they can only differ in translation and 2D rotation • Then we introduced Fourier analysis. • Why? – Because Fourier analysis can help us with scale – Because Fourier analysis can make correlation faster Review: Discrete Fourier Transform • Problem: an image is not an analogue signal that we can integrate. • Therefore for 0 ≤ x < N and 0 ≤ u <N/2: N −1 * # 2πux & # 2πux &- F(u) = ∑ f (x),cos % ( − isin% (/ x=0 + $ N ' $ N '. And the discrete inverse transform is: € 1 N −1 ) # 2πux & # 2πux &, f (x) = ∑F(u)+cos % ( + isin% (. N x=0 * $ N ' $ N '- CS 510, Image Computaon, ©Ross 3/2/14 3 Beveridge & Bruce Draper € 2D Fourier Transform • So far, we have looked only at 1D signals • For 2D signals, the continuous generalization is: ∞ ∞ F(u,v) ≡ ∫ ∫ f (x, y)[cos(2π(ux + vy)) − isin(2π(ux + vy))] −∞ −∞ • Note that frequencies are now two- dimensional € – u= freq in x, v = freq in y • Every frequency (u,v) has a real and an imaginary component. CS 510, Image Computaon, ©Ross 3/2/14 4 Beveridge & Bruce Draper 2D sine waves • This looks like you’d expect in 2D Ø Note that the frequencies don’t have to be equal in the two dimensions. hp://images.google.com/imgres?imgurl=hFp://developer.nvidia.com/dev_content/cg/cg_examples/images/ sine_wave_perturbaon_ogl.jpg&imgrefurl=hFp://developer.nvidia.com/object/ cg_effects_explained.html&usg=__0FimoxuhWMm59cbwhch0TLwGpQM=&h=350&w=350&sz=13&hl=en&start=8&sig2=dBEtH0hp5I1BExgkXAe_kg&tbnid=fc yrIaap0P3M:&tbnh=120&tbnw=120&ei=llCYSbLNL4miMoOwoP8L&prev=/images%3Fq%3D2D%2Bsine%2Bwave%26gbv%3D2%26hl%3Den%26sa%3DG CS 510, Image Computaon, ©Ross 3/2/14 5 Beveridge & Bruce Draper 2D Discrete Fourier Transform N /2 N /2 * # 2π & # 2π &- F(u,v) = ∑ ∑ f (x, y),cos % (ux + vy)( − isin% (ux + vy)(/ x=−N /2 y=−N /2 + $ N ' $ N '.
    [Show full text]