Correlation Theorem for Wigner-Ville Distribution Mawardi Bahri Department of Mathematics, Hasanuddin University Jl. Perintis KM 10 Tamalanrea, Makassar 90245, Indonesia mawardibahri@gmail.com Abstract In this paper we provide an alternative proof of the Moyal’s formula for the Wigner-Ville distribution (WVD). Based on the definition of convolution of the WVD we establish the correlation theorem of the WVD. Keywords: Wigner-Ville distribution, convolution, correlation 1 Introduction The Wigner-Ville distribution (WVD) is first introduced by Eugene Wigner in his calculation of the quantum corrections of the classical statistical mechanics. It was independently derived again by J. Ville in 1948 as a quadratic representation of the local time-frequency energy of a signal. This distribution plays as a major role in the time-frequency signal analysis. A number of useful properties of this distribution are already known, including nonlinearity, shift, modulation, differentiation, and the energy density spectrum [2,3,4]. The most fundamental and important properties of the WVD are convolution and correlation theorems. They are mathematical operations with several applications in pure and applied mathematics such as numerical analysis, numerical linear algebra and the design and implementation of finite impulse response filters in signal processing. In this paper, we consider alternative proof of the Moyal’s formula for the WVD. In the beginning, by introducing the definition of the convolution WVD, we establish the correlation theorem for the WVD, which has not been established in [1, 2, 3, 4]. 2 Preliminary In this section we briefly review the definition of the Fourier transform (FT) and basic properties. For a complex-valued function f defined on the real line β, the complex conjugate of f is the functionπ Μ given by π (Μ π₯) = Μ Μ Μ Μ Μ Μ π(π₯) for every real number x. Definition 2.1 (Fourier Transform) Let f be in πΏ2 (β). Then Fourier transform (FT) of complex function f is defined by ∞ β± {π}(π) = πΜ(π) = ∫−∞ π(π₯)π −πππ₯ ππ₯. (1) In inner product notation, the FT defined in (1) can be also be expressed as πΜ(π) = (π, π πππ₯ ). The norm of the above inner product is given by ∞ (π, π) = ∫−∞|π(π₯)|2 ππ₯ = βπβ2. Theorem 2.2 (Inversion Formula) Suppose that f ∈ πΏ2 (β) and β±{π} ∈ πΏ1 (β), the inverse FT of the function f is given by 1 ∞ β± −π€ [β±{π}](π₯) = π(π₯) = 2π ∫−∞ πΜ(π) π πππ₯ ππ. (2) We first observe that for a special case the function πΜ(π) = 1we obtain the Dirac’s delta function. That is 1 ∞ πΏ(π₯) = 2π ∫−∞ π πππ₯ ππ. (3) In the following we introduce the convolution and correlation definitions and relationship among convolution, correlation and the FT. Definition 2. 3 (Convolution) For two complex functions π, π ∈ πΏ2 (β), the convolution of π and π, denoted π β π, is defined by ∞ (π β π)(π₯) = ∫−∞ π(π‘)π(π₯ − π‘)ππ‘. (4) 2 (β). Theorem 2.4 Suppose that π, π ∈ πΏ Then the FT of convolution of the two complex functions is given by β±{π β π}(π) = β±{π}(π)β±{π}(π). (5) Next, let us examine how the FT behaves under correlations. First, we give the definition of the correlation of two complex functions. Definition 2. 6 (Correlation) For π, π ∈ πΏ2 (β), the correlation of two complex functions is defined by ∞ (π β π)(π₯) = ∫−∞ π(Μ π‘)π(π‘ + π₯) ππ‘. (6) Theorem 2.4 Suppose that π, π ∈ πΏ2 (β). Then the FT of correlation of two complex functions is given by Μ β±{π β π}(π) = β±{π}(−π)β±{π}(π). (7) Or, equivalently, ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ 1 (π β π)(π₯) = β±{π}(π)π πππ₯ ππ. (8) ∫ β±{π}(π) 2π −∞ 3 Wigner-Ville Distribution Here we describe the basic facts of the WVD, which will be needed to derive the main result of the paper. First we give the definition of the cross WVD and auto WVD. Definition 3.1 (Cross Wigner-Ville Distribution) If two signals f, g ∈ πΏ2 (β), the cross Wigner-Ville distribution of f and g is defined by ∞ π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ ππ,π (π‘, π) = ∫−∞ π (π‘ + 2) π (π‘ − 2) π −πππ₯ ππ₯. (9) Roughly speaking, the cross Wigner-Ville distribution ππ,π (π‘, π) can be considered as a function which indicates the distribution of the signal energy over space and frequency. Obviously, if π = π, then ππ,π (π‘, π) = ππ (π‘, π) is called the auto Wigner-Ville distribution. That is ∞ π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ ππ (π‘, π) = ∫−∞ π (π‘ + 2) π (π‘ − 2) π −πππ₯ ππ₯. (10) Often both the cross WVD and the auto WVD are usually referred to simply as the WVD. We should remember that the WVD is a nonlinear time frequency transform because the signal enters integral more than once. It is obvious that if π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ π π,π (π‘, π₯) = π (π‘ + 2) π (π‘ − 2), then the cross WVD is the FT of the function π π,π (π‘, π₯) with respect to π₯, i.e., ππ,π (π‘, π) = β±{π π,π (π‘, π₯)}(π). (11) Applying the Dirac’s delta function defined in equation (3) we get the FT of the WVD with respect to π (see [2, 3]) as π Μπ,π (π‘, π) =2π π (π‘ − π) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π (π‘ + ). 2 2 (12) Some important properties of the WVD are summarized as follows. Let f, g ∈ πΏ2 (β). Denote by ππ is a shift operator defined by ππ π(π₯) = π(π₯ − π) and by ππ0 is a modulation operator defined ππ0 π(π₯) = π ππ0 π₯ π(π₯). 1. Complex conjugation Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ππ,π (π‘, π) = ππ,π (π‘, π). 2. Time marginal 1 ∞ ∫ π (π‘, π)ππ 2π −∞ π,π 3. 4. 5. 6. = |π(π‘)|2 . Frequency marginal ∞ ∫−∞ ππ,π (π‘, π)ππ‘ = |β±{π}(π)|2. Energy Distribution ∞ ∞ 1 ∞ 1 ∞ ∫ ∫ π (π‘, π)ππππ‘ = ∫−∞|π(π‘)|2 ππ‘ = 2π ∫−∞|β±{π}(π)|2 ππ. 2π −∞ −∞ π,π Shift ππππ,πππ (π‘, π) = ππ,π (π‘ − π, π). Modulation πππ π,ππ π (π‘, π) = ππ,π (π‘, π − π0 ). 0 0 7. Inversion formula ∞ 1 π‘ π(π‘) = ∫ ππ,π ( , π) π πππ‘ ππ, Μ Μ Μ Μ Μ Μ −∞ 2 2ππ(0) Μ Μ Μ Μ Μ Μ ≠ 0. provided π(0) To illustrative the application of the properties mentioned above, we shall give an example of the WVD (compare to [6]). Example. Consider a chirp signal which is modulated by a Gaussian envelope π 1/4 −ππ‘ 2 π(π‘) = (π) exp ( −ππ‘ 2 2 + πππ‘ 2 + ππ0 π‘), 2 πππ‘ 2 where exp ( 2 ) is the Gaussian term, exp ( exp(ππ0 π‘) is a frequency-shifting term. 2 ) is the chirp signal, and An application of the WVD definition (10) we may compute the shift of the WVD as ππππ (π‘, π) π 1/2 ∞ −π(π‘ − π + π₯/2)2 ππ(π‘ − π + π₯/2)2 = ( ) ∫ exp ( + + ππ0 (π‘ − π + π₯/2)) π 2 2 −∞ −π(π‘ − π − π₯/2)2 ππ(π‘ − π − π₯/2)2 × exp ( − − ππ0 (π‘ − π − π₯/2)) 2 2 × exp( −πππ₯)ππ₯ π₯2 2 −π ((π‘ − π) + π‘π₯ − ππ₯ + π 1/2 4 ) ππ((π‘ − π)2 + π‘(π₯ − π)) = ( ) ∫ exp ( + ) π 2 2 −∞ ∞ π₯ × exp (ππ0 (π‘ − π + )) ππ₯ 2 π₯2 −π ((π‘ − π)2 + 4 − π‘π₯ + ππ₯) ππ((π‘ − π)2 − π‘(π₯ − π)) × exp ( − ) 2 2 π₯ × exp (−ππ0 (π‘ − π − )) exp( −πππ₯)ππ₯ 2 π 1/2 = (π ) 2 ∞ −ππ₯ 2 π −π(π‘−π) ∫−∞ exp ( 4 ) exp(ππ(π‘ − π)π₯ + ππ0 π₯) exp( −πππ₯) ππ₯. Taking the FT of shift property of the Gaussian function we finally get 2 2 ππππ (π‘, π) = 2 π −π(π‘−π) π −(π−π(π‘−π)−π0 ) /π. Now we provide the alternative proof of the Moyal’s formula as follows. Theorem (Moyal’s formula) For any complex functions π1 , π2 , π1 , π2 ∈ πΏ2 (β), then the following results hold ∞ ∞ 1 Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π ∫−∞ ∫−∞ ππ1 ,π1 (π‘, π) π (13) π2 ,π2 (π‘, π)ππππ‘ = (π1 , π2 )(π1 , π2 ). 2 In particular, we have ∞ 1 ∞ 2 π ∫−∞ ∫−∞|ππ,π (π‘, π)| ππππ‘ = βπβ2 βπβ2 , 2 and 1 2 ∞ ∞ π ∫−∞ ∫−∞ ππ (π‘, π) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ππ (π‘, π)ππππ‘ = |(π, π)|2 . Proof. The Parseval’s formula of the FT applies to π-integral and yields ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∫ ∫ ππ1 ,π1 (π‘, π) π π2 ,π2 (π‘, π)ππππ‘ −∞ −∞ ∞ ∞ = ∫ ( ∫ β±{π π1 ,π1 (π‘, π₯)}(π) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ β±{π π2 ,π2 (π‘, π₯)}(π)ππ) ππ‘ −∞ −∞ ∞ 1 ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = 2π ∫−∞(∫−∞ π π1 ,π1 (π‘, π₯)π π2 ,π2 (π‘, π₯) ππ₯) ππ‘. Therefore, we further get ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∫ ∫ ππ1 ,π1 (π‘, π) π π2 ,π2 (π‘, π)ππππ‘ −∞ −∞ ∞ 1 ∞ π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ = ∫ ∫ π1 (π‘ + ) π1 (π‘ − ) π2 (π‘ − ) π2 (π‘ + ) ππ₯ππ‘, 2π −∞ 2 2 2 2 −∞ π₯ π₯ which is, due to change of variables, y= π‘ + 2 and π§ = π‘ − 2, ∞ ∞ 2 ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∫ ∫ ππ1 ,π1 (π‘, π) ππ2 ,π2 (π‘, π)ππππ‘ = ∫ ∫ π1 (π¦)π 1 (π§)π2 (π§)π2 (π¦)ππ¦ππ§ π −∞ −∞ −∞ −∞ 2 Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = π (π1 , π2 )(π 1 , π2 ). This proves the proof of (13). 4 Correlation Theorem for WVD The convolution is very important in the theory of linear time-invariant (LTI) systems (also known as linear shift-invariant systems for two-and higher dimensional signals). Based on the correlation definition mentioned in (6) we establish two following correlation theorems for the WVD. Theorem 4.1 For any two signals π, π ∈ πΏ2 (β), the following result holds ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ππβπ (π‘, π) = ∫−∞ π (14) π (π’, π) ππ (π’ + π‘, π)ππ’. Proof. By the WVD and correlation definitions we immediately get ∞ π₯ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π₯ ππβπ (π‘, π) = ∫ (π β π) (π‘ + ) (π β π) (π‘ − ) π −πππ₯ ππ₯ 2 2 −∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∞ ∞ Μ Μ Μ Μ Μ Μ ∞ π₯ π₯ = ∫−∞ ∫−∞ π(π) π (π + (π‘ + 2)) ππ ∫−∞ π(π) π (π + (π‘ − 2)) πππ −πππ₯ ππ₯. π π Putting π = π’ + 2 , π = π’ − 2 and π₯ = π − π, we easily get ∞ ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π π₯ ππβπ (π‘, π) = ∫−∞ ∫−∞ ∫−∞ π (π’ + 2) π (π’ + 2 + (π‘ + 2)) π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π₯ × π (π’ − ) π (π’ − + (π‘ − )) π −ππ(π−π) ππππππ’ 2 2 2 ∞ ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π = ∫−∞ ∫−∞ ∫−∞ π (π’ + ) π (π’ − ) π (π’ + π‘ + ) π (π’ − π‘ − ) 2 2 2 2 × π −ππ(π−π) ππππππ’. Applying the definition of the WVD we obtain ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π ππβπ (π‘, π) = ∫ ∫ π (π’ + ) π (π’ − ) π πππ ππ 2 2 −∞ −∞ ∞ π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π × ∫ π (π’ + π‘ + ) π (π’ + π‘ − ) π −πππ ππππ’ 2 2 −∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∞ π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π = ∫ ∫ π (π’ + ) π (π’ − ) π −π€ππ ππ 2 2 −∞ −∞ ∞ π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π × ∫ π (π’ + π‘ + ) π (π’ + π‘ − ) π −πππ ππππ’ 2 2 −∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = ∫−∞ ππ (π’, π) ππ (π’ + π‘, π)ππ’. From the above theorem, it can easily be seen that that the Wigner-Ville distribution of the correlation of the two signals is the correlation in time of their corresponding Wigner-Ville distributions. Theorem 4.2 (Correlation with Respect to Both Variables) For any two signals π, π ∈ πΏ2 (β), the following result holds (ππ β ππ )(π, π) = 2π|(π, Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ππ π−π π)|2 . (15) Proof. It readily follows from the correlation theorem (8) of the classical FT that ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ (ππ β ππ )(π, π) = ∫ π π (π‘, π) ππ (π‘ + π, π + π)ππ = −∞ 1 ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ (π‘, π) π Μπ (π‘ ∫ π 2π −∞ π + π, π)π πππ ππ. With the help of (12) we further obtain ∞ π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π π Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π = 2π ∫ π (π‘ − ) π (π‘ + ) π (π‘ + π − ) π (π‘ + π + ) π πππ ππ, 2 2 2 2 −∞ π which is, due to the change of variables π = π‘ − 2 , ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = 4π ∫ π(π)π(2π‘ − π) π(π + π)π(π + 2π‘ − π)π π2π(π−π‘) ππ. −∞ Integrating the above expression with respect to π‘ yields ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∫ ∫ π π (π‘, π) ππ (π‘ + π, π + π)ππππ‘ −∞ −∞ ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = 4π ∫ π(π) π(π + π)π π2ππ ππ ∫ π(2π‘ − π) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ π(π + 2π‘ − π) π −π2ππ‘ ππ‘ −∞ −∞ which is, by substitution of π’ = 2π‘ − π, ∞ ∞ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ∫ ∫ π π (π‘, π) ππ (π‘ + π, π + π)ππππ‘ −∞ −∞ ∞ ∞ Μ Μ Μ Μ Μ Μ π −πππ’ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = 2π ∫ π(π) π(π + π)π πππ ππ ∫ π(π’) π(π’ + π)ππ’ −∞ −∞ = 2π(π, Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ ππ π−π π)(πΜ , ππ π−π π) Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ Μ = 2π(π, π π π−π π)(π, ππ π−π π). This is desired result. Acknowledgements This work is partially supported by Hibah Penelitian Kompetisi Internal tahun 2013 (No. 110/UN4-.42/LK.26/SP-UH/2013) from the Hasanuddin University, Indonesia. References [1] L. Cohen, Time-Frequency Distribution- A Review, Proceedings of the IEEE, 77 (7), 1989, 941-980. [2] L. Debnath, B. V. Shankara and N. Rao, On New Two-Dimensional Wigner-Ville Nonlinear Integral Transforms and Their Basic Properties, [3] [4] [5] [6] Integral Transforms and Special Functions, 21(3) (2010), 165-174. L. Debnath, Wavelet Transform and Their Applications, Birkhauser, Boston, 2002. L. Debnath, Recent developments in the Wigner–Ville distribution and the time–frequency signal analysis, Proc. Indian Natl. Sci. Acad. , 68A (2002), 35–56. K. Groechenig, Foundation of Time-Frequency Analysis, Birkhauser, Boston, 2001. L. Cohen, Time-Frequency Analysis, Prentice Hall PTR, Englewood Cliffs, New Jersey, 1995.