Characteristic Functions in Radar and Sonar Southeastern Symposium on System Theory March 18,2002 Stephen R. Addison Department of Physics and Astronomy University of Central Arkansas Conway, AR 72035 saddison@mail.uca.edu Presented By John E. Gray Code B-32 Naval Surface Warfare Center Dahlgren Division Dahlgren, VA 22448 grayje1@nswc.navy.mil 1 Outline • FOUR QUESTIONS • WHAT ARE CHARCTERISTIC FUNCTIONS? • SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION. • MULTI-DIMENSIONAL CHARCTERISTIC FUNCTIONS • RAYLEIGH PROBLEM 2 FOUR QUESTIONS å å å Question 1: Given we know the density for x is PÝxÞ, what is the distribution for u = fÝ xÞ? å å Question2: Givenwe knowthe densityfor x is PxÝxÞ and the densityfor y is PyÝyÞ, what is the distributionfor å å å u = x + y? å å Question 3: Given we know the density for x is P x ÝxÞ and the density for y is P y ÝyÞ, what is the distribution for å åå u = x y? å å Question 4: Given we know the density for x is P x ÝxÞ and the density for y is P y ÝyÞ, what is the distribution for å åx u = å? y 3 DEFINTIONS OF CHARCTERISTIC FUNCTION MP ÝSÞ = e jSx = X DEFINTION: 1. 2. 3. 4. PROPERTIES K ?K e jSx PÝxÞ dx MÝ0Þ = 1. MD Ý?SÞ = MÝSÞ. |MÝSÞ| ² 1. |MÝSÞ| ² MÝ0Þ. ALTERNATIVE DEFINITION: K K MP ÝSÞ = X ?K e PÝxÞ dx = X ?K > jSx n K ÝjSxÞ PÝxÞdx n=0 n! => n K ÝjSÞ n=0 n! Öx n × 4 DEFINTIONS OF CHARCTERISTIC FUNCTION TWO METHODS FOR CALCULATION OF MOMENTS K Öx × = X ?K x n PÝxÞ dx n n Öx × = n 1 / MP ÝSÞ /S n jn P S =0 differentiation is easier than integration Fourier transform pair relationship between the PDF and CF: PÝxÞ = 1 2^ K X?K e ?jSx MP ÝSÞdx 5 SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION one would often like to know the density of a new variable, å say û, that is a function of the old variable: å u= fÝ xÞ Probability Density Function: Characteristic Function: Apply CF to PDF: PDF for Transformed variable: 1 2^ PÝuÞ = Mu ÝSÞ = e PÝuÞ = 1 2^ K X?K e ?jSu Mu ÝSÞ dS jSfÝxÞ K K = X ?K e jSfÝxÞ PÝxÞ dx K X?K X ?K e jSfÝxÞ e ?jSx PÝxÞ dS dx K PÝuÞ = X ?K NÝu ? fÝxÞÞ PÝxÞ dx This solves the problem for Exercise 1. 6 SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION Interpretation of Delta Function of Function NÝgÝxÞÞ = >i NÝx ? x iÞ g v Ýx1 Þ | i | NÝfÝxÞ ? uÞ = >i NÝx ? x i Þ fv Ýx1 Þ | PÝuÞ = >i i | PÝxi Þ |fv Ýx iÞ | 7 SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION Translation å å å (Translation): Let u = ax + b, what is PÝuÞ given we know that the PDF of x is PxÝxÞ? Solving for zero of the equation gives x = u ?a b and note dx = 1a , then the distribution is 1 PxÝ u ? b Þ u?b = # PÝxÞ| PÝuÞ = 1 x= a a a a 8 SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION Square Law Detector å å å (Square Law Detector): Let u = x 2 , what is PÝuÞ given we know that the PDF of x is P x ÝxÞ? Answer: Solve for zeros x1 = u and x2 = ? u note 2xdx = du, then * becomes PÝuÞ = å If x is NÝ0, aÞ, then the PDF is 1 P x ÝuÞ| x ,x = 1 PÝ u Þ + PÝ? u Þ 1 2 2 u 2 u P u ÝuÞ = # ? u 1 e 2a2 BÝuÞ. a Ý2^uÞ where B is the unit step function BÝxÞ = 1 Ýx ³ 0Þ = 0 ÝotherwiseÞ # 9 SINGLE VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION The coordinate transform ation y = R sinÝj! Þ a P DF f j Ýj Þ is onto but not one-to-one ov er interv al ß?K, Kà. Thus ref: * has an infinite num ber of zeros. It is m ore conniv ent to determ ine the C F directly, so the transform ation of the P DF is M S Ýg Þ = K X ?K e jgR sin Ýj Þ f j Ýj Þdj . The ex ponential can be w ritten as K > J n Ýg RÞe jn j = e jgR sin Ýj Þ , n =?K so the C F is giv en by K M S Ýg Þ = > n =?K J n Ýg RÞ X K ?K e jn j K fÝj Þdj = > J n Ýg RÞF ÝnÞ; n =?K w hich can be rew ritten as K M j Ýg Þ = J 0 Ýg RÞ + > J n Ýg RÞ ßF ÝnÞ + Ý? 1Þ n F Ý?nÞ à n =1 w here F ÝnÞ is the F ourier transform of the P DF f j Ýj Þ ev aluated at g = n. Depending on the problem , the C F is sufficient, but som etim es it is still useful to k now the P DF . N ow if w e apply the identity (A bram ow itz) n K Þ B Ý1 ? |g |Þ, X ?K e ?jgt J n ÝtÞ dt = 2Ý? jÞ T n Ýg Ý1 ? g 2 Þ w here T n Ýx Þ is the n-th order C hebyshev polynom ials, the P DF of the coordinate transform ation is f y ÝyÞ = w here a n = Ý? jÞ n F ÝnÞ. a 0 + > Kn =1 Ýa n + a ?n ÞT n Ý 2 2 ^ ÝR ? y Þ y R Þ B Ý1 ? y Þ, R # 10 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION Repeating the same procedure for the single dimensional case to the multi-dimensional case yields: K K PÝu, vÞ = X ?K X ?K X X Nßu ? fÝx, yÞàNßv ? gÝx, yÞàPÝx, yÞ dxdy This due to Cohen, is sufficient to allow us to solve the problem of determining many of the two dimensional combinations of random variables such as those in Exercises 2-4. 11 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION Let û = fÝx!,yÞ, the CF is K K MuÝSÞ = X ejSfÝx,yÞ PÝx,yÞ dxdy, X ?K ?K so the density becomes K 1 e?jSuMuÝSÞ dS PÝuÞ = X 2^ ?K K K K 1 ejSßu?fÝx,yÞà PÝx,yÞ dydxdS = X X X 2^ ?K ?K ?K integrating out S gives delta functions K PÝuÞ = X K X ?K ?K Nßu ? fÝx,yÞàPÝx,yÞ dx. # (C) 12 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION EXAMPLE: Translation å å å Let z = x + y, with distributions P x ÝxÞ and P y ÝyÞ (assuming they are uncorelated), then the distribution of û which is denoted by P u ÝuÞ is ref: C P z ÝzÞ = = = K K X ?K X?K Nßz ? x ? yàP xy Ýx, yÞ dxdy K X ?K Pxy Ýx, z ? xÞ dx K X ?K Px ÝxÞP y Ýz ? xÞ dx (uncorrelated) # 13 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION EXAMPLE: Multiplication å åå Let z = x y, with distribution PxyÝx,yÞ or distributions PxÝxÞ and PyÝyÞ (assuming they are uncorelated), then the distribution of û which is denoted by PzÝzÞ is ref: C K PzÝzÞ = X If we note that >i NÝx ? x i Þ K X ?K ?K Nßz ? xyàPxyÝx,yÞ dxdy. 1 , so |f Ýx iÞ| v NÝy ? zx Þ Nßz ? xyà = |x | so K f zÝzÞ = X ?K 1 P z ,x dx. |x| xy x # 14 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION EXAMPLE: Division or Monopulse Ratio å åx Let z = å , with distribution PxyÝx,yÞ or distributions PxÝxÞ and PyÝyÞ (assuming they are y uncorelated), then the distribution of z! which is denoted by PzÝzÞ is K K PzÝzÞ = X X Nßz ? xy àPxyÝx,yÞ dydx. ?K ?K If we note that Nßz ? xy à = NÝy ? zxÞ|x| so K f zÝzÞ = X |x|PxyÝzx,xÞdx. ?K # 15 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION RAYLEIGH PROBLEM Sum of two dimension random amplitudes and phases: X=> N ! x i=1 i => N i=1 â i cosÝS! i Þ Arose in Scattering off Rough Surfaces, Applications in Radar, Sonar, Acoustics, Communications, Physics, etc. 16 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION INSTANCE OF TRACKING PROBLEM Transformations from spherical to C artesian coordinates x! = Ý r! Þ cos S! sinÝ j! Þ , y = Ý r! Þ sin S! sinÝ j! Þ , z! = Ý r! Þ cos Ý j! Þ . Rewritten as: x! = y= !r 2 !r 2 #( sin S! + j! + sin S! ? j! cosÝS! + j! Þ ? cosÝS! + j! Þ 17 MULTI-VARIABLE APPLICATIONS OF CHARCTERISTIC FUNCTION 1 Then the distribution for z is: f z ÝzÞ = X ?1 >Kn=1 Ýan +a ?n ÞT n ÝjÞ P r Ý jz Þ a 0 + ^|j| Ý1?j 2Þ dj. two sum case z = z 1 + z 2 fzÝzÞ K =X ?K 1 X?1 (uncorrelated) = 1 X?1 K X?K Pz Ýz1ÞPz Ýz1 ? zÞ dz1 1 # 2 K Pr1Ý j11 Þ a0 + >n=1 Ýan +a?n ÞTnÝj1Þ z ^|j1 | Ý1 ? j21Þ K 1?z Pr2Ý z2+z j2 Þ a0 + >n=1Ýan + a?n ÞTnÝj2Þ ^|j2 | Ý1? j22Þ dj1 × dj2dz1 Case n is obtained by repeated application of Case 2 18 Conclusions • Demonstrated how to use CF to bypass difficulties in computing combinations of random variables using a method based on Cohen • Using Fourier analysis and application of definition of Dirac delta function, combined PDF is obtained with considerable simplification • Motivation engineering applications that occur in radar/sonar applications that involve combinations of probability density functions • Have demonstrated simple method to obtain PDF for different combinations of random variables • Rayleigh problem for i=2 solved which implies general solution for case i=n 19