Disclaimer! Notations, e.g. or , may vary from book to book, or from exam paper to other.
Even and odd functions: Frequencies, angular frequencies, periods:
Even{x(t)} = 0.5 [x(t) + x(t)] Here fs is sampling frequency (also fT later) Odd{x(t)} = 0.5 [x(t) x(t)] Frequency f , [f ] = Hz = 1/s Roots of second-order polynomial: Angular frequency ax2 + bx + c=0 = 2f = 2/T , [] = rad/s x = (b b2 4ac)/(2a) Normalized angular frequency Complex numbers: = 2/s = 2f /fs , [] = rad/sample i j = 1 = 1j Normalized frequency in Matlab z=p x + jy = r ej fMAT LAB = 2f /fs , [fMAT LAB ] = 1/sample r = x2 + y 2 , = arctan(y/x) Integral transform properties x = r cos(), y = r sin() Here all integral transforms share some basic properties. ej = cos() + j sin() Examples given with CTFT, x[n] X(ej ), x1 [n] N N z = r ej = | N r| ej(+2k)/N , k = 0, 1, 2, . . . , N 1 X1 (ej ), and x2 [n] X2 (ej ) are time-domain signals with corresponding transform-domain spectra. a and b are cons- Trigonometric functions: tants. sinc() = sin()/() Linearity. All transforms are linear. sin()/ 1, when 0; sinc() 1, when 0 ax1 [n] + bx2 [n] aX1 (ej ) + bX2 (ej ) cos() = 0.5ej + 0.5ej Time-shifting. There is a kernel term in transform, e.g., sin() = 0.5jej 0.5jej x[n k] ejk X(ej ) cos2 () + sin2 () = 1 Frequency-shifting. There is a kernel term in signal e.g., 0 /6 /4 /3 ejk n x[n] X(ej(k ) ) sin() 0 0.5 2/2 3/2 Conjugate symmetry. cos() 1 3/2 2/2 0.5 x [n] X (ej ) /2 3/4 /2 If x[n] R, then sin() 1 2/2 0 1 X(ej ) = X (ej ) cos() 0 2/2 1 0 |X(ej )| = |X(ej )|
3.1416, 3/2 0.8660, 2/2 0.7071 X(ej ) = X(ej ) If x[n] R and even, then X(ej ) R and even. Geometric series: P+ n 1 If x[n] R and odd, then X(ej ) purely C and odd. n=0 a = 1 a , |a| < 1 Time reversal. Transform variable is reversed, e.g., PN n 1 aN +1 , |a| < 1 x[n] X(ej ) n=0 a = 1a Differentiation. In time and frequency domain, e.g., Continuous-time unit step and unit impulse func- x[n] x[n 1] (1 ej )X(ej ) tions: ( nx[n] j d d X(ej ) 1, t > 0 Duality. Convolution property: convolution in time domain (t) = 0, t < 0 corresponds multiplication in transform domain (t) = dtd (t), (t) = lim0 (t) x1 [n] x2 [n] X1 (ej ) X2 (ej ) R (t) dt =1 and multipication R property: vice versa R x1 [n] x2 21 X1 (ej )X2 (ej() ) d
(t t0 )x(t) dt = x(t0 ) 2 R Parsevals relation. Energy in signal and spectral compo- In DSP notation 2(t) is computed 2 (t) 1 dt = 2, nents: when t = 0, and = 0 elsewhere. 1 |x[n]|2 = 2 j 2 P R 2 |X(e )| d Discrete-time unit impulse and unit step functions: ( Integral transforms 1, n = 0 Definitions given in first two lines of each type. Some com- [n] = 0, n 6= 0 mon pairs as well as properties are listed. See math reference ( 1, n 0 book for complete tables. [n] = 0, n < 0 Fourier-series P of continuous-time periodic signals: x(t) = k= ak ejk0 t Convolution ak = T1 T x(t) ejk0 t dt R Convolution is commutative, associative and distributive. R y(t) = h(t) x(t) = h( )x(t ) d x(t t0 ) ak ejk0 t0 jM0 t y[n] = h[n] x[n] = + P Re x(t) akM k= h[k]x[n k] x ( )xb (t T a P ) d T ak bk Correlation: xa (t)xb (t) l al bkl P+ d rxy [l] = n= x[n]y[n l] = x[l] y[l] dt x(t) jk0 ak P+ rxx [l] = n= x[n]x[n l] Fourier-series of discrete-time periodic sequences: x[n] = k=hN i ak ejk0 n , x[n] periodic with N0 P Mean and variance of random signal: ak = N1 n=hN i x[n] ejk0 n , ak periodic with N0 R P mX =RE[X] = xpX (x)dx 2 X = (x mX )2 pX (x)dx = E[X 2 ] m2X x[n M ] ak ejk0 M T-61.3010 DSP 2006 (F) 2/2 v. 1.04, 2006-04-25
ejM0 n x[n] akM [n] 1, ROC all z
[n k] z k , all z, except 0 (k > 0) or (k < 0) Continuous-time Fourier-transform (CTFT): 1 R [n] 1z1 1 , |z| > 1 x(t) = 2 X(j) ejt d R jt [n 1] 1z1 1 , |z| < 1 X(j) = x(t) e dt 1 jtk an [n] 1az 1 , |z| > |a| x(t tk ) e X(j) az nan [n] (1az 1 ejk t x(t) X(j( k )) 1 )2 , |z| > |a| xa (t) xb (t) Xa (j)Xb (j) n 1 (n + 1)a [n] (1az1 )2 , |z| > |a| 1 xa (t)xb (t) 2 Xa (j) Xb (j) 1r cos(0 )z 1 d rn cos(0 n)[n] 12r cos(0 )z 1 +r 2 z 2 , |z| > |r| dt x(t) jX(j) r sin(0 )z 1 tx(t) j d d X(j) rn sin(0 n)[n] 12r cos(0 )z 1 +r 2 z 2 , |z| > |r| j0 t e 2( 0 ) LTI filter cos(0 t) [( 0 ) + ( + 0 )] Panalysis Stability h |h[n]| < ; unit cirle belongs to ROC sin(0 t) j[( + 0 ) ( 0 )] Causality h[n] = 0, n < 0; P belongs to ROC x(t) = ( 1 2() n Unit step response s[n] = k= h[k] 1, |t| < T1 Causal transfer function of order max{M, N }: x(t) = 2 sin(T
1) M M 0, |t| > T1 P Q bm z m (1dm z 1 ) H(z) = K Pm=0 N = G Qm=1 N n=0 an z n=1 (1pn z ( n 1 )
sin(W t) 1, || < W Zeros: B(z) = 0; Poles: A(z) = 0; where H(z) = B(z)/A(z)
t X(j) = 0, || > W Frequency, magnitude/amplitude, phase response, z ej j (t) 1 H(ej ) = |H(ej )| ejH(e ) (t tk ) ejtk j H(e ) = H(z)|z=ej 1 eat (t) a+j , where Real{a} > 0 H[k] = H(ej )|=2k/N d Group delay () = d H(ej ) Discrete-time Fourier-transform (DTFT): 1 j Important transform pairs and properties: ) ejn d R x[n] = 2 2X(e X(e ) = n= x[n] ejn , X(ej ) periodic 2 j P a [n k] a ejk a z k an [n] 1/[1 a ej ] 1/[1 a z 1 ] x[n k] ejk X(ej ) a x[n k] a ejk X(ej ) a z k X(z) ejk n x[n] X(ej(k ) ) y[n] = h[n] x[n] Y (z) = H(z) X(z) x1 [n] x2 [n] RX1 (ej ) X2 (ej ) 1 j j() rectangular sinc, sinc rectangular x1 [n] x2 2 2 X1 (e )X2 (e ) d d j LTI filter design (synthesis) nx[n] j dP X(e ) ej0 n 2 P l ( 0 2l) Bilinear transform H(z) = H(s)|s and prewarping cos(0 n) Pl [( 0 2l) + ( + 0 2l) s = k (1 z 1 )/(1 + z 1 ), k = 1 or k = 2/T = 2fT sin(0 n) j P l [( + 0 2l) ( 0 2l) prewarp,c = k tan(c /2), k = 1 or k = 2/T = 2fT x[n] = ( 1 2 l ( 2l) Spectral transformations, c desired cut-off 1, |n| N1 LP-LP z 1 = (z 1 )/(1 z 1 ), where x[n] = sin((N 1 +0.5)) sin(/2) = sin(0.5(c c ))/ sin(0.5(c + c )) 0, |n| > N1 sin(W n) LP-HP z 1 = (z 1 + )/(1 + z 1 ), where W Wn n = sinc( ( ) ... = cos(0.5(c + c ))/ cos(0.5(c c )) 1, 0 || W Windowed(Fourier series method . . . X(ej ) = 1, || < c 0, W < || H(ej ) = ... [n] 1 0, || c [n k] ejk . . . h[n] = sin( c n) = c sinc( c n ) n an [n] 1ae1j , |a| < 1 hF IR [n] = hideal [n] 1 R w[n] Discrete Fourier-transform (DFT): HF IR (ej ) = 2 j Hideal (e )W (e j() ) d Connection to DTFT: X[k] = X(ej )|=2k/N Fixed window functions, order N = 2M , M n M : WN = ej2/N Rectangular w[n] = 1 x[n] = N1 PN 1 kn Hamming w[n] = 0.54 + 0.46 cos((2n)/(2M )) k=0 X[k]WN , 0k N 1 PN 1 kn Hann w[n] = 0.5 1 + cos((2n)/(2M )) X[k] = n=0 x[n]WN , 0 k N 1 Blackman w[n] = 0.42 + 0.5 cos((2n)/(2M )) + Laplace transform: 0.08 cos((4n)/(2M )) Convergence with a certain ROC (region of convergence). Bartlett w[n] = 1 (|n|/M ) Connection to continuous-time Fourier-transform: s = j 1 R +j Multirate systems x(t) = 2j X(s)est ds R j st Upsampling with factor L, L X(s) = x(t) e dt ( x[n/L], n = 0, L, 2L, . . . z-transform: xu [n] = xu [n] = 0, otherwise Convergence with a certain ROC (region of convergence). ConnectionH to discrete-time Fourier-transform: z = ej Xu (z) = X(z L ), Xu (ej ) = X(ejL ) 1 x[n] = 2j X(z)z n1 dz Downsampling with factor M , M P X(z) = n= x[n]z n xd [n] = x[nM ] PM1 k x[n k] z k X(z) Xd (z) = (1/M ) k=0 X(z 1/M WM ), j PM1 j(2k)/M x1 [n] x2 [n] X1 (z) X2 (z) Xd (e ) = (1/M ) k=0 X(e )