36
8 Lecture 2: Wednesday Jan 11, 2017 Lecture: • Determinstic DSP (Chapter 2)

Lecture: • Determinstic DSP (Chapter 2)barry.ece.gatech.edu/6603/lectures/lec2.pdfLecture 2: Wednesday Jan 11, 2017 Lecture: • Determinstic DSP (Chapter 2) 9 Fourier and Inverse

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

  • 8

    Lecture 2: Wednesday Jan 11, 2017

    Lecture:

    • Determinstic DSP (Chapter 2)

  • 9

    Fourier and Inverse Fourier Transforms

    X( ej) =∞Xk= –∞xk e

    –jk

    xk = X( ej)ejk d

    0 k

    For a discrete-time signal xk:

    1 2

    x0

    x1x2

    12------

  • 10

    Example

    X(ej) =∞Xk= 0

    0.8ke–jk

    =∞Xk= 0

    (0.8e–j)k

    = .

    0 5 10

    0.8

    1

    0.64xk = 0.8kuk

    1

    1 0.8e j––---------------------------

  • 11

    Filtering

    Time domain: yk =∞Xi= –∞xihk – i = xk hk

    Freq domain: Y(ej) = X(ej)H(ej)

    Terminology:

    hk = impulse response H(ej) = frequency response|H(ej)| = magnitude response.

    hkxk yk

    FILTER

  • 12

    Pop Quiz:What is the Fourier transform X(ej) of the unit step, xk = uk?

    X(ej) = ?

    0 5 10

    1 ...

    k

  • 13

    Z TransformZ transform:

    H(z) =∞Xk= –∞hkz

    –k ,

    When z = e j, H(e j ) reduces to the DTFT.

    ROC = values of z for which above sum is finite.

    If DTFT exists, ROC includes u. circle.

  • 14

    Example

    X(z) = 6 + 6z–1 + 3z–3

    Pop Quiz: What is the ROC?

    0

    6

    xk = (6, 6, 0, 3, ...):

    6

    0

    3

    1 3

  • 15

    Example

    X(z) =∞Xk= 0

    akz–k =∞Xk= 0

    (az–1)k =

    0 5 10

    a

    1

    a2

    xk = akuk

    a3

    1

    1 az 1––--------------------

    when |a/z|< 1

  • 16

    NotationCausal ⇒ hk = 0 for k < 0

    Anticausal ⇒ hk = 0 for k > 0

    Monic ⇒ {h0 = 1}AND{causal OR anticausal}

    Right-sided ⇒ hk = 0 for k < some K.

    Left-sided ⇒ hk = 0 for k > some K.

    FIR ⇒ both right- and left-sided.

    2-sided ⇒ neither right- nor left-sided.

    0

    0

    0

    1

    K 0

    K0

    ... ...

  • 17

    ROCFIR ⇒ ROC = entire plane, except possibly z = 0 and/or |z| = Right-sided ⇒ ROC = | z | > r (beyond a circle)Left-sided ⇒ ROC = | z | < r (inside a circle)Two-sided ⇒ ROC = r1 < | z | < r2 (a ring)

  • 18

    Stable ROCA filter is bounded-input bounded-output (BIBO) stable if

    ∞Xk= –∞|hk|

  • 19

    Poles and ZerosA rational transfer function

    H(z) = A

    ⇒{b1, ..., bM}arezeros ⇒{a1, ..., aN} arepoles

    • # poles = # zeros [including z = 0 and |z| = ]• ROC is bounded by poles• Causal ⇒ ROC = | z | > | outermost pole|• Causal and stable ⇒ all poles must be inside the unit circle. • Anticausal and stable ⇒all poles must be outside the unit circle.

    • N = 0 ⇒FIR; N > 0 ⇒ IIR

    k 1=M

    z bk–

    k 1=N z ak–

    ---------------------------------

  • 20

    Examples

    H(z) = 1 – 0.5z–1 zero at 1/2, pole at origin.

    H(z) = 1 - 0.5z zero at 2, pole at ∞

  • 21

    Examples

    Stable and 2-sidedStable and left-sidedStable and right-sided

    1

    Imz

    Rez

  • 22

    The Matched FilterA filter that is “matched” to hk has impulse response h–k* :

    k

    k

    hk

    h–k*

    H(ej) H(z)

    H*(ej). Same mag response. H*(1/z*)

    Example:

    The corresponding matched filter:

  • 23

    What happens to poles and zeros?

    b is a zero of H( z ) ⇒ H( b ) = 0 ⇒ H*(1/z*) = 0 when z = 1/b*⇒ 1/b* is a zero of H*(1/z*)

  • 24

    Compare H to its MatchH(z) H*(1/z*)

  • 25

    ExampleH( z ) = has a zero at 1/a*, and a pole at z = a:

    What is magnitude response?

    |H(e j)| = = = 1, for all

    z 1– a*–1 az 1––--------------------

    Stable and right-sided

    a

    1/a*

    e j– a*–1 ae j––----------------------- e j– 1 a*e

    j–1 ae j––-----------------------

  • 26

    All Pass FiltersAn allpass A(z) satisfies

    |A(e j )| = 1

    A(e j)A*(e j) = 1

    ak a*– k = k

    A(z)A*(1/z*) = 1 .

    any zero [or pole] must be accompanied by a matching pole [or zero] at the conjugate-reciprocal location.

    Every rational allpass transfer function has form:

    A(z) = e j , B(z) = 1 + b1z + + bNzN B z

    zN M– B* 1/z* -------------------------------------------

  • 27

    An All Pass

    d

    1/d*

    1/c*

    cRe{ z}

    Im{ z}

  • 28

    Minimum Phase

    H(z) = A

    [Strictly] minimum-phase |bk|, |ak| < 1.

    [Strictly] maximum-phase |bk|, |ak| > 1.

    Loosely minimum-phase |bk| 1, |ak| < 1.

    Loosely maximum-phase |bk| 1, |ak| > 1.

    A minimum-phase transfer function is both causal and stable.

    Why?

    Min ph all poles/zeros inside none at z = H(z) = khkz–k cannotcontain any positive powers of z hk is causal hk is right-sided ROC isoutside the outermost pole. But all poles inside ROC must include the unitcircle stability.

    k 1=M

    z bk–

    k 1=N z ak–

    ---------------------------------

  • 29

    Nonzero at time zeroA min-phase sequence must be nonzero at time zero (prevents zero at z = )

  • 30

    Properties of Min Phase• Causal• Stable• Nonzero at time zero (prevents zero at | z | = )• Of all transfer functions / sequences with a given magnitude response:

    • Only the minimum-phase transfer function can be inverted by a causaland stable filter

    • Only the minimum-phase sequence has its energy maximallyconcentrated in the vicinity of time zero.

  • 31

    ExampleFind a min-phase sequence hk with samemagn response as xk:

    X( z ) = 2z–1 + 3z–2

    = 2 ·

    • zeros at –3/2 and • poles at 0 and 0 A( z ) = · z2

    H( z ) = 2 · = 3 + 2z–1

    1 20k

    xk

    23

    z32---+

    z2------------

    z 1–32---+

    z32---+

    -----------------

    1 20 k

    hk

    23

    z 1–32---+

  • 32

    ExampleFind a min-phase sequence hk with samemagn response as xk:

    X( z ) = 8(z + 3/2)(z + 3/4)

    • zeros at –3/2 and –3/4• poles at and

    A( z ) = · z–1

    H( z ) = 8 · (z–1 + 3/2)(z + 3/4)· z–1

    = 12 + 17z–1 + 6z–2

    –1 0–2 k

    xk

    98

    18

    z 1– 3/2+z 3/2+

    -----------------------

    k–10 –2

    hk

    8

    3

    10

  • 33

    Two Numbers

    A{x, y} =

    G{x, y} = = exp(A{ logx, logy})

    H{x, y} = = = =

    Q1. Drive 800 miles from ATL to NYC. 80 mph there, 40 mph back.

    Average speed? mph = totalmiles/totalhrs = 2(800)/(10 + 20) = 53.3 mph

    Q2: QCOM gains 20%, 60%, and 10% in 3 consecutive years.

    What is the average rate of return?

    (X)(1.2)(1.6)(1.1) = X(1 + r)3

    ⇒ 1 + r = G{1.2, 1.6, 1.1} = 1.283⇒ 28.3%

    x y+2

    ------------

    xy

    21x--- 1

    y---+

    ------------- 1

    A 1x--- 1y---

    --------------------- 2xyx y+------------ G

    2

    A------

  • 34

    OrderingLet S = {x, y}.

    min{S} H{S} G{S} A{S} max{S}

    Example: S = {5, 45}

    min{S} H{S} G{S} A{S} max{S}

    0

    minH

    G

    A

    max

    5 9 15 25 45

  • 35

    Valid PSDA transfer function S(z) is a “valid PSD” when S(e j ) 0 for all

  • 36

    Arithmetic Mean

    The arithmetic mean of S(ej) is

    A{S} = lim iSi =

    -

    1N---- 1

    2------ S ej d

  • 37

    Geometric MeanThe geometric mean is

    G{S} = limi = exp .

    Unproven Fact:

    G{|1 – cej|2} = 1 for any |c| 1.

    Example:

    S(ej ) = 2 – 2cos() ⇒A{S} = 2 and G{S} = 1.

    Remark: G{S} is nonzero despite the zero at = 0.

    Si1 N/ 1

    2------ logS ej d

  • 38

    Harmonic MeanThe harmonic mean is

    H{S} = lim1/i1/Si = .

    Interpret: H = 1/g0, where G( z ) = 1/S( z )

    1

    12------ 1

    S ej ---------------- d

    ---------------------------------------

  • 39

    Inequalitiesmin{S} H{S} G{S} A{S} max{S}

    All inequalities collapse to equality if and only if S is a constant

  • 40

    ExampleS(z) =

    A{S} = 4, G{S} = 3, and H{S} = 12/5 = 2.4.

    3

    1 0.5z 1–– 1 0.5z– -------------------------------------------------------

    A{S} = 4

    G{S} = 3

    H{S} = 2.4

    12

    0 –

  • 41

    Properties1. min{S} H{S} G{S} A{S} max{S} ,

    with equalities if and only if S(e j ) is constant

    2. A{aS bP } = aA{S} + bA{S}

    3. G = .

    4. G{|1 – ce–j|2} = 1 for any |c| 1.

    SPQ

    ------- G S G P

    G Q ----------------------------

  • 42

    What Exactly Limits Data Rate? 3 Key Parameters

    r( t )s( t )n( t )

    H( f )

    bandwidth W [Hz]

    signal power P [W] one-sided

    The 3 combine to give the signal-to-noise ratio: SNR = P

    N0W------------

    noise density N0 [W/Hz]

  • 43

    Solution: Use QAM

    • Bandwidth determines symbol rate, 1/T = W

    • SNR determines size of QAM alphabet, M = 1 +

    ⇒Rb = W log2(1 + )

    Compare to Shannon capacity:

    ⇒C = W log2(1 + SNR)

    DACa s( t )UP

    SNR

    -------------

    SNR

    -------------