Chapter 0
Chapter 0
Chapter 0
Textbook: S.Haykin, Adaptive Filter Theory, 4th Ed., Prentice Hall, 2002
Introduction BACKGROUND REVIEW 2. Discrete-time and random signals 3. Mathematical Tools OPTIMAL LINEAR FILTERS 4. Wiener Filters 5. Linear Prediction ADAPTIVE FILTERING 6. Stochastic Gradient Descent Algorithms 7. Family of LMS Algorithms 8. Method of Least Squares 9. Recursive Least Squares (RLS) Algorithm 10. Square-Root Algorithms 11. LMS and RLS Algorithms: Practical Issues 12. Kalman Filtering (?) APPLICATIONS 13. Spectrum Estimation 14. Array Processing ELE 774 - Adaptive Signal Processing
Definition of filtering
A filter is commonly used to refer to a system that is designed to extract information about a prescribed quantity of interest from noisy data.
Applications
Communications; radar, sonar, Control Systems; navigation, Speech/Image Processing; echo and noise cancellation, biomedical engineering Others; seismology, financial engineering, etc.
Applications
!!! Noise and errors are statistical in nature !!! We will use statistical tools.
ELE 774 - Adaptive Signal Processing
Filter
Linear
A filter is linear if the filtered, smoothed or predicted quantity at the output of the filter is a linear function of the observations applied to the filter input. u(t) u(n) Filter y(t) y(n) Non-linear or
Non-Linear
Otherwise, it is non-linear.
Linear
or
Optimum Filter
Definition: Solution of an optimization problem wrt. a certain criterion with statistical parameters.
Nonlinear: Maximum Likelihood (ML) sense (very difficult to implement) Linear: Minimum Mean Square Error (MMSE) sense
Etc.
Adaptive Filters
-estimation (knowledge of the system) is needed before filtering -inversion of a huge matrix
-computationally inefficient!
If the environment is stationary converges to the Wiener soln. non-stationary tracks the changes.
ELE 774 - Adaptive Signal Processing 9
Rate of convergence (to the optimum Wiener soln.) Misadjustment (deviation from the optimum Wiener soln.) Tracking (the variations in a non-stationary environment) Robustness(to disturbances of small energy) Computational Requirements/Cost Numerical Properties (Numerical stability & accuracy)
10
IIR
11
Transversal Filter
unit-delay element Convolution Sum
multiplier
adder
ELE 774 - Adaptive Signal Processing 12
Transversal Filter
13
Lattice Predictor
14
Lattice Predictor
Backward prediction error m: the mth reflection coefficient Input seq. u(n) is correlated, backward prediction error b(n) is uncorrelated Together with m, b(n) approximates d(n) (innovations process).
ELE 774 - Adaptive Signal Processing 15
Systolic Array
Boundary cell
Internal cell
s
16
Systolic Array
17
IIR Filter
May have stability problems, We will prefer FIR filters for Adaptive filtering.
18
Error
Difference
Filter
y(n) d(n)
(n) : Error + -
Error Squares
19
Cost Function depends on Mean-Square Error criterion (!!! Stochastic, depends on second order statistics !!!) Solution of Wiener-Hopf Equations
Results in Wiener soln. but with an iterative approach Requires Expectations E{.}
Least-Squares Estimation
Cost Function depends on sum of weighted error squares Low computational complexity due to recursive operation
Three
categories
Relies on Matrix Inversion Lemma Numerically unstable, high computational complexity Based on QR-decomposition Numerically stable Exploits certain matrix structures to reduce complexity.
ELE 774 - Adaptive Signal Processing 21
Standard RLS
Applications
Four Classes
Identification
system identification layered earth modeling deconvolution adaptive and blind equalisation linear predictive coding adaptive differential PCM spectrum analysis signal detection noise canceling echo cancellation adaptive beamforming
22
Inverse modeling
Prediction
Interference cancellation
System Identification
Observing the output of a plant(system), given the input signal, tries to estimate the IR of the plant. Filter coefficient are found by an adaptive algorithm.
23
Adaptive Equalization
Removes intersymbol interference (ISI). Filter coefficient are found by an adaptive algorithm.
24
25
26
Echo Cancellation
Coupling due to imperfect balancing in hybrid transformer creates an echo in analog telephone lines. Echo signal can be estimated by an adaptive filter and the subtracted out.
27
Adaptive Beamforming
Multiple sensors (antenna, microphone, etc) used to steer the beam to a specific position. Radar, sonar Commun. systems, Astrophysical exploration, Biomedical signal processing, etc.
28
Historical Notes
Method of least squares, Gauss, 1795 Minimum mean square error estimation, late 1930s Discrete-time Wiener-Hopf equation, Levinson, 1947 Kalman filter, Swerling, 1958 and Kalman, 1960 Stochastic approximation, Robins and Monro, 1951 LMS algorithm, Widrow and Hoff, 1959 Gradient adaptive lattice (GAL) algorithm, Griffiths, 1977-8
29
Historical Notes
Kalman filter, Godard Algorithm, Godard, 1974 Relationship between RLS and Kalman, Sayed and Kailath, 1994 QR decomposition based systolic array, Gentleman & Kung, 1981 Fast RLS algorithm, 1970s, Morf Logical calculus for neural networks, McCulloch and Pitts, 1943 Perceptron, Rosenblatt, 1958 Back-propagation algorithm, Rumelhart, et al., 1986 Radial basis function network, Broomhead and Lowe, 1988
Neural Networks
30
Applications
Zero-forcing equaliser, Lucky, 1965 MMSE equaliser, Gersho, 1969, Proakis&Miller, 1969 Godard Algorithm, Godard, 1974 Fractionally Spaced Equaliser (FSE), Brady, 1970 Decision Feedback Equaliser (DFE), Austin 1967, MMSE, Monsen, 1971. Maximum Likelihood speech prediction, Saito and Itakura, 1966 Linear Predictive Coding (LPC), Atal and Hanauer 1970-1 Adaptive Lattice Predictor, Nakhoul and Cossell, 1981
Speech Coding
31
Applications
Maximum entropy method, Burg, 1967 Method of multiple windows, Thomson, 1982
Intermediate Frequency (IF) sidelobe canceller, Howells, 1950 Control law for adaptive array antenna, Applebaum, 1966 Application of LMS, Widrow et al., 1967 Minimum Variance Distortionless Response (MVDR) beamformer, Capon, 1969
32