EECS0712 Adaptive Signal Processing 1 Introduction to Adaptive Signal Processing Assoc. Prof. Dr. Peerapol Yuvapoositanon Dept. of Electronic Engineering CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Course Outline Introduction to Adaptive Signal Processing Adaptive Algorithms Families: Newton’s Method and Steepest Descent Least Mean Squared (LMS) Recursive Least Squares (RLS) Kalman Filtering Applications of Adaptive Signal Processing in Communications and Blind Equalization CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Evaluation Assignment= 20 % Midterm = 30 % Final = 50 % CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Textbooks CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
http://embedsigproc. wordpress http://embedsigproc.wordpress.com /eecs0712-adaptive-signal-processing/ CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
QR code CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Adaptive Signal Processing Definition: Adaptive signal processing is the design of adaptive systems for signal-processing applications. [http://encyclopedia2.thefreedictionary.com/adaptive+signal+processing] CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
System Identification Let’s consider a system called “plant” We need to know its characteristics, i.e., The impulse response of the system CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Plant Comparison CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Error of Plant Outputs CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Error of Estimation Error of estimation is represented by the signal energy of error CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Adaptive System We can do it adaptively CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
One-weight Adjust the weight for minimum error e CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Error Curve Parabola equation CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Partial diff. and set to zero Partial differentiation Set to zero Result: CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Multiple Weight Plants We calculate the weight adaptively Questions: What is the type of signal “x” to be used, e.g. Sine, Cosine or Random signals ? If there is more than one weight w0 , i.e., w0….wN-1, how do we calculate the solution? CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Plants with Multiple Weight If we have multiple weights CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Two-weight In the case of two-weight CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Input From We construct the x as vector with first element is the most recent CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Plants with Multiple Weight (aka “Transversal Filter”) If we have multiple weights CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Regression input signal vector If the current time is n, we have “Regression input signal vector” CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Convolution Output of plant is a convolution Ex For N=2 CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
We can use a vector-matrix multiplication For example, for n=3 we construct y(3) as For example, for n=1 we construct y(1) as CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Let us stop there to consider Random signal theory first. The error squared is Let us stop there to consider Random signal theory first. CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Review of Random Signals CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Wireless Transmissions Ideal signal transmission 1 1 1 1 1 1 1 Information Information is Random CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Random variable CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Random Variable Random variable is a function For a single time Coin Tossing CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Our signal x(n) is a Random Variable For a series of Coin Tossing CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Coin tossing and Random Variable If random We have random variable X CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Random Digital Signal If the random variable is a function of time, it is called a stochastic process CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Probability Mass Function We need also to define the probability of each random variable CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Probability Mass Function PMF is for Discrete distribution function CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Time and Emsemble CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Probability of X(2) CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Probability Density Function PDF is for Continuous Distribution Function CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Probability Density Function PDF values can be > 1 as long as its area under curve is 1 2 1 1/2 1 CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Cumulative Distribution Function CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Expectation Operator CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Expected Value Expected value is known as the “Mean” CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Example of Expected Value (Discrete) We toss a die N times and get a set of outcomes Suppose we roll a die with N=6, we might get CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Example of Expected Value (Discrete) But, empirically we have Empirical (Monte Carlo) estimate as Expected Value CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Theoretical Expected Value But in theory, for a die CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Ensemble Average 1 ensemble i ensembles CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Ensemble Average CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
I) Linearity CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
II) CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
III) CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Autocorrelation CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Autocorrelation n=m CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Autocorrelation Matrix CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Covariance CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Stationarity (I) I) n1 n2 CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Stationarity (II) II) CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Expected Value of Error Energy Let’s take the expected value of error energy CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Vector-Matrix Differentiation CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Partial diff. and set to zero Differentiation Result: CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
2-D Error surface CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Four Basic Classes of Adaptive Signal Processing I) Identification II) Inverse Modelling III) Prediction IV) Interference Cancelling CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
The Four Classes of Adaptive Filtering CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
System Identification CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Inverse Modelling CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Prediction CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
Interference Canceller CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon
What are we looking for in Adaptive Systems? Rate of Convergence Misadjustment Tracking Robustness Computational Complexity Numerical Properties CESdSP EECS0712 Adaptive Signal Processing http://embedsigproc.wordpress.com/eecs0712 Assoc. Prof. Dr. P.Yuvapoositanon