ECE 6640 Digital Communications Dr. Bradley J. Bazuin Assistant Professor Department of Electrical and Computer Engineering College of Engineering and Applied Sciences Chapter 2 Chapter 2: Deterministic and Random Signal Analysis 17 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 2.10 Bandpass and Lowpass Signal Representation Signal Space Representation of Waveforms Some Useful Random Variables Bounds on Tail Probabilities Limit Theorems for Sums of Random Variables Complex Random Variables Random Processes Series Expansion of Random Processes Bandpass and Lowpass Random Processes Bibliographical Notes and References Problems 18 28 40 56 63 63 66 74 78 82 82 ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 2 Signal Definitions 1, 1 rect t t , 2 0, 1 2 1 t 2 otherwise t t 1, tri t t t t 1 t , 0, 1 t 0 0 t 1 otherwise t 0 1, sinct sin t t , ECE 6640 1, sgn t 0, 1, t0 t0 1, 1 u 1 t , 2 0, t 0 t0 t 0 t0 t0 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 3 Table of Fourier Transforms • Continuous Time and Frequency Fourier Transforms and Properties ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 4 Fourier Transforms Pairs ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 5 Lowpass Signal Representations • The Fourier Transform of a real signal has complex conjugate symmetry – the real part is symmetric about 0 – the imaginary part is anti-symmetric about 0 ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 6 Bandpass Signal Representations • The Fourier Transform of a real signal has complex conjugate symmetry – the real part is symmetric about 0 – the imaginary part is anti-symmetric about 0 • For real signals, the positive spectrum is sufficient to completely describe the spectral response. * X f X f X f X f X f ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 7 Lowpass Equivalent of Bandpass • Complex Signal Baseband representation • The “analytic signal” is the complex representation of a real signal – Based on the Hilbert Transform x t 1 1 1 1 1 xt j xˆ t xt j xt 2 2 2 2 t xl t 2 x t exp j 2 f 0 t xt Rexl t exp j 2 f 0 t ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 8 Errata • Equation 2.1-5 X l f 2 X f f 0 2 X f f 0 u1 f f 0 • Dr. Bazuin would prefer the equation to read – The complex lowpass equivalent has negative frequency components! Therefore, the step function is incorrect and should allow “all positive” frequencies! X l f 2 X f f 0 2 X f f 0 u 1 f ECE 6640 9 Complex Signal Representation • While a complex signal has a real and imaginary part, in communications we typically say: – The in-phase component for the real part – The quadrature-phase component for the imaginary part xl t xI t j xQ t xI t xt cos2 f 0 t xˆ t sin 2 f 0 t xQ t xt sin 2 f 0 t xˆ t cos2 f 0 t – Note also xt xI t cos2 f 0 t xQ t sin 2 f 0 t xˆ t xQ t cos2 f 0 t xI t sin 2 f 0 t ECE 6640 10 Envelope and Phase Representation • We are often interested in magnitude (envelope) and phase rx t xI t xQ t 2 2 x t j x t x t j x t * I Q I Q xQ t x t arctan x t I xl t rx t exp j x t xt Rerx t exp j 2 f 0 t j x t xt rx t cos2 f 0 t x t ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 11 Modulator: Frequency Up-conversion ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 12 Demodulators: Frequency Down-conversion ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 13 Signal Constructs • The RF signal is considered as a bandpass real signal in time. • Both the baseband signal to be modulated (up-converted) and demodulated (down-converted) baseband signal are often more useful as complex signals. – Up- or down-conversion is often a better description, as modulation and demodulation can be applied to signals at baseband before or after “frequency translation”. – By the way, we have so far skipped some very important structures, FILTERS!!! ECE 6640 14 Signal Energy Consideration • The energy of a signal is defined as EX xt dt 2 xt xt dt * • Based on Rayleigh’s Theorems, the Fourier Transform equivalence EX xt dt 2 X f df 2 • Also of note is Parseval’s Theorem xt , y t xt yt dt ECE 6640 * X f Y f df * 15 Signal Correlation • For two non identical signal, a correlation coefficient exists x, y xt , y t Ex E y • If two signal are “orthogonal” x, y 0 – Note that orthogonal baseband imply that bandpass signals will also be orthogonal … but not the other way around (as shown in Example 2.1.-1). ECE 6640 16 Filters • Since this chapter skips a discussion of filters … • See Dr. Bazuin’s filter notes. End Lecture 2 Begin Lecture 3 ECE 6640 17 Signal Space Representation • When transmitting digital symbols, – The symbols can be described in terms of an N-dimensional space with each individual symbol having a vector description. – The “set” of symbols form a “constellation” in the N-dimensional space. – In this space, we can describe the “distance” between the constellation points. This distance and the additive noise will define the probability that we can correctly detect the symbol. – Therefore, when describing symbols this way, you need to review and understand vector mathematics and relationships. ECE 6640 18 Vector Space Concepts • Vectors v v1 v2 vn v1 v11 t v12 v1n t column vectors • Inner Product (dot product) n v1 , v2 v1 v2 v1i v2i v2i v1i * H i 1 • Orthogonal v1 , v2 v1 v2 0 • Normal (Unit Magnitude) v1 • Orthonormal 2 n v1 v1 v1i v1i v1i v1i 1 * H i 1 – All the vectors in the set are (1) orthogonal and (2) normal. – Mathematically, we prefer using a “orthonormal basis set” ECE 6640 19 Vector Space Concepts • Linearly independent vector set – No one vector can be represented as a linear combination of the other vectors. – The vectors can be “created” from and orthonormal set of vectors. • If an orthonormal basis set has been defined … each vector can be defined based on a linear transformation from the basis set v Q v' – for A, a full rank matrix, A can be decomposed into a linear transformation and an orthonormal basis set. A = QR – Alternately, letting A v1 v2 v N Q R – Note: A is mxn, Q is mxn and R is nxn for rank(A)=n and m>=n ECE 6640 20 Gram-Schmidt Procedure • Shown in the textbook as Equ 2.2-10 to 2.2-14 A v1 v2 v N Q R q1 v1 v1 q2 ' v2 v2 , q1 q1 q2 ECE 6640 e2 eK I r1 v1 e1 r2 v2 , q1 e1 q2 ' e2 q2 ' q2 ' q3 ' v3 v3 , q1 q1 v3 , q2 q2 q3 e1 q3 ' q3 ' r3 v3 , q1 e1 v3 , q2 e2 q3 ' e3 Note: R is upper triangular QH Q I 21 Vector Inequalities • Triangle Inequality v1 v2 v1 v2 – Equality condition … orthogonal vectors • Cauchy-Schwarz Inequality v1 , v2 v1 v2 v1 v2 – more easily considered if the vectors are normalized … the dot product of two unit magnitude vectors can not be greater than 1. ECE 6640 22 Signal Space Concepts • vectors as a set of complex valued time samples … x1 t , x2 t x t x t dt 1 2 • signal norm (square root of energy) xt 2 xt xt dt * xt dt E x 2 • Orthogonal signals x1 t , x2 t x t x t dt 0 1 2 ECE 6640 23 Orthogonal Signal Set • Define and Orthonormal Basis Set 1, n t , m t n t m t dt 0, • Estimated (or defined signal) si ,n si t , n t , mn mn n 1,2, , K The image part with relationship ID rId6 was not found in the file. • Signal error K ei t si t sˆi t si t si ,k k t k 1 ECE 6640 24 Signal Set • Minimum mean square approximation error K ei t si t sˆi t si t si ,k k t k 1 Emin ei t si t dt * Emin K * si t si ,k k t si t dt k 1 mn mn 1, 0, n t , m t n t m t dt Emin ECE 6640 K K si t dt si ,k Es sk ,i 2 k 1 2 i 1 2 25 Orthogonal Basis Set • The Fourier Series Coefficients are orthonormal … 2 k t 2 k t s t ak cos bk sin T T k 0 • Set of functions for the interval [0,T] k t ECE 6640 k 1 exp j 2 t T T 26 Time Domain Gram-Schmidt • symbol waveforms can be decomposed into for description by an orthogonal -orthonormal basis set …. ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 27 Example 2.2-3 • Three orthonormal basis for the 4 signals can be defined. • The signals can then be described as vectors 2 ,0,0 0, 2 ,0 2 ,0,1 2 ,0,1 s1 s2 s3 s3 ECE 6640 t t t t Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 28 Signal Generation and Vector Analysis • Vectors and basis generators to form symbols • Orthogonal integrations to determine received symbol vectors ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 29 Example 2.2-4 • Vector representation of the signal space 2 ,0,0 0, 2 ,0 2 ,0,1 2 ,0,1 s1 • The vector magnitudes can be computed – and the energy derived Ek sk 2 • The Euclidean distances between symbols can also be computed s2 s3 s3 t t t t s1 2 s2 2 s3 3 s4 3 – defines how easy they are to detect … ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 30 Orthonormal Basis • In general, low pass signal can be represented by an Ndimensional complex vector, and the corresponding bandpass signal can be represented by 2N-dimensional real t vectors. sml sml1 , sml 2 , , smlN Baseband N sml t ss ln nl t , m 1,2, , M n 1 N sm t Re ss ln nl t exp j 2 f 0 t , n 1 m 1,2, , M Bandpass sm t Re ss ln nl t exp j 2 f 0 t , m 1,2, , M n 1 1 r i r i r i t sm sml1 , sml1 , sml 2 , sml 2 , , smlN , smlN 2 N ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 31 Example 2.2-6 • Describing M bandpass signal sm t Re Am g t exp j 2 f 0 t sml t Am g t n t 2 Renl t exp j 2 f 0 t t g t Eg 2 g t cos2 f 0 t Eg ˆt Bandpass 2 gˆ t g t sin 2 f 0 t E Eg g sml t Am E g t sm t Am g t cos2 f 0 t Am g t sin 2 f 0 t r i Bandpass symbols ECE 6640 32 Random Variables • Probability and Statistics Review • See Review Notes from ECE3800 – Cooper and McGillem Textbook – ECE3800_Review_1to8 – ECE3800_Review_9 ECE 6640 33 Communication RV: Bernoulli • Bernoulli Random Variable: Example flipping a coin S X 0,1 p1 p p0 1 p q m X EX p X 2 VARX p 1 p p q ECE 6640 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 34 Communication RV: Binomial (1) • Binomial Random Variable: Coin flip sequence S X 0,1,2, , n n k nk p k p 1 p k m X EX n p X 2 VARX n p 1 p ECE 6640 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 35 Communication RV: Binomial (2) • Binomial Random Variable: Example bit/symbol errors S X 0,1,2, , n n k nk p k p 1 p k m X EX n p X 2 VARX n p 1 p ECE 6640 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 36 Communication RV: Uniform • Uniform Random Variable: Received signal phase 1 , a xb fU x b a 0, x 0 and x b ba E U 2 ECE 6640 x0 0, x a , a xb FU x b a xb 1, 2 b a VARU 12 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 37 Communication RV: Gaussian/Normal • Gaussian Random Variable: Noise, Law of large number f X x x m exp 2 2 2 1 2 xm FX x EX m 1 2 xm t2 exp 2 dt VARX 2 • Normal Random Variable: zero mean and unit variance x t2 1 x exp dt 2 2 x2 1 exp x 2 2 xm FX x ECE 6640 1 2 xm t2 exp 2 dt Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 38 Communication RV: Gaussian/Normal • The Q-function is derived from the Gaussian – a single sided Gaussian tail integration Q x 1 x t2 exp 2 x 2 1 dt x t2 exp 2 2 1 x dt xm 1 F x Q ECE 6640 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 39 Matlab Functions • The error and complementary error functions are built-in to MATLAB. – see Q_fn.m and Q_fninv.m erf x 2 Q x x exp u 2 du u 0 1 x 1 erf 2 2 x X 1 FX x 1 1 erf 2 2 ECE 6640 erfc x 1 erf x Q x 1 x erfc 2 2 1 1 x X erf 2 2 2 x X 1 FX x 1 erfc 2 2 40 Communication RV: Rayleigh • Rayleigh Random Variable: Two dimensional Gaussian R f R r r2 , exp 2 2 2 0, r E R ECE 6640 E X E Y 0 X 2 Y2 2 for 0 r for r 0 FR r VARX VARY 2 r2 , 1 exp 2 2 0, for 0 r for r 0 VARR 2 2 2 Alberto Leon-Garcia, “Probability, Statistics, and Random Processes For Electrical Engineering, 3rd ed.”, Pearson Prentice Hall, 2008 41 More • Exponential – the modeling of the time between occurrence of events • Chi-Square – sum of squared Gaussian R.V. • Rician – two dimensional Gaussian, non-zero mean • Maxwell – three dimensional Gaussian, zero mean • Geometric – number of failures before the first success (coins or bits) • Poisson – in counting the number of occurrences of an event in a certain time period or in a certain region in space ECE 6640 42 Properties of RV: A Summary ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 43 Jointly Gaussian Random Variables • Multi-dimensional Gaussian R.V. where the vector values are not independent. ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 44 Jointly Gaussian, n=2 ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 45 Probability Bounds • Often in communications, the exact computation for performance analysis is very difficult or even “unreasonable”. • For these cases we seek to bound the performance, particularly for bit/symbol error rates. P X • Markov Inequality • Chernov Bound ECE 6640 46 Markov Inequality • The Markov inequality gives an upper bound on the tail probability of nonnegative random variables. Let us assume that X is a nonnegative random variable, i.e., p(x) = 0 for all x < 0, and assume α > 0 is an arbitrary positive real number. The Markov inequality states that: P X ECE 6640 EX Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 47 Chernov Bound • • The Chernov bound is a very tight and useful bound that is obtained from the Markov inequality. Unlike the Markov inequality that is applicable only to nonnegative random variables, the Chernov bound can be applied to all random variables. Let X be an arbitrary random variable, and let δ and ν be arbitrary real numbers (ν <>0). Define random variable Y by Y = eνX and constant α by α = eνδ. Obviously, Y is a nonnegative random variable and α is a positive real number. Applying the Markov inequality to Y and α yields Pe ECE 6640 vX e v E e vX v E e v X e 48 Limit Theorems Theorems related to the running average of random variables: • Law of Large Numbers: 1 n Xi E X j , n i 1 for E X j and X j i.i.d . • Central Limit Theorem: for X j i.i.d . and 2 VAR X j and E X j 1 n Xi m n i 1 N 0,1 Gaussian with zero mean and unit variance n ECE 6640 49 MATLAB Signals and SNR • Review Matlab for analog signal performance – Computing signal power for SNR – Generating and Applying Filters – Interpolation and decimation primitives used • interp function • decimation using (1:M:end) – Saving and Loading simulation data ECE 6640 50 Complex Random Variables • Complex numbers and variables can be expressed as a vector or matrix. Z X i Y ~ X Z Y ~ Z 1 i Z • Means and covariance can be described. E Z E X i E Y VARZ E Z E Z Z E Z H VARX VARY • If and when we need this stuff we will come back to it. ECE 6640 51 Wide Sense Stationary • The RV mean is note dependent on time. • And, the autocorrelation function is not dependent in time. E Z t Const. RZZ t1 , t 2 E Z t1 Z t 2 RZZ , for t 2 t1 RZZ E Z t Z t . • Numerous simplification in computations result, including: RZZ conj RZZ • Signal Power PZ E Z t RZZ 0 2 S f df ZZ ECE 6640 52 WSS More Properties • For two RV that are WSS RXY conj RYX • For Z=aX+bY RZZ a RXX a b* RXY a * b RYX b RYY 2 2 S ZZ f a S XX a b* S XY a * b SYX b SYY 2 2 S XY f conj SYX f S ZZ f a S XX 2 Re a b* S XY b SYY 2 ECE 6640 2 53 White Noise Process • For a noise process … from probability notes. RNN N0 2 N k T S NN f 0 2 2 – where k is Boltzmann’s constant, k=1.38 x 10-23 J/K (text error) and T is temperature in Kelvin. ECE 6640 54 Cyclostationary Random Process • The mean and autocorrelation are periodic functions. – example, a cosine wave • The autocorrelation based on the signal periodicity T can be described as RXX t T , t T RXX t , t – an average autocorrealtion is defined as T0 RXX 1 RXX t , t dt T0 0 – with the power spectral density S XX f RXX ECE 6640 55 Example 2.7-1: Random Symbols ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 56 2.7-3 p. 71 Errata • Text • Errata • Comment: when performing a covariance computation, the mean must be subtracted. The original would be OK if Z is a zero-mean process. ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 57 Markov Chains • Markov chains are discrete-time, discrete-valued random processes in which the current value depends on the entire past values only through the most recent values. In a j th order Markov chain, the current value depends on the past values only through the most recent j values, i.e., • There are numerous “differential encoding” schemes for communications where the “difference” between successive bits define the symbol transmitted. The encoding processes is a 1st order Markov chain. AS we study encoding and decoding of signals we will be studying trellis codes and trellis based signal detection. These are Morkov chains of higher orders. Therefore some background is required … • ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 58 Markov Chains (Cont) • In this book, the finite-state Markov chain is required. • The symbols received transition from one state to the next based on the symbol history and new symbols transmitted and received. • We will come back to this material at a alater date. ECE 6640 59 Sampling Theorem/Digital Processing • Continuous-time, continuous frequency X F xt exp j 2 F t dt xt X F exp j 2 F t dF • Discrete-time, continuous frequency X w xn exp j w n n xn 1 X w exp j w n dw 2 xn 1 N 1 k n X k exp j 2 N N k 0 • Discrete-time, discrete frequency N 1 k n X k xn exp j 2 N n0 ECE 6640 60 Sampling • Continuous Time Signal Relationships for aperiodic signal xa(t) X a F x a t xa t exp j 2 F t dt • • For Discrete Time Sampling X F • X F expi 2 F t df a x a n T x t t n t n a xn xa n T , for n x t t n T exp j 2 F t dt a n Note: as spectral replication has now occurred, due to sampling, we use X(F) instead of Xa(F) X F x t t n T exp j 2 F t dt a n • But the integral involving a delta function is just the value of the function at the time … X F x n T exp j 2 F n T n ECE 6640 X F a xn exp j 2 F n T n 61 Sampling (cont) • • The value of the function at the time … T Letting X F • • X F xn exp j 2 F n T n 1 Fs F n s xn exp j 2 F n 2 F F w f allowing or F F Results in a relationship between the sampling rate and the DTCF Fourier transform, which can then be written as s s X w xn exp j w n n Xf or xn exp j 2 f n n ECE 6640 62 Reconstruction x a t X F exp j 2 F t dF a • the band limiting can be accounted for as Fs x a t 2 X F exp j 2 F t dF a Fs 2 • using the relationship for a sampled time transform Fs x a t 2 X a F exp j 2 F t dF • we have Fs X F Fs 2 Fs x a t 2 Fs 2 X F k F k a s 1 X F exp j 2 F t dF Fs • and Fs x a t 2 Fs 2 ECE 6640 1 F xn exp j 2 n exp j 2 F t dF Fs n Fs 1 x a t xn Fs n Fs j 2 F t n dF exp F F s s 2 2 n sin Fs t Fs x a t xn n n Fs t Fs 63 Reconstruction (cont.) n sin Fs t Fs x a t xn n n Fs t Fs • From • or using the sampled period T sin t n T T t n T x a t xn xn sinc T n n t n T T • Therefore, for band-limited sampled signals, perfect reconstruction is performed as the sum of sinc functions centered on each of the sampled data points. x a t ECE 6640 xn sinc T t n T n 64 Reconstruction (cont.) x a t ECE 6640 xn sinc T t n T n Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 65 2.9 Bandpass and Lowpass Random Processes • In general, bandpass and lowpass random processes can be defined as WSS processes X(t) for which the autocorrelation function RX(τ ) is either a bandpass or a lowpass signal. – Recall that the autocorrelation function is an ordinary deterministic function with a Fourier transform which represents the power spectral density of the random process X(t). • • Therefore, for a bandpass process the power spectral density is located around frequencies± f0, and for lowpass processes the power spectral density is located around zero frequency. To be more specific, we define a bandpass (or narrowband) process as a real, zero mean, and WSS random process whose autocorrelation function is a bandpass signal. ECE 6640 66 Example 2.9-1 White Gaussian Noise ECE 6640 Notes and figures are based on or taken from materials in the course textbook: J.G. Proakis and M.Salehi, Digital Communications, 5th ed., McGraw-Hill, 2008. 67