EE445S Real-Time Digital Signal Processing Lab Spring 2014 Matched Filtering and Digital Pulse Amplitude Modulation (PAM) Slides by Prof. Brian L. Evans and Dr. Serene Banerjee Dept. of Electrical and Computer Engineering The University of Texas at Austin Lecture 14 Outline • Transmitting one bit at a time • Matched filtering • PAM system • Intersymbol interference • Communication performance Bit error probability for binary signals Symbol error probability for M-ary (multilevel) signals • Eye diagram 14 - 2 Transmitting One Bit • Transmission on communication channels is analog • One way to transmit digital information is called 2-level digital pulse amplitude modulation (PAM) x0 (t ) receive ‘0’ bit y0 (t ) ‘0’ bit input Tb t x(t) -A x1 (t ) A Tb ‘1’ bit t Additive Noise Channel output y(t) How does the receiver decide which bit was sent? Tb t -A receive ‘1’ bit y1 (t ) A Tb t 14 - 3 Transmitting One Bit • Two-level digital pulse amplitude modulation over channel that has memory but does not add noise x0 (t ) ‘0’ bit input Tb t x(t) -A x1 (t ) A Tb ‘1’ bit t output LTI Channel Model channel as LTI system with impulse response c(t) y(t) receive ‘0’ bit Th Th+Tb y1 (t ) receive ‘1’ bit t -A Th c (t ) A Th 1 Th y0 (t ) t Th Th+Tb Assume that Th < Tb 14 - 4 t Transmitting Two Bits (Interference) • Transmitting two bits (pulses) back-to-back will cause overlap (interference) at the receiver c (t ) x(t ) * A y (t ) = 1 Th+Tb 2Tb Tb t Th Tb t t -A Th ‘1’ bit ‘0’ bit Assume that Th < Tb ‘1’ bit ‘0’ bit • Sample y(t) at Tb, 2 Tb, …, and threshold with threshold of zero • How do we prevent intersymbol interference (ISI) at the receiver? Intersymbol interference 14 - 5 Preventing ISI at Receiver • Option #1: wait Th seconds between pulses in transmitter (called guard period or guard interval) c (t ) x(t ) * A y (t ) = 1 Th+Tb Tb Th+Tb t Th Th Tb t t -A Th ‘1’ bit ‘0’ bit Assume that Th < Tb ‘1’ bit ‘0’ bit Disadvantages? • Option #2: use channel equalizer in receiver FIR filter designed via training sequences sent by transmitter Design goal: cascade of channel memory and channel equalizer should give all-pass frequency response 14 - 6 Digital 2-level PAM System ak{-A,A} s(t) bi bits PAM Clock Tb g(t) pulse shaper x(t) c(t) h(t) Decision Sample at Maker w(t) matched AWGN N(0, N0/2) Transmitter y(t) y(ti) Channel t=iTb 1 0 bits Threshold l filter Clock T b p N lopt 0 ln 0 4 ATb p1 Receiver • Transmitted signal s(t ) ak g (t k Tb ) k • Requires synchronization of clocks p0 is the probability bit ‘0’ sent between transmitter and receiver 14 - 7 Matched Filter • Detection of pulse in presence of additive noise Receiver knows what pulse shape it is looking for Channel memory ignored (assumed compensated by other means, e.g. channel equalizer in receiver) g(t) Pulse signal x(t) h(t) y(T) y(t) t=T Matched filter w(t) Additive white Gaussian noise (AWGN) with zero mean and variance N0 /2 T is the symbol period y(t ) g (t ) * h(t ) w(t ) * h(t ) g0 (t ) n(t ) 14 - 8 Matched Filter Derivation • Design of matched filter Maximize signal power i.e. power of g 0 (t ) g (t ) * h(t ) at t = T Minimize noise i.e. power of n(t ) w(t ) * h(t ) • Combine design criteria max , where is peak pulse SNR T is the symbol period | g 0 (T ) |2 instantane ous power 2 E{n (t )} average power g(t) x(t) h(t) y(T) y(t) t=T Pulse signal w(t) Matched filter 14 - 9 Power Spectra • Deterministic signal x(t) w/ Fourier transform X(f) Power spectrum is square of absolute value of magnitude response (phase is ignored) Px ( f ) X ( f ) X ( f ) X * ( f ) 2 • Autocorrelation of x(t) Rx (t ) x(t ) * x* (t ) Maximum value (when it exists) is at Rx(0) Rx(t) is even symmetric, i.e. Rx(t) = Rx(-t) x(t) Multiplication in Fourier domain is convolution in time domain Conjugation in Fourier domain is reversal & conjugation in time X ( f ) X * ( f ) F x(t ) * x* (t ) 1 0 Rx(t) Ts t Ts -Ts Ts t 14 - 10 Power Spectra • Two-sided random signal n(t) Pn ( f ) F Rn (t ) Fourier transform may not exist, but power spectrum exists * Rn (t ) E n(t ) n (t t ) n(t ) n* (t t ) dt Rn (t ) E n(t ) n* (t t ) n(t ) n* (t t ) dt n(t ) * n* (t ) For zero-mean Gaussian random process n(t) with variance s2 Rn (t ) E n(t ) n* (t t ) 0 when t 0 2 * 2 P ( f ) s Rn (t ) E n(t ) n (t t ) s (t ) n • Estimate noise power spectrum in Matlab N = 16384; % finite no. of samples gaussianNoise = randn(N,1); plot( abs(fft(gaussianNoise)) .^ 2 ); approximate noise floor 14 - 11 Matched Filter Derivation g(t) Pulse signal x(t) N0 2 t=T w(t) Matched filter N0 E{ n (t ) } S N ( f ) df 2 AWGN 2 2 | H ( f ) | df j 2 H ( f ) G ( f ) e f t df j 2 H ( f ) G ( f ) e Matched filter G0 ( f ) H ( f )G( f ) | g 0 (T ) |2 | N0 | H ( f ) |2 2 • Signal g 0 (t ) g (t ) * h(t ) f S N ( f ) SW ( f ) S H ( f ) • Noise n(t ) w(t ) * h(t ) g 0 (t ) y(T) y(t) h(t) Noise power spectrum SW(f) fT df |2 T is the symbol period 14 - 12 Matched Filter Derivation • Find h(t) that maximizes pulse peak SNR | H ( f ) G( f ) e N0 2 j 2 f T df |2 a 2 | H ( f ) | df b • Schwartz’s inequality For vectors: T a b T * | a b | || a || || b || cos || a || || b || 2 For functions: ( x) 1 - * 2 ( x) dx ( x) 1 - 2 dx ( x) 2 2 dx - upper bound reached iff 1 ( x) k 2 ( x) k R 14 - 13 Matched Filter Derivation Let 1 ( f ) H ( f ) and 2 ( f ) G * ( f ) e j 2 f T | H ( f ) G ( f ) e j 2 f T df |2 | H ( f ) |2 df - | G ( f ) |2 df | H ( f ) G ( f ) e j 2 f T df |2 - N0 2 | H ( f ) |2 df 2 N0 | G ( f ) |2 df T is the symbol period 2 2 max | G ( f ) | df , which occurs when N 0 H opt ( f ) k G * ( f ) e j 2 f T k by Schwartz ' s inequality Hence, hopt (t ) k g * (T t ) 14 - 14 Matched Filter • Impulse response is hopt(t) = k g*(T - t) Symbol period T, transmitter pulse shape g(t) and gain k Scaled, conjugated, time-reversed, and shifted version of g(t) Duration and shape determined by pulse shape g(t) • Maximizes peak pulse SNR 2 Eb 2 2 2 2 max | G ( f ) | df | g (t ) | dt SNR N 0 N 0 N0 Does not depend on pulse shape g(t) Proportional to signal energy (energy per bit) Eb Inversely proportional to power spectral density of noise 14 - 15 Matched Filter for Rectangular Pulse • Matched filter for causal rectangular pulse shape Impulse response is causal rectangular pulse of same duration • Convolve input with rectangular pulse of duration T sec and sample result at T sec is same as First, integrate for T sec Sample and dump Second, sample at symbol period T sec Third, reset integration for next time period • Integrate and dump circuit h(t) = ___ t=nT T 14 - 16 Digital 2-level PAM System ak{-A,A} s(t) bi bits PAM Clock Tb g(t) pulse shaper x(t) c(t) h(t) Decision Sample at Maker w(t) matched AWGN N(0, N0/2) Transmitter y(t) y(ti) Channel t=iTb 1 0 bits Threshold l filter Clock T b p N lopt 0 ln 0 4 ATb p1 Receiver • Transmitted signal s(t ) ak g (t k Tb ) k • Requires synchronization of clocks p0 is the probability bit ‘0’ sent between transmitter and receiver 14 - 17 Digital 2-level PAM System • Why is g(t) a pulse and not an impulse? Otherwise, s(t) would require infinite bandwidth s(t ) ak (t k Tb ) k We limit its bandwidth by using a pulse shaping filter • Neglecting noise, would like y(t) = g(t) * c(t) * h(t) to be a pulse, i.e. y(t) = p(t) , to eliminate ISI y (t ) ak p(t kTb ) n(t ) where n(t ) w(t ) * h(t ) k y (ti ) ai p(ti iTb ) ak p(i k )Tb n(ti ) k ,k i actual value (note that ti = i Tb) intersymbol interference (ISI) p(t) is centered at origin noise 14 - 18 Eliminating ISI in PAM • One choice for P(f) is a rectangular pulse W is the bandwidth of the system Inverse Fourier transform of a rectangular pulse is is a sinc function p (t ) sinc (2 W t ) 1 2 W ,W f W P( f ) 0 , | f | W 1 f P( f ) rect ( ) 2W 2W • This is called the Ideal Nyquist Channel • It is not realizable because pulse shape is not causal and is infinite in duration 14 - 19 Eliminating ISI in PAM • Another choice for P(f) is a raised cosine spectrum 1 P( f ) 4W 1 2W 1 sin (| f | W ) 2W 2 f 1 0 0 | f | f1 f1 | f | 2W f1 2W f1 | f | 2W • Roll-off factor gives bandwidth in excess f1 1 W of bandwidth W for ideal Nyquist channel • Raised cosine pulse p(t ) sinc t cos2 2 W2 t 2 Ts 1 16 W t has zero ISI when Nyquist channel dampening adjusted by sampled correctly idealimpulse response rolloff factor • Let g(t) and h(t) be square root raised cosine pulses 14 - 20 Bit Error Probability for 2-PAM • Tb is bit period (bit rate is fb = 1/Tb) s(t) r(t) r(t) rn s(t ) ak g (t k Tb ) k h(t) Sample at Matched filter r (t ) s(t ) w(t ) t = nTb r(t) = h(t) * r(t) w(t) w(t) is AWGN with zero mean and variance s2 • Lowpass filtering a Gaussian random process produces another Gaussian random process Mean scaled by H(0) Variance scaled by twice lowpass filter’s bandwidth • Matched filter’s bandwidth is ½ fb 14 - 21 Bit Error Probability for 2-PAM • Noise power at matched filter output v(nT ) h(t ) w(nT t )dt Filtered noise T = Tsym 2 E v (nT ) E h(t ) w(nT t )dt 2 Noise power E{ gT (t 1 ) w(nT t 1 ) gT (t 2 ) w(nT t 2 )dt 1dt 2 } g T (t 1 ) gT (t 2 ) E{w(nT t 1 ) w(nT t 2 )}dt 1dt 2 s2 (t1–t2) 1 s h (t )dt s 2 2 2 2 sym / 2 H sym / 2 2 ( )d s2 T 14 - 22 Bit Error Probability for 2-PAM • Symbol amplitudes of +A and -A • Rectangular pulse shape with amplitude 1 • Bit duration (Tb) of 1 second • Matched filtering with gain of one (see slide 14-15) Integrate received signal over nth bit period and sample n 1 Prn (rn ) rn r (t ) dt n n 1 A w(t ) dt n A vn - A rn 0 A Probability density function (PDF) 14 - 23 Bit Error Probability for 2-PAM • Probability of error given that Tb = 1 transmitted pulse has amplitude –A vn A P(error | s (nTb ) A) P( A vn 0) P(vn A) P s s • Random variable PDF for vn / s vn s is Gaussian with zero mean and variance of one 0 PDF for N(0, 1) A/s 1 vn A P(error | s(nT ) A) P e s s A 2 v2 2 A dv Q s s Q function on next slide 14 - 24 Q Function • Q function 1 y2 / 2 Q( x) dy e 2 x • Complementary error function erfc 2 t 2 erfc ( x) e dt • Relationship x 1 x Q( x) erfc 2 2 Erfc[x] in Mathematica erfc(x) in Matlab 14 - 25 Bit Error Probability for 2-PAM • Probability of error given that transmitted pulse has amplitude A Tb = 1 P(error | s(nTb ) A) Q( A / s ) • Assume that 0 and 1 are equally likely bits P (error) P ( A) P (error | s (nTb ) A) P ( A) P (error | s (nTb ) A) 1 A 1 A A Q Q Q Q 2 σ 2 σ σ A2 2 where, SNR 2 1 e s Q( ) • Probability of error exponentially 2 decreases with SNR (see slide 8-16) for large positive 14 - 26 PAM Symbol Error Probability • Set symbol time (Tsym) to 1 second • Average transmitted signal power 1 PSignal E{an2 } 2 | GT ( ) |2 d E{an2 } 3d d d -d d GT() square root raised cosine spectrum • M-level PAM symbol amplitudes M M i 1, . . . , 0, . . . , 2 2 li d (2i 1), • With each symbol equally likely 1 PSignal M M 2 2 l M M i 1 2 i M 2 d2 d (2i 1) ( M 1) 3 i 1 2 3 d 2-PAM 4-PAM Constellation points with receiver decision boundaries 2 2 14 - 27 PAM Symbol Error Probability • Noise power and SNR PNoise 1 2 sym / 2 sym / 2 N0 N0 d 2 2 two-sided power spectral density of AWGN SNR PSignal PNoise 2( M 2 1) d 2 3 N0 • Assume ideal channel, i.e. one without ISI rn an vn channel noise after matched filtering and sampling • Consider M-2 inner levels in constellation Error only if | vn | d where s 2 N 0 / 2 Probability of error is d P(| vn | d ) 2 Q s • Consider two outer levels in constellation d P ( vn d ) Q s 14 - 28 PAM Symbol Error Probability • Assuming that each symbol is equally likely, symbol error probability for M-level PAM Pe M 2 d 2 Q M s M-2 interior points 2 d 2( M 1) d Q Q M M s s 2 exterior points • Symbol error probability in terms of SNR 1 2 PSignal M 1 3 d2 2 Pe 2 Q 2 SNR since SNR M 1 2 M PNoise 3s M 1 14 - 29 Visualizing ISI • Eye diagram is empirical measure of signal quality g (nTsym kTsym ) x(n) ak g (n Tsym k Tsym ) g (0) an ak g (0) k k k n • Intersymbol interference (ISI): D ( M 1) d g (nTsym kTsym ) k , k n g (0) ( M 1) d g (kTsym ) k , k n g (0) Raised cosine filter has zero ISI when correctly sampled 14 - 30 Eye Diagram for 2-PAM • Useful for PAM transmitter and receiver analysis and troubleshooting Sampling instant M=2 Margin over noise Distortion over zero crossing Slope indicates sensitivity to timing error Interval over which it can be sampled t - Tsym t t + Tsym • The more open the eye, the better the reception 14 - 31 Eye Diagram for 4-PAM 3d d -d -3d Due to startup transients. Fix is to discard first few symbols equal to number of symbol periods in pulse shape. 14 - 32