ECE 6640 Digital Communications Dr. Bradley J. Bazuin Assistant Professor Department of Electrical and Computer Engineering College of Engineering and Applied Sciences Chapter 4 Chapter 4: 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 4.10 4.11 ECE 6640 Optimum Receivers for AWGN Channels Waveform and Vector Channel Models Waveform and Vector AWGN Channels Optimal Detection and Error Probability for Band-Limited Signaling Optimal Detection and Error Probability for Power-Limited Signaling Optimal Detection in Presence of Uncertainty: Noncoherent Detection A Comparison of Digital Signaling Methods Lattices and Constellations Based on Lattices Detection of Signaling Schemes with Memory Optimum Receiver for CPM Signals Performance Analysis for Wireline and Radio Communication Systems Bibliographical Notes and References Problems 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. 160 160 167 188 203 210 226 230 242 246 259 265 266 2 Waveform and Vector Channel Models • The additive white Gaussian noise (AWGN) channel model is a channel whose sole effect is addition of a white Gaussian noise process to the transmitted signal. r t sm t nt vl t ECE 6640 I m m g t m T r t I m m g t m T nt 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 Symbol Detection • The receiver observes the received signal r (t) and, based on this observation, makes the optimal decision about which message m, 1 ≤ m ≤ M, was transmitted. • An optimal decision rule results in minimum error probability or “disagreement” between the transmitted message m and the detected message . Pe Pmˆ m 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 Optimal Detection Assumption • Oservation prompts us to view the waveform channel r(t)=sm(t)+n(t) in the vector form r = sm + n where all vectors are N-dimensional and components of n are i.i.d. zero-mean Gaussian random variables with variance N0/2. r t sm t nt • If we assume optimal signal processing and sampling z t sm t hc t hs t nt hs t r t s m t hc t n t z n T z t t n T s m t r sm n n t r sm 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. 5 Baseband Vector • The mathematical model for the AWGN vector channel is r sm n given by • We study a more general vector channel model in this section which is not limited to the AWGN channel model. • In our model, vectors sm are selected from a set of possible signal vectors {sm, 1 ≤ m ≤ M} according to prior or a priori probabilities Pm and transmitted over the channel. The received vector r depends statistically on the transmitted vector through the conditional probability density functions p(r|sm). 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 Decision Function • The receiver observes r and based on this observation decides which message was transmitted. • The decision function determines and estimate of the symbol m. g r mˆ • The probability of a correct decision becomes. Pcorrect _ decision | r Pmˆ sent | r Pcorrect _ decision Pmˆ sent | r pr dr • Our goal is to design an optimal detector that minimizes the error probability or, equivalently, maximizes P [correct decision]. 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 Maximization Pcorrect _ decision Pmˆ sent | r pr dr • Maximization is equivalent to maximizing the probability term Pmˆ sent | r mˆ g opt r arg max Pm | r 1 m M • The optimal detection scheme described simply looks among all P [m |r ] for 1 ≤ m ≤ M and selects the m that maximizes P [m |r ]. The detector then declares this maximizing m as its best decision. 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 Equivalent Maximum • Note that since transmitting message m is equivalent to transmitting sm, the optimal decision rule can be written as either mˆ g opt r arg max Pm | r 1 m M mˆ g opt r arg max Psm | r 1 m M 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. 9 MAP and ML Receivers • Detection methods based on probability and estimation of the symbol sent – MAP: maximum a-posteriori probability – ML: maximum likelihood • MAP detection mˆ g opt r arg max 1 m M Pm p r | sm pr mˆ g opt r arg max Pm pr | sm 1 m M – as p(r) is independent of m, the second function is valid – Pm is the a-priori probability of m • For equally probably messages: mˆ g opt r arg max p r | sm 1 m M – this is referred to as the maximum-likelihood 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. 10 Decision Regions • Any detector partitions the output space RN into M regions denoted by D1, D2, . . . , DM such that if r ∈ Dm, then = g(r) = m, i.e., the detector makes a decision in favor of m. • The region Dm, 1 ≤ m ≤ M, is called the decision region for message m; and Dm is the set of all outputs of the channel that are mapped into message m by the detector. • If a MAP detector is employed, then the Dm’s constitute the optimal decision regions resulting in the minimum possible error probability. 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 Error Probability • An error occurs when the received r is not in Dm. • The symbol error probability is given by M M m 1 m 1 Pe Pm Pr Dm | sm Pm Pe|m M Pe Pm m 1 pr | s dr 1 m ' M Dm ' m ' m m • This equation gives the probability that an error occurs in transmission of a symbol or a message and is called symbol error probability or message error probability. 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 Preprocessing • The concepts defined are applicable whether “preprocessing” is performed or not. – Preprocessing in fact is usually, if not always, performed. The signal will always be filtered based on the required signal bandwidth prior to detection. ECE 6640 13 Preprocessing (cont) • Preprocessing forms – Noise power reduction filtering. Goals: (1) reduce noise power, (2) minimize any signal effects. – Matched filter for signal of interest. Goal: (1) maximize signal power received for detection, (2) filter the noise power. – Whitening filter. Goal: (1) “Whiten” the signal spectrum to remove interference (make the noise power flat) • Critical filter considerations – How does the filter change the spectral response of the signal-ofinterest? Does it cause ISI or signal power loss? – What is the noise-equivalent bandwidth of the filter? N0Beq noise power. ECE 6640 14 Developing Error Rates • The following slides are going to follow an alternate methodology and derivations. • It looks more specifically at bit detection of a binary signal in Gaussian noise. – A non-vector approach prior to the more theoretical vector approach taken in the textbook. – The following can be reviews in either Carlson or Sklar textbooks. ECE 6640 15 Baseband Binary Receiver Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Figure 11.2-1 yt a k pt kT h t n in t h t k y t k ak n t k • Synchronous Time sampling of maximum filter output Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 16 Regeneration of a unipolar signal Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. (a) signal plus noise (b) S/H output (c) comparator output: Figure 11.2-2 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 17 Unipolar NRZ Binary Error Probability • Hypothesis Testing using a voltage threshold – Hypothesis 0 • The conditional probability distribution expected if a 0 was sent pY yk | H 0 pY ak n t k | ak 0 pY n t k pY yk | H 0 p N yk – Hypothesis 1 • The conditional probability distribution expected if a 1 was sent pY yk | H1 pY ak n t k | ak A pY A n t k pY yk | H1 p N yk - A Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 18 Decision Threshold and Error Probabilities Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Conditional PDFs Figure 11.2-3 y t k ak n t k V Pe1 PY V p Y y | H1 dy Pe 0 PY V p Y y | H 0 dy V • Use Hypothesis to establish a decision rule – Use threshold to determine the probability of correctly and incorrectly detecting the input binary value Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 19 Average Error Probability • Using the two error conditions: – Detect 1 when 0 sent – Detect 0 when 1 sent Perror PH 0 Pe 0 PH1 Pe1 • Selecting an Optimal Threshold PH p V | H PH p V 0 Y opt 0 1 Y opt | H1 • For equally likely binary values PH 0 PH1 1 2 Perror 1 Pe 0 Pe1 2 pY Vopt | H 0 pY Vopt | H1 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 20 Threshold regions for conditional PDFs Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Figure 11.2-4 P H 0 P H1 1 2 Vopt A 2 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 21 For AWGN • The pdf is Gaussian y2 exp p Y y | H 0 p N y 2 2 2 2 1 for 2 1 Q x exp d 2 x 2 V A Pe 0 PY V p N y dy Q Q 2 V V for AV A Pe1 PY V p N y A dy Q Q 2 P H 0 P H 1 1 2 Vopt A 2 A V A Pe1 Q Q Pe 2 2 22 Modification for Polar NRZ Signals (+/- A/2) • Hypothesis Testing using a voltage threshold – Hypothesis 0 • The conditional probability distribution expected if a 0 was sent A A pY yk | H 0 pY ak n t k | ak pY n t k 2 2 A pY yk | H 0 p N yk 2 – Hypothesis 1 • The conditional probability distribution expected if a 1 was sent A A pY yk | H1 pY ak n t k | ak pY n t k 2 2 A pY yk | H1 p N yk - 2 Vopt A A 0 2 2 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 23 Modification for Polar NRZ Signals (+/- A/2) • Determining the error probability A V A Pe 0 PY V p N y dy Q 2 2 V A V A Pe1 PY V p N y dy Q 2 2 V Q A 2 Q A 2 • Notice that the error is the same as Unipolar NRZ – The distance between the expected signal values is the same Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 24 Modification for Bipolar NRZ Signals (+/- A) • Hypothesis Testing using a voltage threshold – Hypothesis 0 • The conditional probability distribution expected if a 0 was sent pY yk | H 0 pY ak n t k | ak A pY A n t k pY yk | H 0 p N yk A – Hypothesis 1 • The conditional probability distribution expected if a 1 was sent pY yk | H1 pY ak n t k | ak A pY A n t k pY yk | H1 p N yk - A Vopt A A 0 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 25 Modification for Bipolar NRZ Signals • Determining the error probability AV A Pe 0 PY V p N y A dy Q Q V V AV A Pe1 PY V p N y A dy Q Q • Notice that the error has been reduced – The distance between the expected signal values may be twice as large as the unipolar case (using +/- A) Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 26 Relationship to signal power • Defining the average received signal power – Unipolar NRZ – Polar NRZ SR SR – Bipolar NRZ 1 2 A , 2 1 2 A , 4 S R A2 , 0, A A A 2 , 2 T2 2 1 S R E lim xc t dt T T T 2 A, A • In terms of SNR 1 S 2 2 N 2 A R A 4 N R S 2 N R for Unipolar for Polar 2 A 2 S A N R N R for Bipolar 27 Probability of error • The probability of detecting a transmitted symbol correctly is dependent upon the received signal-to1 noise ratio …. assuming PH PH 0 – Unipolar NRZ (orthogonal) 1 S A Pe Q Q 2 2 N R 1 2 2 A2 1 S A 4 NR 2 N R 2 – Polar NRZ (antipodal) S A Pe Q Q 2 N R 2 A2 A S 4 N R N R 2 – Bipolar NRZ (antipodal) S A Pe Q Q N R 2 A2 S A N R N R 28 Power Ratio vs. Bit Energy • For continuous time signals, power is a normal way to describe the signal. • For a discrete symbol, the “power” is 0 but the energy is non-zero – Therefore, we would like to describe symbols in terms of energy not power • For digital transmissions how to we go from power to energy? – Power is energy per time, but we know the time duration of a bit. Noise has a bandwidth. 1 S R Eb Tb N R N 0 W SR S ? N R N R 29 SNR to Eb/No • For the Signal to Noise Ratio – SNR relates the average signal power and average noise power (Tb is bit period, W is filter bandwidth) 1 Eb R b Tb E b 1 S N N 0 W N 0 Tb W N 0 W Eb – Eb/No relates the energy per bit to the noise energy (equal to S/N times a time-bandwidth product) Eb S W S Tb W N0 N R b N 30 Relationship to Eb/No • Defining the energy per bit to noise power ratio for a time-bandwidth product of W T R T 1 b b 2 b 2 2 – Unipolar Eb A2 1 S A 4 N R 2 N R N0 2 – Polar 2 Eb A2 A S 4 N R N R N0 2 – Bipolar 2 2 2 Eb A2 S A N R N R N0 31 Relationship to Bit Error Probability • Defining the binary bit error probability 1 for a time-bandwidth product PH PH 0 – Unipolar 1 2 Eb A Perror Q Q 2 N0 2 Eb A Q Q 2 N0 – Polar Perror – Bipolar 2 Eb A Perror Q Q N0 32 Bit Error Rate Plot Classical Bit Error Rates 0.5 Orthogonal Antipodal 0.45 0.4 Bit Error Rate 0.35 EbNo=(0:10000)'/1000; 0.3 % Q(x)=0.5*erfc(x/sqrt(2)) 0.25 Ortho=0.5*erfc(sqrt(EbNo)/sqrt(2)); 0.2 Antipodal=0.5*erfc(sqrt(2*EbNo)/sqrt(2)); 0.15 semilogx(EbNo,[Ortho Antipodal]) 0.1 ylabel('Bit Error Rate') xlabel('Eb/No') 0.05 title('Classical Bit Error Rates') legend('Orthogonal','Antipodal') 0 -3 10 10 -2 -1 10 Eb/No 10 0 10 1 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 33 BER Performance, Classical Curves log-log plot 10 10 Bit Error Rate 10 10 10 10 10 10 Classical Bit Error Rates 0 Orthogonal Antipodal -1 -2 -3 -4 -5 -6 -7 -1 0 1 2 3 4 5 6 7 8 Eb/No 9 10 11 12 13 14 15 34 Antipodal and Orthogonal Signals • Antipodal – Distance is twice “signal voltage” – Only works for one-dimensional signals d 2 Eb T 1 1 zij si t s j t dt E 0 1 for i j for i j • Orthogonal – Orthogonal symbol set – Works for 2 to N dimensional signals d 2 Eb T 1 1 z ij s i t s j t dt E 0 0 for i j for i j 35 Returning to Textbook Notes ECE 6640 36 4.2 AWGN Channels • This sections proves that the noise contribution to the sampled signal vector has the expected properties • The noise is zero mean (Eq. 4.2-10) • The Noise covariance is described as (Eq. 4.2-11) N 0 Cov ni n j 2 0 for i j for i j • Noise is uncorrelated with other noise components (Eq. 4.2-12) Covni n2 t 0 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. Optimal Detection • Based on the MAP detector and AWGN noise mˆ arg maxPm pr | sm 1 m M for r sm n mˆ arg maxPm pn r sm 1 m M 1 mˆ arg max Pm N 1 m M 0 r sm exp N0 N 2 This function should be computed for all M symbols and the “arg max” determined in order to make a detection decision. 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. 38 Optimal Detection (cont.) 1 mˆ arg max Pm N 1 m M 0 r sm exp N0 N 2 • As there are elements of this computation that are identical for every potential symbol, the text “drops” the common elements to focus on what is unique for different symbols. 2 r sm mˆ arg max ln Pm N 0 1 m M 1 2 N mˆ arg max 0 ln Pm r sm 2 1 m M 2 1 N mˆ arg max 0 ln Pm Em r sm 2 1 m M 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. 39 Detection Interpretations 1 N0 mˆ arg max ln Pm Em r sm 2 1 m M 2 • A symbol decision region based on terms related to noise and symbols N0 1 nm lnPm Em 2 2 • For equally probably symbols, a simplified detection A nearest-neighbor or criteria mˆ arg min r sm 1 m M minimum distance detection – the smallest distance between the received values one of M possible 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. 40 Decision Regions 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. 41 Binary Antipodal Signaling • In a binary antipodal signaling scheme s1(t) = s(t) and s2(t) = −s(t). • The probabilities of messages 1 and 2 are p and 1 − p, respectively. s1 Es s2 E s • Es is energy in each signal and is equal to Eb. • The decision regions based on MAP becomes N N 1 1 D1 r : r Eb 0 ln p Eb r Eb 0 ln 1 p Eb 2 2 2 2 1 p N 0 1 p N0 r : r ln ln D1 r : r 2 Eb 2 4 Eb p p 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. 42 Binary Antipodal Signaling • Defining a detection threshold rth D1 r : r rth 1 p N0 ln 4 Eb p and D2 r : r rth • For equally probable symbols, p=0.5 and ECE 6640 rth 0 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 Bit Error Probability • To derive the error probability for this system, use Eq. 4.1–15. M Pe Pm m 1 pr | s dr 2 Pe Pm m 1 m 1 m ' M Dm ' m ' m pr | s dr m 1 m ' 2 Dm ' m ' m Pe p p r | sm Eb dr 1 p p r | sm Eb dr D2 rth Pe p pr | s D1 m Eb dr 1 p p r | sm Eb dr rth • Assuming AWGN N N Pe p P N Eb , 0 rth 1 p P N Eb , 0 rth 2 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. 44 Bit Error Probability (cont) N N Pe p P N Eb , 0 rth 1 p P N Eb , 0 rth 2 2 • Recognizing the tail of the Gaussians E r E r b th b th 1 p Q Pe p Q N0 N0 2 2 • for equally probably symbols p=-.5 2 Eb Pe Q N0 ECE 6640 45 Error Based on Symbol Distance • For equiprobable symbols, the regions are separated by the perpendicular bisector of s1 and s2. • By symmetry, error probabilities are equal, therefore Pb = P[error |s1 sent ]. d12 s2 s1 n r s1 ECE 6640 n s2 s1 d12 Pb P 2 d12 d12 2 2 d 2 d12 12 Q 2 Pb P n s2 s1 Q 2 N0 2 N0 d12 2 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. 46 Error Based on Symbol Distance (cont) • For the distance defined as d12 s2 s1 2 d12 s2 t s1 t 2 2 d12 s t dt 2 2 2 s1 t s2 t s1 t dt 2 s2 t dt 2 s1 t s2 t dt s1 t dt 2 2 d12 E1 E2 2 s1 t , s2 t 2 • Based on equal symbol energy and the cross correlation d12 2 Eb 2 Eb 2 Eb 1 2 1 1 ECE 6640 47 Error Based on Symbol Distance (cont) • With distance defined as d12 2 Eb 1 2 d 2 12 Pb Q 2 N0 Q Eb 1 N0 – For antipodal signals =-1 – For orthogonal signals =0 ECE 6640 2 Eb Pb Q N0 Eb Pb Q N0 48 Bit Error Rate Probability • The reference comparisons for all digital communication systems. – The focus on Eb/N0 – The interest in antipodal or orthogonal 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. 49 The Correlation Receiver • A receiver typically operates on the time domain waveform r(t). Therefore computation of the time domain is needed based on options. – (1) Resolve the basis components to determine the r vector. r j r t j t dt , 1 j N – (2) Directly correlate r(t) with every possible sm(t) r , sm r t sm t dt , 1 m M mˆ arg maxnm r , sm 1 m M ECE 6640 50 Correlation with Basis Set r j r t j t dt , ECE 6640 1 j N mˆ arg maxnm r , sm 1 m M 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. 51 Correlation with Signal r , sm r t sm t dt , ECE 6640 1 m M mˆ arg maxnm r , sm 1 m M 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. 52 Matched Filter • Matched Filter hm t u t sm L T t * t z m t r t hm t r hm t d • For the correct “matched” received symbol m t z m t sm sm L T t d * 0 z m L T LT sm sm L T L T d 0 * LT sm sm d 0 * LT sm d 2 0 Note: LT used for filter length. ECE 6640 53 Optimum binary detection Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. (a) parallel matched filters (b) correlation detector: Figure 14.2-3 Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 54 Frequency Domain Matched Filter • The equivalent noise bandwidth of the “matched filter” must be known. Note: Single symbol SNR, not a detection region 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. 55 Bandpass binary receiver Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Figure 14.2-1 • Using superposition of the “parallel matched filters”, the BPF is the difference of the two filters. hBPF t h1 t h0 t • This results in an optimal binary detector Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 56 Binary Receiver hBPF t h1 t h0 t • OOK h1 t K s1 T t h0 t K s0 T t hBPF t h1 t K s1 T t cos2 f c T t • BPSK hBPF t h1 t h0 t 2 h1 t 2 K s1 T t cos2 f c T t • BFSK hBPF t h1 t h0 t K s1 T t K s0 T t hBPF t cos2 f c f d T t cos2 f c f d T t hBPF t 2 sin 2 f c T t sin 2 2 f d T t 57 Correlation receiver for OOK or BPSK Copyright © The McGraw-Hill Companies, Inc. Permission required for reproduction or display. Figure 14.2-4 • Since both optimal filters consist of cosine waveforms, mix and integrate instead of filter an optimally sample. – Note that the integrator can be a rectangular window filter that is optimally sampled. (Provides functionality near synchronization as well.) Notes and figures are based on or taken the course textbook: A. Bruce Carlson, P.B. Crilly, “Communication Systems, 5th ed.”, McGraw-Hill, 2010. 58 Optimal Parallel Matched Filter Receiver Error Analysis 2 z1 z0 2 max T 2 E s1 t s0 t dt 0 2 N0 • Evaluating the expected value T T T T 2 2 2 E s1 t s0 t dt E s1 t dt 2 E s1 t s0 t dt E s0 t dt 0 0 0 0 T 2 E s1 t s0 t dt E1 2 E10 E0 0 Eb E1 E0 2 2 2 Eb 2 E10 Eb E10 z1 z0 2 2 N N0 max 0 59 Optimal Parallel Matched Filter Receiver Error Analysis E10 Eb T Eb E s1 t s0 t dt E1 E0 0 2 • OOK • PSK • FSK E10 0 Eb z1 z0 2 max N 0 E10 1 Eb 2 Eb z1 z0 N0 2 max E10 0 2 2 E z1 z0 b 2 max N 0 60 Generalized Probability of Error • Using the optimal BPF filter and sampling for each symbol, the relationship will be based on: Eb E10 Eb 1 z1 z0 N0 N0 2 max 2 • The BER is then based on E 1 z z Pe Q 1 0 Q b N 2 0 • Therefore picking arbitrary symbols is possible, but the symbol correlation coefficient will drive the BER performance. 61 Generalized FSK s0 t Ac cos2 f c f d t s1 t Ac cos2 f c f d t T E10 Ac E cos2 f c f d t cos2 f c f d t dt 0 2 T A E10 c cos2 2 f c t cos2 2 f d t dt 2 0 2 T E 1 E10 b expi 2 2 f d t exp i 2 2 f d t dt T 2 0 Eb Eb 1 expi 2 2 f d T exp i 2 2 f d T T 2 i 2 2 f d i 2 2 f d f Eb sin 2 2 f d T Eb sinc4 f d T Eb sinc 4 d T 2 2 f d rb k 2 f d f step There are multiple “orthogonal” tone separations. 2T Eb • • The correlation coefficient can go negative! The minimum occurs at approximately sinc(1.22) = -0.166 62 MATLAB Coherent Receivers • BASK example code • BPSK example code • BFSK example code ECE 6640 63 4.2-3 Bounds on Probability of Error • For multiple symbol constellation, it is important to be able to estimate and/or bound the expected BER performance. • The text approach is to develop a union bound of error for maximum likelihood detection. 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. 64 Union Bound (cont) • For equally likely symbols • In AWGN • For very few constellations, decision regions Dm’ are regular enough that the integrals in the last line of Equation 4.2–63 or Equation 4.2–62 can be computed in a closed form. For most constellations these integrals cannot be put in a closed form. In such cases it is convenient to have upper bounds for the error probability. • 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. 65 Union Bound (cont) • • There exist many bounds on the error probability under ML detection. The union bound is the simplest and most widely used bound which is quite tight particularly at high signal-to-noise ratios. First we note that in general the decision region Dm’ under ML detection can be expressed as Note: m are true symbols, m’ are ‘generalized’ symbol (worse case?) • Define • Note that Dmm’ is the decision region for m’ in a binary equiprobable system with signals sm and sm’. Note that the decision regions of Dm’ are a subset of Dmm’. • Therefore 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. 66 Union Bound (cont) • For • • Note that the right-hand side of this equation is the error probability of a binary equiprobable system with signals sm and sm’ when sm is transmitted. We define the pairwise error probability, denoted by Pm→m’ as • The derivation proceeds as 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 Union Bound (cont) • Equations 4.2–70 is the union bound for a general communication channel. • In the special case of an AWGN channel, we know from Equation 4.2–37 that the pairwise error probability is given by • • For Q bounded as This becomes 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. 68 Union Bound (cont) • • The derived union bound is uses the distance between symbols in the constellation to determine the bit error bound. While a more exact expression will use specific “enumerated” distances, approximations may either use the minimum distance or functions base on the enumerated distances. Let us define dmin, the minimum distance of a constellation, as • Substituting in Equation 4.2–70 results in • Equation 4.2–78 is a looser form of the union bound in terms of the Q function and dmin which has a very simple form. • 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. 69 Union Bound (cont) • The union bound clearly shows that the minimum distance of a constellation has an important impact on the performance of the communication system. A good constellation should be designed such that, within the power and bandwidth constraints, it provides the maximum possible minimum distance; i.e., the points in the constellation should be maximally separated. • In Summary • If Q is approximated as • This becomed 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. 70 Example 4.2-2: 16 QAM Close observation of this constellation shows that from a total of 16 × 15 = 240 possible distances between any two points in the constellation, 48 are equal to dmin, 36 are equal to √2 dmin, 32 are 2dmin, 48 are √5 dmin, 16 are √ 8 dmin, 16 are 3dmin, 24 are √ 10 dmin, 16 are √13 dmin, and finally 4 are √18 dmin. Note that each line connecting any two points in the constellation is counted twice. Taking just the minimum distance (term 1 in T(X)) assuming a high enough SNR to ignore the other terms and using the Q(x) approx, 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. 71 Example 4.2-2: 16 QAM BER vs. Eb/No The bound derived was The dmin bound previoulsy derived as a simplified form of the union bound The exact expression (derived in Example 4.3-1) is 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. 72 Probability Lower Bound • See the derivation in the textbook. • The result suggested is – where Nmin is the number of the points in the constellation that are at the distance from dmin from at least one other point in the constellation. • If we take the conservative upper bound and this lower bound, the probability of error is bounded as 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. 73 4.3 Optimal Detection and Error Probability for Band-Limited Signaling • These are for “lower bandwidth”, low dimensionality signaling types. • This section is an excellent reference for some of the primary signal types discussed. • Explicit BER vs. Eb/No equations are derived based on the previous material presented. – An assumption of equally likely symbols is made for each derivation. – 4.3-1 Derives ASK or PAM – 4.3-2 Derives MPSK – 4.3-3 Derives QAM 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. 74 MASK 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. 75 MASK 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. 76 MASK Performance 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. 77 MPSK • The marginal probability density function for a symbol can be defined as • The pdf is a function of the average symbol energy • The higher the number of symbols, the tighter the symbol decision regions must become and more errors can be expected. 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. 78 MPSK • In general, the integral of p(θ) does not reduce to a simple form and must be evaluated numerically, except for M = 2 and M = 4. • For M=2 • For M=4 • For other M (M large and SNR large) 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. 79 MPSK M=2 M=4 M other 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. 80 QAM QAM is dependent upon the symbol constellation selected. • Default to square constellations of 4, 16, 64, & 256 • Numerous others are possible with potentially better system performance • The optimal detector uses 2 basis symbols to resolve the in-phase and quadrature components 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. 81 Square Constellation QAM • This case appears as two dimensional ASK/PAM 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. 82 Square Constellation QAM 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. 83 Comparing QAM and MPSK • Looking at the ratio of the Q(x) arguments – At M=4 the systems are equivalent, but for higher M QAM has better performance. 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. 84 Demodulation and Detection of BandLimited Signals • Matched filter involve the basis form of the signals. Note: Filter are matched to basis, not matched to 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. 85 4.4 Optimal Detection and Error Probability for Power-Limited Signaling • These are for “wider bandwidth”, higher dimensionality signaling types. • BER vs. Eb/No equations are derived based on the previous material presented. – An assumption of equally likely symbols is made for each derivation. – 4.4-1 Orthogonal FSK – 4.4-2 Biorthogonal – 4.4-3 Simplex ECE 6640 86 Orthogonal Signals - MFSK • For equiprobable, equal-energy orthogonal signals, the optimum detector selects the signal resulting in the largest cross-correlation between the received vector r and each of the M possible transmitted signal vectors {sm}, i.e., d min 2 E ECE 6640 87 Orthogonal Signals (cont) • The probability of correct symbol detection can be described as • assuming Gaussian noise elements where the elements are independent and identically distributed (IID) • with an individual dimension represented as ECE 6640 88 Orthogonal Signals (cont) • The integral becomes • The error is the complement, therefore • In general, Equation 4.4–10 cannot be made simpler, and the error probability can be found numerically for different values of the SNR. To determine bit errors, let us assume that s1 corresponds to a data sequence of length k with a 0 at the first component. The probability of an error at this component is the probability of detecting an sm corresponding to a sequence with a 1 at the first component. Since there are 2k−1 such sequences, we have • ECE 6640 89 FSK Signaling Another Union Bound • The orthogonal case is easier than the previous derivation as all symbols have the minimum distance. Taking the result • For orthogonal signaling • Using M = 2k and Eb = E/k, we have • Note that if • Then Pe 0 as k ∞ (not Pe infinite like the text says!!) ECE 6640 d min 2 E Note: a necessary and sufficient condition for reliable comm. that is slightly lower is derived in Chap. 6. It is called the Shannon limit. 90 Orthogonal Signaling ECE 6640 91 4.4-2 Biorthogonal Signaling d min 2 E d other 2 E ECE 6640 92 Biorthogonal Signaling ECE 6640 93 ECE 6640 94 COMMENT ON SIMULATIONS ECE 6640 95 Correlation Receiver Design • Maximizing Design (maximum symbol correlation) – Create a matched filter based on every one of the symbols being transmitted. * sm L T , for 1 m M – Determine the correct symbol sampling time. (Proportional to LT) – The maximum of the sampled correlates is selected as the output symbol. max s t nt sm L T t * 1 m M • @ t k T filter _ delay Minimizing Design (minimum distance from signal to symbol) – Create a matched filter based om every one of the basis symbol set. * rn t s t nt n t , for 1 n N – Determine the correct symbol sampling time. – Find the symbol vector, sm, that is a minimum distance. ECE 6640 min rn k sm 1 m M for t k T filter _ delay 96 Design Considerations • If the dimensionality of the symbols is small, the minimization approach using the basis set for the symbols will require less hardware/processing. – For ASK N=1 – For PSK N=2 • If the dimensionality of the symbols is large, the maximization approach using the matched filter of the symbols will be a likely choice – For FSK N=M ECE 6640 97 Matlab Minimization for MPSK % RECEIVER ---------------------------------------------% Process the array of symbol plus noise values % Detect the correlator output using an optimal threshold corrc= (ccarrier/sqrt(NSampSym))'*symbol_noise; corrs=-(scarrier/sqrt(NSampSym))'*symbol_noise; csym = corrc + sqrt(-1)*corrs; Note: Correlator perfectly matched to symbol time % map to the closest symbol for the symbol detection/decision % The first column is noise only [ee ind]=min(abs(sym_map-csym(1))); sd(:,1)= [real(sym_map(ind));imag(sym_map(ind))]; ECE 6640 98 Matlab Maximization for MFSK % RECEIVER ---------------------------------------------Note: Correlator % Process the array of symbol plus noise values perfectly matched to % Detect the correlator output using an optimal threshold symbol time for nnn=1:M FcorrM=Fstart+(2*(nnn-1)*Fstep); corrsym=sqrt(2)*(-1)^(n-1)*sin(2*pi*(FcorrM)*time_symbol)/sqrt(NSampSym); corr(nnn,:) = corrsym'*symbol_noise; end [val received_sym]=max(corr); bits_rcvd=Sym_bits(received_sym',:)'; ECE 6640 99 Matlab Matched Filter for MASK, QAM & MPSK %% % Symbol plus scaled filter and SNR noise RX_samples = TX_samples + sqrt(1/SNR)*input_noise; %% % Received signal baseband processing % % firrcos matched filter receiver pred_samples =filter(hsqnyq,1,RX_samples); … % optimal symbol time sampling pred_sym = pred_samples(Tdelay-1:expand:end); num_pred_sym = length(pred_sym); … ECE 6640 %% % Symbol and Bit Error Analysis % % map to the closest symbol for the symbol detection/decision eepower = 0; sd = zeros(num_pred_sym,2); ygsym = zeros(num_pred_sym,1); % minimum distance computations for jj=1:num_pred_sym [ee,ind]=min(abs(complex_const_map/sigma_s-pred_sym(jj))); sd(jj,:)= [real(complex_const_map(ind)) imag(complex_const_map(ind))]; ygsym(jj) = ind-1; eepower = eepower + ee^2; end 100 Useful Matlab Study • Compare BER performance based on using different filters. – Using different forms of matched filters. • rectangular, square-root Nyquist, truncated sinc functions, etc. • must have appropriate nulls to minimize ISI – Using “unmatched” filters. • rect TX with Nyquist filter receive • Section 4.5 is on Non-coherent Detection ECE 6640 101