ECE 6640 Digital Communications Dr. Bradley J. Bazuin Assistant Professor Department of Electrical and Computer Engineering College of Engineering and Applied Sciences Chapter 9 Chapter 9: 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8 ECE 6640 Digital Communication Through Band-Limited Channels Characterization of Band-Limited Channels Signal Design for Band-Limited Channels Optimum Receiver for Channels with ISI and AWGN Linear Equalization Decision-Feedback Equalization Reduced Complexity ML Detectors Iterative Equalization and Decoding—Turbo Equalization 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. 597 598 602 623 640 661 669 671 673 674 2 Chapter Content • In this chapter, we consider the problem of signal design when the channel is bandlimited to some specified bandwidth of W Hz. Under this condition, the channel may be modeled as a linear filter having an equivalent lowpass† frequency response C( f ) that is zero for | f | > W. • The first topic that is treated is the design of the signal pulse g(t) in a linearly modulated signal, represented as vt I n g t n T n that efficiently utilizes the total available channel bandwidth W. • The second topic that we consider is the design of the receiver in the presence of intersymbol interference and AWGN. The solution to the ISI problem is to design a receiver that employs a means for compensating or reducing the ISI in the received signal. The compensator for the ISI is called an equalizer. ECE 6640 3 Band Limited Channels • For our purposes, a bandlimited channel such as a telephone channel will be characterized as a linear filter having an equivalent lowpass frequency-response characteristic C( f ). Its equivalent lowpass impulse response is denoted by c(t). Then, if a signal of the form s t Revt exp j 2 f c t is transmitted over a bandpass telephone channel, the equivalent lowpass received signal is r t v ct d nt where c(t) is a filter representing a complex channel response. • Equalization will be about compensating for (or whitening) the channel. – whitening is an attempt to make the channel uniform in power spectral density. similar to the psd of white noise. ECE 6640 4 Frequency Limitation • The first assumption is that only a limited frequency band is available and can be used by the signal of interest. – For a low pass band C f 0, for f W – For a bandpass signal 0, for 0 f f c W 0, for f c W f C f ECE 6640 5 Channel Frequency Response • Within the channel, there is a complex frequency response that can be represented in magnitude and phase. – For low pass signals 0, for f W C f exp j f , for f W C f – For bandpass signals 0, C f C f exp j f , 0, ECE 6640 for 0 f f c W for f c W f f c W for f c W f 6 Ideal, Non-distorting Channel • The amplitude is a constant across the frequency band. – Based on RF path loss at center of frequency band. – Not path loss is frequency/wavelength dependent. Therefore this is an ideal assumption! • The phase must be a linear function. – A time delay exists in the signal. This directly defines a linear phase delay in frequency. C f t Td exp j 2 f t dt exp j 2 f Td ECE 6640 f 2 f Td – “Envelope delay” is defined as 1 f Td 2 f 7 Distorted Channels • The magnitude will vary with frequency. – amplitude distortion • The phase will not be linear. – phase distortion ECE 6640 8 Channel - Linear Envelop Delay PAM Nyquist Pulse Channel Input: Zeros at adjacent symbol locations ECE 6640 Linear Delay Channel Channel Output: Distortion will cause ISI 9 Medium Range Telephone Channel Wired network with switching Usable Bandwidth: 300Hz to 3000 Hz Impulse response duration: 10 msec ECE 6640 Telephone lines also have 50-60 Hz power line interference present. 10 Additional Impairments • Impulse Noise – additive disturbances from impulsive sources (e.g. lightening, switching power transients) – Popcorn noise or Laplacian noise • Thermal Noise – Gaussian or white noise from thermal noise sources • Phase Jitter – FM caused by low-frequency harmonic interference • Frequency Offset – carrier frequency mismatch between transmitter, relay equipment, and receiver • Multipath ECE 6640 – wireless signals being reflected by near-path objects 11 For More on Channels • Wikipedia – https://en.wikipedia.org/wiki/Channel_%28communications%29 • PDF Presentation: V. Arun, Associate Professor, Department of Computer Science, University of Massachusetts Amherst, CS653: Advanced Computer Networks – https://people.cs.umass.edu/~arun/653/home.html – https://people.cs.umass.edu/~arun/653/lectures/channel_models.pdf • GA Tech Ph.D. Thesis, Chirag S. Patel, “Wireless Channel Modeling, Simulation, and Estimation”, 2006. – https://smartech.gatech.edu/bitstream/handle/1853/10480/patel_chirag_s_ 200605_phd.pdf • IEEE 802.16 Broadband Wireless Access Working Group, “Channel Models for Fixed Wireless Applications”, 2001-07-17. – http://www.ieee802.org/16/tg3/contrib/802163c-01_29r4.pdf ECE 6640 • Web search provides other examples 12 Section 9.2 Signal Design • The Nyquist criteria defines how to achieve zero ISI for a “simple” channel. – Nyquist (p. 607 & Equ. 9.2-26) and square root Nyquist filters are typically used. – This is one of the filters you applied in Exam #1. The tri2 and sinc filters caused ISI which is why the performance degraded! • This criteria can be used to design “optimal” transmit and receive filters for a “simple” channel. ECE 6640 13 Filtering Comm. Symbols • General Filter Concept st d n ht n T n The scaling terms d(n), are selected from a small finite alphabet such as for BPSK {-1, +1} or for ASK {-1,-1/3, +1/3, +1} in accord with a specified mapping scheme between input symbol (bits) and output levels. The signal s(t) is sampled at equally spaced time increments identified by a timing recovery process in the receiver to obtain output samples as shown in Eqn. (4.2). sm T d n hm T n T n ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 14 Filter Concept We can partition this sum as shown in Eqn. (4.3), to emphasize the desired and the undesired components of the measurement. Here the desired component is d(m) and the undesired component is the remainder of the sum which if non-zero, is the ISI. sm T d m h0 d n hm T n T nm How do we eliminate the intersymbol interference (ISI) ? Let the time/sample representation of the filter be. 0, hn T 1, ECE 6560 n0 n0 “Perfect time sampling is implied” Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 15 Possible Filters “Function/filter must be zero at all integer values except n = 0” • They could be multiple symbols in length if they are zero at all integer values except n = 0 • Do we already know of a filter with this characteristic? (What about a time domain Sinc ?!) ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 16 Sinc as a Zero ISI Filter Considering the spectral and time domain requirements, we can also use t sin 2 2 T ht t 2 2 T ECE 6560 n T sin 2 2 T sin n hn T n T n 2 2 T Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 17 Sinc Function and “Reconstruction” • The “convolution” of the sinc function with sampled time waveform “impulse samples” is how perfect band-limited signal reconstruction is performed. – The continuous time sinc is the time-domain transform of the perfect frequency-domain “brick-wall low pass filter” • For symbols, we only need to “reconstruct” the symbol value without ISI at one time instant during the symbol period. – Nominally select the center of the symbol. – The “reconstructed” continuous time signal need not look like the original symbol waveform (they have significantly different frequency spectra and bandwidth!) ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13146511-2. 18 Rough Example • See SincEye.m amd SincEyev2.m – Each Symbol represented by a multi-cycle sinc function – The nulls of the sinc function occur at the “optimal” symbol sample point. All other sample points would be required to sum the signals levels from the other symbols (symbol interference). – Therefore, to limit ISI, you must 1. Properly filter 2. Properly (perfectly) sample in time ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 19 Optimal Filter for Pulse Detection (1) • If we want to detect a transmitted pulse with maximum SNR, the following applies SNR PSignal PNoise E s t N 0 BEQ 2 filtered to s t nt so t no t so t no t h s t nt d 0 2 E h st d 0 SNRout 1 N o ht 2 dt 2 0 ECE 6560 Notes Methods Notesand andfigures figuresare arebased basedon onorortaken takenfrom frommaterials materialsininthe thecourse coursetextbook: textbook:Probabilistic fredric j. harris, of Signal and System Analysis (3rd ed.) by George R. Cooper and Clare D. McGillem; Oxford Press, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-131999. ISBN: 0-19-512354-9. 146511-2. 20 Optimal Filter for Pulse Detection (2) • Applying Schwartz’s Inequality to the output SNR 2 2 2 h s d h d s d 0 0 0 • The upper bounds on the SNR may be defined as SNRout 2 h d E s t 2 d 2 2 0 0 E s t d No 0 1 2 N o ht dt 2 0 • But we can also define a condition for “equality” ECE 6560 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods and figures areAnalysis based on(3rd or taken from materials in the and course textbook: fredric Oxford j. harris,Press, ofNotes Signal and System ed.) by George R. Cooper Clare D. McGillem; Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-131999. ISBN: 0-19-512354-9. 146511-2. 21 Optimal Filter for Pulse Detection (3) • For equality to exist 2 2 2 h s t d h d s t d 0 0 0 • A possible solution is h K st u • This is an “optimal inverse-time filter” – The filter is the inverse time response of the transmitted pulse s(t)! ECE 6560 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods and figures areAnalysis based on(3rd or taken from materials in the and course textbook: fredric Oxford j. harris,Press, ofNotes Signal and System ed.) by George R. Cooper Clare D. McGillem; Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-131999. ISBN: 0-19-512354-9. 146511-2. 22 Optimal Filter for Pulse Detection (4) • Continuing for completeness – The desired impulse response is simply the time inverse of the signal waveform at time t, a fixed moment chosen for optimality. If this is done, the maximum filter power (with K=1) can be computed as 0 h 2 d st 2 d 0 t s 2 d t • And the maximum output SNR becomes maxSNRout 2 t No • The filter is commonly called a “Matched Filter” ECE 6560 Notes and figures are based on or taken from materials in the course textbook: Probabilistic Methods and figures areAnalysis based on(3rd or taken from materials in the and course textbook: fredric Oxford j. harris,Press, ofNotes Signal and System ed.) by George R. Cooper Clare D. McGillem; Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-131999. ISBN: 0-19-512354-9. 146511-2. 23 An Approach to Generating Filters 1. Defined the desired/required/stuck-with symbol spectrum or “time pulse” with a finite duration (less than or equal to the symbol period). 2. Multiply by the sinc in the time domain – – – Convolve in the frequency domain Infinite time /non-causal nature still a problem This enforces the h(nT) requirement! 3. Apply a time domain window after spectral filter design – ECE 6560 Modify passband and stopband ripple and edges as needed Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 24 Spectral Convolutions Zero Time Requirements Infinite Sinc for ISI Window Function Spectral Convolution Symbol Time Window, finite time, small BW penalty /T for 0.1<<0.5 Windowed, zero ISI filter Note: frequency domain shown as two-sided spectrum widths ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 25 Using Previously Defined Windows • Start with the time-domain sinc function – Determine the filter length based on the number of sinc cycles to be maintained (null-to-null samples from +/-1st, +/-2nd, +/-3rd, etc. – The Fourier transform has a sin(n)/sin() shape with frequency periodicity based on the number of sinc cycles. • Generate a window of the same number of samples – Multiply in time domain, convolve in frequency domain. – Removes the Gibbs phenomenon peaks and reduce the passband ripple. • Is there a preferred window/filter? (Yes, raised Cosine) ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 26 Web References • Wikipedia – InterSymbol Interference (ISI) • http://en.wikipedia.org/wiki/Intersymbol_interference – Nyquist ISI Criterion • http://en.wikipedia.org/wiki/Nyquist_ISI_criterion – Inter Symbol Interference (ISI) and Raised cosine filtering • http://complextoreal.com/wp-content/uploads/2013/01/isi.pdf • From C. Langton “Complex to Real” web site – A windowed sinc function will be used for ISI • The window often applied is the raised cosine ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 27 Nyquist Filtering with Raised Cosine • The Nyquist pulse is the wave shape required to communicate over band-limited channels with no ISI. – It is generated as a raised cosine frequency spectrum window • Even symmetric spectral window. • Finite frequency width that is a fraction of the perfect reconstruction width. (i.e. /T) – Preference to limit the time response to a length 4T/ – Truncated window (window length) and infinite sinc • With convolution in the frequency domain, the spectrum becomes a width of (1+)/T ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 28 Spectral Convolutions (1) • The term is called the roll-off factor and is typically on the order of 0.1 to 0.5 with many systems using values of = 0.2. – The transition bandwidth caused by the convolution is seen to exhibit odd symmetry about the half amplitude point of the original rectangular spectrum. – This is a desired consequence of requiring even symmetry for the convolving spectral mass function. When the windowed signal is sampled at the symbol rate 1/T Hz, the spectral component residing beyond the 1/T bandwidth folds about the frequency ±1/2T into the original bandwidth. – This folded spectral component supplies the additional amplitude required to bring the spectrum to the constant amplitude of H(f). ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 29 Spectral Convolutions (2) • We also note that the significant amplitude of the windowed wave shape is confined to an interval of approximate width 4T/ so that a filter with = 0.2 spans approximately 20T, or 20 symbol durations! – We can elect to simply truncate the windowed impulse response to obtain a finite support filter, and often choose the truncation points at ± 2T/ or 10 symbols. – A second window, a rectangle, performs this truncation. The result of this second windowing operation is a second spectral convolution with its transform. This second convolution induces pass-band ripple and out-of-band side lobes in the spectrum of the finite support Nyquist filter. (Nothing is perfect ….) ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 30 Symbol Periods in Communications “Let’s do the math” A communication system can be modeled most simply by the signal flow shown in Figure 4.4. Here d(n) represents the sequence of symbol amplitudes presented at symbol rate to the shaping filter h1(t). Perfect time sampling provides the detected symbol output …. the equivalent of filter-decimation! ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 31 Symbol Periods Math (1) st 1 m Filter received signal y t r t h2 t y t r t h2 d r t s t nt d m h t m T st nt h d 2 y t d m h1 t m T h2 d nt h2 d m y t d m h1 t m T h2 d n2 t m ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 32 Symbol Periods Math (2) Filter received signal y t d m h1 t m T h2 d n2 t m Define the convolution of the transmitter and receiver filters g t h t h d 1 2 The received signal is then y t d mg t m T n t m ECE 6560 2 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 33 Symbol Periods Math (3) Minimally sampling the symbol output t n T y n T d mg n m T n n T 2 m Expanding y n T dˆ n d n g 0 d m g n m T n2 n T mn ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 34 Discrete Samples at the Symbol Rate Interpretation y n T d n g 0 d m g n m T n2 n T m n 1. Desired Signal (convolved filter, prefer a matched filter) 2. Band limited Noise (n2(t)) filtered by h2(t) (typically filtered white noise, for power use BWeqn 3. Combined ISI (remove with Nyquist g(t)) We want g(nT) to be a Nyquist filter to remove ISI !! G w H1 w H 2 w H Nyquist w e ECE 6560 j wTdelay Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 35 Transmit and Receive Filters • We want the result to be a Nyquist Filter with time delay H1 w H 2 w H Nyquist f exp j w Tdelay • Using a matched filter for H1 and H2 should provide maximum outputs. A Square-root Nyquist filter! ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 36 One Approach: Square Root Nyquist Concept To maximize the signal-to-noise (SNR) in (4.10), the receiver filter must be matched to the transmitter-shaping filter. The matched filter is a timereversed and delayed version of the shaping filter, which is described in the frequency domain as shown in (4.11). Let H 2 w conjH1 w exp j w Tdelay H1 w H 2 w H1 w exp j w Tdelay H Nyquist w exp j w Tdelay 2 H1 w H Nyquist w 2 H1 w H Nyquist w ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 37 Square Root Nyquist ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 38 Second Approach: Square Root Nyquist Concept Use the equivalent of a ZOH output …. h1 t 1 Tsymbol t rect T symbol H1 f sinc f Tsymbol H1 f H 2 f exp j 2 f Tdelay H Nyquist f exp j 2 f Tdelay Note that the Nyquist filter is formed from a sinc basis, therefore the nulls appear at the same locations as the Nyquist filter! Let, H2 f ECE 6560 H Nyquist f H1_ compensation f exp j 2 f Tdelay Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 39 Nyquist and Square Root Nyquist • Well defined time and spectral responses … • Wikipedia Raised-Cosine Filter – http://en.wikipedia.org/wiki/Raised-cosine_filter ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 40 Nyquist Filter The most common spectral mass selected for communication systems is the half cosine of width *fSYM. The half cosine convolved with the spectral rectangle forms the spectrum known as the cosine-tapered Nyquist pulse with roll-off . The description of this band-limited spectrum normalized to unity passband gain is presented in (4.14). w 1 1 for H Nyq w 0.5 1 cos 2 0 wSym w w 1 for 1 1 w w Sym Sym w for 1 wSym sin f Sym t cos f Sym t hNyq t f Sym 2 f t 1 2 f t Sym Sym ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 41 Discrete Time Filter • • Let fsample = M*fsymbol so that fsymbol*t is replaced by fsymbol*n/(M * fsymbol) or n/M. It is common to operate the filter at M= 4 or 8 samples per symbol hNyq n n cos f Sym sin f Sym M f Sym M f Sym n f Sym 2 n f sample n f Sym 1 2 f M f Sym Sym M f Sym n sin 1 M hNyq n M n M ECE 6560 n cos M 2 n 1 2 M Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 42 Discrete Time Filter • The filter described in Eqn. (4.16) has a two-sided bandwidth that is approximately 1/Mth of the sample rate. A digital filter exhibits a processing gain proportional to the ratio of input sample rate to output bandwidth, in this case a factor of M. The 1/M scale factor in Eqn. (4.16) cancels this processing gain to obtain unity gain. When the filter is used for shaping and up sampling, as it is at the transmitter, we remove the 1/M scale factor since we want the impulse response to have unity peak value rather than unity processing gain. n sin 1 M hNyq n M n M ECE 6560 n cos M 2 n 1 2 M Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 43 Square Root Nyquist Filter • The square root of the cosine-tapered Nyquist filter results in a quarter cycle cosine tapered filter. This description is normally confined to square-root raised cosine or root raised cosine Nyquist filter. The description of this band-limited spectrum normalized to unity pass-band gain is shown in (4.17). H Sqrt Nyq w cos 4 1 for w 1 w Sym w wSym 1 for 1 for 1 0 w wSym 1 w wSym Textbook sign error! 4 f Sym t cos 1 f Sym t sin 1 f Sym t hSqrt Nyq t f Sym 2 1 4 f Sym t f Sym t ECE 6560 Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 44 Discrete Time Filter • Let fsample = M*fsymbol so that fsymbol*t is replaced by fsymbol*n/(M * fsymbol) or n/M. hSqrt Nyq n n n 4 f sin 1 f Sym cos 1 f Sym Sym M f M f M f n Sym Sym Sym f Sym 2 f Sample n n 1 4 f Sym f Sym M f Sym M f Sym 4 n 1 M hSqrt Nyq n M ECE 6560 n n cos 1 sin 1 M M 2 n n 1 4 M M Textbook errors! Sign and extra n. Notes and figures are based on or taken from materials in the course textbook: fredric j. harris, Multirate Signal Processing for Communication Systems, Prentice Hall PTR, 2004. ISBN 0-13-146511-2. 45 Design with Controlled ISI • Partial Response Signaling • If ISI can not be avoided, define a signaling system hat allows successive symbols to be combined and correctly decoded. A “controlled” or deterministic amount of ISI. – The “wave shape” can be used to define successive symbols. – By decoding symbol pairs successively, correct symbol detection can be performed. – A trellis structure can be used! • Examples and concept presented in textbook. – Signal Generation Section 9.2-2: p. 609-612 – Signal Detection Section 9.2-3: p. 613-619 ECE 6640 46 Section 9.3 • When the channel is not simple, the following model applies • The received filter seeks to be an optimally matched filter for the transmission AND remove the channel effects introduced by the effective channel filter. – It is a combination of the previously defined filter and an equalizer, where the equalizer is equivalent to a reciprocal filter that “flattens the channel” so that the channel and equalizer act like the “simple” channel (time delay only). ECE 6640 47 Section 9.3-1 Optimum Maximum-Likelihood Receiver • Following up on the concept of partial response signaling. • If ISI is going to be present, can a Viterbi like decoder use the ISI that exists over a limited number of symbols to perform detection? ECE 6640 48 Discrete-Time Model for Channel • • • In dealing with band-limited channels that result in ISI, it is convenient to develop an equivalent discrete-time model for the analog (continuous-time) system. Since the transmitter sends discrete-time symbols at a rate of 1/T symbols/s and the sampled output of the matched filter at the receiver is also a discrete-time signal with samples occurring at a rate of 1/T per second, it follows that the cascade of the analog filter at the transmitter with impulse response g(t), the channel with impulse response c(t), the matched filter at the receiver with impulse response h∗(−t), and the sampler can be represented by an equivalent discrete-time transversal filter having tap gain coefficients {xk}. Consequently, we have an equivalent discrete-time transversal filter that spans a time interval of 2LT seconds.. ECE 6640 49 Discrete Time Model Concern • • The major difficulty with this discrete-time model occurs in the evaluation of performance of the various equalization or estimation techniques that are discussed in the following sections. The difficulty is caused by the correlations in the noise sequence {νk} at the output of the matched filter. While many systems ignore this in simulation, a technique to eliminate the correlation is developed in Section 9.3-1. ECE 6640 50 Discrete Time Channel Characteristics ECE 6640 51 General Equalizer Discussion • Wikipedia – https://en.wikipedia.org/wiki/Equalization_%28communications%29 • See Digital Equalizer Types, in particular – MMSE equalizer, • Does not completely remove ISI – Zero Forcing Equalizer • Minimizes ISI but may increase noise • Exclusively discussed in textbook – Decision Feedback Equalizer • previously mentioned based on an adaptive system • Feedback attempts to remove ISI from previous symbols – Blind Equalizer ECE 6640 • based on symbol statistics alone 52 Decision Feedback Equalizer ECE 6640 53 Equalization Notes from ECE 6950 • The following was derived for the adaptive systems class. • Adaptive System Demonstration – Matlab Adaptive_TB, chap5, equalizer – Other examples developed for exam2 and the final. ECE 6640 54