This action might not be possible to undo. Are you sure you want to continue?
4. ISI & Equalization
Chapter 4 Intersymbol Interference and Equalization
The all-pass assumption made in the AWGN (or non-dispersive) channel model is rarely practical. Due to the scarcity of the frequency spectrum, we usually ﬁlter the transmitted signal to limit its bandwidth so that efﬁcient sharing of the frequency resource can be achieved. Moreover, many practical channels are bandpass and, in fact, they often respond differently to inputs with different frequency components, i.e., they are dispersive. We have to reﬁne the simple AWGN (or non-dispersive) model to accurately represent this type of practical channels. One such commonly employed reﬁnement is the dispersive channel model1:
where Ù´Øµ is the transmitted signal, with power spectral density Æ¼ by the linear ﬁlter
Ù £ ´Øµ · Ò´Øµ
´Øµ is the impulse response of the channel, and Ò´Øµ is AWGN
In essence, we model the dispersive characteristic of the channel
The simplest dispersive channel is the bandlimited channel for which the
channel impulse response
´Øµ is that of an ideal lowpass ﬁlter.
This lowpass ﬁltering smears the
transmitted signal in time causing the effect of a symbol to spread to adjacent symbols when a sequence of symbols are transmitted. The resulting interference, intersymbol interference (ISI), degrades the error performance of the communication system. There are two major ways to mitigate the detrimental effect of ISI. The ﬁrst method is to design bandlimited transmission pulses which minimize the the
For simplicity, all the signals considered in this chapter are real baseband signals. All the developments here can, of
course, be generalized to bandpass signals using either the real bandpass representation or the complex baseband representation.
Wong & Lok: Theory of Digital Communications
4. ISI & Equalization
effect of ISI. We will describe such a design for the simple case of bandlimited channels. The ISIfree pulses obtained are called the Nyquist pulses. The second method is to ﬁlter the received signal to cancel the ISI introduced by the channel impulse response. This approach is generally known as equalization.
4.1 Intersymbol Interference
To understand what ISI is, let us consider the transmission of a sequence of symbols with the basic waveform
Ù´Øµ. To send the Òth symbol Ò , we send Ò Ù´Ø ÒÌ µ, where Ì is the symbol interval.
Therefore, the transmitted signal is
ÒÌ µ ÒÌ µ · Ò´Øµ
Based on the dispersive channel model, the received signal is given by
where Ú ´Øµ
Ò Ú ´Ø
Ù £ ´Øµ is the received waveform for a symbol. If a single symbol, say the symbol ¼ , Ú ´ Øµ and then sample the matched ﬁlter output at
is transmitted, the optimal demodulator is the one that employs the matched ﬁlter, i.e., we can pass the received signal through the matched ﬁlter Ú ´Øµ time Ø
¼ to obtain the decision statistic. When a sequence of symbols are transmitted, we can still
ÑÌ to obtain the decision statistic for the symbol ÞÑ
Ñ . At Ø
employ this matched ﬁlter to perform demodulation. A reasonable strategy is to sample the matched ﬁlter output at time Ø of the matched ﬁlter is
ÑÌ , the output
£ Ú´ÑÌ ÒÌ µ · ÒÑ Ú ¾· Ò Ú £ Ú ´ÑÌ ÒÌ µ · ÒÑ
ÒÚ Ò Ñ
where ÒÑ is a zero-mean Gaussian random variable with variance
Æ¼ Ú ¾ ¾. The ﬁrst term in (4.4)
is the desired signal contribution due to the symbol Ñ and the second term contains contributions from the other symbols. These unwanted contributions from other symbols are called intersymbol interference (ISI).
ISI & Equalization 1. The effect of ISI is to cause a reduction in the eye opening by reducing the peak as well as causing ambiguity in the timing information. the demodulation strategy above can be interpreted as matched ﬁltering for each symbol.3 .. One common way to observe and measure (qualitatively) the effect of ISI is to look at the eye diagram of the received signal.Wong & Lok: Theory of Digital Communications 4.5 5 10 15 20 25 30 Figure 4.e.5 0 −0.5 1 0. Ú £ Ú´ÑÌ ÒÌ µ Ì . respectively. and there is no ISI. Therefore. Ú ´Øµ ¼ except for ¼ ¼ except for Ì Ø Ì . Such For illustration.2. The effect of ISI and other noises can be observed on an oscilloscope displaying the output of the matched ﬁlter on the vertical input with horizontal sweep rate set at multiples of a display is called an eye diagram. Then it is easy to see that ¼ for all Ò Ñ.5 −1 −1. Ú ´Øµ and. let us consider the basic waveform Ù´Øµ is the 4. Unfortunately. Therefore.1 and 4. i. Ø As a result. a timelimited waveform is never bandlimited. The eye diagrams for the cases where the channel is all-pass (no ISI) and lowpass (ISI present) are shown in Figures 4. rectangular pulse ÔÌ ´Øµ and binary signaling is employed.1: Eye diagram of a BPSK signal with no ISI Suppose Ú £ Ú´Øµ Ú ´Øµ were timelimited. in general. Ú £ Ú ´Øµ are not timelimited and hence ISI is. hence. ½ Ì . for a bandlimited channel. present.
e.5 −1 −1.5 0 −0. ÞÑ Ñ Ü´¼µ · Ò Ñ Ò Ü´ÑÌ ÒÌ µ · ÒÑ ¼ ¼ (4. To do so. The Nyquist condition in this form is not very helpful in the design of ISI-free pulses.4) as: Ú £Ú´Øµ.Wong & Lok: Theory of Digital Communications 4. ISI & Equalization 1.6) where is some constant and. we can set ½. It turns out that it is more illustrative to restate the Nyquist condition in frequency domain.7) .2: Eye diagram of a BPSK signal with ISI 4. without loss of generality. the basic pulse shape Ù´Øµ and/or the channel is bandlimited.2 Nyquist Pulses A careful observation on (4.5) There is no ISI if the Nyquist condition is satisﬁed: Ü´ÒÌ µ ¼ Ò for Ò for (4. i.4) reviews that it is possible to have no ISI even if the Ú ´Øµ is bandlimited.4 (4..5 1 0. letting Ü´Øµ we can rewrite the decision statistic ÞÑ in (4. ﬁrst let ÜÆ ´Øµ ½ Ò ½ Ü´ÒÌ µÆ ´Ø ÒÌ µ 4.5 5 10 15 20 25 30 Figure 4. More precisely.
11) . The folded spectrum ½ ½ ´ Ò ´ µ Ì for for ¼ Ü´Øµ The corresponding time domain function is the sinc pulse sinc´ 4.6) is equivalent to the condition ÜÆ ´Øµ Æ ´Øµ or Æ ´ µ ½ in frequency domain.. i. ¾Ï .8) ½ Ò Ì Ì (4.10) ¯ ½ Suppose that the symbol rate is slower so that Ì È their neighbors. Æ´ µ ¾Ï : non-overlapping spectrum ½ ½ Ì Ò ½ where ´ µ is the Fourier transform of Ü´Øµ.5 Ø Ìµ (4.8). Now.9) This is the equivalent Nyquist condition in frequency domain. It says that the folded spectrum of Ü´Øµ has to be ﬂat for not having ISI. Then.Wong & Lok: Theory of Digital Communications 4. ´ µ looks. the Nyquist condition has ¯ ½ Suppose that the symbol rate is so high that Ì Ò Ì µ looks like the one in Figure 4. When the channel is bandlimited to Ï Hz. the folded spectrum È½ ½ ´ Ò There are gaps between copies of ´ µ. the Nyquist condition cannot be satisﬁed and ISI is inevitable. the following implications: ´ µ ¼ for Ï .3: Case Ì Taking Fourier transform.3. copies of ´ µ can just touch Ò Ì µ is ﬂat if and only if Ï otherwise (4. by employing (4.e. ISI & Equalization ½ Figure 4. The Nyquist condition in (4. No matter how ¾Ï . we get ½ Ò Ò Ì (4. Then.
it reduces to the sinc function.Wong & Lok: Theory of Digital Communications 4. it is not physically ¾Ï above which ISI is unavoidable is known as the Nyquist rate. the raised cosine spectrum gives a pulse that is much ´ µ × Ò´ Ø Ì µ Ó×´ «Ø Ì µ Ø Ì ½ « ¾ Ø¾ Ì ¾ (4. Ü´Øµ decays as ½ Ø¿ while for ¼. ½·« ¾Ì Ï . The folded spectrum of È½ ´ µ.5): ´ µ where Ì Ì for ¼ ¼ « ¾ ½ · Ó× «Ì ´ ½ « µ ¾Ì ¼ for for ½ « ¾Ì ½·« ¾Ì ½ « ¾Ì ½·« ¾Ì (4. is. we can design an ISI-free pulse shape Ò ½ ¾Ï . Therefore. When the symbol rate is below the Nyquist rate. The critical rate Ì ¯ ½ Suppose that the symbol rate is even slower so that Ì their neighbors. ISI & Equalization ½ Figure 4. a widely used ISI-free spectrum is the raised-cosine spectrum (Figure 4. The Ì corresponding time domain function (Figure 4. A truncated and delayed version is used as an approximation.6 . An example is shown in Figure 4.12) It determines the excess bandwidth2 beyond ¾½ . Hence. We note that for « ¼.4. ½ realizable.13) Another way to interpret this is that to ﬁt into a channel bandlimited to Ï . copies of ´ µ overlap with Ò ´ Ì µ can be ﬂat with many different choices which gives a ﬂat folded spectrum. Therefore.4: Case Ì ¾Ï : overlapping spectrum We note that the sinc pulse is not timelimited and is not causal. the larger Ì has to be and the slower is the symbol rate.6) is: ½ is called the roll-off factor. Then. Ü´Øµ (sinc pulse) decays as ½ Ø. Ü´Øµ When « 2 « ¼. Therefore. the larger « 4.
5 0.Wong & Lok: Theory of Digital Communications 4.4 −3 −2 −1 0 t/T 1 2 3 Figure 4.5: Raised-cosine spectrum 1 0.4 0.9 0.2 0 0.2 frequency/symbol rate 0.6 −0.5 α=0 −0.6 X(f) 0.7 0.4 −0.2 0 1 0.4 0.2 α=0 0.5 1 0 −1 −0.1 0.4 x(t) 0.3 0.8 0.2 −0.6 0.8 −0.6: Time domain function of the raised-cosine spectrum 4.7 .8 0. ISI & Equalization Raised Cosine Spectrum 1 0.6 0.8 1 Figure 4.
we can simply also µ to be ´ µ. ISI & Equalization HC ( f ) AWGN Figure 4.7.14) Î ´ µ. but they also introduce distortions in their passbands. the whole system can be described the conceptual model 4. This is the dispersive channel model we describe before. such as telephone lines. ´ µ where Í ´ µ. For a communication system system employing a linear modulation. respectively. Of course. not only are they bandlimited. the channel does not introduce any distortion within its passband. Mathematically.. if the raised-cosine spectrum is chosen. the resulting ISI-free pulse Ù´Øµ is called the square-root raised-cosine pulse.3 Equalization For many physical channels. In general. truncation and delay is required for realization.7: Dispersive channel model less sensitive to timing errors than the sinc pulse. and À ´ µ are the Fourier transform of Ù´Øµ. Then the Fourier transform the transfer function of the matched ﬁlter is ´ µ. For example. ISI is often introduced.8 . Just like all other bandlimited pulses. Given that an ISI-free spectrum choose Í ´ Õ Õ ´ µ is chosen. Such a channel can be modeled by an LTI ﬁlter followed by an AWGN source as shown in Figure 4. recall that Ü´Øµ is the overall response of the transmitted pulse passing through the ban- dlimited channel and the receiving ﬁlter. Ú ´Øµ. such as BPSK.14) to obtain the simple case of a ban- dlimited channel.Wong & Lok: Theory of Digital Communications 4. suitable truncation and delay are required for physical realization. and ´Øµ. Î´ µ ¾ À ´ µÍ ´ µ ¾ (4. 4. Ü´Øµ from the raised cosine spectrum is not timelimited. we can employ (4. i. Finally. Therefore. through a dispersive channel.e.
ISI & Equalization Ik HT ( f ) HC ( f ) HR ( f ) Decision every T Impulses AWGN Figure 4. the dispersive channel ´ µ Á and ÀÌ ´ µ. respectively. The Nyquist condition for no ISI developed in Section 4. This is the Nyquist pulse design method described in Section 4. As for the previous method.9. we have to know the channel response À ´ µ in advance to construct the transmission and receiving ﬁlters.Wong & Lok: Theory of Digital Communications 4. Ò Ì Ì ÀÌ ´ µÀ ´ µÀÊ ´ µ. it is also difﬁcult to build the appropriate ÀÊ ´ µ to eliminate ISI.15) One method to achieve the Nyquist condition is to ﬁx the receiving ﬁlter to be the matched ﬁlter. the receiving ﬁlter ÀÊ ´ µ is chosen to be the complex conjugate of ÀÌ ´ µÀ ´ µ. called equalizer. This approach to remove ISI is usually known as equalization. i. we may choose to build a simpler (practical) ﬁlter ÀÊ ´ µ.2 can be easily generalized to the above communication system. notice that what we want eventually are the samples at intervals Ì at the receiver.. 4 Of course. in which the sequence of information symbols is denoted by and ÀÊ ´ µ are the transfer functions of the transmission (pulse-shaping) ﬁlter.15). at the output to eliminate ISI analog ﬁlter as shown below in Figure 4..2. Therefore. 3 If the matched ﬁlter demodulation technique is employed. Moreover.15) is satisﬁed. and put a digital ﬁlter. 4. the ½ Ò ½ (4. However. and the receiving ﬁlter3 . is constant for all frequencies. À ´ µ. The main advantage of this approach is that a digital ﬁlter is easy to build and is easy to alter for different equalization schemes. take samples at intervals Ì . we should choose one that can be easily built. i.e.8: Continuous-time communication model over a dispersive channel in Figure 4. and choose the transmission ﬁlter so that (4.9 . as well as to ﬁt different channel conditions. set ÀÊ ´ µ £ ÀÌ ´ µÀ £ ´ µ.e. An alternative method is to ﬁx the transmission ﬁlter4 and choose the receiving ﬁlter shaping method is that it is in general difﬁcult to construct the appropriate analog ﬁlters for ÀÌ ´ µ and ÀÊ ´ µ to satisfy the condition in (4. Letting condition for no ISI is that the folded spectrum ´ µ. The major disadvantage of this pulse- ÀÊ ´ µ in practice.8.
9: Communication system with equalizer Ik HT ( f ) HC ( f ) HR ( f ) Equalizer every T Decision Impulses colored Gaussian noise Figure 4.3. we can consider an equivalent colored noise is the output of ÀÊ ´ Figure 4.9 to an equivalent discrete-time model that is easier to work with.16) . The ¯ We input a bit or a symbol to the communication system every Ì seconds. equivalent colored Gaussian noise being added after ÀÊ ´ µ due to AWGN. À ´ µ and ÀÊ ´ µ are LTI ﬁlters. they can be combined and represented by an equivalent digital ½ . The LTI ﬁlter. we translate the continuous-time communication system model in Figure 4. and get back a sample at the output of the sampler every Ì seconds.10 ·Ò Á ·Ò (4. To do so. in which the output sequence Á ¼ is given by Á¼ Á Á ¼· 4.10: Equivalent communication system with colored Gaussian noise 4. The resulting model is shown in µ when we analyze the system. The following steps describe the translation process: ¯ Instead of considering AWGN being added before the receiving ﬁlter ÀÊ ´ µ.11. Since ÀÌ ´ µ.10 from the information source to the sampler as a digital ﬁlter. Denote its transfer function by À ´Þ µ and its impulse response by ½ result is the discrete time-linear ﬁlter model shown in Figure 4.10.1 Equivalent discrete-time model Our goal is to design the equalizer which can remove (or suppress) ISI. Therefore.Wong & Lok: Theory of Digital Communications 4. we can represent the communication system in Figure 4. ISI & Equalization Ik HT ( f ) HC ( f ) HR ( f ) Equalizer every T Decision Impulses AWGN Figure 4.
¼ for some ¼.11: Equivalent discrete-time communication system model with colored noise HW ( z ) EQZ Figure 4.11 ´Þµ. Ò consists of samples of the colored Gaussian noise (AWGN ﬁltered by ÀÊ ´ µ).12: Typical equalizer In general.17) where is the impulse response corresponding to the transfer function 4. we can consider the equivalent discrete-time model shown in Figure 4. a noise-whitening digital ﬁlter ÀÏ ´Þ µ and an equalizing circuit that equalizes the noise-whitened output as shown in Figure 4.12.13. The ÀÏ ´Þ µ is to “whiten” the noise sequence so that the noise samples are uncorrelated. ISI & Equalization Ik H(z) Ik ’ Equalizer Decision nk Figure 4. the equalizer consists of two parts.Wong & Lok: Theory of Digital Communications 4. in which Ò is an effect of AWGN sequence. ISI is present. ¯ Let ´Þµ À ´ÞµÀÏ ´Þµ. Notice that ÀÏ ´Þ µ depends only on ÀÊ ´ µ.14. equivalently. the noise sequence is white. The output sequence Á is given by Á Á ·Ò Á ¼· Á ·Ò (4. At the output of ÀÏ ´Þ µ. and is not white in general. The communication system from the information source to the output of the noise whitening ﬁlter can now be represented by the discrete-time white-noise linear ﬁlter model in Figure 4. and can be determined a prior according to our choice of ÀÊ ´ µ. and Ò is an . Notice that the noise sequence ¯ Usually. Therefore. Therefore. namely.
¯ Finally. Then the output of the equalizer is given by Á Á (4.13: Equivalent discrete-time communication system model with white noise Ik G(z) ~ Ik EQZ Decision ~ nk Figure 4. Suppose that the equalizer is also an LTI ﬁlter with transfer function À ´Þ µ and corresponding impulse response . Á Á for all if noise is not present.18) Ideally.. i. Á contains only contributions from the current symbol Á and the AWGN sequence with small variance. ISI & Equalization Ik H(z) HW ( z ) ~ Ik EQZ Decision ~ nk Figure 4. i.Wong & Lok: Theory of Digital Communications 4.2 Zero-forcing equalizer First. we employ an LTI ﬁlter with transfer function À ´Þ µ as the equalizing circuit.. 4. 4. This can be achieved by simply setting the transfer function À ´Þ µ ½ ´Þ µ.3. This method is called zero-forcing equalization since the ISI component at the equalizer output is forced to zero. The simplest way to remove the ISI is to choose À ´Þ µ so that the output of the equalizer gives back the information sequence.14: Equivalent discrete-time white-noise linear ﬁlter model AWGN sequence. let us consider the use of a linear equalizer.e.e. the equalizing circuit (we simply call it the equalizer from now on) attempts to remove ISI from the output of ´Þµ. We will work with this discrete-time model in all the following sections. The focus of our coming discussion is the design of this equalizer.12 .
noise is always present. ISI & Equalization In general. then the PSD of the signal component is 4.19) In this case. Suitable We note that the effect of the equalizing ﬁlter on the noise is neglected in the development of the zero-forcing equalizer above.21) Hence. To see this. we choose the zero-forcing ﬁlter À ´Þµ as ´ ½ ¾ Ìµ Õ À ¾ Ì À´ ¾ ½ Ìµ (4. the noise-whitening ﬁlter ÀÏ ´Þ µ can be chosen as ÀÏ ¾ Ì Õ À´ ¾ ½ Ìµ (4.13 ½ and hence the signal ..20) and the PSD of the colored Gaussian noise samples Ò in Figure 4.e. there may be a chance that the equalizing ﬁlter will greatly enhancing the noise power and hence the error performance of the resulting receiver will still be poor. the overall digital ¾ Ì Õ À ¾ Ì À Ï À´ ¾ Ìµ (4.Wong & Lok: Theory of Digital Communications 4.24) Since the zero-forcing ﬁlter simply inverts the effect of the channel on the original information symbols Á . can be an inﬁnite length sequence.23) Now. let us evaluate the signal-to-noise ratio at the output of the zero-forcing equalizer when the transmission ﬁlter ÀÌ ´ ﬁxed and the matched ﬁlter is used as the receiving ﬁlter. µ is ÀÊ ´ µ £ ÀÌ ´ µÀ £ ´ µ (4. the corresponding impulse response truncation and delay is applied to get an approximation.11 is given by ¨Ò ¾ Ì ¬ Æ¼ ½ ¬ ¬ ¾Ì Ò ½ ¬ÀÌ (4.14 is ¾ Ì Ò is simply Æ¼ ¾. As a result. In reality. i. it is easy to see that the digital ﬁlter À ´Þ µ is given by À ¾ Ì Ì ½ Ò ½ ½ ¬ ¬ ¬ ¬ ÀÌ Ò Ì À Ò Ì À Ò ¬¾ Ì ¬¬¬ Ò Ì ¬¾ ¬ ¬ ¬ (4.22) and then the PSD of the whitened-noise samples ﬁlter ´Þµ in Figure 4. Although the ISI component is forced to zero. the signal component at its output should be exactly Á . If we model the Á as iid random variables with zero mean and unit variance.
28) (4. Assume that the information sequence Á is WSS. Deﬁning the SNR as the ratio of the signal energy to the noise energy. Á beforehand. If there is a certain region in the folded spectrum with very small magnitude. we have SNR Æ¼ Ì ¾ ½ ¾Ì ½ ¾ ½ ¾Ì Ò ½ ¬ ¬ ¬ ¬ ÀÌ Ò Ì À Ò ¬¾ ½ Ì ¬¬¬ ½ (4. then the SNR can be very poor.15 as the equalizer. On the other hand. ¾¼ We note that a delay of Ä symbols is incurred at the output of the FIR ﬁlter. ISI & Equalization energy at the output of the equalizer is just ½ ¾Ì component at the output of the equalizer is Ê ½ ¾Ì output is ½ ¾Ì Æ¼ ¾ ¬ ¬ ¬ À ¬ ¾ Ì ¬¾ ¬ ¬ Æ¼ ¬ ¾ ¬ À ¬ ¾ Ì ¬¾ .27) Á Á ·Ä Ä 4.29) .26) Without knowing the values of the information symbols Let us employ the FIR ﬁlter of order ¾Ä · ½ shown in Figure 4. Then MSE E E where Á Á ÁÌ Ä Ä Á ¾ ½¾ ¿ (4.25) Notice that the SNR depends on the folded spectrum of the signal component at the input of the receiver. may not give the best error performance for the communication system because it does not take into account noises in the system.Wong & Lok: Theory of Digital Communications Ê ½ ¾Ì 4.14 Á Ä Ä Ì Ì (4. Hence the noise energy at the equalizer ¬ ½ Ì . A different equalizer that takes noises into account is the minimum mean square error (MMSE) equalizer.3. we model each symbol Á as a random variable. We choose a linear equalizer À ´Þ µ to minimize the MSE between the original information symbols Á and the output of the equalizer Á : MSE E ¾ E ´Á Á µ¾ (4. the PSD of the noise . although removes ISI. 4.3 MMSE equalizer The zero-forcing equalizer. It is based on the mean square error (MSE) criterion.
Differentiating with respect to each Á ÁÌ ¼ (4. then the MMSE equalizer can be found by solving the linear matrix equation (4.Wong & Lok: Theory of Digital Communications 4. we get Ä .-L hE.31). . In our notation. The gradient of the MSE with respect to gives the direction to change for the largest increase of the MSE...32) (4.33) ÁÁ If Ê and are available. notice that the MSE is a quadratic function of . the gradient is ¾´ 4. Ä . The linear MMSE equalizer can also be found iteratively.30) Ê where (4. To decrease the MSE. we get E Á Rearranging.31) Ê E Á ÁÌ E (4. It can be shown that the signal-to-noise ratio at the output of the MMSE equalizer is better than that of the zero-forcing equalizer. ISI & Equalization ~ I k z -1 z -1 .15: FIR ﬁlter as an MMSE equalizer We want to minimize MSE by suitable choices of and setting the result to zero. we . First.15 Ê µ. z -1 h E.L Σ ^ I k-L Figure 4.
we can actually minimize the sum of the square errors. we can replace Ê and at each step in the steepest descent algorithm by the rough estimates Á ÁÌ and Á Á . the “data” sequence.16 ÁÁ .39) ´Ã µ 4. the training sequence is known to the equalizer. Suppose that the known sequence lasts for Ã symbols. which is a statistical average. we get Ê´Ã µ ´Ã µ (4. respectively. With a training sequence. the transmitter can transmit In many applications.35) This is a stochastic steepest descent algorithm called the least mean square (LMS) algorithm.Wong & Lok: Theory of Digital Communications 4. ISI & Equalization can update in the direction opposite to the gradient. Instead of minimizing the MSE. Alternatively. the vector ´ µ is updated as ´µ ´ ½µ · Ê ´ ½µ (4.34) where is a small positive constant that controls the rate of convergence to the optimal solution. we do not know Ê and a training sequence that is known a priori by the receiver.36) Differentiating with respect to ´Ã µ and setting the result to zero. However. Then the sum of the square errors is given by Ã Ã ¾ Ã ½ ½ ´Á Á µ¾ Á ÁÌ ´Ã µ ¾ ´Ã µ Ã Ã (4.37) This time. This is called the least squares (LS) criterion.. in advance. the receiver can estimate Ê and . This is the steepest descent algorithm: At the th step. i. with a training sequence.3. The algorithm becomes: ´µ ´ ½µ · Á ÁÌ ´ ½µ Á (4.4 LS equalizer In the training period for the MMSE equalizer. Ê´Ã µ ½ ½ Á ÁÌ (4. 4.e.38) (4.
45) In general. However. ´ Of course. In many cases. ´ µ to be able to do so. With decision feedback..17 . there is a more efﬁcient approach. and solve the matrix equation all over again. it is possible to use previously decided symbols (output from the decision device) provided that we have made correct decisions on them.41) · ½µ ´Ã · ½µ Using the matrix inversion lemma5. an obvious approach to eliminate ISI would be to subtract their effects off6 . ´Ã µ Ê ½ ´Ã µ ´Ã µ.44) The current symbol we want to determine is Á . we can think of the equalizer to contain two parts — a feedforward part and a ½ ½ ½ ½ ½ ¡ ½ ½ 5 6 Assuming all the required invertibilities.40) (4. 4. If we had known the other symbols exactly.42) (4. we get ´Ã µ · ÁÃ ·½ÁÃ ·½ Ê ½ ´Ã · ½µ Ê ½ ´Ã µ ´Ã · ½µ Ê ½ ´Ã µÁÃ ·½ÁÌ ·½Ê ½ ´Ã µ Ã Ì Ê ½ ´Ã µÁ ½ · ÁÃ ·½ Ã ·½ Ì ·½ ´Ã µ · ÁÃ ·½ ÁÃ ½ ´Ã µ Ê ½´Ã µÁÃ ·½ ½ · ÁÌ ·½Ê ´Ã µÁÃ ·½ Ã (4. However. This approach is called decision feedback equalization.5 Decision feedback equalizer Recall from the equivalent discrete-time model in Figure 4. we have to recalculate Ê´Ã and ´Ã ·½µ. actually. i.e.3. Assuming Ê´Ã µ is non-singular. the equalizer would give Á Á Á (4.Wong & Lok: Theory of Digital Communications 4. we need to have knowledge of Þ µ · .14 that Á Á ¼· Á ·Ò (4. Notice that Ê´Ã · ½µ Ê´Ã µ · ÁÃ ·½ÁÌ ·½ Ã (4. Apparently. 4.43) The procedure is called the recursive least squares (RLS) algorithm. we do not know all the symbols that are affecting the reception of the current symbol. the RLS algorithm converges much faster than the steepest descent algorithm at the expense of more complex computation. ISI & Equalization Suppose that we are given one more training symbol.
. we get ÁÌ µ µ µ¾ (4. Then Á ¼ Ä½ Á · Ä¾ ½ Á (4.53) (4. We seek the ﬁlters and that minimize the MSE given by E ´Á Á µ¾ and E Á E E ´Á ÁÌ .49) ¼ Ä¾ Ì Ì (4. Suppose that the feedforward ﬁlter is of order Ä½ · ½ and the feedback ﬁlter is of order Ä¾ .54) Notice that E Á Á and ¼ and E Á ÁÌ E Á ÁÌ ÁÄ¾ ¢Ä¾ . Again.52) Differentiating with respect to ´Á ÁÌ Á ´Á ÁÌ ÁÌ ÁÌ ¼ ¼ (4. Á Á .47) where Á Á Á ·Ä½ Á ·Ä½ ½ Á ½ Á ¾ Ä½ ½ ¾ Ä½·½ Á Á Ä¾ Ì (4. signiﬁcant improvement over linear equalizers can be obtained with the decision feedback equalizer. In general. i.55) (4.46) can be found by minimizing the where Á are the decided symbols. The equations for optimal reduce to · E Á ÁÌ E Á ÁÌ · 4.56) ¼ .18 EÁ Á (4. Consider a DFE with a feedforward ﬁlter of order Ä½ ·½ and a feedback ﬁlter of order Ä¾ .e. the ﬁlter coefﬁcients MSE.e.48) (4. the identity matrix. Assume perfect decision feedback. Then Á ÁÌ · ÁÌ Ì (4.50) (4. ISI & Equalization feedback part.. i.Wong & Lok: Theory of Digital Communications 4.51) Further assume that the data symbols Á are zero-mean unit-variance iid random variables.
but some long segments of it.58) E Á ÁÌ Similar to the case of the MMSE equalizer. As a matter of fact.4 Maximum Likelihood Sequence Receiver Whenever it is not practical to construct ISI-free transmission pulses. Using the whole received signal. we can employ the MAP principle to develop the optimal symbolby-symbol detector. we can use an equalizer to eliminate (or reduce) ISI. The development is rather involved and the optimal symbol-by-symbol detector is usually too complex to implement. If we do not know the expectations of the matrices above a priori. it is intuitive that the optimal receiver should observe not only the segment of received signal concerning the desired symbol. we can consider to decide the whole transmitted symbol sequence simultaneously from the received signal. we can also solve for the feedforward and feedback ﬁlters using the steepest descent approach. In fact. but the whole received signal instead. In these cases. and decision-feedback equalizers. to minimize the average symbol error probability. we are not strictly observing the whole received signal.57) (4. we can send a training sequence to facilitate the estimation of them. we can employ the ML principle to achieve our goal. and then make a decision on the current symbol based on the equalizer output. 7 We assume the use of FIR ﬁlters for the MMSE. we opt for another possibility.Wong & Lok: Theory of Digital Communications 4.19 . Because of the fact that the effect of a symbol is spread to other symbols. ISI & Equalization Solving these equations. which decides one transmitted symbol at a time. In this way. it turns out that all the equalization methods discussed in the previous section are not optimal. this strategy is also employed in the equalization techniques described previously7. 4. Instead of deciding a transmitted symbol at a time. With this sequence detection approach. we aim at minimizing the probability of choosing the wrong sequence of symbols instead of the average symbol error probability. Although this approach is simple and practical. we have E Á ÁÌ EÁ ½ ÁÌ E Á ÁÌ EÁ Á (4. LS. 4. The resulting “optimal” receiver is referred to as the maximum likelihood sequence (MLS) receiver. Here. we have no idea whether it is optimal in terms of minimizing the average symbol error probability.
60) up to time ´Ã · ½µÌ is Ã Ö Ö ¾ Ã ½ ÑÒ Ö´Øµ Ø Ò Ú ´Ø ÒÌ µ Ã ½ Ò ¼ Ã ½ ½ ½ Ã Ú´Ø ÒÌ µ ¾ Ø ÑÜ Ö´Øµ Ò Ú ´Ø ÒÌ µ Ø ¾ ½ Ò ¼ Ò Ã ½ Ò ¼ (4.e. Assuming all the bit sequences are equally probable.61) Deﬁning ÝÒ Ö £ Ú ´ÒÌ µ and ÜÒ Ü´ÒÌ µ Ã Ò Ú £ Ú ´ÒÌ µ.59) as our desired signal9.3.Wong & Lok: Theory of Digital Communications 4. although our goal is to combat ISI. ISI & Equalization Now. From the results in Section 2. i.5. decides Ã Ö Ñ Ü Ã´ Ãµ Ã (4.59) This is simply a rewrite of (4.20 .3) with the above assumptions applied. This is the problem we have solved. Ironically. 4. there is no interference in this interpretation. for Ø ½Ì Ò Ò ½ ¼ symbols per second and the transmission and channel ﬁlters are both causal. we try to decide the sequence of transmitted symbols ½µÌ 8. we can write the metric as ¼ Ò ÝÒ Ã´ Ãµ ½ ¾ Ã Ò Ã ¼Ñ ¼ Ò Ñ ÜÒ Ñ (4. let us develop the MLS receiver. we treat the whole summation on the right hand side of (4. Each bit sequence is treated as an Å -ary “symbol”. The re- ´Ã · ½µÌ .. we can employ the ML receiver to minimize the probability of making an error when determining the transmitted sequence. As mentioned above. we are equivalently considering an AWGN channel with Å Ã Ò Ò ¼ based on our observation Ö´Øµ up to time ´Ã · Å -ary communication system over the ¾Ã exhausting all the possible transmitted bit sequences.16. we know that the ML estimator of the bit sequence Ã Ã Ã ½ ¼ (4. Suppose we transmit a sequence of information bits with rate ceived signal up to time ´Ã · ½µÌ . To do so. is ¼ Ò Ú ´Ø Ö´Øµ Ã Ò ÒÌ µ · Ò´Øµ (4.63) In order to make the decision. With this point of view. the MLS receiver has to check all the ¾Ã sequences to ﬁnd the one 8 9 We cannot decide the future symbols until we have observed them.62) Then the MLS receiver. which is depicted in Figure 4.
67) Recall our goal is to maximize Ã ´ Ã µ. even when ÜÒ is not strictly of ﬁnite-support.68) ×Ã µ This is generally not true for a channel with a bandlimited response unless some form of transmission pulse design is applied. When ÜÒ is of ﬁnite-support10..16: Maximum likelihood sequence receiver that gives the largest metric. ÜÒ the metric in the following way: ¼ ¼ for Ò Ä. From (4. we can update Ã ½ Ò Ã Ä Ã´ Ãµ Ã ½ ´ Ã ½ µ · Ã ÝÃ ½ Ò ÜÒ Ã Ã Ü¼ ¾ ½ (4. (4. 4.66) where Ô´ÝÃ Ã ×Ã µ Ã ÝÃ ½ Ò ÜÒ Ã Ã Ü¼ ¾ Ã Ä ½ µ · Ô´ÝÃ Ã (4. Hence its complexity increases exponentially with time. a signiﬁcant simpliﬁcation can be applied to the calculation of the maximum metric rendering the MLS receiver practical.65) ×Ã Ã Ä ½ . i.64) can be rewritten as Ã Ä ½ µ · Ô´ÝÃ Ã ½ Ò Ã Ä Ã Ã´ Ãµ Ã ½ ´×Ã ¼ ×Ã µ ½ (4.64) to simplify the maximization of the metric.21 . However. This is clearly impractical and this is why the relatively simple equalization techniques described in the previous section are more popular in practice.66) Ñ Ü Ã´ Ãµ Ã 10 ÑÜ Ã ½ ´×Ã Ñ ×Ü ÑÃ ´×Ã µ · Ô´ÝÃ Ã ×Ã Ã Ä ½ Ã Ã Ã ×Ã µ (4.64) We are going to make use of two important observations from (4. First.e. The ﬁrst observation is that the updating part (the second term on the right hand side) depends only on ×Ã If we decompose Ã into Ã Ã Ã ½ Ã ¾ Ã Ä (4.Wong & Lok: Theory of Digital Communications 4. ISI & Equalization @t=nT r(t) ~ v(t) Calculate metrics Select largest ^ bK Figure 4. the simpliﬁcation here can still be applied approximately provided ÜÒ decays fast enough.
70) to update and store Ã Ä ½ that maximizes it: Ã Ä ½ ´×Ã µ 11 Ö Ñ Ü ÑÃ ´×Ã µ Ã Ä ½ The important observation here is that Ô´ÝÃ Ã Ã µ does not depends on Ã Ä ½ . this Ã ¿ task can be efﬁciently accomplished by an iterative algorithm known as the Viterbi algorithm.69) is the sequence that maximizes Ã ´¡µ. all the receiver needs to know is the value of ÑÃ ´×Ã µ ÑÃ ´×Ã µ for each state ×Ã . We need to calculate let each possible pattern of ×Ã ½ system at the time instant Ø arrives at time Ø tant simpliﬁcation in the maximization of Ã ´¡µ that we do not need to test all the ¾Ã patterns of Ã This contradicts with our original claim that Ã maximizes Ã ´¡µ. suppose that there is another segment ¼Ã Ä ½ such that Ã ½ ´×Ã Ã ´ Ã µ. This observation reveals the impor- ¼ 11 Ã Ä ½ µ. Fortunately. There are altogether ¾Ä states. which does not grow with time.22 . there is still a problem needed to be solved before we can beneﬁt from this large complexity reduction.68). the system changes state from ×Ã ½ to ×Ã Ã ½ Ã ¾ Ã Ä .16. where Ã Ã Ä ½ must be the one that maximizes ¼ Ã Ä ½ µ Ã ×Ã ¼ Ã Ä ½ . × 4. as long as we know the values of ÑÃ ´×Ã µ for all the ¾Ä patterns of ×Ã . ISI & Equalization where ÑÃ ´×Ã µ Ã Ä ½ ÑÜ Ã ½ ´×Ã Ã Ä ½ µ Ã (4. Hence.Wong & Lok: Theory of Digital Communications 4. However. When the new input bit Ã Ã ¾ ÃÌ .70) Ã Ä ½ Ñ Ü Ñ Ü Ã ¾´×Ã ½ Ã Ä ¾µ · Ô´ÝÃ ½ Ã Ä ¾ Ñ Ü ÑÃ ½´×Ã ½µ · Ô´ÝÃ ½ Ã ½ ×Ã ½µ ÑÃ ´×Ã µ and the Therefore. ÑÃ ´×Ã µ for each of the ¾Ä patterns of ×Ã . But Ã Ä ½ Ã Ä ¾ Ã Ä ½ Ñ Ü Ã ½´ Ã Ä ½ ÑÜ Ã ½ µ Ã ¾ ´×Ã ½ Ã Ä ¾ µ · Ô´ÝÃ ½ Ã ½ Ã ½ ×Ã ½ µ ×Ã ½µ (4. during the transition from state ×Ã ½ to state ×Ã . First. The order of complexity of the maximization reduces dramatically from ¾Ã to ¾Ä . the receiver calculates Ã ´ Ã µ making use of (4. Then it is clear from (4. ´Ã Ã Ä ½ represents a state of the communication ½µÌ . When the received sample ÝÃ is available at the output of the matched ﬁlter in Figure 4. then the segment Ã ½ ´×Ã Ã ½ ´×Ã The second equality above is due to our second important observation that if Ã ×Ã ¼ Ã Ä ½ ¡µ. we can use (4.66) that Ã ´ Ã µ Indeed.
-1] [-1.Wong & Lok: Theory of Digital Communications 4.23 .71) with Ã Ä ¾ ´×Ã ½ µ as the sequence Ã Ä ¾ that maximizes ÑÃ ½ ´×Ã ½ µ.17: Trellis diagram for the case of Ä ¾ Ã Ä ¾ ´×Ã ½ µ Ö Ñ Ü ÑÃ ½´×Ã ½µ · Ô´ÝÃ ½ Ã Ä ½ Ã ½ ×Ã ½ µ (4.+1] [+1.-1] SK-2 SK-1 SK bn = -1 bn = +1 Figure 4. Therefore.+1] [-1.73) The update relations of the states are depicted by the trellis diagram. then Ã Ä ½ ´×Ã µ needed values of ÑÃ ´×Ã µ for the determination of Ã ½ Ã Ä ¾ ´×Ã ½ µ Ã Ä ¾ ´×Ã ½ µ . Otherwise. The trellis diagram for the case of Ä ¾ is given in Figure 4. we know that the transition can only come from two previous states. This update is usually performed in a “backward” manner. ISI & Equalization SK-3 [+1. The other previous information can be discarded. For each state ×Ã namely ×¼Ã ½ Ã ½ Ã ¾ Ã Ä . We perform this update for all the ¾Ä states ×Ã at state transition to make the ½ ¼ Ö Ñ ×Ü ÑÃ ´×Ã µ · Ô´ÝÃ Ã Ã Ã ×Ã µ ÑÃÜ × Ã Ä ½ ´×Ã µ (4.72) If ×¼Ã ½ is the one chosen above.17. Ã Ä Ã ¾ Ã ¿ Ò ½ and ×¼¼ ½ Ã Ã ½ Ã ¾ Ã ¿ Ã Ä ½ . ÑÃ ´×Ã µ Ñ Ü ÑÃ ½´×¼Ã ½µ · Ô´ÝÃ ½ ¼ Ó ×¼Ã ½µ ÑÃ ½ ´×¼¼ ½µ · Ô´ÝÃ ½ Ã ½ ×¼¼ ½µ Ã Ã (4. Ã Ä ½ ´×Ã µ . 4.
First. although the MLS receiver minimizes only the sequence error probability.Wong & Lok: Theory of Digital Communications 4. its symbol error performance is usually better than that of the equalization techniques.24 . ISI & Equalization Let us end our discussion on the MLS receiver by making two remarks. the Viterbi algorithm described above can be easily generalized to the cases of non-binary symbols. 4. Second.
This action might not be possible to undo. Are you sure you want to continue?