Vous êtes sur la page 1sur 16

Department of Electrical and Computer Engineering Faculty of Engineering and Architecture American University of Beirut

EECE 442L Communications Laboratory

Experiment on

Pulse Amplitude Modulation over InterSymbol Interference Channels: Part II

Version: August 2009

Pulse Amplitude Modulation over InterSymbol Interference Channels: Part II


OBJECTIVES Understand the building blocks of the receiver structure for PAM over ISI channels. Demonstrate and analyze demodulation using matched filters. Investigate various equalization techniques to combat ISI.

THEORETICAL BACKGROUND In part I of this experiment, we have investigated the transmitter side of the PAM system over linear Gaussian channels. In Part II, we will analyze the basic parts which constitute the receiver. In addition, we will discuss various techniques that can be implemented at the receiver side to improve the performance of the communication system. Figure 1 presents the block diagram of the entire system.

10101101

PAM Modulator {I n , Am , g (t )}

s (t )
Channel

c (t )

N (t )
$n I
1010 Detector Equalizer

Yk

Y (t ) Whitening Filter Ts

Matched Filter

h * (t )

R (t )

Figure 1: Block diagram of the PAM communication system where In is the discrete informationbearing sequence of symbols, g(t) is the pulse shaping filter, c(t) is the channel impulse response, N(t) is the additive noise term and h(-t) is the matched filter impulse response.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page1

A. THE RECEIVER The function of the receiver is to recover the sequence of transmitted symbols {I n } . The information pertaining to this sequence is embedded in the received signal R (t ) along with the effect of the channel (noise, delay, fading, ISI) and the modulator (carrier frequency, pulse-shaping filter, ISI). The Receiver usually consists of a chain of diverse components that include: Matched Filter, Whitening Filter, Sampler, Equalizer, and Detector. B. ADDITIVE WHITE GAUSSIAN NOISE (AWGN) Noise modeled as white has a constant power spectral density of value No 2 over the whole spectrum. In the time domain, this means the autocorrelation function of the noise is an impulse at t = 0 . This noise model simplifies the mathematical and statistical properties of the communication system. However, this model is idealistic and nonrealizable because of its infinite spectrum. C. THE MATCHED FILTER AND SAMPLER The matched filter to a signal h (t ) is a filter whose impulse response is h (T t ) where

h (t ) is assumed to be confined to the time interval 0 t T . The time shift T is to


ensure causality. The signal h (t ) is defined to be the response of the channel with impulse response c (t ) when excited with a pulse g (t ) :

h (t ) = g (t ) c (t ) =

g ( )c (t )d

Note: When the signal h (t ) is introduced at the input, the response of the matched filter is

a (t ) = h (t ) h (T t ) = h ( ) h (T t + )d
0

which is basically the autocorrelation function of the signal h (t ), shifted to the right by

T . Since the non-shifted autocorrelation function is an even function peaking at 0 ,


PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page2

therefore, the response of the matched filter a (t ) will be symmetric over the interval
0 t 2T and with a peak at t = T . More importantly, the output of the matched filter is

a projection on the signal space of the h (t ) and thus it is considered sufficient statistics by the theorem of irrelevance for AWGN channels. This property results in a maximized signal-to-noise ratio (SNR) making the matched filter attractive to use at the demodulation stage in a communication system. In the PAM system shown in figure 1, the response of the matched filter to the received signal R(t ) is

Y (t ) = T (t ) h (TS t )
= In h (t nTS ) h (TS t ) + N (t ) h (TS t )

= I n a (t nTS ) + Z (t )
n

Therefore, the output of the matched filter Y (t ) is equivalent to a signal formed from the discrete information-bearing sequence of symbols {I n } modulated with a pulse shaping filter a (t ) . The sampling is done at Ts seconds. The output of the sampler is a is a discrete sequence

{Y k}

Y (kTS ) = I n a (kTS nTS ) + Z (kTS )


n

Y k = I n a k n + Zk = a 0 I k + I n a k n + Zk
n nk

where a n represents the samples of the autocorrelation function of h (t ) , taken periodically at 1 TS and zk denotes the additive noise sequence of the output of the matched filter. The output of the matched filter at k -th sampling instant contains the desired information symbol I k corrupted by the ISI component colored noise term zk .
PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page3

Inak n n k

and by the

D. THE WHITENING FILTER


At the output of the sampler,

Y k =

n = L

I n a k n + Zk

Consequently, the communication system of Figure 1 can be replaced with an equivalent discrete channel model (see Figure 2).

{I n }

Yk
A (z )

Zk
Figure 2: Equivalent discrete channel model with colored additive noise.

In this model, the noise sequence terms Zk are correlated (i.e the noise is colored). Since it is more convenient to deal with a white noise sequences we whiten the noise sequence

{Zk } . It can be shown that this may be achieved by passing the sequence {yk }into a filter
that has a z -transform equal to 1 F (z 1) where

A(z ) = F (z )* F * (z 1 ) ,
referred to as spectral factorization of A( z) . The equivalent discrete channel model then becomes as shown in Figure 3.

{I n }

yk
A (z )

Noise Whitening Filter 1 F (z 1)

vk

zk
Figure 3: Equivalent discrete channel model with a noise-whitening filter.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page4

Consequently, the output of the digital filter 1 F (z 1) is a sequence {vk } that can be expressed as:

vk = fn I k n + k
n =0

Where {k } is a white Gaussian noise sequence, { fk } is a set of tap coefficients of an equivalent discrete-time transversal filter having a transfer function F (z ) and L represents the memory of the FIR filter { fk } . In summary, the cascade of the transmitting filter g (t ) , the channel c (t ) , the matched filter h * (t ) , the sampler, and the discrete-time noise whitening filter 1 F (z 1) can be represented as an equivalent discrete time transversal filter having the set { fk } as its tap coefficients as shown in Figure 4.

{I n }

z 1

z1

z1

z 1

f0

f1

fL1

fL

{v k }
{ k }
Figure 4: Equivalent discrete-time model of inter-symbol interference channel with WGN.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page5

E. EQUALIZERS
The output of the equivalent discrete-time white-noise filter model of Figure 3 contains ISI. One approach is to use equalizers to compensate for the interference effect. The output of the equalizer is the estimated sequence of transmitted symbols {I n } . There are two main types of equalizers, namely, Linear equalizers and Decision-feedback (nonlinear) equalizers.

E.1 LINEAR EQUALIZERS


Linear equalizers (LE) consist of a linear filter followed by a symbol-by-symbol detection (see Figure 5).

{k }
{I n }

F (z )
{v k }

L (z )

{ Wn }

{I n }

Wn = I n + n where n is a noise term


Figure 5: General scheme of a linear equalizer.

Different approaches exist as to the assignment of L (z ) leading to different types of LE. We will present two types of LE: the Zero-Forcing LE (ZF-LE) and the Minimum MeanSquare Error LE (MMSE-LE).

E.1.1 Zero-Forcing Linear Equalizer (ZF-LE)


The concept of the ZF-LE lies in choosing L (z ) so that zero ISI occurs. Therefore we choose to have

L (z ) =

1 F (z )

One of the advantages of the ZF-LE is that it does not depend on the noise statistics, however, its performance is poor whenever the channel frequency response exhibits a null or near null since the ZF-LE will amplify the noise when trying to eliminate ISI .
PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page6

E.1.2 Minimum Mean-Square Error Linear equalizer


In a Minimum Mean-Square Error Linear Equalizer (MMSE-LE), L (z ) is chosen to minimize the mean-square error defined by

E [W n I n ]
The expression of the filter that minimizes the mean square error is given by:

L (z ) =
where

F *(z 1 ) F (z ) + N 0 E s
2
2

Es = [ I k ]
and

N 0 = [ k ]
Compared to the ZF-LE, the MMSE-LE takes into account the noise statistics (by introducing the noise variance term N 0 ). Yet, when the SNR is high, the filter L (z ) asymptotically approach 1 F (z ) and consequently, the MMSE-LE will be acting as ZFLE. The major problem when using ZF-LE is commonly referred to as noise enhancement. Since a ZF-LE essentially inverts the channel, it can result in infinite noise power spectral densities after the equalizer whenever the channel frequency response exhibits a null. With the MMSE-LE, the ISI is not necessarily removed. The resulting noise is further colored. To resolve the issue of noise enhancement, a feedback mechanism is employed to implement the L (z ) filter as shown in Figure 6.

{I n }

F (z )

{v k }
+ -

{ Wn }

{I n }

{ k }

F (z ) 1

Figure 6: ZF-LE implement with a feedback mechanism. PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page7

E.2 DECISION-FEEDBACK EQUALIZERS (DFE)

If the feedback mechanism was initiated at the output of the detector, a nonlinear equalizer called a decision-feedback equalizer is obtained. In such equalizers, the purpose of the feedback is to eliminate ISI by subtracting the interference caused by previously detected symbols before detecting the next symbol. In the experiment, we will consider zero-forcing decisionfeedback equalizers (ZF-DFE).
E.2.1 Zero-Forcing Decision Feedback Equalizer (ZF-DFE)

In a Zero-Forcing Decision Feedback Equalizer (ZF-DFE), not only the threshold detector is placed inside the feedback loop, but the loop filter is chosen to cancel out the effect of the FIR filter F ( z) (see Figure 7).

{I n }

F (z )

{v k }
+ -

{ Wn }

{I n }

{ k }

F (z ) 1
Figure 7: Implementation of the ZF-DFE: The threshold detector is placed inside the feedback loop.

E.2.3 ZF-DFE versus ZF-LE with Feedback

In both cases of the ZF-LE with feedback and the ZF-DFE, we need Wn to be close to I n to achieve the best equalization possible, i.e. Wn should be equal to I n plus a certain noise term:

Wn = I n + n
In the case of ZF-LE with feedback (i.e. the detector outside the feedback loop), we have

W n = v n g 1W n 1 ... g L W n L
PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page8

Consequently, the noise is being accumulated leading to noise enhancement. This is due

W n } s which are not exactly equal to {I n } s. to the fact that we are subtracting {
However, in the case of the ZF-DFE (i.e. the detector is placed inside the feedback loop), the feedback filter uses the estimated sequence {I n } to perform the subtraction

W n = v n g 1 I n 1 ....... g L I n L = I n + n
Therefore, providing that the decisions are correct, the noise enhancement is avoided, the ISI is removed and the only remaining noise is the original noise n inserted by the channel. The ZF-DFE has however several disadvantages: Non-linear operation: variation in the input will not result in corresponding linear variations at the output. Error propagation: The ZF-DFE is based on the decisions of the detector. Any error made by the detector will propagate since we will keep subtracting the wrong value of I n from v n . It is worth noting that if the detector is ideal, the ZF-DFE is a very good detector. It is actually implemented in the telephone-line modem.

For more background information on PAM receiver structures over ISI channels including equalization, check Sections 8.1, 8.3, and 8.6 in [1], Sections 7.7-7.9 in [2], and/or Sections 11.1-11.7 in [3].

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page9

EXPERIMENT DESCRIPTION GENERAL RULES

If you open a VI and are not asked to do any changes in it, then close it without saving changes by clicking on Defer decision.

Save VIs as [GroupID]_name of VI.vi.

PART I: WHITE NOISE

In this part, you will study the spectral and time domain properties of the additive white Gaussian noise.
A. THE AWGN VI Q.1 Q.2 What is a White process? Calculate the Autocorrelation of White Noise.

SW ( f )} HINT: RW ( ) = F 1 {
In a new VI, generate a Gaussian White Noise sequence. Use Gaussian White
Noise.vi with standard deviation of 0.001 and 1000 samples. Q.3 Based on Q.2, suggest a way to verify that the noise is white. HINT: You need to generate a graph plot in the time domain.

Implement your idea in Q.3 on the white noise sequence in your VI.
Q.4 Provide a plot of the graph obtained. Is the noise white?

PART II: THE RECEIVER

In this part, you will examine the basic structure of a receiver. You will go over the operation of the matched filter and the noise-whitening filter.
A. THE MATCHED FILTER VI

Open the Block Diagram of MatchedFilter.VI.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page10

Q.5

Explain how the matched filter is implemented. What is the role of Reverse 1D Array function (marked with a frame)?

Briefly going over the sampler.


Q.6 A Sampler usually follows a matched filter. At what rate should the sampling occur? Q.7 What is the relation between the spectrums of the sampler output and the sampler input (i.e. the output of the matched filter)? B. THE WHITENING FILTER VI

Open Whitening.VI. Add spectrum plots before and after the Whitener.
Q.8 Q.9 What is the role of a whitening filter? Provide plots of the two spectra.

PART III: THE EQUALIZERS

In this part, you will study the operation of equalizers within a complete communication system. Two types of equalizers will be used, the ZF-LE and the ZF-DFE.

A. THE SYSTEM VI

Notes: [1] The Channel used in this lab has the following causal impulse response:
c ( t ) = 5 ( t ) 2 ( t TS )

[2] The whitening filter was not included in the system. Build the VI shown in Figure 8 and save it as [GroupID]_SystemLE.VI. Replace the ZF-LE.VI equalizer with the ZF-DFE.VI equalizer and re-save it (SAVE AS not SAVE OVER) as [GroupID]_SystemDFE.VI. Delete ZF-LE.VI equalizer and connect directly the output of the sample to the detector. Re-save it (SAVE AS not SAVE OVER) as [GroupID]_SystemNoEq.VI.
PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page11

Figure 8: Complete communications system with equalization.

The VIs used in Figure 8 are the following:


RootRaisedCosinePulseShape.vi.VI

Generates the Root Raised Cosine Pulse Shaping Filter Coefficients.


PAMmodulate.VI

Performs PAM modulation of a binary sequence using the pulse shape coefficients input into Filter Coefficients.
Channel.VI

Represents a linear Gaussian channel with AWGN noise.


MatchedFilter.VI

Performs matched filtering based on the channel and pulse shaping coefficients.
Sampler.VI

Performs sampling at rate RS.


PAMdetect.VI

Does PAM detection and outputs the detected bitstream.


BER.VI

Computes the BER by comparing the detected bitstream with the original bitstream.
ZF-LE.VI

Zero-Forcing Linear Equalizer.


ZF-DFE.VI

Zero-Forcing Decision Feedback Equalizer.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page12

Prepare all three VIs to be used as subVIs. Connect Nb of bits and N 0 ( dB ) as inputs and
BER as an output.

These three VIs will be used later in this experiment as subVIs on the same block diagram to compare their BER over a span of noise levels. To tell them from each other, you need to edit their icons and write LE, DFE and NoEq accordingly. Make sure to save your work (both [GroupID]_SystemLE.VI and
[GroupID]_SystemDFE.VI)

Open [GroupID]_SystemLE.VI and add waveform graphs to:


1. s (t ) : The PAM Waveform output of PAMmodulate.VI 2. r (t ) : The Output with Noise output of Channel.VI 3. a (t ): The Output of MatchedFilter.VI

[n] : The Equalizers Output. 4. I


Set N0 ( dB ) to 10 and Nb of bits to 10 and run.

Q.10 When do we need an equalizer in a communication system? Q.11 Try the following range of parameters: Nb of bits = 50,

N0 = 10,0,10, 20,30. Provide plots of all 4 graphs for the cases of


lowest and highest BER.

Close [GroupID]_SystemLE.VI without saving.


Q.12 Write a(t ) in terms of s(t ) , c(t ) (channel impulse response), n(t ) (noise impulse response), m(t ) (matched filter impulse response). Q.13 Open the SubVIs ZF-LE equalizer and ZF-DFE equalizer. Comment on the main difference, what really happens inside both equalizers.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page13

Q.14 Open PAMdetect.VI. How is the detector implemented? Q.15 Open BER.VI. Explain the method used to calculate the bit error rate in this VI. B. BER VS. N0 GRAPHS

Now, we will build a VI to draw BER vs Noise Level graphs. As a reminder: You need two nested loops.
o The inner loop runs with certain parameters many times to calculate many

values for the BER that will be averaged in the outer loop.
o The outer loop scans over many values of Noise Level.

Use Figure 9 as a starting point and apply the reminders above to replace the box labeled Loop Structure.

The inner loop must contain all three subVIs you created before, namely
[GroupID]_SystemLE.vi, [GroupID]_SystemNoEq.vi. [GroupID]_SystemDFE.vi,

In choosing the number of times the inner loop runs and the number of bits to pass to the systems keep your chosen values below 100 or the run will take too long. Suggested values are 10 for the number of runs of the inner loop and 100 for the number of bits. Note that the VI will take some time so be patient.

Figure 9

Save your VI as [GroupID]_BERvsN0.vi.


PAMoverInterSymbolInterferenceChannels:PartIIAugust2009 Page14

Q.16 Provide a plot of the XY Graph. Compare the performance of both equalizers and the no-equalizer case and comment why one is better than the other.

REFERENCES

[1] J. Proakis and M. Salehi, Communication Systems Engineering. Prentice-Hall, 2nd edition, 2002. [2] S. Haykin, Communication Systems. John Wiley & Sons, 3rd edition, 1994. [3] A. Goldsmith, Wireless Communications. Cambridge University Press, 2005.

PAMoverInterSymbolInterferenceChannels:PartIIAugust2009

Page15

Vous aimerez peut-être aussi