Linear Equalizers

Zero Forcing and Mean Square Error

 Objectives

At the end of this session, you

will know about
the concept of Linear Equalizers
Zero forcing
Mean square error
 Session Outcome
At end of this session learners can
Understand the concept of Linear Equalizers
Why Zero forcing equalizers are not preferred
always?
Classification of Equalizers
 Linear Equalizers -Introduction
• Linear equalizers are simple linear filter structures
that try to invert the channel in the sense that the
product of the transfer functions of channel and
equalizer fulfills a certain criterion.
• This criterion can either be achieving a completely
flat transfer function of the channel–filter
concatenation, or minimizing the mean-squared
error at the filter output.
• A transmit sequence {ci} is sent over a dispersive,
noisy channel, so that the sequence {ui} is
available at the equalizer input.
 Linear Equalizers -Introduction

Linear equalizer in the time domain (a)

Time-discrete equivalent system in the z-transform domain (b).
 Linear Equalizers -Introduction

We now need to find the coefficients of a Finite Impulse Response

(FIR) filter (transversal filter, Figure 16.3) with 2K + 1 taps. This filter
should convert sequence {ui } into sequence { ^ci }:
Linear Equalizers -Introduction
Linear equalizer-Principle
 Zero-forcing equalizer
• The ZF equalizer can be interpreted in the frequency
domain as enforcing a completely flat (constant) transfer
function of the combination of channel and equalizer by
choosing the equalizer transfer function as E(z) = 1/F (z).
• In the time domain, this can be interpreted as minimizing
the maximum ISI - (peak distortion criterion).
• Peak distortion is simply defined as the worst case Inter
symbol Interference at the output of the equalizer.
• The minimization of this performance index is called
Peak distortion criterion.
• Consider the minimization of the peak distortion assuming
that the equalizer has an infinite number of taps.
 Zero-forcing equalizer
The zero forcing equalizer is designed to remove
the ISI completely
 Linear equalizer-Zero Forcing
Equalizer
• A serious problem with the zero-forcing equalizer
is the noise enhancement, which can result in
infinite noise power spectral densities after the
equalizer.
• The noise is enhanced (amplified) at frequencies
where the channel has a high attenuation.
• Another, related, problem is that the resulting
noise is colored, which makes an optimal detector
quite complicated.
• By applying the minimum mean squared-error
• The ZF equalizer is optimum for elimination of ISI. However,
channels also add noise, which is amplified by the equalizer. At
frequencies where the transfer function of the channel attains small
values,
• the equalizer has a strong amplification, and thus also amplifies the
noise. As a consequence, the noise power at the detector input is
larger than for the case without an equalizer
Linear equalizer- MSE Equalizer
 Linear equalizer- MSE Equalizer
• The MSE equalizer removes the most problematic noise
enhancements as compared to the ZF equalizer. The noise
power spectral density cannot go to infinity any more.
• This improvement from a noise perspective comes at the
cost of not totally removing the ISI.
• The ultimate goal of an equalizer is minimization, not of
the ISI, but of the bit error probability.
• Noise enhancement makes the ZF equalizer ill-suited for
this purpose. A better criterion is minimization of the
Mean Square Error (MSE) between the transmit signal and
the output of the equalizer.
 Adaptation Algorithms for Mean Square Error Equalizers
In order to find the optimum equalizer weights, we can directly solve
Eq. . However, this requires on the order of (2K + 1)3
complex operations. To ease the computational burden, iterative
algorithms have been developed. The quality of an iterative algorithm
is described by the following criteria:
• Convergence rate: how many iterations are required to “closely
approximate” the final result? It is usually assumed that the channel
does not change during the iteration period. However, if an algorithm
converges too slowly, it will never reach a stable state – the channel
has changed before the algorithm has converged.
• Misadjustment: the size of deviation of the converged state of the
iterative algorithm from the exact MSE solution.
• Computational effort per iteration.
In the following, we discuss two algorithms that are widely used –
1. The Least Mean Square (LMS) and
 The Recursive Least Squares Algorithm
• In most cases, the LMS algorithm converges very slowly. Furthermore, the use of
this algorithm is justified only when the statistical properties of the received signal
fulfill certain conditions.
• The general Least Squares (LS) criterion, on the other hand, does not require such
assumptions. It just analyzes the N subsequent errors εi , and chooses weights such
that the sum of the squared errors is minimized.
• This general LS problem can be solved by a recursive algorithm as well – known as
RLS
 Comparison of Algorithms
There are two classes of algorithms for the determination of
equalizer coefficients:
1. direct implementation (Wiener filter, LS criterion) and
2. Iterative methods (LMS, RLS).
• For the Wiener filter, we first have to determine the
correlation matrix; this can be the major part of the
numerical effort especially if the number of weights is
small.
• The actual inversion of the matrix requires (2K + 1)3
operations. Alternatively, we can use the data matrix
directly, and invert it (LS algorithm). Construction of the
data matrix requires less numerical effort than the
correlation matrix;
• When comparing iterative algorithms, we find that the LMS algorithm usually
converges too slowly.
• The RLS algorithm converges faster, but has a larger residual error. Figure shows

• the typical example of the MSE for a Digital Enhanced Cordless

Telecommunications (DECT) cordless telephone. We see that the RLS algorithm
has converged after 10 bits, while the LMS algorithm requires almost 300 bits.
• Due to the temporal variance of the channel, as well as spectral efficiency
 Summary
• Zero forcing equalizer
• Minimum mean square error Equalizer
• Disadvantages of Zero forcing
equalizer.
• The Least Mean Square (LMS)
• The Recursive Least Square (RLS).
 Test your Understanding
1. Explain the basic structure of Linear equalizer
and compare ZF and MMSE in terms of noise
2. Illustrate the classification of equalizer.
3. What is noise whitening filter in equalizer
 References
• “Wireless Communications – Andreas F. Molisch”