Channel Equalization in Digital Communication

Systems
Kananelo Chabeli
Department of Electrical Engineering
University of Cape Town
Cape Town, South Africa
[email protected]

Abstract—Channel equalization is the process for mitigating the overall system design is presented together with detailed
distortions introduced by digital channels with the intent of description of adaptive algorithm used. Furthermore, sections
improving the transmission performance. The major challenges V and VI presents the simulation results and their analysis.
facing digital communication systems are inter-symbol inter-
ference and multi-path interference. This paper presents and The last section draws conclusions based on the results
implementation of adaptive channel equalizer and simulates it on found.
software to validate its performance. The results show that the
overall bit error rate and speech quality of the equalized output II. L ITERATURE R EVIEW
at the receiver are much better than that of the unequalized.
Index Terms—component, formatting, style, styling, insert A. Introduction to Literature Review

I. I NTRODUCTION In this section, a brief review of the literature on channel

equalization and digital communication systems is provided.
Digital communication systems have significantly The section starts by looking into digital systems, the pros and
outperformed their analog counterparts in terms of efficiency cons, and then looks in to equalizer systems that have been
and reliability for data transmission,[2]. However, as the proposed in literature.
demand for higher data rates escalates, these systems face
challenges such as signal distortion, increased bit-error
B. Digital Communication Channels
rate(BER),and reduced signal-to-noise ratio mainly because
of inter-symbol interference (ISI). ISI arises significantly from A typical communication channel consists of a transmitter,
the band-limited nature of a digital channel and, in context channel and a receiver. The purpose of a digital communi-
of cellular networks, from multi-path propagation.[2],[4]. ISI cation system is to transmit data or information from the
is defined as the phenomenon in which signals (symbols) are transmitter to the receiver at lowest possible latency . Factors
smeared out such that pulses belonging to different symbols that need to be considered when designing communication
are indistinguishable[3] systems include type of signals to be transmitted(analog or
Channel equalizers play a pivotal role in this context by digital), receive power, and error performance [1].
counteracting the channel distortions and recovering the 1) Digital Channel Models: The most widely used channel
transmitted information. Ideally, the impulse response of the models in literature are additional noise channel and band-
channel equalizer system should be an inverse of the channel limited FIR filter channel models [7]. An additive channel is
frequency response, ensuring that cascade effect of channel one which is completely modelled by white Gaussian Noise.
and equalizer results in the system function of z δ , where δ is On the other-hand, filter channel is one modelled with a filter,
some delay, but effectively restoring signal clarity [3]. usually low-pass FIR filter added to additive Noise. These
This study investigates the impact of an adaptive equalizer on models are shown in the fig.1.
signals transmitted through noisy, linear, and time-invariant 2) Challenges facing Digital Communication Systems:
digital channels. It is being hypothesized that applying an The most commonly occurring challenges are inter-symbol
adaptive equalizer reduces the bit error rate,and improves interference and non-linearity of physical channels. To get a
speech quality of audio effectively mitigating channel sense of what ISI is, consider a digital signal x[n] transmitted
distortion effects. over a digital channel characterised by a linear finite-impulse
response filter h[n] and additive noise v[n]. the received
The rest of the paper is organised as follows: section II signal,r[n] is given by r[n] = h[n] ∗ x[n] + v[n]. which can
presents a review of the literature in digital communications be written as:
and channel equalization. Next, section III provides the
method of investigation taken in the project. It highlights r[n] = x[n] ∗ h[0] + Σ∞
k=−∞,k̸=0 h[k]x[n − k] + v[n] (1)
the procedure followed, simulation tools and how the system
was tested to verify the proposed hypothesis. In section IV The term Σ∞
k=−∞,k̸=0 h[k]x[n − k], is is the ISI [5].
 individual code cells. Python, known for its extensive range
of signal processing packages, was utilized within the Jupyter
Notebook environment to enhance the development and testing
phases.
B. System Implementation Procedure
(a) Additive noise channel model [1]. The implementation began with the modeling of the digital
channel, which was represented as a Finite Impulse Response
(FIR) filter with additive zero-mean noise. To enable the
easy variation of noise intensity, the noise variance was kept
adjustable. The channel coefficients were generated using the
(b) Filter Channel Model pyFDA tool, chosen for its ease of use and availability at no
cost.
Fig. 1: Common Channel Models
The simulation was structured into two stages to efficiently
gather results:
Training Stage: This involved transmitting a known signal
C. Channel Equalization
at the receiver to learn the channel characteristics, allowing
1) Zero-forcing Channel Equalizer: The Zero-Forcing for the proper adjustment of the equalizer coefficients.
(ZFE) equalizer stands as one of the simplest and earliest Decision-Directed Stage: Following the adjustment of the
channel equalization techniques, characterized by its approach equalizer coefficients, this stage entailed applying the channel
of inversely mirroring the communication channel’s frequency to various signals and subsequently measuring the resulting
response to counteract distortions [1]. This is achieved by bit error rates and Perceptual Evaluation of Speech Quality
transmitting a noise-like training sequence rich in spectral (PESQ) scores.
content through the channel, thereby learning its impulse
response to accurately derive the Finite Impulse Response C. Approaches of System Validation
(FIR) of the equalizer [1][7].Despite its foundational status, Validation of the system was performed using two principal
the ZFE equalizer harbors significant drawbacks, including metrics: Bit-Error Rates (BER) and Perceptual Evaluation of
attenuation of signals at higher frequencies and a tendency to Speech Quality (PESQ) for the transmitted audio signals. The
amplify channel noise due to its disregard for additive noise BER was calculated by comparing the bit representations of
present in the channel [1]. transmitted, received, and equalized signals under different
2) Adaptive Equalization Techniques: An adaptive equal- channel conditions, achieved by altering the noise variance.
izer combats inter-symbol interference by dynamically learn- For the PESQ evaluation, the pesq Python package was
ing channel characteristics and adjusting its filter coefficients. employed to compute scores for the audio signals. This
It comprises two phases: the filtering process, which cleans approach provided a quantitative assessment of the speech
the corrupted input signal, and the adaptive process, adjusting quality as perceived by listeners, further affirming the system’s
coefficients to minimize a cost function. This facilitates two effectiveness in practical scenarios.
operational modes: training mode, utilizing a known signal
to learn channel characteristics, and decision-directed mode, IV. S YSTEM D ESIGN
applying adjustments for current conditions, thus adapting to A. Introduction
the channel’s changing nature [5] [3] [6]. Common adaptive This section presents the detailed system design specifi-
equalizers are the Least Mean Square (LMS) and Recursive cations and implementation of the algorithm. A thorough
Least Squares (RLS) algorithms. LMS minimizes mean square derivation of the LMS is also presented. The sections starts
error between desired and actual outputs, making it easy to by specifying the metrics tested by the project as per the
implement but may have slow convergence. Conversely, RLS hypothesis. Next process of simulation set-up and some of
accelerates convergence by including historical error values in the python functions used are presented.
its computation, enhancing efficiency.
B. System Performance Metrics
III. M ETHODOLOGY
The system performance metrics, derived directly from he
A. System Design Choices hypothesis are as follows:
The design process commenced with the selection of an • Bit-Error-Rate(BER): Computed by representing sig-
appropriate equalization algorithm. The Least Mean Square nals in their raw bits and counting the number of cor-
(LMS) Adaptive equalizer was chosen for its simplicity, facil- responding bits that are not equal and dividing by total
itating a detailed analysis of the hypothesis metrics. bits.
For the implementation and simulation phase, the selection • Speech Quality: Evaluated by using the Perceptual
of suitable tools was critical. Jupyter Notebook was selected Evaluation of Speech Quality(PESQ). This as achieved
due to its flexibility, notably its feature that allows executing by using python package pesq. The function takes in
 reference and degraded audios and return the speech python function that takes in a signal data symbol and filters
quality score. it is shown below:

C. Channel Model def fir_filter(in_val, current_ins, fir_c):

output = 0.0;
The digital channel used for simulating the Least Mean
if len(current_ins)!= len(fir_c):
Square adaptive algorithm is modelled as a low-pass FIR filter
return 1
with additive zero-mean white Gaussian noise. The low-pass
#shift the current inputs:
response was chosen specifically to reflect the band-limited
N= len(current_ins)
nature of practical communication channels. The following are
for i in range(N-1,-1,-1):
the filter specification for FIR filter used:
current_ins[i]= current_ins[i-1]
• Must be minimum-order low-pass FIR filter designed #add thre current input at index 0
using equiripple algorithm. current_ins[0] = in_val;
• Must have cut-off frequency of ωc = 1/3π, 0.1dB ripple #now find the output value
for ω < 0.8ωc and at least 30dB attenuation for ω > for i in range(N):
1.1ωc . output+=current_ins[i]*fir_co[i]
The filter coefficients were generated using pyFDA tool and return output;
were fixed for the entire simulation. Channel characteristics
As the code snippet shows, the FIR filter is implemented to
were varied through changing of noise variance.
reflect how practical filter would be implemented in hardware.
D. Least Mean Square (LMS) Algorithm This approach gave much flexibility in adding some delays in
between the symbols effectively realizing a real communica-
The Least Mean Square (LMS) algorithm’s implementation
tion system.
in filter design involves adaptive adjustment of equalizer
coefficients (bk [n]) to minimize mean square error (MSE). F. Equalizer Implementation
Key notations include x[n] for the transmitted data symbol The equalizer system was realized by two functions:
at time nTs , f [n] for the received symbol, bk [n] for the k th equalizer f ilter() and update coef f icients(). The latter
equalizer coefficient, and y[n] for the equalizer output. The was used during training and the former during decision-
error signal is defined as e[n] = x[n] − y[n] and the MSE directed stage. Furthermore, the function equalizer f ilter(),
objective function is formulated as M ean|x[n] − y[n]|2 = uses f ir f ilter() function with received symbols and learnt
M ean|x[n] − ΣN 2
k=0 b[k]f [m − k]| . To minimize MSE, the equalizer coefficients, passing each symbol one by one into
algorithm employs gradient descent, adjusting bk [n] iteratively the filter, and recording the output.
based on the equation bk [n+1] = bk [n]+µe[n]f [n−k], where
µ denotes the step-size, crucial for balancing convergence V. S IMULATION R ESULTS
speed and stability. Typically, µ = 0.01 is used. This process A. Training Results
dynamically adapts the filter coefficients to the changing
Channel FIR coefficients were generated using pyFDA, and
channel conditions, aiming for optimal signal reconstruction.
the resulting taps are:
This is shown in the figure below.
channel_coefficients =
np.array([0.014639903439961355,
-0.016192490896558002,-0.00472899759069931,
0.0610807859152884,-0.1421777303627756,
0.21559737364151998,0.7547096850970777,
0.21559737364151998,-0.1421777303627756,
0.0610807859152884,-0.00472899759069931,
-0.016192490896558002,0.014639903439961355])
The training signal used in this simulation was a synthesized
sinusiod generated as follows:
Fig. 2: LMS Equalizer Block Diagram.
n = np.arange(0,100)
training = 2*np.cos(np.pi/25*n)
E. Simulation Set-up + np.sin(np.pi/4*n)
The system was simulated in a Jupyter notebook. Unlike This sequence was used as a training sequence to learn the
conventional discrete-time signal processing, here x[n] rep- channel coefficients. It was ’transmitted through the channel’
resents signal amplitude at time nT s, reflecting real-world by applying the channel filter and adding additive noise. Figure
filtering processes. This approach diverges from theory, which below shows the plot of the resulting equalizer output.
assumes knowledge of all signal symbols for convolution. The
 Fig. 3: Channel Impulse Response Filter Fig. 6: Channel Coefficient Plot learnt by LSM algorithm
.

channel filter were applied to this signal and additive noise,

with same statistical properties as for training stage, were
added to obtain the channel output. This output were then
filtered with the ’learnt’ equalizer coefficients and the figure
.7 shows the plot of the resulting signals.

Fig. 4: Plot of the training Sequence.

Fig. 7: Plot of signals obtained from training sequence.

2) Real Audio Signal Results: To get an approximate analy-

sis of how the system performs with real signal,audios samples
were used . The plot of the simulation with one of the audios
is shown below.

Fig. 5: Plot of signals obtained from the training

These coefficients were then used to filter a test signal passed

through the channel.

B. Testing Results
For testing stage, synthesized and real audio signals were
used.
1) Synthesized Test Signal Results: The synthesized test Fig. 8: Testing Results with Audio(zoomed in)
signal was a sinusoidal of different frequency. This was
generated by the code: C. System Performance Results
test1 = 3*np.cos(np.pi/60*n)+ The system performance was measured using Bit-Error Rate
4*np.sin(np.pi/10*n) and Perceptual Evaluation of Speech Quality. The simulation
was run several times with different channel characteristics(
different noise variance), and the BER and PESQ were mea-
sured in each iteration.
TABLE I: BER data for Synthesized Signals
Noise Standard Deviation Equalized BER Unequalized BER
0.001 0.486 0.525
0.05 0.498 0.519
0.1 0.497 0.507
0.15 0.5 0.522
0.2 0.493 0.53
0.25 0.5 0.505
0.3 0.499 0.505
0.35 0.506 0.509
0.4 0.494 0.511 Fig. 10: PESQ Scores of Equalized and Unequalized audio
0.45 0.509 0.524 outputs
0.5 0.506 0.507

signal, as shown in Fig.4. The channel outputs and the

corresponding errors are depicted in Fig.7. As illustrated by
these plots, the discrepancy between the desired signal and
the equalized output significantly diminishes, demonstrating
the system’s capability to reduce channel distortions on the
transmitted signal effectively.
Moreover, a channel was applied to a test signal and the
results are shown in fig.7. These plots reflect that the system
error is high for a different signal.

Furthermore, it was hypothesized that the equalizer system

could lower the bit error rate (BER) and improve the quality
of audio signals. This was examined by comparing the BER
Fig. 9: Noise standard deviation vs BER for synthesized test
of channel outputs, with and without equalization, under
signals.
various noise conditions. The resulting BERs, as a function
of noise standard deviation, are tabulated in Table I and
1) BER Performance Test: illustrated in Fig.9. The figure clearly shows that the BER
2) PESQ Performance Test : To quantify how the system for the equalized output is consistently lower than that of
performs with real audio, audio signals were transmitted and the unequalized output. Additionally, Fig.9 indicates that
equalized with the coefficients estimated during training stage. the overall BER of the output escalates with an increase
Python package pesq were used to obtain PESQ score for in noise standard deviation, which aligns with expectations:
equalized and equalized outputs and these are recorded in the higher standard deviation implies more noise, causing more
table below: distortions and, consequently, an increase in the overall BER.
TABLE II: PESQ scores for audio data
The system was also tested using real audio signals,
Noise Standard Deviation Equalized PESQ Unequalized PESQ
0.001 4.401 4.081
with the Perceptual Evaluation of Speech Quality (PESQ)
0.05 1.98 1.54 scores of both equalized and unequalized outputs recorded
0.1 1.58 1.316 in Table II and illustrated in Fig.10. These results further
0.15 1.449 1.271 corroborate the efficacy of the equalizer, as indicated by the
0.2 1.37 1.236
0.25 1.284 1.202 higher PESQ scores for the equalized outputs compared to
0.3 1.283 1.199 those of the unequalized signals. The findings also reveal
0.35 1.269 1.188 an exponential decline in PESQ scores with an increase in
0.4 1.268 1.193
noise level, reinforcing the negative impact of noise on signal
0.45 1.258 1.181
quality.

VI. A NALYSIS OF R ESULTS VII. C ONCLUSIONS

This paper describes the development and implementation This project looked into implementing digital equalizer
of an adaptive channel equalizer aimed at mitigating the system to combat the effect inter-symbol interference and
effects of noise and inter-symbol interference. The equalizer’s noise in digital communication. The hypothesis investigated
effectiveness was evaluated using a synthesized sinusoidal was that application of equalizer systems at the receiver of
communication system could reduce bit-error rate and improve
speech quality for voice communication. The report detailed
the design procedure of the system and presented simulation
results. From these results, it has been shown that equalizing
the channel at the receiver reduces bit-error-rate and improves
speech quality. However, adaptive system realized has shown
to over-fit the training sequence resulting is significantly huge
test errors. Also, the overall system performed poorly when
transmitting audio signals. These limitations of system call
for further improvements. The system can be improved by
incorporating modulation techniques, so that both training and
test signal have common format.
R EFERENCES
[1] Kazi Muhammed et al. “Channel Equalization in Digital
Transmission”. In: International Journal of Computer
Science and Information Security 4.1 2 (2009).
[2] Liaqat Mahmood et al. “Adaptive Filtering Algorithms
for Channel Equalization in Wireless Communication”.
In: Indian Journal of Science and Technology 8.17
(2015). DOI: 10.17485/ijst/2015/v8i17/57805.
[3] Garima Malik and Amandeep Singh Sappal. “Adaptive
equalization algorithms: an overview”. In: International
journal of advanced computer science and applications
2.3 (2011).
[4] Praneet Soni Agya Mishra. “Kalman Filter Based Chan-
nel Equalizer: A Literature Review”. In: International
Journal of Eme rging Technology and Advanced Engi-
neering 4 (5 2014), pp. 165–173.
[5] Cristiano M Panazio et al. “Channel equalization tech-
niques for wireless communications systems”. In: Opti-
mizing Wireless Communication Systems. Springer, 2009,
pp. 311–352.
[6] Jalali Sammuel. “Wireless Channel Equalization in Dig-
ital Communication Systems”. In: Claremont, Claremont
Graduate University Theses and Disseratations, Califor-
nia (2012).
[7] Bernard Sklar. “Defining, designing, and evaluating dig-
ital communication systems”. In: IEEE Communications
magazine 31.11 (1993), pp. 91–101.