Proceedings of the 2023 International Conference on Machine Learning and Automation

DOI: 10.54254/2755-2721/35/20230358

Application of matrix in signal processing

Haotian Du
G social Radley College, Abingdon, UK, OX14 2HR

[email protected]

Abstract. Signal processing, a foundational discipline in modern technology, encompasses a

diverse array of applications, ranging from audio and image processing to communication
systems and medical imaging. This review investigates how matrix-based techniques are widely
used to advance signal processing methodologies. In order to discretize continuous-time signals
for digital processing, which occurs in the first section of the paper, matrices play a crucial role
in signal sampling. A key principle, the Nyquist-Shannon Sampling Theorem, directs appropriate
sampling rates to prevent aliasing, with matrices permitting effective signal representation. The
effectiveness of matrix-based filtering methods for frequency modulation and noise reduction,
such as convolution and correlation, is then investigated. By utilising matrix operations, these
methods enable real-time signal processing. The Fourier Transform and Wavelet Transform are
also featured in matrix-driven signal transformation, providing insights into frequency analysis
and non-stationary signal characterization. By reducing noise components, matrix-based
approaches, particularly Singular Value Decomposition (SVD) denoising, are essential for
improving signal quality. Additionally, image compression employs SVD. Matrix-based
compressive sensing revolutionises signal recovery from sparse data and results in data-efficient
reconstruction. Signal processing has been transformed by matrix-based approaches, which have
enabled previously unheard-of levels of efficiency, accuracy, and adaptability. The review
highlights their significant influence on several signal processing fields.

Keywords: signal processing, matrix-based techniques, signal sampling, signal transformation.

1. Introduction
Matrices, rectangular arrays of numbers, symbols, or expressions organized in rows and columns.
Although the idea of matrices has been around since antiquity, it wasn't until the 19th and 20th centuries
that matrix theory was formally developed as a separate field of mathematics. Persian mathematician al-
Khwarizmi employed them during the Islamic Golden Age [1]. Cayley introduced matrix algebra in the
nineteenth century, describing operations such as addition, scalar multiplication, and matrix
multiplication [2]. The notion of determinants was introduced by Gauss [3], and matrix rank was defined
by Sylvester [4]. Łukasiewicz, Gelfond, and subsequent mathematicians furthered the formalization and
applications of matrices. As the foundation of linear algebra today, matrix theory offers a strong
framework for data representation and manipulation, enabling effective algorithms and techniques in
many areas, including signal processing.
Signal processing is an interdisciplinary field that encompasses a wide range of techniques for
manipulating and analyzing signals to extract meaningful information. These signals can come from
different sources such as images, audio, video, radar, sonar, biomedical sensors, etc. In order to process

© 2023 The Authors. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0
(https://creativecommons.org/licenses/by/4.0/).

41
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

and interpret these signals effectively, a rigorous mathematical foundation is essential. With its
versatility and well-defined operations, matrices have become an indispensable tool in the field of signal
processing. The use of matrices in signal processing dates back to the early developments in the field.
In the mid-20th century, with the advent of computers and digital signal processing technology, the
processing of signals became more efficient and accurate. Matrices provide a natural representation of
discrete signals and allow efficient computation of signal processing operations [5]. The pioneering
work of researchers such as Wiener laid the foundation for the application of matrix-based techniques
in signal processing [6].
The capacity of matrices to effectively describe and alter signals in both the time and frequency
domains is one of the fundamental benefits of employing them in signal processing [7]. In the time
domain, signals are represented as sequences of samples, and matrices enable the concise representation
and processing of these samples. The employment of matrices in addition, subtraction, and
multiplication operations allows for the implementation of both linear and nonlinear signal
transformations. Matrix-based signal analysis and transformation are essential in the frequency domain.
For instance, the Fourier transform, which enables the breakdown of a signal into its individual
frequency components, can be visualised as a matrix operation. This frequency representation provides
valuable insights into the spectral content of a signal, allowing for various applications such as spectral
analysis, filtering, and modulation [7].
The intrinsic capability of matrices to handle vast volumes of data efficiently is another benefit of
employing them in signal processing. Matrices allow for parallel computation and enable the
exploitation of hardware acceleration techniques, such as GPUs, to process signals in real-time or near
real-time scenarios [8]. Matrix-based signal processing techniques are appropriate for a variety of
applications due to their scalability, including small embedded systems and large-scale data processing
and communication systems. Additionally, matrices provide a structured framework for expressing and
resolving signal processing issues. By formulating signal processing tasks as matrix operations,
researchers and practitioners can leverage well-established mathematical techniques and algorithms.
This results in effective implementations and the possibility of optimisation, enabling signal processing
algorithms to operate more quickly and accurately [9].
In conclusion, the invention and use of matrices are integral to the history of signal processing. The
use of matrices in signal processing has become quite popular as a result of its benefits, including
effective representation, manipulation in the time and frequency domains, scalability, and methodical
problem-solving. By harnessing the power of matrices, researchers and practitioners continue to advance
signal processing techniques, enabling innovative applications across various domains. Then, start to
elaborate on the "Signal Sampling" portion, referencing the Nyquist-Shannon Sampling Theorem,
pertinent works, and correctly notating the equation in matrix form.

2. Method

2.1. Signal sampling

In order to process continuous-time signals digitally, a crucial step in signal processing is signal
sampling. Due to their ability to effectively represent and reconstruct signals from discrete samples,
matrices are essential to this process [10]. The Nyquist-Shannon sampling theorem serves as a guiding
principle for proper sampling rates to prevent aliasing.
A key conclusion in signal processing is the Nyquist-Shannon theorem, sometimes referred to as the
Nyquist Sampling Theorem or the Sampling Theorem. It demonstrates that a continuous-time signal
may be accurately reconstructed from its discrete samples if the sampling frequency is at least twice the
highest frequency present in the signal. The sampling rate must be greater than or equal to double the
signal's bandwidth in order to prevent aliasing.
From a series of real numbers, a continuous-time bandlimited function can be created using the
Whittaker-Shannon interpolation formula or sinc interpolation [11].

42
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

!"#$
𝑥(𝑡) = ∑%
#&"% 𝑥[𝑛]𝑠𝑖𝑛𝑐 - $
. (1)
Mathematically, this theorem can be represented using matrices as follows:
(
𝑥(𝑡) → 𝑥[𝑛] = 𝑥(𝑛𝑇' ) = (𝑇' ∙ 𝑛) 𝑤ℎ𝑒𝑟𝑒 𝑇' = )* (2)
!

Here, 𝑥(𝑡) is the continuous-time signal, 𝑥[𝑛] is the discrete-time signal obtained through sampling,
𝑇' is the sampling interval, 𝑓' is the sampling frequency, and 𝑛 represents the discrete time index.
In this equation, we can represent the signal sampling process using matrices. The continuous-time
signal 𝑥(𝑡) can be converted into its discrete-time counterpart 𝑥[𝑛] by sampling at regular intervals 𝑇'
[12]. The matrix notation [𝑥[0]𝑥[1]𝑥[2] ⋯ ] represents the discrete samples of the signal 𝑥(𝑡) at
discrete time indices 𝑛 = 0, 1, 2, ⋯.

2.2. Signal filtering

Matrix-based filtering techniques are widely used to process signals by modifying their frequency
content or attenuating unwanted noise and interference [13]. Commonly used techniques include
convolution, correlation, and adaptive filtering, all of which may be properly expressed and put into
practise using matrices.
Techniques for filtering and analysing signals that are essential include convolution and correlation.
One signal (the input) is convolved or correlated with another (the kernel or template) in these
procedures, which are theoretically described as matrix operations. These procedures may be carried out
more effectively thanks to matrices, which makes them computationally possible for real-time signal
processing [14].

2.3. Signal transformation

A potent tool for examining the frequency content of signals is the Fourier Transform. It breaks down a
signal into its individual sinusoidal parts. Modern signal processing heavily relies on the Fast Fourier
Transform (FFT), an effective approach for performing the Discrete Fourier Transform (DFT), which
uses matrix operations [15]. Noise reduction is essential in signal processing to improve the quality and
accuracy of signals in noisy environments. In order to reduce undesired noise while maintaining signal
integrity, matrix-based approaches offer effective denoising algorithms and filters [16]. SVD is a matrix
factorization technique widely used in denoising applications. Singular values, right singular vectors,
and left singular vectors are the three components that make up a signal matrix. To effectively minimise
noise components, small isolated values can be removed or attenuated [17]. The technique can also be
used to compress and simplify signals.

2.4. Signal recovery

Signal recovery involves retrieving signals from incomplete or degraded observations . By taking
advantage of the inherent sparsity of signals, matrix-based approaches, particularly in the area of
compressive sensing, have revolutionised signal recovery [18].

3. Process and result

Signal sampling is a critical process that converts continuous-time signals into discrete-time
representations, enabling digital processing. The Nyquist-Shannon Sampling Theorem, a cornerstone of
signal processing, establishes the necessary conditions for accurate signal sampling. The importance of
proper signal sampling cannot be overstated. It forms the basis for the accurate representation and
faithful reconstruction of continuous-time signals in the discrete domain, forming the foundation for
various digital signal processing tasks. Proper sampling is particularly crucial in applications such as
audio and image processing, communication systems, and medical imaging, where the preservation of
signal integrity is paramount. Numerous literature achievements have contributed to the exploration of
optimal sampling strategies, extending beyond the traditional uniform sampling.

43
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

3.1. Seminal works in signal sampling

The Nyquist-Shannon Sampling Theorem, which forms the bedrock of signal sampling, can be traced
back to Claude Shannon's groundbreaking paper "Communication in the Presence of Noise" published
in 1949 [12]. Shannon's work established the concepts of the theorem, showing that continuous signals
may be precisely reconstructed from their discrete samples if the sampling frequency is at least twice
the highest frequency contained in the signal. This work served as the cornerstone for information theory.
With its solid structure for signal capture and reconstruction, this essential article revolutionised
communication engineering and digital signal processing.
Another key figure in the development of sampling theory is the Swedish-American engineer Harry
Nyquist. Nyquist's work in the 1920s laid the theoretical groundwork for Shannon's theorem,
contributing the concept of the Nyquist rate and underscoring the need for adequate sampling to avoid
aliasing [19]. Nyquist’s work on signal-to-noise ratio and minimum sampling rates has been pivotal in
shaping modern signal processing practices.
Beyond the Nyquist-Shannon theorem, modern signal sampling has been significantly influenced by
breakthroughs in compressed sensing. David Donoho, a prominent mathematician, and statistician, has
made remarkable contributions to compressed sensing theory. Donoho’s work on sparsity-based signal
reconstruction algorithms and the concept of compressed sensing in high-dimensional data recovery has
opened new possibilities for efficient sampling and signal processing [20]. Compressed sensing
leverages the sparsity of signals in certain domains, and it involves the use of measurement matrices to
acquire fewer samples and reconstruct the original signal. The matrix formulation of compressed sensing
algorithms allows for the efficient and accurate recovery of signals from a reduced set of non-uniform
samples.
Moreover, the contributions of Tao and Candes are notable in advancing the understanding of the
potential of sparsity-based sampling. Their work on mathematical theory and the design of robust sparse
signal recovery algorithms has enabled compressed sensing to be applied successfully in various fields
[21].

3.2. Practical applications of proper signal sampling

The concept of proper signal sampling finds ubiquitous applications in numerous practical domains. In
audio processing, for instance, digital audio signals are sampled at rates higher than the Nyquist rate to
capture the complete audio spectrum without distortion, the process involves matrix multiplications to
obtain the discrete samples. This ensures that high-frequency components, essential for audio fidelity,
are preserved during digital conversion. The principles of proper sampling are fundamental to the
success of modern digital audio formats and high-quality sound reproduction [22].
In image processing, pixel sampling plays a pivotal role in preserving the original image details
during digitization. Pixel sampling is achieved through matrix operations, ensuring that the original
image details are preserved during the digitization process. Properly sampling image pixels is crucial in
applications such as digital photography, medical imaging, and video processing [23]. Image sensors in
digital cameras and medical imaging devices use proper sampling techniques to capture high-resolution
images that accurately represent the visual information.

3.3. Thechinque used for signal filtering

Signal filtering is a fundamental aspect of signal processing, matrix-based techniques have proven to be
highly effective in various filtering applications due to their efficiency and versatility. Two common
filtering methods, convolution and correlation, are expressed as matrix operations, where one signal is
convolved or correlated with another.
Convolution is a widely used technique for smoothing or modifying a signal's frequency
characteristics. It involves sliding a filter kernel over the input signal and computing the weighted sum
of the overlapping elements. The operation can be efficiently represented using matrix multiplication,
making it suitable for real-time signal processing applications [24]. Correlation, on the other hand, is
essential for signal matching and pattern recognition. It measures the similarity between two signals by

44
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

sliding one signal over another and computing the dot product at each position. Like convolution,
correlation can be efficiently implemented using matrix operations, enabling rapid and accurate signal
analysis [24].
To illustrate the effectiveness of matrix-based filtering, Figure 1 displays a wave graph show the
comparison. The noisy signal contains random noise that obscures the underlying pattern. However,
after convolving the signal with a properly designed filter, the noise is significantly reduced, and the
original signal is more discernible [25].

Figure 1. A 3 Hz sine wave that has been contaminated by noise in the top graph, followed by a
comparison of the output from a three-point moving average filter with the original (actual) signal.

For instance, consider a scenario where a continuous-time signal represents a sound wave variation
over time. The raw data might contain various sources of noise, including measurement errors,
interference, or random fluctuations. As shown in Figure 1, the raw audio data appears to have a
discernible pattern, but the noise makes it challenging to analyze and extract meaningful information
from the signal.
To enhance the quality of the data and highlight the underlying pattern, a convolution-based filtering
technique can be applied. The filtering process involves using a convolution kernel, designed to smooth
the signal while preserving essential features. In matrix-based convolution, the filter kernel is
represented as a matrix, and the convolution operation is performed through matrix multiplication.
As a result, the noisy sound signal is convolved with the filter kernel, effectively reducing the impact
of noise and enhancing the clarity of the original pattern. The resulting filtered signal displays a smoother
and more accurate representation of the sound variation, making it easier for analysts to interpret and
extract valuable insights from the data.
Matrix-based filtering techniques, such as convolution, offer computational efficiency, making them
suitable for real-time signal processing applications, including weather monitoring systems, audio
processing and image enhancement. The ability to represent filtering operations as matrix
multiplications contributes to the versatility and effectiveness of matrix-based signal processing
techniques in a wide range of applications.

3.4. Signal transformation

Signal transformation techniques enable the conversion of signals between several domains, such as
from time to frequency or from spatial to transform domains. A crucial technique in frequency analysis,
the Fourier Transform may break down a signal into its individual sinusoidal components. It delivers
useful details on the frequency content of a signal and is used in a range of industries, including audio

45
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

processing, image analysis, and communication systems. The Fast Fourier Transform (FFT), a matrix-
based technique that quickly computes the Discrete Fourier Transform (DFT), is a key component of
modern signal processing [26]. By employing matrices for computation, the FFT significantly reduces
the computational complexity of the DFT, making it applicable for real-time and large-scale signal
analysis tasks. Another powerful transformation technique is the Wavelet Transform, which has gained
widespread popularity in signal processing. The Wavelet Transform represents signals as a combination
of wavelets of varying scales and positions, providing both frequency and time localization. This
property makes it particularly useful for analyzing non-stationary signals and detecting transient events.
Both the Discrete Wavelet Transform (DWT) and Continuous Wavelet Transform (CWT) leverage
matrix-based operations for wavelet decomposition and reconstruction [27]. These matrix
representations allow efficient computation of the transform coefficients and facilitate various
applications, such as image compression, signal denoising, and feature extraction.
Noise reduction techniques are indispensable for enhancing signal quality and accuracy, especially
in noisy environments or during signal acquisition processes. Matrix-based methods offer efficient
denoising algorithms and filters that are widely used in various applications. Singular Value
Decomposition (SVD) denoising is one such technique that has gained prominence. SVD is a matrix
factorization method that decomposes a signal matrix into its singular components, namely the left
singular vectors, singular values, and right singular vectors. In the context of noise reduction, SVD is
used to identify and attenuate small singular values, which often correspond to noise components in the
signal. By removing or dampening these noise-related singular values, SVD denoising effectively
reduces the impact of noise while preserving the important signal features [16].
The SVD of a matrix can be used for image compression. Modern technology frequently uses
digitised images, which are comparable to a matrix representing the value of each pixel's level of grey.
Pictures made from natural sources typically have fairly huge size. For instance, the image in Figure 2
corresponds to a picture with large matrix size [28].

Figure 2. A picture with a matrix on the order of 204 × 290 = 59610 pixels.

We wish to represent the image with less numerical values than the original number of pixels in order
to avoid the issue of transmitting or storing the numerical values of such large images. Calculating the
singular value decomposition of an image and then reconstructing it using an estimate of reduced rank
is one method of compressing an image. This method is seen in Figure 3, which a rank 25 approximation
rebuilt almost exactly (to the human sight). This results in an 80% compression ratio [28].

46
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

Figure 3. The image in Figure 2 originated using 25 pairs of singular vectors with 80% compression
rate.

3.5. Signal recovery

Signal recovery involves the retrieval of signals from incomplete or degraded observations. Matrix-
based techniques, particularly those used in compressive sensing, have revolutionized signal recovery
by exploiting the inherent sparsity of signals. Compressive sensing is based on the principle that a signal
can be accurately recovered from a small number of non-uniform samples, provided the signal is sparse
or compressible in some domain [29]. This breakthrough technique has significant implications for data
acquisition and storage, enabling the reconstruction of signals from a fraction of the original data.
The matrix-based approach in compressive sensing involves the use of sparse matrices for signal
representation and recovery. The signal is first represented using a sparse matrix, which captures the
essential information while reducing redundancy [30]. Then, the recovery process utilizes matrix
operations to reconstruct the original signal from the sparse measurements efficiently. Compressive
sensing has found applications in various fields, including imaging systems, wireless communication,
and sensor networks, where it offers significant advantages in terms of data reduction, energy efficiency,
and robustness to noise and data loss. By leveraging matrices in signal transformation, noise reduction,
and signal recovery, modern signal processing techniques have attained unprecedented efficiency and
accuracy. The versatility and power of matrix-based methods have enabled their widespread adoption
in numerous applications, revolutionizing the way signals are analyzed, processed, and utilized in
various domains.

4. Conclusion
In the realm of signal processing, the utilization of matrix-based techniques has played an instrumental
role in transforming the landscape of signal analysis, manipulation, and application. This review has
delved into the profound impact of matrices across various signal processing domains, highlighting their
pervasive presence and pivotal contributions.
The inception of matrix-based signal sampling has fundamentally shaped the way we acquire and
represent signals in digital form. Guided by the Nyquist-Shannon Sampling Theorem, proper signal
sampling has become a cornerstone, preventing aliasing and ensuring accurate signal reconstruction.
Matrix-based signal filtering techniques, exemplified by convolution and correlation, have demonstrated
exceptional efficacy in frequency modification and noise reduction. Signal transformation, a process
facilitated by matrices, has empowered the shift between diverse signal domains, uncovering essential
information vital to various industries. The pinnacle of matrix-based contributions lies in signal
recovery, where compressive sensing has revolutionized the reconstruction of signals from sparse data.
In conclusion, the matrix-driven paradigm in signal processing has ushered in an era of
unprecedented precision, efficiency, and adaptability. From sampling to filtering, transformation to
noise reduction, and recovery to application, matrices have indelibly transformed signal processing into
a dynamic, data-driven discipline. The marriage of mathematical elegance and practical utility positions
matrix-based techniques as a cornerstone of modern signal processing, offering a myriad of tools that

47
 Proceedings of the 2023 International Conference on Machine Learning and Automation
DOI: 10.54254/2755-2721/35/20230358

empower signal analysts and practitioners to unlock the full potential of signals across diverse
applications.

References
[1] Berggren J L 2007 Mathematics in medieval Islam. History of Indian Mathematics: A Source
Book. Springer.
[2] Cayley A 1858 A memoir on the theory of matrices. Philosophical Transactions of the Royal
Society of London, 148, 17-37.
[3] Gauss C F 1801 Theoria combinationis observationum erroribus minimis obnoxiae (Theory of
the combination of observations least subject to error). Commentationes societatis regiae
scientiarum Gottingensis recentiores.
[4] Sylvester J J 1852 On a theory of the syzygetic relations of two rational integral functions,
comprising an application to the theory of Sturm's functions; with numerous examples and
applications to the investigation of quantities algebraical and transcendental. McGraw-Hill.
[5] Chen M, Zheng C, Zhang Z 2020 Matlab function and instance quick check manual. People's Post
and Telecommunications Publishing House.
[6] Wiener N 1949 The theory of prediction. In Modern mathematics for the engineer: Second series.
McGraw-Hill.
[7] Padgett W T, et al. 2010 Discrete-Time Signal Processing. Pearson.
[8] Cai Y 2000 Research on magnetic resonance image reconstruction based on hardware
acceleration. University of the Chinese Academy of Sciences.
[9] Boyd S, Vandenberghe L 2004 Convex optimization. Cambridge University Press.
[10] Sun X B, Bao Z 1993 Intermediate frequency signal sampling and orthogonal coherent detection.
Systems Engineering and Electronic Technology (5), 1-9.
[11] Shannon C E 1949 Communication in the presence of noise. Proceedings of the IRE, 37(1), 10-
21.
[12] Proakis J G, Manolakis D G 2007 Digital Signal Processing: Principles, Algorithms, and
Applications. Pearson.
[13] Wang C B 2021 Research on the filtering and stability of two-dimensional systems with stochastic
parameter matrices. Pearson.
[14] Bahri M et al. 2013 Convolution Theorems for Quaternion Fourier Transform: Properties and
Applications. Abstract and Applied Analysis, 1–10.
[15] Good I J 2021 The Relationship Between Two Fast Fourier Transforms. in IEEE Transactions on
Computers, 310-317.
[16] Han X M 2021 Matrix singular value decomposition algorithm and application research. Working
paper.
[17] Qian Z W, et al. 2011 Signal noise reduction method using singular value decomposition.
Working paper.
[18] Zhang J C 2014 Research on sparse multi-band signal compression sampling method [D]. Harbin
Institute of Technology.
[19] Nyquist H 1928 Certain Topics in Telegraph Transmission Theory. in Transactions of the
American Institute of Electrical Engineers, 617-644.
[20] Donoho D L 2006 Compressed sensing. IEEE Transactions on Information Theory, 52(4), 1289-
1306.
[21] Candès E J, Tao T 2006 Near-optimal signal recovery from random projections: Universal
encoding strategies? IEEE Transactions on Information Theory, 52(12), 5406-5425.
[22] Li J F 2013 Research and design of centralized digital audio matrix. University of Electronic
Science and Technology.
[23] Luo Y P 2013 Design of image display system based on subpixel sampling. University of
Electronic Science and Technology.

48