1. Introduction
Offset Linear Canonical Transform (OLCT) [
1,
2,
3,
4] is a mathematical tool widely employed in signal processing, optics, and various scientific and engineering domains. It extends the capabilities of Linear Canonical Transform (LCT). Moshinsky and Quesne’s foundational work in quantum mechanics laid the formulation for LCT [
5]. In contemporary signal processing research, LCT has emerged as a prominent area of study. This heightened interest has spurred substantial efforts to elucidate its mathematical foundations, leading to the establishment of numerous theorems pertaining to critical concepts such as sampling theorems [
6,
7,
8], convolution theorems [
9,
10], and uncertainty principles [
11]. The field of signal processing, in particular, has experienced a notable surge in investigations centered on LCT [
12]. Originally, LCT was recognized for its ability to encompass fundamental transforms such as Fourier transform (FT) [
13,
14], fractional Fourier transform (FrFT) [
15,
16,
17] and Fresnel transform [
18]. These transformations have long been cornerstones in signal processing and optics, enabling the analysis and manipulation of signals and images in different domains.
Defined by a set of six parameters
, OLCT, also known as special affine Fourier transform (SAFT) [
19], represents the time-shifted
and frequency-modulated
version of the LCT [
20], which has four parameters
. The additional parameters enhance the generality and flexibility of OLCT, making it applicable to a wide range of electrical and optical signal systems. OLCT has seen a rise in applications in various fields, such as optics and signal processing, in recent years [
2,
3,
21]. In signal processing, FT represents signals widely in both temporal and spatial domains. While short-time Fourier transform (StFT) has addressed some limitations of Fourier transform, it struggles with signals featuring high frequencies for a small duration and low frequencies for a large duration because the analysis window width is fixed. Continuous wavelet transform (WT) [
22,
23] was introduced to overcome this limitation by offering greater adaptability. However, WT has its own drawbacks, such as a lack of phase information and poor directionality.
LCT faces challenges when analyzing signals with poorly concentrated energy in the frequency domain and non-stationary characteristics. Windowed linear canonical transform (WLCT) involves segmenting the signal before LCT spectral analysis to address these issues. However, the fixed window width restricts its practical application. Gupta et al. [
24] introduced linear canonical wavelet transform (LCWT) as a solution, overcoming the limitations of LCT, WT, and WLCT.
To further address these challenges, Stockwell et al. [
25] introduced Stockwell transform (ST) by merging advantages of the StFT and wavelet transform; the classical Stockwell transform provides accurate information about the local behavior of a signal in temporal–spatial analysis [
26]. To analyse signals with multi-angle, multi-resolution, multi-scale, and temporal localization capabilities, researchers have proposed extensions of the Stockwell transform, such as fractional Stockwell transform (FrST) [
27,
28,
29,
30,
31] and linear canonical Stockwell transform (LCST) [
32,
33,
34].
In recent years, several expansions of Stockwell transform (ST) have emerged, integrating it with other relevant transforms such as FrFT, LCT, and Dunkl transforms, and others [
27,
32,
34,
35,
36]. These extensions, such as Dunkl–Stockwell transform, fractional Stockwell transform (FrST), and linear canonical Stockwell transform (LCST) facilitate the analysis of signals with capabilities for multi-scale, multi-resolution, and temporal localization, etc. Consequently, they are adept at handling chirp-like signals, which are pervasive in various domains. Delving into the expansive realms of Stockwell transforms (STs) and recognizing the significant utility of the OLCT, there is a compelling drive to integrate their distinct advantages into an innovative transformation termed offset linear canonical Stockwell transform (OLCST). The main benefit of this suggested OLCST is that it inherits some interesting aspects from the classic ST and OLCT in addition to the good mathematical qualities of the former, especially when it comes to concurrently showing both temporal and OLCST domain information. The proposed transform offers increased degrees of freedom, improving its versatility in non-stationary signal processing applications, among other fields.
In 1981, Mikusinski [
37] presented Boehmians as sequence quotients to generalize functions and distributions. Following the significant work by Mikusiński on the convergence of Boehmians [
38], numerous studies [
39,
40,
41,
42,
43,
44,
45] have been conducted by various researchers from different viewpoints. In this work, we apply OLCST to the space of square-integrable Boehmians. The space contains an example of a Boehmian that does not match to any distribution, as shown in [
38]. Hence, defining OLCST on square-integrable Boehmians is a reasonable generalization of OLCST on
.
The structure of the paper is as given:
Section 2 offers the necessary background, including OLCST. In
Section 3, we extend OLCST to the space of square-integrable Boehmians and the definition of OLCST in this space serves as a proper generalization of OLCST on
Section 4 discusses the discrete iteration of the new transform and derives its reconstruction formula. In
Section 5, we expand our study to explore the application of the novel transform to almost periodic functions (APFs). In
Section 6, we provide the potential applications, and finally,
Section 7 provides the conclusions of our work.
3. Boehmian Space (B-S)
In this part, we revisit the construction of B-S and proceed to establish the B-S . This B-S encompasses the OLCSTs of square-integrable Boehmians.
Consider a semi-group (commutative) vector space , and , meeting the subsequent conditions:
- (1)
, and .
- (2)
, and .
- (3)
, and .
- (4)
If and in , then in
A set of all sequences originating from denoted by , that adhere to the following conditions:
if belongs to , then ;
if in and , then in .
All Boehmians are expressed in the form
, where
and
, satisfying the following condition:
The collective set of all Boehmians is denoted as
. Two Boehmians,
and
in
, are considered equal if
The space is mapped to a subset of via the mapping , where is arbitrary. In the sense of , properties of operation ★ are defined as follows:
Addition:
Scalar Multiplication:
Two convergence notions for B-Ss are described as follows.
Definition 3. A Boehmian sequence is δ-converged to X if , so that and , and for each k and n that approaches ∞, in Γ.
Definition 4. A Boehmian sequence is Δ-converged to X if , so that , and as n approaches ∞ in Γ.
The construction of
-Boehmians, as detailed in [
46] for
, focuses on recalling the definition of Boehmians that are square-integrable. The space
denotes the collection of square integrable Boehmians and can be represented as
. This representation arises by considering
as
. The semi-group (commutative)
is specified as
, where
denotes the Schwartz space on
. In this context,
denotes the convolution operation functions, and the space is defined by
where
and ∗ is classical convolution.
Here, ★ denotes the identical convolution operation , and represents the collection denoted as , encompassing all sequences originating from that adhere to the following properties:
- (1)
;
- (2)
for some ;
- (3)
For the given , such that supp , where supp is the support of .
Subsequently, we formulate the B-S denoted as by introducing the operation and establishing its essential properties through rigorous proof.
Definition 5. For and , define Lemma 1. If and then , and hence .
Proof. For
, the lemma follows trivially. Therefore, we proceed under the assumption that
. By applying Jensen’s property and Fubini’s theorem, we obtain
Given that
forms a probability measure on
, Jensen’s inequality can be applied:
Hence, it is proven. □
The subsequent lemma directly results from Lemma 1.
Lemma 2. If in as and , then in as .
Lemma 3. If and , then
- (1)
.
- (2)
.
- (3)
.
Proof. Given that the first two identities follow directly, we focus our attention solely on demonstrating the validity of the third identity.
using the change of variable
Hence, in . □
Lemma 4. If in as and , then in as .
Proof. Applying Lemma 3, we obtain
Using Lemma 1 and property (2) of
on the right-hand side of (
6), we observe that the initial term is bounded by
, which diminishes as
. Our subsequent objective is to demonstrate the convergence of the second term towards zero.
Let We opt for , such that , leveraging the density property of the space , which encompasses all continuous functions with compact support on , in . For any given , if we define for all , then , signifying that g is uniformly continuous on . Consequently, there exists such that whenever and . As supp is as , we can find such that supp is encompassed within the closed ball with the center at the origin and radius in , for all . Now, we have the following:
If supp for some and a compact subset of , then , for every and for every .
From property (2) of , we have , for .
Therefore, by employing Jensen’s property, for
, we obtain
Here,
and
represent the Lebesgue measure of
. Utilizing Lemma 1 and the calculated approximation, we obtain
Thus, the lemma is concluded. □
Theorem 1. If a function and , then .
Proof. Let
be arbitrary:
Hence, the theorem follows. □
Definition 6. We define the extended Stockwell transform by The extension of OLCST onto is clearly well defined. To elaborate, suppose , and then Applying OLCST on both sides, we obtainand hence . Furthermore, if , then we have Using OLCST on both sides once more, we find that This demonstrates that remains unaffected by the choice of representative for X.
Theorem 2. The extended OLCST is consistent with .
Proof. Consider any arbitrary function
. The Boehmian representing f in
is expressed as
, where
is chosen arbitrarily. Consequently,
which is the Boehmian in
, which represents
.
□
Theorem 3. The extended OLCST is linear.
Proof. The demonstration of the above theorem is obvious. □
Theorem 4. The extended OLCST is one–one.
Proof. Let
be such that
, that is,
. Therefore, it follows that
and therefore, according to Theorem 1, we obtain
Since
is one–one, we obtain
Thus, . □
Theorem 5. The range of the extended OLCST is Proof. If
, then
such that
. It is evident that
works as an appropriate representation for
X. Alternatively, consider
such that
for all
and
such that
. Our claim is that
. From
, we derive
Due to the fact that
is injective, we obtain
Thus, and . □
Theorem 6 (Generalized convolution theorem). If and , then .
Proof. The demonstration that supports this theorem is clearly deduced from Theorem 1. □
Theorem 7. The extended OLCST maintains continuity associated with both δ-convergence and Δ-convergence.
Proof. Suppose
as
in
. Then,
so that
and
as
in
for each fixed
. Given the continuity of the OLCST
, we deduce that
for every fixed
. By employing Theorems 2 and 6, we obtain
which tends to
as
in
, for
. Therefore,
as
.
Next, let us assume that
as
. Consequently,
such that
for all
, and
converges to 0 as
in
. By leveraging Theorems 2, 3 and 6, we have
Additionally, utilizing the continuity of
on
, we obtain
This shows that as . □
4. Discrete OLCST
The development of discrete OLCST relies on discretizing the parameters
a and
b. Mathematically, we achieve this by discretizing the continuous parameters
a and
b in Equation (
3) as integral values. To effectively discretize the window functions, we opt for
and
, where
and
are constant positive values. This chosen approach results in a well-defined family of discretized window functions. Embracing this methodology not only simplifies the procedure but also enhances the accuracy of our analysis. Therefore, the set of discretized window functions is given as
In this context, the parameters
m and
n play pivotal roles in dilation and translation, respectively. To simplify computations, we can set
and
, resulting in a binary dilation of
and a dyadic translation of
. With these choices, the simplified form of the discretized family (
7) is
The precise definition of discrete OLCST is provided as follows.
Definition 7. For and , the discrete OLCST of f
is defined bywhere is given by (8).
Now, let us consider an example to illustrate discrete OLCST better.
Example 1. Assume the Gaussian function of unit amplitude as , where . The discrete OLCST of f
concerning the rectangular function with a bounded width T
is given by Using integer translations and dyadic dilations of ψ, we have Therefore, the discrete OLCST is computed for f(x)
as Here, ‘erfi(·)’ represents the imaginary error function.
After formulating discrete OLCST, our primary objective is to reconstruct the original signal from the transformed one. The inversion process involves working with coefficients in the form of
. This reconstruction follows two fundamental principles [
47]: characterizing random functions from
through the sequences from
, and ensuring this characterization by numerical stability.
It is important to note that functions can be uniquely identified by their coefficients
. Specifically, if
, then
. Likewise, if
, then f = 0. Achieving this relies on ensuring that whenever f and g are close, the coefficient
sequences
and
are close. In this context, closeness is defined using
the standard
l2-norm in the space
. The condition holds for a constant
, which is
free from f, so that
The practical interpretation of numerical stability in characterizing functions in
through their sequences of coefficients is as follows: when the coefficient sequences of two functions are close in
, it indicates that they are close in
. Hence, if
is lower, then
is also lower. More precisely, there should be a finite constant
, such that if
it implies that
In simpler terms, when the sum of squared coefficients is bounded, it indicates that the norm of the original function is also bounded. This concept ensures stability in the numerical representation of functions in
. For
. Subsequently, we can easily confirm that
, ensuring that
. This indicates that the squared norm of the transformed signal, denoted as
, is bounded
by
ϖ. Hence,
For a certain positive constant
, then the numerical stability of characterizing
in terms of
is satisfied. This condition stipulates the existence
of constants
, such that
Now, let us introduce the inversion formula for (
9).
Theorem 8. Consider , the discrete OLCST of , as defined by (9). Assuming the
stability condition (10), let us define linear operator T asand Proof. Condition (
10) makes it easy to verify that
T is a bounded linear operator with
injective properties. This implies that for g =
T(f), the following relationship holds:
By employing the Cauchy–Schwarz property, we obtain
The proof of Theorem 8 is now complete. □
5. OLCST for Almost Periodic Functions (APFs)
In this section, we aim to broaden the application scope of OLCST to encompass
persistent signals, with a particular emphasis on APFs. APFs, extending the concept of
traditional periodic functions to the real line, have been extensively investigated across
various mathematical, physical, and engineering disciplines [
48,
49,
50]. Recent research
efforts have focused on representing APFs using WFT [
51], WT [
52], FrFT [
53], Gabor transform [
54], etc. Now, we will try to extend the OLCST introduced in Equation (
2) to the
realm of APFs. This extension holds the potential to unveil new insights and enhance our
understanding of these persistent signals.
Let the set of trigonometric polynomials (TPs)
defined on
be given by
where the norm is
. By completing
with this norm, we arrive at the space of APFs, denoted as
. Alternatively,
can be regarded as the closed subspace of
, comprising functions of form
, where
. It is noteworthy that all APFs exhibit uniform continuity and boundedness.
An alternative norm on a set of TPs
is established as
with inner product relation
The set of TPs
is characterized by the norm (
12), and it is important to note that this space is neither complete nor separable. The Hilbert space
(where
serves as a subset of
) of Besicovitch APFs is obtained by completing
with respect to this norm [
48]. It is worth emphasizing that all APFs display uniform continuity and boundedness when considered with the aforementioned norm.
Consider the set
of functions
given as
In this context, L represents a natural number, and ranges over real values for , where . Consider a generalized trigonometric polynomial (GTP) on as where and .
Definition 8. For every , a function possess a robust limit power if there exists a GTP , so that The the inner product of collection possessing a robust limit power, is defined as Hence, it is evident that is a subset of , and the norms and are equivalent and both and are closed subspaces of [55]. Now, we will explore OLCST (
2) in the space of APFs.
Theorem 9. If f is an APF, then the OLCST of f concerning with emerges as a function with a robust limit power.
Proof. Let us initiate the proof by considering the case where f is a TP; that is,
The integral part adopts the following expression
This expression represents a GTP in ξ. Moreover, for an APF f, there is a sequence of
TPs such that it uniformly converges to f. Hence, it is enough to show that the limit of tends to 0 as n approaches infinity, and then the limit of also tends to 0.
The proof of Theorem 9 is now complete. □
The following theorem shows that the presumption is sufficient to ensure that OLCST represents a function with a robust limit power.
Theorem 10. Consider a window function . Then, for any APF , the OLCST is characterized as a function with a robust limit power in b.
Proof. Let f(t) be a TP defined as
Therefore, the OLCST of f is
With
, we have
and here
Hence, for a fixed a, we infer that the OLCST represents a GTP in b. Furthermore, if f is an APF, a sequence of GTPs fn converges uniformly to f. Therefore,
it is enough to demonstrate that whenever the limit
approaches 0, the limit of also tends to 0.
Hence, we observed that the OLCST of an APF f represents a function with
a robust limit power in b. □
Now, we will develop a frame decomposition for APFs using OLCST.
Theorem 11. For any APF invariables so that Proof. For TP
we obtain
With
we have
where
Moreover, we observe that
As
Equation (
14) simplifies to
Rest
and the summation includes all pairs
, where
is a non-zero multiplication of
and
Now, for Rest (f), we apply the Cauchy–Schwarz property:
where
Similar to the assumptions formulated by Daubechies [
47] when establishing wavelet frames within
, we will embrace the same hypothesis, that
This gives rise to the inequality (
13), which is applicable to TPs. Through the use of a standard approximation argument, we can infer that the same conclusion holds for APFs. □