Two‐dimensional sparse fractional Fourier transform and its applications
Introduction
The Fourier transform (FT) is an important tool in communication, signal processing, and other fields [1], [2], [3], [4]. However, for non-stationary signals, the FT cannot analyze their time-frequency characteristics completely. The non-stationary signals, especially the chirp signals, are common in radar [5], [6], gravitational waves [7], etc. As a generalized FT, the fractional Fourier transform (FRFT) [8], [9], [10], [11] has the advantage of processing non-stationary signals [12], [13], [14], [15]. It decomposes the signals by chirp orthonormal basis instead of sine basis, and the chirp signals are energy aggregated in the FRFT domain. The FRFT has been applied to optical systems [16], communication [17], signal and image processing [18], [19], [20], and other fields [21], [22], [23], [24] successfully.
Efficiently numerical algorithms for FRFT are necessary for engineering. Many definitions [25], [26], [27], [28], [29], [30], [31] of discrete fractional Fourier transform (DFRFT) have been proposed and can be classified as sampled DFRFT, eigendecomposition DFRFT, and linearly weighted DFRFT. The linearly weighted DFRFT is used rarely since it cannot approximate the continuous FRFT. The eigendecomposition DFRFT is not closed and there is no effective fast algorithm. So, it is difficult to be used in real-time engineering. For applications that do not require rotational additivity, the sampled DFRFT is the favorite and corresponding fast algorithms [30], [31], [32], [33], [34], [35], [36], [37] are studied. Pei [31] proposes currently the most efficient definition of DFRFT with the runtime complexity . Based on Pei-type DFRFT and the sparse nature of some real signals, Liu [32] proposes the SFRFT algorithm and reduces the runtime complexity to . Where denotes the signals’ sparsity. Subsequently, variants of SFRFT are investigated and applied to radar [33], [34], [35], [36]. However, the study of SFRFT is limited to one dimension. The two-dimensional (2D) DFRFT [38] is important and more computational. Thus, fast algorithms for 2D DFRFT are needed.
Based on the advantages of FRFT, 2D DFRFT has been applied to image processing [39], [40], [41], [42], [43], remote sensing [44], radar [18], artificial intelligence [45], and so on. However, little literature is devoted to the efficient implementation of 2D DFRFT. Generally, the 2D DFRFT is obtained by cascading the 1D DFRFT of the rows and columns of the signal. Utilizing the fast algorithm of 1D DFRFT, the runtime complexity of 2D DFRFT is times that of 1D DFRFT. It is worth noting that SFRFT requires the signal to be sparse after the transformation. Thus, the SFRFT algorithm with low runtime complexity cannot replace two sets of 1D DFRFTs. To summarize, 2D DFRFT is time-consuming. Moreover, all discrete points of the signal are used, making data acquisition, storage, and transmission challenges. It is necessary to study the SFRFT algorithm for two dimensions signals.
With low computational and sampling costs, the SFT [46], [47], [48], [49], [50] are efficient algorithms to estimate the discrete Fourier transform (DFT). And the 2D SFT algorithms [51], [52], [53], [54], [55], [56] are proposed. Among them, the downsampled method [51] performs only a few 1D DFTs. The algorithm has a simple structure and low complexity for noiseless signals. However, it may be locked if the signals are not appropriate. To address this limitation, different schemes [52], [53], [54] are investigated. And by randomly downsampled slices, the algorithm in Wang et al. [54] has the strongest degree of freedom and the lowest runtime complexity. Moreover, the sparsity does not have to be pre-given, which fits the real applications. However, 2D SFT cannot be used for non-stationary signals. Therefore, we are willing to propose algorithms to estimate the 2D DFRFT based on sparsity and randomly downsampled slices. In addition, for the case of noise, the voting method in Wang et al. [54] is only suitable for high SNR. We expect to extend the range of SNR in the novel algorithm estimating 2D DFRFT.
To estimate 2D DFRFT with low sample and runtime complexity, we will propose a two-dimensional sparse fractional Fourier transform (2D SFRFT) algorithm. The signals’ 2D FRFT domain are assumed to be sparse, which is common in radar, magnetic resonance images, optical images, and other engineering. The 2D SFRFT algorithm will be implemented by chirp products and downsampled 2D SFT. The sparsity can be unknown and only a few slices of data will be leveraged. Thus the 2D SFRFT algorithm can achieve an unprecedented low sample and runtime complexity. The localization errors caused by noises will be analyzed and the significant frequency position estimation method of the 2D DFRFT algorithm will be improved to robust. The robust 2D SFRFT algorithm can achieve 2D DFRFT estimation at low SNR with great probability. The specific contributions of this paper are as follows.
- •
Firstly, based on the Pei-type DFRFT, a definition of 2D DFRFT is given, which is the target to be estimated.
- •
Secondly, based on randomly downsampled slices, we establish the 2D SFRFT algorithm for noiseless signals. The 2D DFRFT can be fully estimated with sample complexity and runtime complexity . Where and is the number of iterations.
- •
Thirdly, we analyze the effect of noises on the 2D SFRFT and upgrade 2D SFRFT to be robust. With inner loops, the sample and runtime complexity are , respectively. The correctly estimated probabilities are also analyzed.
- •
Finally, we verify the performance and convergence of the proposed algorithms by simulation. Multi-component 2D chirp signal parameter estimation is achieved and precision is far superior to that of 2D SFT. In addition, the robust 2D SFRFT algorithm is applied to image fusion and SAR radar successfully.
The rest of this paper is organized as follows. Section 2 describes the definition of Pei-type DFRFT and 2D DFRFT. In Section 3, the 2D SFRFT algorithm is introduced. Robust improvements are provided in Section 4. The relevant simulations and applications are presented in Sections 5 and 6. Section 7 concludes the paper.
Section snippets
Preliminaries
In this section, we will describe the definition of the Pei-type DFRFT and 2D DFRFT.
Proposed 2D SFRFT algorithm
In this section, we will present the two-dimensional sparse fractional Fourier transform (2D SFRFT) algorithm for a noiseless situation [61]. It will reduce the runtime and sample complexity when signals are sparse in the fractional Fourier domain.
Based on the Pei-type DFRFT, the 2D DFRFT of -point signal is defined as (8)if and . For and
Robust 2D SFRFT algorithm
This section will present the robust two-dimensional sparse fractional Fourier transform (robust 2D SFRFT) algorithm for noisy signals. In signal processing, complex Gaussian noises are common. In communication theory, narrow-band white Gaussian noises can be represented by a cyclic symmetric complex Gaussian process when represented by an equivalent low-pass representation. Assume that the sparse signal with size is polluted by additive noise which obeys a cyclic symmetric
Simulations
In this section, to verify the advantages of the 2D SFRFT algorithm, the efficiency and error will be compared with the conventional methods. Then, the position error threshold in the robust 2D SFRFT algorithm will be analyzed. Subsequently, the convergence of the robust 2D SFRFT algorithm will be analyzed, specifically in terms of the number of iterations, the number of inner loops, and the probability of strict estimation. Finally, the sample complexity, the runtime complexity, and the
Application
In this section, to illustrate the effectiveness of the algorithm, we will implement parameter estimation of multicomponent 2D chirp signals, the fusion of multi-focused images, and parameters estimation for complex maneuvering targets in SAR radar.
Conclusion
In this paper, we explore the robust 2D SFRFT algorithm in theory and applications. First of all, a low-complexity 2D SFRFT algorithm for noiseless signals is proposed. Then, based on the analysis of noises, the spectral estimation method for noisy signals is proposed. Finally, algorithm performance simulations and applications are given. Compared with the decomposition and Pei method, 2D FRFT is optimal in terms of efficiency and accuracy, regardless of signal sizes. The simulation also shows
CRediT authorship contribution statement
Deyun Wei: Conceptualization, Formal analysis, Funding acquisition, Methodology, Writing – review & editing. Jun Yang: Conceptualization, Data curation, Formal analysis, Investigation, Software, Validation, Writing – original draft.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgment
This work was supported in part by the National Natural Science Foundation of China under Grant 61971328.
References (61)
- et al.
Analysis of a-stationary random signals in the linear canonical transform domain
Signal Process.
(2018) - et al.
Multiple color image fusion, compression, and encryption using compressive sensing, chaotic-biometric keys, and optical fractional Fourier transform
Opt. Laser Technol.
(2022) - et al.
ISAR imaging of target with complex motion associated with the fractional Fourier transform
Digit. Signal Process.
(2018) - et al.
Fractional Fourier transform as a signal processing tool: an overview of recent developments
Signal Process.
(2011) New Wigner distribution and ambiguity function based on the generalized translation in the linear canonical transform domain
Signal Process.
(2016)- et al.
Analysis and comparison of discrete fractional Fourier transforms
Signal Process.
(2019) - et al.
Computation of an eigendecomposition-based discrete fractional Fourier transform with reduced arithmetic complexity
Signal Process.
(2019) - et al.
Optimized sparse fractional Fourier transform: principle and performance analysis
Signal Process.
(2020) - et al.
Sparse fractional Fourier transform and its applications in radar moving target detection
2018 International Conference on Radar (RADAR)
(2018) - et al.
Sparse discrete linear canonical transform and its applications
Signal Process.
(2021)
Two dimensional discrete fractional Fourier transform
Signal Process.
Performance evaluation and parameter optimization of sparse Fourier transform
Signal Process.
Sparse high-dimensional FFT based on rank-1 lattice sampling
Appl. Comput. Harmon. Anal.
Frequency diverse array radar: new results and discrete Fourier transform based beampattern
IEEE Trans. Signal Process.
Fractionalisation of an odd time odd frequency DFT matrix based on the eigenvectors of a novel nearly tridiagonal commuting matrix
IET Signal Process.
Novel tridiagonal commuting matrices for types I, IV, V, VIII DCT and DST matrices
IEEE Signal Process. Lett.
Binary discrete Fourier transform and its inversion
IEEE Trans. Signal Process.
A new method for radar high-speed maneuvering weak target detection and imaging
IEEE Geosci. Remote Sens. Lett.
New chirp sequence radar waveform
IEEE Trans. Aerosp. Electron. Syst.
Observation of gravitational waves from a binary black hole merger
Phys. Rev. Lett.
The Fractional Fourier Transform with Applications in Optics and Signal Processing
Two-dimensional fractional Fourier transform and some of its properties
Integral Transf. Spec. Funct.
Sliding 2D discrete fractional Fourier transform
IEEE Signal Process. Lett.
Linear Canonical Transforms: Theory and Applications
Detection and parameter estimation of multicomponent LFM signal based on the fractional Fourier transform
Sci. China Ser. F
The fractional Fourier transform and time-frequency representations
IEEE Trans. Signal Process.
Linear canonical Wigner distribution based noisy LFM signals detection through the output SNR improvement analysis
IEEE Trans. Signal Process.
FRFT-based interference suppression for OFDM systems in IoT environment
IEEE Commun. Lett.
Recent developments in FRFT, DFRFT with their applications in signal and image processing
Recent Pat. Eng.
Generalized sampling expansions with multiple sampling rates for lowpass and bandpass signals in the fractional Fourier transform domain
IEEE Trans. Signal Process.
Cited by (6)
Fractional Fourier transform associated with polar coordinates
2024, International Journal of Wavelets, Multiresolution and Information ProcessingFractional Fourier transform in spherical polar coordinates
2023, Signal, Image and Video ProcessingFast Band-Limited Sparse Signal Reconstruction Algorithms for Big Data Processing
2023, IEEE Sensors JournalTexture Direction Detection for Side of Stacking Solar Cells
2023, 2023 2nd International Conference on Smart Grids and Energy Systems, SGES 2023Time-frequency Representation Optimization using InfoGAN Latent Codes
2022, 2022 30th Telecommunications Forum, TELFOR 2022 - Proceedings