ReviewCubic phase function: A simple solution to polynomial phase signal analysis
Introduction
Engineers in many fields often encounter non-stationary signals including biological, speech and music signals, radio signals in wireless communications and radars, and dispersive seismic signals [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23], [24], [25], [26], [27], [28], [29], [30], [31], [32], [33], [34], [35], [36], [37], [38], [39], [40], [41], [42], [43], [44], [45], [46], [47], [48], [49], [50], [51], [52], [53], [54], [55], [56], [57]. The conventional Fourier transform (FT), a popular tool to bridge between time and frequency, is considered to be inadequate to analyze such real-life signals [1], [2], [4]. In contrast, joint time-frequency (TF) analysis is an efficient way to reveal frequency contents of signals evolving over time, alternatively known as the instantaneous frequency (IF).
One particularly interesting model of non-stationary signals is the polynomial phase signal (PPS) model. The last 25 years have witnessed tremendous developments in the area of PPS parameter estimation, driven by applications originated in radars, sonars, biomedicine, machine engine testing, etc. [58], [59], [60], [61], [62], [63], [64], [65], [66], [67], [68], [69], [70], [71], [72], [73], [74], [75], [76], [77], [78], [79], [80], [81], [82], [83], [84], [85], [86], [87], [88], [89], [90], [91], [92], [93], [94], [95]. The maximum likelihood (ML) estimator has limited application due to a required multi-dimensional search over the parameter space. Early developments for the PPS parameter estimation are based on high-order ambiguity function (HAF) and its product form (PHAF) [91], [96]. The HAF-based estimation procedure consists of phase order decrementing by the process known as the phase differentiation (PD) until obtained signal is a sinusoid (the PPS of the first order). Then, the highest-order phase parameter is estimated using the fast algorithm, i.e., by an one-dimensional (1-D) search over the parameter space. This strategy is efficient but with numerous shortcomings. Firstly, in each stage of the procedure, the PD (performed by the auto-correlation function) reduces signal length and increases the number of noise-related terms in the resulted signal. These effects increase the signal-to-noise-ratio (SNR) threshold and estimation mean squared error (MSE). The auto-correlation also introduces cross-terms when multicomponent signals are considered. Finally, after estimation of the highest-order phase parameter, the same procedure is performed on the dechirped signal. Dechirping procedure causes error propagation from higher- to lower-order phase parameters. Some negative effects of the HAF, in particular cross-terms, are mitigated using the PHAF obtained as the multiplication of several HAFs calculated with different lag sets sharing the same product. The alternative technique is the integrated generalized ambiguity function (IGAF) [97]. It is accurate but, unfortunately, with heavy computations. Specifically this technique requires integrations over multi-dimensional lag space. Large number of integrals (or sums), i.e., calculation complexity, limits application of the IGAF to lower-order PPSs. Nevertheless, the IGAF is known to enhance the signal term and suppress the oscillating cross-terms and noise from the (coherent) lag integrations. In addition, two highest-order parameters are estimated at once meaning that the effect of error propagation is also reduced.
Aside from the IF, instantaneous frequency rate (IFR) or chirp-rate (CR) provides additional insights into the signal's frequency changing rate [1], [2], [98], [99], [100] and has received significant attention after O'Shea's seminal paper of [88]. At first, O'Shea proposed the original cubic phase function (CPF) for the parameter estimation of a third-order PPS, i.e., a cubic phase (CP) signal [88], [89]. The CPF-based procedure requires only one PD resulting in significantly better performances with respect to the (P)HAF-based alternatives for the CP signal. Later, numerous researchers strive for its extensions to higher-order PPSs [69], [70], [101], [102], [103], [104]. Besides, the CPF maps a signal to a 2-D joint time-chirp (frequency) rate domain. The time-CR domain and representations are still not well understood compared to the TF domain and representations. Meanwhile, there exist considerable interests in generalization of this transform to a 2-D PPS [67].
The aim of this paper is to show how this, at the first glance, simple modification of the CPF can motivate significant developments in the field of the PPS estimation and in more general nonstationary signal analysis. These developments resulted in significant improvement in the PPS estimation performance with respect to the state-of-the-art techniques. This overview article is devoted to the CPF since this transform and related approaches still reverberate in the community working with both theoretical and practical developments in the PPS estimation and with general nonstationary phase signals. The remaining of this paper is organized as follows. Theoretical background on the CPF is given in Section 2 with basic performance analysis of this technique both in parametric and nonparametric estimation. Various extensions of the CPF are presented and compared in Section 3. 2-D PPSs are considered in Section 4, while Section 5 brings literature overview of some practical applications where the CPF is used or where CP signals appear.
Section snippets
Signal model
Consider a frequency modulated (FM) signal
where A is the amplitude, and is the signal phase. The first derivative of is defined as the IF, , while its the second derivative is commonly referred to as the CR (or IFR) . Assume that the observed signal x(t) is corrupted by the complex zero-mean white Gaussian noise with variance σ2
The discrete version of signal (2) is obtained by sampling y(t) with a sampling interval :
Extensions of the CPF
The CPF is initially proposed for monocomponent CP signals, corrupted by Gaussian noise motivated by problems in the passive radar surveillance and echolocation. This limitation motivated researchers to actively look for the CPF modification that is able to address some of the following issues: higher-order PPS, multicomponent signals, non-Gaussian noise, etc. In the following subsections we summarize developments along these lines.
In Section 3.1 extensions of the CPF with higher-order
Two-dimensional PPS
The application of the CPF to 2-D and multidimensional signals is not quite straightforward due to issues related to search dimensions and calculation complexity. In addition, in many practical applications, the 2-D PPS has fast variations only along single direction while in the other direction signal changes are relatively slow meaning that parameters can be estimated along single line (1-D signal) and summed or interpolated along the other. Here, we consider cases when signal phase is fast
Applications
In the following, we provide a brief literature overview with applications of the CPF and its variants. The most important application appears to be radar signal processing given in Section 5.1, while application to joint estimation of the signal parameters and direction (angle)-of-arrival (DOA) to sensor array systems is presented in Section 5.2. Section 5.3 summarizes other fields where the PPS appears to be a valid signal model.
Conclusion
This paper has reviewed recent progresses in the PPS parameter estimation and in related fields motivated and inspired by the CPF. In less than 15 years, the CPF has attracted significant attention and numerous upgrades that improve significantly standard PD-based techniques in the PPS estimation. One of the aim of this paper is to demonstrate that such a simple modification has ability to advance a research field and to clear the major obstacles of the state-of-the-art methods.
It should be
References (270)
- et al.
Aliasing detection and resolving in the estimation of polynomial-phase signal parameters
Signal Process.
(2012) - et al.
Are genetic algorithms useful for the parameter estimation of FM signals?
Digit. Signal Process.
(2012) - et al.
Blind deconvolution of polynomial-phase signals using the high-order ambiguity function
Signal Process.
(1996) - et al.
Estimation of multicomponent polynomial phase signals of mixed orders
Signal Process.
(2001) - et al.
Polynomial-phase signal analysis using stationary moments
Signal Process.
(1996) - et al.
STFT-based estimator of polynomial phase signals
Signal Process.
(2012) - et al.
Estimation of the instantaneous amplitude and frequency of non-stationary short-time signals
Signal Process.
(2008) - et al.
Instantaneous frequency estimation at low signal-to-noise ratios using time-varying notch filters
Signal Process.
(2008) - et al.
Marginal bayesian analysis of polynomial-phase signals
Signal Process.
(2001) - et al.
Performance analysis on Lv distribution and its applications
Dig. Signal Process.
(2013)