- Open Access
Filtering in the joint time/chirp-rate domain for separation of quadratic and cubic phase chirp signals
© Özgen; licensee Springer. 2012
Received: 26 October 2011
Accepted: 8 June 2012
Published: 8 June 2012
This article investigates the possibility and convenience of a filtering operation in the joint time/chirp-rate (TCR) domain, and proposes a novel linear TCR filter for decomposing multicomponent signals into their quadratic and/or cubic phase chirp components with monotonic instantaneous chirp-rate (ICR) laws only. The TCR domain mask of the filter is selected on a display of a TCR representation of an input signal to isolate the desired chirp component. Projecting the input signal onto the phase signal associated with the TCR mask and approximating the phase difference in this projection operation in terms of ICR values result in the proposed TCR filter that recovers the selected component. Simulations illustrate the proposed filtering in recovery of undersampled cubic phase signals and in resolving back-to-back objects from in-line holograms for which cases it is easier to design filter masks in the TCR domain than in the time-frequency domain.
Multicomponent nonstationary signals are widely encountered in many applications including radar, sonar, communications and optics. Parametric methods mostly based on polynomial phase modeling may be used to analyze and estimate such signals and separate them into their components; such as nonlinear least squares techniques [1, 2], a maximum likelihood algorithm , an expectation-maximization based method , an array processing approach based on state estimation via an extended Kalman filter , a cyclic moment based method for polynomial phase signals with independent random amplitudes , techniques using transforms like high-order ambiguity function [7–9] and time-frequency (TF) Hough transform [10–12], and an approach for chirplet approximation , among other such methods.
The above methods require the number of components in the analyzed signal and/or orders of their polynomial phases to be known or estimated beforehand, although some of them are able to estimate these parameters along the way . Besides, for some applications it is sufficient to decompose the analyzed signal into its chirp components, as in object reconstruction from in-line Fresnel holograms , without much need for signal model parameters. Nonparametric signal separation methods may be more suitable for such applications, such as a periodicity-based algebraic separation algorithm  and an automatic signal separation method based on difference equation representation of chirp signals . The first method requires the number of signal components and relies on inequality of component periodicities . The second one has been reported to give better performance in instantaneous frequency (IF)/amplitude estimation when applied to monocomponent signals especially for low SNR cases, and, has been suggested to be used after signal component separation by TF filtering in such cases . Hence, TF filtering is still indispensible for many signal separation applications, as reviewed in .
There are various linear TF filter types; such as Zadeh , Weyl [19, 20] and generalized Weyl filters [21, 22] encompassing these two, TF projection filters [23–25], short-time Fourier transform (FT) filter by means of an analysis-masking-synthesis procedure [26, 27], local polynomial FT filter , S-transform filters again based on analysis-masking-synthesis approaches [29–31], and a method for chirp signal reconstruction from ridges of Gabor and wavelet transforms of the analyzed signal [32–34], among others.
As maintained in [35, 36], if TF support region of a signal component is nearly disjoint from those of other components and the background noise in an input signal, then, TF transfer function of the Wiener filter that optimally estimates that component reduces to the indicator function of its TF support region. Hence, TF transfer function or pass region of such a filter is selected on a display of a TF representation of the input signal to isolate the desired component .
Kozek and Hlawatsch  compares linear TF filters to nonlinear TF analysis-masking-synthesis methods based on the Wigner distribution (WD) and the smoothed WD, with prescribed TF pass regions, in TF signal separation problems, and finds that TF filters, in general, yield improved performance with reduced computational cost. Indeed, our simulations indicate that especially Weyl and TF projection filters separate chirps with excellent accuracy. They usually give several percents of error in the noiseless case, where percentage error is defined as the energy of the deviation of the filter output from the desired chirp component normalized by the energy of that chirp.
Despite the good performance and convenience of linear TF filters in chirp separation applications, it may still be more convenient to prepare the mask function of a time-varying separating filter in the joint time/chirp-rate (TCR) domain, rather than in the TF domain, for some of those applications. One such application is reconstruction of back-to-back objects from in-line Fresnel holograms . Each such object is represented by a pair of lines with opposite slopes in an associated space/spatial-frequency (or TF) representation obtained from the hologram, intersecting at the object coordinate. Magnitude of the slopes is inversely proportional with the object depth, i.e., the distance of the object to the hologram plane . Thus, linear tracks associated with back-to-back objects overlap in the space-frequency (SF) domain, making it tedious to design SF filter mask functions to resolve such objects. In the space/chirp-rate (SCR) plane, those objects are represented by distinct horizontal strips corresponding to different slopes or depths. Hence, if the mask function of a separating filter can be prepared in the SCR plane to isolate such strips, that would further ease the filter design task.
Motivated by the above application, we propose a novel linear time-varying filter in the TCR domain, reminiscent to TF filtering, for decomposing multicomponent signals to reconstruct their chirp components of the form a(t) exp(j 2πφ(t)), where φ(t) is a quadratic or cubic phase with a monotonic instantaneous chirp-rate (ICR) law (given by its second derivative). Aside from the mentioned problem, it is also more beneficial to use the proposed TCR filter with its mask function prepared in the TCR plane to recover undersampled quadratic or cubic phase chirps if their ICR curves change more slowly than their IF curves. We present simulations illustrating separation and reconstruction of severely undersampled cubic phase signals with IF curves traversing the discrete TF plane many times within the signal duration while their ICR laws vary much more slowly and exhibit single linear tracks in the TCR plane. For such signals, it is almost impossible to design a TF mask function for a separating TF filter, but a TCR mask can be easily prepared for the proposed TCR filter.
The idea of filtering in the joint TCR domain is novel. Filtering schemes based on domains other than frequency and TF domains have been developed before; however, they are not directly related to the TCR domain. As a previous work of this kind, [38, 39] have proposed an extended FT (EFT) matching a known IF function and have developed a time-varying filter for reconstruction of signals with that known IF, by masking in the EFT domain and then taking the inverse EFT. Similarly, a filtering operation in the frequency modulation (FM) rate parameter domain has been performed for suppressing interference chirp signals with nonlinear phase functions in direct sequence spread spectrum communication systems . A linear transform with a kernel that matches these phase functions maps these signals to impulses in the FM rate domain. Then, undesired chirps can be masked out and desired chirp components or the spread spectrum sequence can be recovered after an inverse matched signal transform . Both methods employ an analysis-masking-synthesis approach.
Although mask design in the parameter domain is time-invariant in these methods, they yield time-varying filters with suitable TF transfer functions for separation of selected signal components. The former method requires a positive IF function [38, 39], whereas the later does not have this restriction . Both of them require the IF function of the signal to be reconstructed to be known up to a scaling constant. Another filtering approach has been developed in  to filter dispersive guided wave signals, based on unitary operators matching this kind of signals; however, proposed TF filters are specifically tailored for and limited to these Pekeris guided waves .
Unlike the two methods above [38–40], our proposed TCR filter does not employ an analysis-masking-synthesis procedure. Instead, a TCR domain mask function H(t, α) is prepared to enclose the TCR signature of the desired signal component, where α denotes the chirp-rate (or frequency-rate) parameter. Then, this mask function is transformed to yield a time-varying impulse response, as TF transfer functions are transformed to obtain time-varying impulse responses of linear Weyl and Zadeh TF filters . To enable such a TCR filtering operation, firstly, a TCR representation of an input signal should be computed and displayed so that a desired chirp component of it can be identified on the TCR display and can be selected by a TCR mask that encloses its linear ICR trace.
Our proposed filter is more flexible than the above ones in that it does not require the knowledge of the IF or ICR functions of the desired component up to a constant but works for any linear strip-shaped TCR pass region. However, it can separate only quadratic or cubic phase signals with monotonic ICR laws exhibiting single linear tracks, as will be verified in the article. Piecewise linear ICR components can be recovered with repeated use of the filter for each linear segment.
Several TCR representations that can facilitate a TCR filter can be found in the literature. A generalized WD that serves as a joint time-phase derivatives representation for monocomponent, constant-amplitude polynomial phase signals has been proposed in , based on decomposition of polynomial derivatives in terms of shifted versions of the involved polynomial. O'Neill and Flandrin  has presented a quartic, shift-invariant TCR representation. O'Shea [44, 45] have proposed the cubic phase function (CPF) for estimating phase parameters of cubic phase signals. A product CPF has been proposed for multicomponent chirps, in , to eliminate spurious peaks appearing in the CPF when applied to such signals. Extended versions of the CPF have been developed [45, 47, 48] to estimate polynomial phase signals with higher order phases. Finally, a class of joint time-phase derivatives distributions highly concentrated along phase derivative curves has been derived in .
Among the above, [43, 49], beyond estimation of phase parameters, have also used their transforms as joint TCR representations or distributions in the form of two-dimensional (2-D) images that display ICR curves of analyzed signals.
In our article, we employ the CPF [44, 45], the quartic TCR distribution of , a bilinear TCR distribution in , and a shifted version of a quadratic local polynomial periodogram [50–52] to obtain our TCR displays on which desired signal components are identified and masked.
Section 2 derives the proposed TCR filter by approximating the phase difference in terms of the second derivative of the phase, i.e., ICR values, while projecting an input signal onto the phase signal associated with the TCR pass region of the filter. One of the terms in the derived time-varying impulse response requires an approximate knowledge of the IF value of the desired signal component at a reference time instant, in the form of an IF distribution at that time instant. It should be selected on a TF display of the input signal. Hence, the proposed TCR filter is based on joint use of a TCR representation with a TF representation displayed as images.
Section 3 derives the equivalent Weyl TF transfer function for the filter with an infinitesimally narrow linear pass region in the TCR domain, and verifies that it correctly recovers the corresponding quadratic or cubic phase signal. An expression for the noise power at the filter output is also presented in this section. Section 4 addresses discrete implementation of the proposed TCR filter and its computational cost. Section 5 presents simulations that illustrate this filtering scheme in separation or recovery of quadratic and cubic phase signals, including how to resolve back-to-back particles from in-line Fresnel holograms. Separation performance of the proposed filter is compared with those of Weyl and TF projection filters. Section 6 concludes the article.
2 Derivation of the proposed TCR filter
can be viewed as the approximate TCR mask function of a separating filter. α denotes the chirp-rate (or frequency-rate) parameter, and, φ(2)(t) is the second derivative of the phase of s(t) yielding its ICR curve. In the above, we assume that the ICR curve of s(t) is correctly and accurately read on a TCR display of the input signal x(t) and is taken as the TCR domain mask of the filter.
in order to recover s(t). Hence, TCR domain filtering for signal separation consists of (i) displaying a TCR representation of the input signal x(t), (ii) selecting the TCR mask function of the filter so as to isolate the TCR signature of the desired component s(t) on this display, as in Equation (1), (iii) and obtaining the impulse response of the filter given in Equation (4) from the selected mask function in Equation (1). Then, obtaining Equation (4) from Equation (1) reduces to estimating the phase difference φ(t) − φ(t') associated with desired s(t) from the second derivative of its phase function φ(2)(t).
where t0 is a reference time instant.
where H f (t0, f ) accounts for an estimate of the IF value of the desired signal s(t) at the reference time t = t0, f (t0) = φ'(t0).
H f (t0, f ) in Equation (8) serves as a reference IF distribution around the given IF value. If it is taken as H f (t0, f ) = δ(f − φ'(t0)), then, substitution of it and Equation (1) into Equation (8) gives Equation (4) as the impulse response where the phase difference is replaced by its approximation given in Equation (7). This reference IF distribution of the desired component is indispensable in our proposed TCR filter given by Equations (8) and (2).
The proposed filtering procedure is given as follows:
Step. 1. A TCR representation and a TF representation of the input signal x(t) are displayed as 2-D images.
Step. 2. The TCR mask function H(t, α) is prepared on the TCR display to isolate the ICR strip of the desired component s(t). This is idealized by Equation (1).
Step. 3. TF display of the input x(t) is examined and a convenient reference point (t0, f (t0)) is selected on the IF curve or in the TF support region of the desired s(t). Then, a reference IF distribution H f (t0, f) is prepared around the value f (t0) at the reference time t0. This is idealized as H f (t0, f ) = δ(f − f (t0)).
Step. 4. The TCR mask H(t, α), its slice at t0, and the reference IF distribution H f (t0, f ) are substituted into Equation (8) to obtain the filter impulse response h(t, t').
Step. 5. Time-varying impulse response h(t, t') is applied to the input signal x(t) by Equation (2) to yield an estimate of the desired component s(t).
Higher order derivatives in Equation (5) could also be retained and approximated by differences of second derivatives evaluated at different time points. This leads to alternative forms of the TCR filter in place of Equation (8). For example, third derivative in Equation (5) can be approximated by a difference of second derivatives. The remaining terms can be discarded. Alternatively, the integral in Equation (6) can be approximated at three time points t0, (t0 + t')/2 and t', instead of two. Both approaches lead to time varying impulse responses with four product terms in them.
These alternative filters can also successfully recover quadratic and cubic phase signals with monotonic ICR laws exhibiting single linear tracks, as the one proposed in Equation (8) does. This can be verified by showing that their equivalent Weyl TF transfer functions are also concentrated around IF curves of desired signals, as we show for the proposed TCR filter in the next section.
However, our simulations indicate that their performances in chirp signal recovery are worse than that of the proposed one, since their equivalent TF transfer functions exhibit more severe peaks near the origin of the TF plane. Moreover, their discrete implementations require more than one discrete TCR mask functions to be prepared and used, each for a different product term in the filter impulse response. This further complicates their discrete implementations. Our proposed filter in Equation (8) has the best separation performance and is easiest to implement, among them.
3 Equivalent Weyl TF transfer function and output noise power
3.1 Equivalent Weyl TF transfer function
Equivalent Weyl TF transfer function of the proposed TCR filter in Equation (8) will be derived below for a linear TCR pass region approximated as a line impulse as given by Equation (1) and a rectangular pulse-shaped IF distribution at a reference time. For a more realistic case of a linear strip-shaped TCR pass region, we could not evaluate the resulting complicated integral to obtain an analytical expression for the TF transfer function.
is assumed to be the cubic phase of the desired signal s(t) or that of the phase signal underlying the filtering operation. We have to verify that the TF transfer function of the filter given by Equations (9) and (10) is concentrated along the IF curve f = φ'(t) of s(t), in the TF plane, so that this filter will recover the desired s(t).
which is the FT of the above CPF. It is concentrated around the IF curve f = 3at2 +2bt+c of the desired signal, as required for its recovery.
where J1/3 and J−1/3 are Bessel functions of the first kind with given orders.
as the Weyl TF transfer function for the case of an impulsive reference IF distribution.
with sign(a) denoting the sign of a.
for a < 0.
The slope range above is obtained from Equation (22) by substituting φ''(t) = 3at2 and into it. B f ≥ 2/T should be maintained above.
a line impulse along the linear IF law of the desired quadratic phase s(t), in the case of an impulsive reference IF distribution H f (t0, f ) = δ(f − φ'(t0)).
as the Weyl TF transfer function for this more realistic assumption.
following from Equation (26). B f ≥ 2/T should hold for recovery of s(t), for the signal duration −T/2 ≤ t ≤ T/2.
For higher order polynomial phase signals, where Equation (6) is neither exact nor a good approximation, equivalent TF transfer function of the proposed filter does not capture the correct IF curve. This is also the case for segmented quadratic or cubic phase signals for which Equation (6) is again not valid for the whole signal duration. Components of such a signal should be recovered one by one by repeated use of our filter with a different TCR mask each time. Exactness of the trapezoidal approximation in Equation (6) is the key to our proposed TCR filter. It is valid only for a single, linear pass region in the TCR plane corresponding to a quadratic or cubic phase signal with a monotone ICR curve.
3.2 Output noise power
where the phase term above is as given in Equation (9).
Let x(t) = s(t) + w(t) be a noisy input signal for the proposed filter with the impulse response given in Equation (29), where s(t) is the desired signal component and w(t) is additive, zero-mean, white noise with power spectral density S w (f) = η.
The output noise power given in Equation (31) does not depend on the phase, φ(t), of the desired component or that of the phase signal associated with the TCR pass region of the filter. It is determined by TCR and reference IF bandwidths, B α and B f , respectively, current and reference time values t and t0, and input noise power only, regardless of the phase being quadratic or cubic.
4 Discrete implementation
The second expression above is obtained by writing the first one as a sum of integrals over intervals of length 2, changing the variable in each integral and using the fact that the discrete complex exponential above is periodic in the chirp-rate variable α with period 2. n0 denotes the discrete reference time of the filter.
The term in square brackets in Equation (32) is sampled in the time variable and is periodic in the chirp-rate variable α with period 2. Hence, it can be viewed as the discrete-time TCR mask of the filter, denoted as H d (n, α).
where H d (n, α), 0 ≤ n ≤ N − 1, 0 ≤ α ≤ 2 is the discrete-time TCR mask function of the filter, associated with the term in Equation (32). α denotes the discrete chirp-rate parameter above with an overuse of notation. N is the length of the discrete-time input signal x(n), 0 ≤ n ≤ N − 1, of the filter.
for 0 ≤ n ≤ N −1. M = 4N, 8N, 16N, etc., give good results in the simulations presented below.
Computational cost: The proposed TCR filter can be efficiently implemented by means of inverse fast Fourier transform (IFFT) algorithms. Equation (34) can be evaluated by an M-point IFFT for each m value and by a subsequent index-finding among the stored values using the periodicity of the complex exponential kernel with period M. Equations (35) and (37) require an M-point IFFT and an N-point IFFT, respectively, and subsequent index finding stages. Including the multiplication and addition operations in Equation (38), the computational cost of the proposed filter becomes complex operations per output sample, close to those of Weyl and Zadeh TF filters, which are per output sample .
However, the main computational expense of our filtering scheme results from computing a TCR representation on which the filter TCR mask is selected. Such a TCR representation either has a quadratic phase kernel function, as the CPF [44, 45], and thus can not be computed by fast algorithms or it requires interpolation by irrational factors, as in [43, 49]. Hence, its computational cost is operations per output sample, instead of per output sample required for TF representations. Overall, our proposed filtering scheme has an increased computational cost of approximately operations per output sample, as compared to operations per output sample required by conventional TF filtering; which is approximately M/(2 log N) times larger.
5.1 Reconstruction of cubic phase signals
5.1.1 Noiseless input case
We first explain steps of the proposed TCR filtering scheme when there is no noise in Equation (39).
A quadratic IF curve that starts at zero frequency and increases with time can be identified on the left part in Figure 2a. That IF curve belongs to the first cubic phase component in Equation (39), which is also given in Equation (40) as the desired component to be reconstructed at the filter output.
Two quadratic IF curves that start at frequencies of π and 2π radians and decrease with time can also be identified on the left part in Figure 2a. They belong to the second cubic phase component in Equation (39). The quadratic IF curve that starts at π/2 radians and increases belongs to the third component. The quadratic phase component is represented by the IF line starting at zero and ending at π/2 values in Figure 2a.
Each quadratic IF curve traverses the discretized TF plane several times, in this simulated signal scenario; hence, they become difficult to identify as time increases, on the right part of Figure 2a. Figure 2b reveals these quadratic curves more clearly, on the right part.
In particular, it is difficult to identify the IF curve of the desired component in Equation (40) and prepare a TF mask to isolate it. TF filtering is difficult to use for this signal separation task involving rapidly changing IF curves of undersampled signal components. However, proposed TCR filter can handle it more easily.
0 ≤ n ≤ N − 1, 0 ≤ k ≤ M − 1. Discrete radian chirp-rate range is taken to be [0, 4π) above to match that of the discrete TCR mask H d (n, 2k/M), 0 ≤ k ≤ M − 1, of the proposed filter, since the mask is prepared based on a display of the TCR distribution given in Equation (41). Thus, two periods of the discrete bilinear TCR distribution in the chirp-rate variable are computed and displayed. M = 8N is taken.
Figure 2c-e show segments of the absolute value of the bilinear TCR distribution in Equation (41) computed for the input signal in radian chirp-rate ranges [0, π/2], [π/2, π] and [2π, 3π], respectively. Horizontal lines are ICR lines of the quadratic phase chirp in Equation (39), and, oblique lines with positive and negative slopes are ICR lines of cubic phase input components in Equation (39), in two periods of the modulus of the TCR distribution.
Instantaneous chirp-rate lines with a larger positive slope, in these figures, belong to the desired component given by Equation (40). Those with a smaller positive slope represent the third cubic phase component. ICR lines with a negative slope that start at chirp-rate values of π and 3π radians correspond to the second cubic phase component. Figure 2e shows all these ICR traces together, in the second period.
(iii) Preparing the TCR mask: Figure 2f shows a segment of the prepared filter TCR mask H d (n, 2k/M), for 0 ≤ n, k ≤ N − 1, corresponding to the radian chirp-rate range [0, π/2], that isolates the ICR line with the larger positive slope in Figure 2c belonging to the desired signal in Equation (40). This linear mask is chosen to be 1 sample wide vertically.
The replica of this line in Figure 2e, located in the range [2π, 4π], should not be included in the TCR mask, since its inclusion would result in an additional, undesired pass region in the TF plane for the equivalent TF transfer function, in addition to the desired TF pass region. If the second cubic phase component were desired at the filter output, then the ICR line with negative slope in Figure 2e, located in the range [2π, 4π], would be selected by the filter TCR mask, but its replica in Figure 2d located in the range [0, 2π] would be left out.
(iv) Selecting the reference IF distribution: Figure 2a indicates that the quadratic IF curve of the desired component in Equation (40) starts from zero frequency at time zero. Hence, we choose n0 = 0 as the reference time point and Hf,d(0, k/N) = 1 for 0 ≤ k ≤ 5, and zero otherwise, as the discrete reference IF distribution of the filter around the zero frequency value. The width of the distribution is determined by a search to maximize the separation performance.
The percentage NMSE of our proposed TCR filter according to Equation (42) is 15.7655% for this noiseless simulation example, as the best result achieved by our filter. Best NMSE figures achieved by interpolated half-band Weyl filter  and full-band suboptimal TF projection filter  are 25.9042 and 20.8597%, respectively, for this case. They give worse results than our proposed filter for this simulation example, because these filters with TF masks that enclose IF traces of the desired component traversing TF plane eight times as shown in Figure 3a are no longer underspread systems. Their performances deteriorate when they become overspread [17, 22], as in this case.
5.1.2 Noisy input case
Percentage NMSEs in recovery of a cubic phase component from a noisy input signal: Percentage NMSEs averaged over 1,000 independent trials that are achieved by the proposed TCR, interpolated half-band Weyl and full-band suboptimal TF projection filters in recovery of the cubic phase component in Equation (40)
The proposed TCR filter gives significantly better results for this signal separation example except for low SNR values, since the desired signal and the filters designed to recover it are overspread. However, NMSE figures for low SNR cases indicate that the TCR filter is less robust against noise than others. At and below SNR = 0 dB, the TCR filter seems to break down, and is surpassed by the TF projection filter. This may be due to the first and second terms in the integral in Equation (31) that might have caused an effect of amplifying the output noise power especially pronounced when the reference IF bandwidth B f is kept larger to obtain a broader TF pass region for capturing the rapidly varying cubic phase component in Equation (40).
5.1.3 Resolution for cubic phase signals
where the first component is the desired signal s(n) given by Equation (40). When s(n) is passed through the TCR filter described above only, the NMSE is 1.9841%. Its slope parameter is a = 1/(48N).
obtained from the increment in the upper bound in Equation (24) by substituting t = N = 128 there and by taking B f /2 = 6/N for the designed TCR filter, the NMSE becomes 30.4978%. Since the second component falls inside the TF pass region of the filter given by Equation (22), for the whole signal duration, the TCR filter could not separate it from the desired component.
When t = N/2 = 64 is taken in the slope difference Δ, the NMSE reduces to 18.2503%, since later parts of these two components are resolved in this case. These results suggest that Equation (24) accurately gives the resolution limit of the proposed filter in the slope parameter.
5.1.4 Recovery of a more slowly varying component
Percentage NMSEs in recovery of a half-band cubic phase component from a noisy input signal: Average NMSE values in percentage achieved by the proposed TCR, interpolated half-band Weyl and full-band suboptimal TF projection filters in recovery of the cubic phase component given by Equation (43), listed versus input SNR values
The TCR mask is prepared to be 1 sample wide along the ICR line of the desired component given by Equation (43), with the number of chirp-rate bins taken as M = 512N for this example to fully expose this ICR line on the underlying TCR pattern, and Hf,d(0, k/N) = 1 for k = 0, 1, N − 1, and zero otherwise, as the discrete reference IF distribution of the TCR filter around the zero frequency value, to yield best results achieved by the TCR filter as listed in Table 2.
For high and medium input SNR values, NMSE figures of the TCR filter are slightly worse than those of the Weyl filter, and slightly better than those of the TF projection filter, as seen in Table 2. But differences are not significant since all filters are underspread systems for this case of a more slowly varying desired component.
At and below SNR = 0 dB, results of the TCR filter deteriorate again due to output noise amplification caused by the first two terms in the integral of Equation (31). However, this deterioration, i.e., the increase in percentage error as input SNR is increased, is about half of the increase observed in Table 1 for the rapidly varying desired signal in Equation (40). This is because the reference IF bandwidth B f is chosen to be 3 samples wide for the desired signal in Equation (43), whereas it was 6 samples wide for the one in Equation (40).
5.2 Reconstruction of quadratic phase signals
5.2.1 Noiseless input case
To recover the quadratic phase component in Equation (39), the TCR mask of the proposed filter is selected to isolate the horizontal ICR line shown in Figure 2c, in the radian chirp-rate range [0, 2π], corresponding to this signal component. The replica of this line in Figure 2e, in the range [2π, 4π], is again left out of the TCR mask. The mask is chosen to be 1 sample wide vertically for the best separation performance.
Figure 2a indicates that the linear IF curve of the desired component also starts from zero frequency at time zero, hence we choose n0 = 0 as the reference time point again, and Hf,d(0, k/N) = 1 for k = 0 and zero otherwise, as the discrete reference IF distribution of the filter as determined by a search to achieve the maximum separation performance.
Figure 3d displays the alias-free WD (AF-WD)  of the output signal of our filter with the above TCR mask and reference IF distribution, demonstrating that the quadratic phase component is separated by the proposed filter. Figure 3e,f plot real parts of the desired quadratic phase component in Equation (39) and the normalized filter output, respectively.
The percentage NMSE of the proposed TCR filter for this separation problem, defined by Equation (42), is 0.2793%, whereas best NMSE results achieved by interpolated half-band Weyl filter and full-band suboptimal TF projection filter are 0.2795 and 0.9992%, respectively.
A noteworthy observation is that the filter TCR mask should be kept as narrow as possible for a good separation performance. While recovering the quadratic phase component in Equation (39), if a neighboring chirp-rate bin is also included in the horizontal TCR mask making it two samples wide, then the NMSE falls down to 14.5101%. Inclusion of the other neighbor for a three-sample TCR mask results in a NMSE value of 17.6307%.
Hence, the chirp-rate value of the desired quadratic phase signal component should be determined accurately from the underlying TCR representation of the input signal, and, the filter TCR mask should match this value in order to recover that component with high accuracy. Quadratic phase signals with chirp-rates differing less than 4π/M radians are not resolvable by the filter.
5.2.2 Noisy input case
Percentage NMSEs in recovery of a quadratic phase component from a noisy input signal: Average NMSE values in percentage achieved by the proposed TCR, interpolated half-band Weyl and full-band suboptimal TF projection filters in recovery of the quadratic phase component in Equation (39), listed versus input SNR values
The proposed TCR filter gives slightly better results for this example except for SNR = 5, 0 and −3 dB cases, but differences are not significant since all filters are underspread for this case. TF projection filter gives slightly worse results here only because its pass region could not be chosen as narrow as those of others, since it could not capture the desired component for such a narrow pass region due to implementation matters.
The TCR filter does not break down for low SNR cases here, since the reference IF bandwidth B f is only one sample wide in this case, limiting noise leakage.
5.3 Recovery of back-to-back objects from in-line Fresnel holograms
Formation of in-line Fresnel holograms can be modelled by a linear, shift-invariant system with a quadratic phase impulse response . Hence, such a hologram can be roughly viewed as a linear combination of chirp signals centered at object plane locations of objects encoded in it.
where X is the spatial sampling period in hologram plane coordinates, N is the discrete hologram image size in samples taken to be N = 256 in our presented simulations, z is the actual object depth, and λ is the illumination wavelength.
Figure 4c displays the AF-WD of the signal in Figure 4b, as a SF pattern. The horizontal axis is the discrete space axis; x = 0, 1, ..., 255. The vertical axis is the discrete spatial radian frequency axis; w m = (2π/256)m, m ∈ [−128, 127]. Each object is represented by a pair of lines with opposite slopes that intersect on the spatial axis at the object coordinate. Magnitudes of the slopes are inversely proportional with real object depths. As can be seen in the figure, these pairs of linear tracks associated with the two objects overlap in the SF plane, making it tedious to design SF filter mask functions to resolve them . However, objects are represented by distinct horizontal strips corresponding to different slopes or depths in the SCR plane. Hence, it is easier to design SCR mask functions isolating such strips and to use our proposed filter to separate and recover these overlapping objects.
Figure 4d-f display modulus of the CPF for the hologram signal in Figure 4b in chirp-rate ranges taken near the zero value, around 2π, and near 4π, respectively. Longer horizontal lines in these figures correspond to slope values of the line pair in Figure 4c with smaller slope magnitude, for the object with normalized depth . Shorter, less visible horizontal lines located symmetrically above or below longer ones correspond to the object with normalized depth . We want to recover the object pulse located at the normalized depth by separating its hologram component from the two-component hologram signal in Figure 4b by our proposed filter. For this purpose, filter SCR masks should be designed to isolate the longer ICR lines in these SCR patterns.
Figure 4c indicates that IF lines of the desired object pass through the zero-frequency level at the intersection point with spatial coordinate n0 = 101. Hence, reference spatial point for our filter is chosen as n0 = 101, and the discrete reference IF distribution for our filtering operations is taken as Hf,d(101, k/N) = 1 for 0 ≤ k ≤ 3 and N − 4 ≤ k ≤ N − 1 around the zero frequency value, and zero elsewhere.
Figure 5c shows a segment of the second SCR mask that selects the longer line in Figure 4f close to the chirp-rate value of 4π. Equivalent Zadeh transfer function of the proposed filter, with this SCR mask and the same reference IF distribution, is shown in Figure 5d. It isolates the IF line in Figure 4c with the larger negative slope. The input hologram signal in Figure 4b is, then, passed through this SCR filter.
The hologram component encoding only the desired object is estimated as the sum of these two filter output signals obtained with different SCR masks given in Figure 5a,c, respectively. Real part of the sum signal is plotted in Figure 5e, as this separated component. The desired object pulse with normalized depth is finally estimated by applying a Fourier synthesis algorithm  to the separated hologram component. The recovered object is plotted in Figure 5f.
In Figure 6d, two back-to-back objects are represented by two pairs of lines with opposite slopes that intersect on the spatial axis (zero-frequency axis) at the common object coordinate n0 = 130. Hence, this intersection point is chosen as the reference spatial point for our filter and its reference IF distribution at this point is, again, chosen as a rectangular pulse centered at the zero-frequency value: Hf,d(130, k/N) = 1 for 0 ≤ k ≤ 7 and N − 8 ≤ k ≤ N − 1, and zero elsewhere.
Figure 6e,f display modulus of the shifted LQP in Equation (45) for the DC-levelled hologram row, with a rectangular window of width 141 samples, displayed in chirp-rate ranges taken near the zero value and 2π, respectively. The lower smeared line in Figure 6e and the upper smeared one in Figure 6f correspond to the square shaped object. The upper horizontal line in Figure 6e and the lower one in Figure 6f are ICR lines corresponding to the triangular object, in the SCR plane.
Similarly, Figure 7d,e show filter SCR masks that isolate horizontal traces in Figure 6e,f, respectively, for the triangular object. Rows of the DC-levelled hologram image are passed through the proposed filter with the described IF distribution and these different SCR masks given in Figure 7d,e, separately. The two filter output images obtained with these two masks are summed, and the triangular object is recovered from the sum image, as shown in Figure 7f.
Object reconstruction procedure described in this subsection does not require prior knowledge of object depths, or a manual search for them, in order to recover the encoded objects from their holograms. It automates this task.
We propose a novel linear time-varying filtering scheme in the joint TCR domain for decomposition of multicomponent signals into their quadratic and/or cubic phase components, in this article. It is valid only when the approximation involved in Equation (6) becomes exact, i.e., it can only separate quadratic or cubic phase signal components with single linear pass regions in the TCR plane corresponding to monotonic ICR laws. This is verified by deriving approximate expressions for equivalent Weyl TF transfer functions of the proposed filter and by checking whether they contain IF curves of such signal components or not, in Section 3.
Multicomponent quadratic or cubic phase signals can be recovered by repeated use of our filter with a different TCR mask, each time, designed to separate a particular signal component. Simulation results presented in Table 1 suggest that our filter decomposes multicomponent quadratic and cubic phase chirps with rapidly varying IF curves with significantly improved accuracies over Weyl and TF projection filters for high and medium input SNR values, albeit with increased computational cost. The reason for this improvement is that filters designed to capture such signals can no longer be underspread systems and hence, Weyl and TF projection filters can not perform so well in recovery of such signals.
However, for low SNR cases, the proposed TCR filter is less robust against noise in recovery of rapidly varying signals since broader TF pass regions required to capture such signals leak more noise and the first and second terms in the integral in Equation (31) further amplify the output noise power.
Average percentage errors listed in Tables 2 and 3 show that the proposed TCR, Weyl and TF projection filters give comparable accuracies in separation of more slowly varying signal components, at least for high and medium input SNR values, since all filters are underspread in such cases.
The idea of filtering in the joint TCR domain is novel. However, since the proposed filter is linear with a time-varying impulse response; it has an equivalent TF transfer function and it can also be designed in the TF domain in theory. The proposed TCR filter intends to ease the filter design task for some applications where it is more convenient to design the filter mask function in the TCR domain than in the TF domain; such as reconstruction of back-to-back objects from in-line holograms for which SF masks have to overlap, whereas SCR masks do not.
Another application involves separation and recovery of undersampled chirps with rapidly varying IF curves that are difficult to identify and enclose in the TF plane. If ICR lines vary more slowly and hence are more readable in the TCR plane, then it is more convenient to perform the filtering operation in the TCR plane by our proposed filter. For example, received radar signals from rapidly maneuvering targets can be modeled as multicomponent cubic phase signals . They can be decomposed by the proposed TCR filter first, for improved estimation of phase parameters of each component to obtain focused SAR or ISAR images of targets as presented in .
We thank the referees for their comments that have significantly improved the manuscript of this work.
- Chen M, Viberg M: Long-range channel prediction based on nonstationary parametric modeling. IEEE Trans Signal Process 2009, 57(2):622-634.MathSciNetView ArticleGoogle Scholar
- Pham DS, Zoubir AM: Analysis of multicomponent polynomial phase signals. IEEE Trans Signal Process 2007, 55(1):56-65.MathSciNetView ArticleGoogle Scholar
- Friedlander B, Francos JM: Estimation of amplitude and phase parameters of multicomponent signals. IEEE Trans Signal Process 1995, 43(4):917-925. 10.1109/78.376844View ArticleGoogle Scholar
- Pham DS, Zoubir AM: Estimation of multicomponent polynomial phase signals with missing observations. IEEE Trans Signal Process 2008, 56(4):1710-1715.MathSciNetView ArticleGoogle Scholar
- Adjrad M, Belouchrani A: Estimation of multicomponent polynomial-phase signals impinging on a multisensor array using state-space modeling. IEEE Trans Signal Process 2007, 55(1):32-45.MathSciNetView ArticleGoogle Scholar
- Shamsunder S, Giannakis GB, Friedlander B: Estimating random amplitude polynomial phase signals: a cyclostationary approach. IEEE Trans Signal Process 1995, 43(2):492-505. 10.1109/78.348131View ArticleGoogle Scholar
- Peleg S, Friedlander B: Multicomponent signal analysis using the polynomial-phase transform. IEEE Trans Aerosp Electron Syst 1996, 32(1):378-386.View ArticleGoogle Scholar
- Barbarossa S, Scaglione A, Giannakis GB: Product high-order ambiguity function for multicomponent polynomial-phase signal modeling. IEEE Trans Signal Process 1998, 46(3):691-708. 10.1109/78.661336View ArticleGoogle Scholar
- Barbarossa S, Petrone V: Analysis of polynomial phase signals by the integrated generalized ambiguity function. IEEE Trans Signal Process 1997, 45(2):316-327. 10.1109/78.554297View ArticleGoogle Scholar
- Cirillo L, Zoubir A, Amin M: Parameter estimation for locally linear FMsignals using a time-frequency Hough transform. IEEE Trans Signal Process 2008, 56(9):4162-4175.MathSciNetView ArticleGoogle Scholar
- Barbarossa S, Lemoine O: Analysis of nonlinear FM signals by pattern recognition of their time-frequency representation. IEEE Signal Process Lett 1996, 3(4):112-115.View ArticleGoogle Scholar
- Barbarossa S: Analysis of multicomponent LFM signals by a combined Wigner-Hough transform. IEEE Trans Signal Process 1995, 43(6):1511-1515. 10.1109/78.388866View ArticleGoogle Scholar
- Greenberg JM, Wang Z, Li J: New approaches for chirplet approximation. IEEE Trans Signal Process 2007, 55(2):734-741.MathSciNetView ArticleGoogle Scholar
- Özgen MT: Analysis of in-line Fresnel holograms via chirp separation. Opt Eng 2006, 45(3):035801-1-035801-11.View ArticleGoogle Scholar
- Santhanam B, Maragos P: Multicomponent AM-FM demodulation via periodicitybased algebraic separation and energy-based demodulation. IEEE Trans Commun 2000, 48(3):473-489. 10.1109/26.837050View ArticleGoogle Scholar
- Kayhan AS: Difference equation representation of chirp signals and instantaneous frequency/amplitude estimation. IEEE Trans Signal Process 1996, 44(12):2948-2958. 10.1109/78.553470View ArticleGoogle Scholar
- Matz G, Hlawatsch F: Linear Time-Frequency Filters: On-Line Algorithms and Applications. In Applications in Time-Frequency Signal Processing. Edited by: Papandreou-Suppappola A. CRC Press, Boca Raton; 2003:205.Google Scholar
- Zadeh LA: Frequency analysis of variable networks. Proc IRE 1950, 76: 291-299.View ArticleGoogle Scholar
- Kozek W: Time-frequency signal processing based on the Wigner-Weyl framework. Signal Process 1992, 29(1):77-92. 10.1016/0165-1684(92)90101-2MathSciNetView ArticleMATHGoogle Scholar
- Shenoy RG, Parks TW: The Weyl correspondence and time-frequency analysis. IEEE Trans Signal Process 1994, 42(2):318-331. 10.1109/78.275605View ArticleGoogle Scholar
- Kozek W: On the generalized Weyl correspondence and its application to time-frequency analysis of linear systems. In Proceedings of the IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis. Victoria, Canada; 1992:167-170.View ArticleGoogle Scholar
- Matz G, Hlawatsch F: Time-frequency transfer function calculus (symbolic calculus) of linear time-varying systems (linear operators) based on a generalized underspread theory. J Math Phys 1998, 39(8):4041-4070. 10.1063/1.532495MathSciNetView ArticleMATHGoogle Scholar
- Hlawatsch F, Kozek W: Time-frequency projection filters and time-frequency signal expansions. IEEE Trans Signal Process 1994, 42(12):3321-3334. 10.1109/78.340770View ArticleGoogle Scholar
- Matz G, Hlawatsch F: Time-frequency projection filters: on-line implementation, subspace tracking, and application to interference excision. In Proceedings of the IEEE International Conference on Acoustics, Speech, Signal Processing. Orlando, FL; 2002:1213-1216.Google Scholar
- Zhang Y, Amin MG: Array processing for nonstationary interference suppression in DS/SS communications using subspace projection techniques. IEEE Trans Signal Process 2001, 49(12):3005-3014. 10.1109/78.969509View ArticleGoogle Scholar
- Portnoff MR: Time-frequency representation of digital signals and systems based on short-time Fourier analysis. IEEE Trans Acoust Speech Signal Process 1980, 28: 55-69. 10.1109/TASSP.1980.1163359View ArticleMATHGoogle Scholar
- Nawab SH, Quatieri TF: Short-time Fourier transform. In Advanced Topics in Signal Processing. Edited by: Lim JS, Oppenheim AV. Prentice Hall, Englewood Cliffs; 1988:289.Google Scholar
- Djukanović S, Daković M, Stanković L: Local polynomial Fourier transform receiver for nonstationary interference excision in DSSS communications. IEEE Trans Signal Process 2008, 56(4):1627-1636.MathSciNetView ArticleGoogle Scholar
- Ventosa S, Simon C, Schimmel M, Dañobeitia JJ, Mànuel A: The S-transform from a wavelet point of view. IEEE Trans Signal Process 2008, 56(7):2771-2780.MathSciNetView ArticleGoogle Scholar
- Simon C, Ventosa S, Schimmel M, Heldring A, Dañobeitia JJ, Gallart J, Mànuel A: The S-transform and its inverses: Side effects of discretizing and filtering. IEEE Trans Signal Process 2007, 55(10):4928-4937.MathSciNetView ArticleGoogle Scholar
- Schimmel M, Gallart J: The inverse S-transform in filters with time-frequency localization. IEEE Trans Signal Process 2005, 53(11):4417-4422.MathSciNetView ArticleGoogle Scholar
- Carmona RA, Hwang WL, Torrésani B: Multiridge detection and time-frequency reconstruction. IEEE Trans Signal Process 1999, 47(2):480-492. 10.1109/78.740131View ArticleGoogle Scholar
- Carmona RA, Hwang WL, Torrésani B: Characterization of signals by the ridges of their wavelet transforms. IEEE Trans Signal Process 1997, 45(10):2586-2590. 10.1109/78.640725View ArticleGoogle Scholar
- Delprat N, Escudie B, Guillemain P, Kronland-Martinet R, Tchamitchian P, Torrésani B: Asymptotic wavelet and Gabor analysis: extraction of instantaneous frequencies. IEEE Trans Inf Theory 1992, 38(2):644-664. 10.1109/18.119728View ArticleMathSciNetMATHGoogle Scholar
- Hlawatsch F, Matz G, Kirchauer H, Kozek W: Time-frequency formulation, design, and implementation of time-varying optimal filters for signal estimation. IEEE Trans Signal Process 2000, 48(5):1417-1432. 10.1109/78.839987View ArticleMATHGoogle Scholar
- Stanković LJ, Stanković S, Djurović I: Space/spatial-frequency analysis based filtering. IEEE Trans Signal Process 2000, 48(8):2343-2352. 10.1109/78.852015MathSciNetView ArticleMATHGoogle Scholar
- Kozek W, Hlawatsch F: A comparative study of linear and nonlinear time-frequency filters. In Proceedings of the IEEE-SP International Symposium on Time-Frequency and Time-Scale Analysis. Victoria, Canada; 1992:163-166.View ArticleGoogle Scholar
- Lee H, Bien Z: Reconstruction of signals with known instantaneous frequency using linear time-varying filter. Electron Lett 1998, 34(24):2312-2313. 10.1049/el:19981605View ArticleGoogle Scholar
- Lee H, Bien Z: A variable bandwidth filter for estimation of instantaneous frequency and reconstruction of signals with time-varying spectral content. IEEE Trans Signal Process 2011, 59(5):2052-2071.MathSciNetView ArticleGoogle Scholar
- Shen H, Papandreou-Suppappola A: Wideband time-varying interference suppression using matched signal transforms. IEEE Trans Signal Process 2005, 53(7):2607-2612.MathSciNetView ArticleGoogle Scholar
- Le Touzé G, Nicolas B, Mars JI, Lacoume J-L: Matched representations and filters for guided waves. IEEE Trans Signal Process 2009, 57(5):1783-1795.MathSciNetView ArticleGoogle Scholar
- Benidir M, Ouldali A: Polynomial phase signal analysis based on the polynomial derivatives decompositions. IEEE Trans Signal Process 1999, 47(7):1954-1965. 10.1109/78.771044MathSciNetView ArticleMATHGoogle Scholar
- O'Neill JC, Flandrin P: Virtues and vices of quartic time-frequency distributions. IEEE Trans Signal Process 2000, 48(9):2641-2650. 10.1109/78.863070View ArticleGoogle Scholar
- O'Shea P: A new technique for instantaneous frequency rate estimation. IEEE Signal Process Lett 2002, 9(8):251-252.View ArticleGoogle Scholar
- O'Shea P: A fast algorithm for estimating the parameters of a quadratic FM signal. IEEE Trans Signal Process 2004, 52(2):385-393. 10.1109/TSP.2003.821097MathSciNetView ArticleGoogle Scholar
- Wang P, Yang J: Multicomponent chirp signals analysis using product cubic phase function. Digit Signal Process 2006, 16: 654-669. 10.1016/j.dsp.2006.09.002View ArticleGoogle Scholar
- Farquharson M, O'Shea P, Ledwich G: A computationally efficient technique for estimating the parameters of polynomial-phase signals from noisy observations. IEEE Trans Signal Process 2005, 53(8):3337-3342.MathSciNetView ArticleGoogle Scholar
- Wang P, Djurović I, Yang J: Generalized high-order phase function for parameter estimation of polynomial phase signal. IEEE Trans Signal Process 2008, 56(7):3023-3028.MathSciNetView ArticleGoogle Scholar
- Cornu C, Stanković S, Ioana C, Quinquis A, Stanković L: Generalized representation of phase derivatives for regular signals. IEEE Trans Signal Process 2007, 55(10):4831-4838.MathSciNetView ArticleGoogle Scholar
- Boashash B: Estimating and interpreting the instantaneous frequency of a signal--Part 2: Algorithms and applications. Proc IEEE 1992, 80(4):540-568. 10.1109/5.135378View ArticleGoogle Scholar
- Katkovnik V: A new form of the Fourier transform for time-varying frequency estimation. Signal Process 1995, 47(2):187-200. 10.1016/0165-1684(95)00107-7MathSciNetView ArticleMATHGoogle Scholar
- Xia X-G: Discrete chirp-Fourier transform and its application to chirp rate estimation. IEEE Trans Signal Process 2000, 48(11):3122-3133. 10.1109/78.875469MathSciNetView ArticleMATHGoogle Scholar
- Gradshteyn IS, Ryzhik IM: Table of Integrals, Series and Products. Elsevier, Burlington; 2007:411.MATHGoogle Scholar
- Abramowitz M, Stegun IA: Handbook of Mathematical Functions, with Formulas, Graphs, and Mathematical Tables. Dover Publications, New York; 1965:446-447.MATHGoogle Scholar
- Nuttall AH: Alias-free Wigner distribution function and complex ambiguity function for discrete-time samples. 1989. NUSC Technical Report 8533, Naval Underwater Systems CenterGoogle Scholar
- Onural L, Scott PD: Digital decoding of in-line holograms. Opt Eng 1987, 26(11):1124-1132.View ArticleGoogle Scholar
- Wang Y, Jiang Y-C: ISAR imaging of ship target with complex motion based on new approach of parameters estimation for polynomial phase signal. EURASIP J Adv Signal Process 2011. doi:10.1155/2011/425203Google Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.