This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Miloˇ s Dakovi´ c, Igor Djurovi´ c∗ , LJubiˇ sa Stankovi´ c Elektrotehnicki Fakultet, University of Montenegro, 81000 Podgorica, Montenegro, Yugoslavia. ∗ on leave at the Department of Mechanical and System Science, Kyoto Institute of Technology, Matsugasaki, Sakyo, Kyoto, 606-8585 Japan. e-mail: {milos, igordj, ljubisa}@cg.ac.yu

ABSTRACT Adaptive local polynomial Fourier transform (ALPFT) is proposed in this paper. For multicomponent FM signals with parallel instantaneous frequencies (IF) the ALPFT is chosen as the local polynomial Fourier transforms (LPFT) which produces maximal concentration measure from a set of the LPFTs. For other forms of the FM signals the ALPFT is determined as a weighted sum of the LPFTs. The weighting coeﬃcients are determined as a function of the concentration measure. The proposed transforms produce highly concentrated time-frequency signal representations. 1 INTRODUCTION 2 LOCAL POLYNOMIAL FOURIER TRANSFORM

**The LPFT is deﬁned by Katkovnik in [10] for the IF estimation of polynomial phase signals: LP F T (t, ω ~ ) = LP F T (t, ω1 , ..., ωM ) = = Z
**

∞

x(t + τ )w∗ (τ )e−j ω1 τ −jω2 τ Z

∞

2

/2−...−j ωM τ M /M !

dτ =

−∞

=

x(t + τ )w∗ (τ )e−j

−∞

PM

i=1

ωi τ i!

i

dτ ,

(1)

In order to get highly concentrated and accurate adaptive TF representations, measures of the time-frequency (TF) distributions concentration have been used in [1][7]. Many of the concentration measures have the origin in the information theory [8]. However, problems in the TF representations concentration measuring can be quite diﬀerent than those in the information theory. Thus, the concentration measures developed in the information theory should be carefully used in TF analysis (usually with necessary modiﬁcations). Analysis of some concentration measures in the TF analysis can be found in [8]. Adaptive harmonic fractional Fourier transform has been proposed in [4]. This transform is based on the concentration measure. It has been successively applied to the speech signals. Unfortunately, this method cannot be used in a straightforward manner to other types of FM signals. In this paper we started with similar assumptions as in [4]. The adaptive local polynomial Fourier transform (ALPFT) is deﬁned as a weighted sum of the local polynomial Fourier transforms (LPFT). The adaptive weighting coeﬃcients are obtained by using a concentration measure. Two schemes for weighting coeﬃcients determination are developed. The paper is organized as follows. Deﬁnition and properties of the LPFT are given in Section II. The ALPFT is introduced in Section III. Numerical examples are given in Section IV.

where ~ ω = (ω1 , ω2 , ..., ωM ), and w(τ ) is the window function. The main advantage of LPFT over other TF representations is in its linearity. The local polynomial periodogram (LPP) is used in order to avoid the complex nature of the LPFT: LP P (t, ω ~ ) = |LP F T (t, ω ~ )|2 . (2)

**For ω2 = ω3 = ... = ωM = 0 the LPFT is reduced to the short-time Fourier transform (STFT): LP F T (t, ω1 = ω , 0, ..., 0) = ST F T (t, ω ) = = Z
**

∞

x(t + τ )w∗ (τ )e−j ωτ dτ .

(3)

−∞

In this case the LPP is equal to the spectrogram SP EC (t, ω ) = |ST F T (t, ω )|2 . The basic drawback of the LPFT is in increase of dimensionality, i.e., increase of the calculation complexity. We will restrict our analysis to the case of M = 2, when: LP F Tα (t, ω) = Z

∞

x(t + τ )w∗ (τ )e−j ωτ −j ατ

2

/2

dτ . (4)

−∞

This form of the LPFT is closely related to the fractional Fourier transform (FRFT) [9]. Properties of the LPFT and the LPP in the IF estimation are analyzed in details [10].

(14) where g(·) is concentration measure of the TF representation calculated for the entire TF plane. when the LPFT is ideally concentrated along the IF for all signal components ω = at + bi . Calculate LP F Tα (t. R∞ |LP F Tα (t. t) will be used in the examples [4]: R∞ |LP F Tα (t. This is the reason for an alternative deﬁnition of the ALPFT as the weighted sum of the LPFTs. ω ). that there are several other measures whose values decrease when the TF representation concentration increases [8]. ω ) for p → ∞. α). The STFT. ω )|2 dω H (α. (7) where f (·) > 0 is an increasing function x1 > x2 → f (x1 ) ≥ f (x2 ). ω) are shown in Figure 1. where p > 0. ω). (15) (8) For signals with time-varying chirp rate a the adaptive LPFT can be calculated for each time-instant: H (t. The adaptive LPFT is the one from the considered set of the LPFTs which produces the highest concentration. (5) (10) The LPFT of this signal is: LP F Tα (t. Procedure for determination of the adaptive LPFT can be summarized as follows. Note. ω ) where α ˆ (t) = arg max H (t. ˆ (t) are shown in Figure 2. ω ) and ALP F T 00 (t. ALP F T 00 (t. ALP F T 000 (t. The concentration measure H (α) is shown in Figure 1d. ω ) is equal to ALP F T 00 (t. 3. ω )). The Hamming window of the width N = 512 is used. α Ai ejat 2 /2+jbi t . The chosen LPFT would have the value α as close as possible to the value of a. t) f (H (α. ω ) = ALP F T 00 (t. ω ) = LP F Tα ˆ (t. ω). ω ) = LP F Tα ˆ (t) (t. Note that in this case holds ALP F T 0 (t. Concentration can be measured by using some of the concentration measures presented in [1]-[4]. Measures that increase their values when the concentration of the TF representation increases its value are used in this paper. ω ) = α where W (ω) is the Fourier transform of the window function w(t). Example 2: Consider the sum of three parallel sinusoidal modulated FM signals: x(t) = exp(j 64 cos(2πt))[1 + 2 cos(512πt)]. t) will be used in numerical examples. the adaptive TF representation can be obtained by considering the set of the LPFT with various values of α ∈ Λ. ω) is close to the optimal one. with the sampling interval ∆t = 1/1024.1 Parallel Instantaneous Frequencies Consider the sum of linear FM signals with parallel instantaneous frequencies: x(t) = XQ i=1 where gt (·) is concentration measure for the considered instant t. 3/4]. The adaptive LPFT is obtained as: ALP F T 0 (t. α within t ∈ [−3/4. ω) for α ∈ Λ. ω ) where α ˆ = arg max H (α). α) = gt (LP F Tα (t. ω)|dω −∞ Example 1: Consider the signal: x(t) = exp(j 256πt2 )[1 + 2 cos(256πt)]. Note that ALP F T 000 (t. (9) Considered time and sampling intervals are the same as in the previous example. (11) ALP F T 000 (t. for all α ∈ Λ. t) = ³ −∞ (13) ´3/2 . It can be seen that ALP F T 00 (t. while α ˆ (t) is given with expression (10). ω ) produces the ideal TF concentration while ALP F T 000 (t. The following function: µ ¶p H (α. t))LP F Tα (t. The absolute values of the STFT. (6) −a For multicomponent signals that could not be represented as a sum of the signals with parallel IFs. 1. the proposed procedure would produce the ideal TF representation of the strongest signal’s component while other components would remain low-concentrated. ω )). 2. The adaptive LPFT can be calculated for the considered instant as: ALP F T 00 (t. Thus. Calculate the concentration measure: H (α) = g(LP F Tα (t. The LPFT (6) achieves maximal concentration for α = a. The weighting coeﬃcients are calculated according to the concentration measure: X f (H (α. t)) = (12) H (ˆ α(t). ω ) = = Z ∞ −∞ XQ ∞ i=1 Ai eja (t+τ )2 2 +jbi (t+τ ) w∗ (τ )e−j ωτ −j α τ2 2 dτ = = ejat Z ejatτ +jaτ Q X i=1 2 2 /2 /2+jbi τ XQ i=1 Ai ejbi t × 2 w∗ (τ )e−j ωτ −j ατ /2 dτ = −∞ 2 = ejat /2 · ¸ q (ω−at−b)2 π (α−a) Ai ejbi t W (ω ) ∗ω α2 e . 4 NUMERICAL EXAMPLES The following form of the concentration measure H (α.3 ADAPTIVE LOCAL POLYNOMIAL FT 3. ALP F T 000 (t. ω ) and α .

tm ). ω ). Example 4: Consider the sum of three FM signals (linear modulated. Example 5: Consider a frequency coded signal with elementary signals of the form [11]: ¯m )2 /2)+ x(t) = A exp(j 8π · a1 (t)(t − t ¯m )2 /2) for t ∈ [tm−1 . ω ). c) ALP F T 00 (t. ω) outperforms ALP F T 00 (t. ω ). c) ALP F T 00 (t. ω). a2 (t) = −am for t ∈ [tm−1 . ω ) is ideally concentrated along the IF of the ﬁrst component exp(j 256π t2 ). ω ) are shown in Figure 4. d) α ¯m is the middle point of the interval of elemenwhere t ¯m = (tm + tm−1 )/2. ω ) and ALP F T 00 (t. sinusoidal modulated and chirp pulse): x(t) = exp(j 512πt2 + j 192 sin(2πt)) + exp(j 512πt2 )+ + exp(−64(t − 0. ω). b) ALP F T 000 (t.Figure 3: TF representations of three non-parallel FM signals: a) STFT. ALP F T 000 (t. Figure 3b. while other two components remain spread. ALP F T 000 (t. ω ). (16) The STFT. b) ALP F T 000 (t. ω ) and ALP F T 00 (t. The trade-oﬀ between concentration of all components is achieved by using ALP F T 000 (t. tm ).2)2 − j 128πt2 − j 512πt). ω ) of signal (16) are shown in Figure 3. b) ALP F T 000 (t. (18) +A exp(−j 8π ·a2 (t)(t−t Figure 2: TF representations of three parallel sinusoidal FM signals: a) STFT. ALP F T 00 (t. c) ˆ (t). It can be seen that ALP F T 000 (t. Figure 3c. ω). The values of the chirp tary signal t rate a(t) represent coded message: a1 (t) = am . It can be seen that ALP F T 00 (t. Example 3: Consider the sum of three linear FM signals with nonparallel IFs: Figure 1: TF representations of three parallel FM signals: a) STFT. (19) . (17) The STFT. ω). d) H (α). x(t) = exp(j 256πt2 ) + exp(j 128πt2 + j 512πt)+ + exp(j 384π t2 − j 384πt).

Proc. Brown. MTS/IEEE Conf.Sig..42. Stankovi´ c: ”A measure of some time-frequency distributions concentration”..J.169-198.10.997-1000. W. No. IEEE Trans. Baraniuk. pp. Chen.8. Federal Republic of Germany.Jour.Sig. Sept. Hero: ”Uncertanity. information and time-frequency distributions”. Katkovnik: ”Local polynomial periodogram for time-varying frequency estimation”. Djurovi´ c: ”Time-frequency analysis of frequency-coded signals”. M.Let. No. pp.J. HI.Sig.L. [2] T.144-156. [8] LJ. Vol. Vol. IEEE Sig. 5 CONCLUSION Figure 5: TF representations of frequency coded signals: a) STFT. No. pp. c) ALP F T 00 (t. [5] H. SPIE. Jones: ”Improved instantaneous frequency estimation using an adaptive short-time Fourier transform”. Cakrak. Vol.29.ICASSP. c) ALP F T 00 (t. Loughlin: ”Multiple window timevarying spectral analysis”.Proc. No. Cakrak. The adaptive LPFT based on the concentration measure is proposed.2964-2972. Work of Dr Igor Djurovi´ c is supported by the Japan Society for the Promotion of Science. 2001. Two diﬀerent types of this transformation are proposed and analyzed. 1995. ω ) are shown in Figure 5. From this ﬁgure we can easily see the improvement in TF representation that is achieved by using ALP F T 000 (t. ω) and ALP F T 00 (t. D.1589-1602. Zhang. D. Vol. pp. The procedure for determination of the adaptive LPFT is presented.1993.281-283.1656-1666. Almeida: ”The fractional Fourier transform and time-frequency representations”. [6] F. b) ALP F T 000 (t. [4] F. 7. Nov. Stankovi´ c.6.48. IEEE Trans. ω ).81. Vol. Oceans 2001. 5. Vol. Loughlin: ”Multimodal timevarying spectrum with instantaneous bandwidth and frequency constraints”. 10. 3. Williams: ”R´ enyi information and signal dependent optimal kernel design”.Proc.. 6 and 1. No.A. nel design”. 1991.11. In all cases highly concentrated representations are achieved. ALP F T 000 (t.Proc. 1995. ω ).J. ω).41.1999. 8.L.G. pp. pp. The signal is corrupted by the complex Gaussian noise with variance 2σ2 = 1..Proc. 9.621-631. [11] O. Proc.K.4. ω ).Figure 4: TF representations of three non-parallel FM signals: a) STFT. Jones: ”A signal dependent time-frequency representations: Optimal ker- . Kwok.Proc. Feb. P.11.2. IEEE Trans.2001. P. where am takes the values 2. Stat.Sig. G. LJ. [7] F. References [1] W. b) ALP F T 000 (t. South. Vol. 6 ACKNOWLEDGMENT This work is supported by the Volkswagen Stiftung.H. pp. No. Signal Processing.2000. IEEE Trans. Vol. including nonlinear FM signals and signals whose components are not parallel in the TF domain. Theory is illustrated on several numerical examples. A. pp. No. Aug. March 2001.49. Bi: ”Adaptive harmonic fractional Fourier transform”. The STFT. Sang.. Rogozovskyi. [3] R. Y.Sig.B. I. pp.1994.L. Apr. Nov. pp. Williams.3. 4.Afr. [9] L.. Vol.2. USA.49.Q.448-453.2001.1566.. [10] V. Vol. IEEE Trans.30843091. Oct.O.J. ω). Honolulu.

tf analiza

tf analiza

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd