Elsevier

Signal Processing

Volume 81, Issue 1, January 2001, Pages 197-202
Signal Processing

IF and GD estimation from evolutionary spectrum

https://doi.org/10.1016/S0165-1684(00)00199-7Get rights and content

Abstract

In this paper, we present estimators of instantaneous frequency (IF) and group delay (GD) obtained from dual evolutionary spectral estimators, the evolutionary periodogram and the transitory evolutionary periodogram. We obtain the IF and the GD estimators using the first moments. We show that IF and GD may be estimated using a sample of evolutionary time and spectral correlation functions, respectively. Examples illustrating the performances of the estimators are presented.

Introduction

There are two basic approaches for estimation of instantaneous frequency (IF) from time–frequency distributions (TFD): peak detection and moment methods [1]. Performance of peak detection method is good at high signal-to-noise ratios (SNR). However, it deteriorates at low SNRs or when IF is a nonlinear function. The moment method may be used for both IF and group delay (GD) estimation from TFDs.

In this work, we use two estimators of the evolutionary spectrum for IF and GD estimation. The estimators are the evolutionary periodogram (EP) and the transitory evolutionary periodogram (TEP). We use the first moments of the EP and the TEP for the estimation of IF and GD. It is shown that IF and GD may be obtained from a sample of evolutionary autocorrelation and spectral correlation of a signal, respectively.

The paper is organized as follows. In Section 2, we present a review of the evolutionary spectral estimators, the EP and the TEP. We consider IF estimation and GD estimation in 3 IF estimation, 4 GD estimation, respectively. In Section 5, we propose modifications to the estimators to improve their performances. Finally, in Section 6, we present experimental results.

Section snippets

Evolutionary spectral estimation

According to the Wold–Cramer decomposition, a nonstationary process {x[n]} may be represented as [5], [7]x[n]=ππH(n,ω)ejωndZ(ω),where Z(ω) is an incrementally orthogonal process and H(n,ω) is a slowly time-varying function. In the above equation, {x[n]} is represented as the integral sum of complex exponentials with time-varying, complex, and random amplitudes.

Then, the (oscillatory) evolutionary spectral density function may be defined as [5], [7]S(n,ω)=12π|H(n,ω)|2.

For a signal x[n], 0⩽nN

IF estimation

Although there are many methods for IF estimation, we will consider only the moment method involving TFDs. The moment method uses the first moment of a given TFD, S[n,l],0⩽n,l⩽N−1, at each time instant [1], [2], [8]f[n]=12πargl=0N−1ej2πl/NS[n,l]mod2π.Since the discrete-time TFDs are periodic in frequency, circular moment is used instead of linear moment in the above equation [2], [8].

In order to estimate IF from the EP, (4) is substituted into (11) and the result is (see Appendix A)f[n]=12πarg

GD estimation

Group delay may be estimated by taking the first moment of TFDs along the time axis [1]τ[l]=n=0N−1nS[n,l]n=0N−1S[n,l].However, both the EP (with Fourier functions) and the TEP may be considered as periodic in n because of their special form. In this case, we can define the normalized GD estimator as the first circular moment of S[n,l],0⩽n,l⩽N−1 asτ[l]=12πargn=0N−1ej2πn/NS[n,l]mod2π.

GD estimate from the EP may be obtained by substituting (3) into (15):τ[l]=12πargi=−(M−1)/2(M−1)/2X

Modified estimators

All of the above estimators involve terms which may be classified as forward or backward correlations. In order to improve their performances, they may be replaced with central correlations. Thus, the modified estimators may be defined as follows.

The modified EP–IF estimator:f[n]=14πargk=1N−2xn[k−1]xn[k+1]mod2π.

The modified TEP–IF estimator:f[n]=14πargi=−(M−1)/2(M−1)/2x[n+i−1]x[n+i+1]mod2π.

The modified EP–GD estimator:τ[l]=14πargi=−(M−1)/2(M−1)/2X[l−i+1]X[l−i−1]mod2π.

The modified TEP–GD

Experimental results

In this section, we present some experimental results. We first consider a constant-amplitude linear FM chirp signal. The signal length, N=128 and the instantaneous frequency, f[n], increases linearly from 0 to 0.5. The instantaneous frequency is estimated using the equations presented in the previous sections for both the EP and the TEP with M=11. The Fourier functions are used in the simulations but other basis functions produce similar results. In Fig. 1(a), we show the IF estimates at 30dB

Conclusions

We presented estimators of IF and GD obtained from the first moments of two evolutionary spectral estimators, the evolutionary periodogram and the transitory evolutionary periodogram. We showed that IF estimators could be obtained from a sample of the evolutionary autocorrelation and GD estimators from a sample of the evolutionary spectral correlation. Finally, we presented experimental results illustrating performances of the estimators.

References (8)

  • B. Boashash

    Estimating and interpreting the instantaneous frequency of a signal-parts 1 and 2

    Proc. IEEE

    (April 1992)
  • B. Boashash et al.

    Instantaneous frequency and time–frequency distributions

  • C. Detka, A. El-Jaroudi, Transitory evolutionary spectrum, ICASSP, Vol. IV, 1994, pp....
  • S. Kay

    A fast and accurate single frequency estimator

    IEEE Trans. Acoust., Speech, Signal Process.

    (December 1989)
There are more references available in the full text version of this article.

Cited by (7)

  • Improved evolutionary spectrum estimation using short time analytic discrete cosine transform with modified group delay

    2022, Mechanical Systems and Signal Processing
    Citation Excerpt :

    If the GR can be reduced by preserving the best FR, then STFT can be a better ES estimator. For estimation of the spectrum, modified group delay (MGD) function has been proposed and applied [10,11] to get rid of the GR in the absence of any smoothing window and this retains the best FR of the boxcar window [11]. The GR, the white noise input and the white noise present along with the signal produce zeros near to the circle of radius unity.

  • HHT and a new method for removing white noise

    2002, Proceedings of the 2002 International Conference on Noise and Vibration Engineering, ISMA
View all citing articles on Scopus
View full text