Frequency estimator of sinusoid by interpolated DFT method based on maximum sidelobe decay windows
Introduction
Frequency estimation of sinusoid is a fundamentally important problem in digital signal processing and can be applied in many areas such as communications, measurement, instrumentation, radar, sonar and power systems. Many sinusoidal frequency estimators have been proposed and they can be categorized into time-domain methods [1], [2], [3], [4], [5], [6], [7], [8], [9], [10], [11], [12] and frequency-domain methods [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23]. Time-domain methods estimate sinusoidal frequency by the maximum likelihood methods [1], [2], [3], autocorrelation methods [4], [5], [6], [7], [8], [9], linear prediction methods [10], [11], [12] and so on. These methods generally have low computational efficiency and are not suitable for real-time applications. Frequency-domain methods based on the Discrete Fourier Transform (DFT) of the sampled sinusoid, on the other hand, exhibit low computational complexity and good anti-interference performance. Therefore, they are often adopted in real-time applications. Typically, DFT-based frequency estimators are implemented in two stages. In the first stage, FFT is performed and the maximum spectral line in the DFT spectrum is found to obtain a coarse estimate. In the next stage, a fine estimate is obtained by interpolating the DFT spectral lines adjacent to the maximum spectral line. The positions of the spectral lines employed by DFT-based estimators can be fixed [13], [14], [15], [16], [17], [18], [19], [20], [21], [22] or arbitrary [23]. Candan estimator utilizes the maximum spectral line and its two neighbors to estimate the frequency [19], and its estimation bias is the best among the non-iterative methods experimentally [20]. AM method uses the amplitudes or the complex values of two DFT samples located exactly halfway between the maximum spectral line and its two neighbors [21], and it can approach Cramer–Rao lower bound (CRLB) in two iterations. In [23], a more general form of frequency estimator utilizes two auxiliary spectral lines located at arbitrary position within the main lobe in the fine estimation.
All the above DFT-based methods [13], [14], [15], [16], [17], [18], [19], [20], [21], [22], [23] estimate the frequency in white noise background and are based on rectangular window. In many engineering applications, there are interfering signals besides wide band noise. For example, due to the existence of non-linear loads, harmonic interfering signals are ubiquitous in power systems. Under this situation, when the spectral leakage effect of interfering signals prevails over the wideband noise, nonrectangular windowing methods are often used to reduce the effect of the interfering signals on estimated parameters of desired signal [24], [25], [26], [27], [28], [29], [30], [31], [32], [33]. When the signal is weighted by a non-rectangular window, the curvature around the peak of the DTFT spectrum changes significantly [24]. Several methods have been proposed to generalize the estimators based on rectangular window to the case when non-rectangular windows are adopted [24], [25], [26]. Candan extends the estimator in [19] to the case of non-rectangular windowing [24]. In [25], Jacobsen [18] and AM estimator [21] are generalized to the case when the acquired signal is multiplied by a generic cosine window.
In this paper, the frequency estimator in [23] based on rectangular window is generalized to the case when the signal is weighted by MSD windows in order to reduce the effect of spectral leakage from interfering signals on estimated frequency of the desired signal. In the fine estimation stage, the proposed estimator utilizes two auxiliary spectral lines located at arbitrary position (symmetrically located about the maximum spectral line) within the main lobe of the frequency spectrum of the signal multiplied by MSD windows. An accurate formula for the estimator MSE due to additive white noise is derived. As the frequency intervals between the spectral lines are arbitrary, when non-rectangular windows are adopted, the magnitude spectrum correlation function of complex sinusoid plus complex white noise is needed for the derivation of MSE formula. Thus, this magnitude spectrum correlation function is given. Iterative procedure is used to improve the estimation performance. In order to verify the performance and the MSE formula, simulation experiments are conducted in the case of noisy or noisy and harmonically distorted complex sinusoid. The influence of the interval length between the maximum spectral line and the auxiliary spectral lines on the MSE is analyzed through theoretical analysis and simulation experiments.
The rest of the paper is organized as follows. In Section 2, the expression of the proposed estimator is derived. In Section 3, an accurate formula for the estimator MSE due to additive white noise is derived, and the iterative procedure is described. In Section 4, the accuracy of the MSE formula and the performance of the algorithm are numerically verified. Furthermore, the proposed estimator is compared with state-of-art estimators and the CRLB. Finally, conclusions are drawn in Section 5.
Section snippets
The proposed estimator
The signal is modeled as [22]:where is the noise-free complex sinusoid and is complex white noise with variance . The parameters , and are the amplitude, the frequency and the initial phase respectively. is the sampling frequency, and N is the number of samples. The signal-to-noise ratio (SNR) is defined as .
In the absence of noise, the weighted signal can be expressed as
MSE and iterated application
In this section, we analyze the MSE of according to the estimation formulas given by (9) and (10) in additive white noise background. The observed signal sequence of in (3) is expressed aswhere .
The N-point DFT of is expressed as
According to Appendix B, the MSE of the proposed estimator is
Simulations
In order to verify the frequency estimation performance and the MSE formula, the proposed estimator is simulated and its performance is analyzed in the case of noisy or noisy and harmonically distorted complex sinusoid in this section. In the experiments, the parameters are used as follows: , , and is uniformly distributed between 0 and .
Conclusions
The frequency estimator proposed in the literature [23] is generalized to the case when the signal is weighted by a suitable MSD window in order to reduce the effect of spectral leakage from the interfering signals on estimated frequency of the desired signal. The main difference between the proposed estimator and the other competing estimators is that the two auxiliary spectral lines can be arbitrarily located within the main lobe of the weighted signal spectrum rather than in fixed positions.
CRediT authorship contribution statement
Lei Fan: Conceptualization, Methodology, Software, Writing - original draft. Guoqing Qi: Methodology, Writing - review & editing, Validation. Jinyu Liu: Software, Data curation. Jiyu Jin: Data curation, Writing - review & editing. Li Liu: Validation, Writing - review & editing. Jun Xing: Validation, Visualization.
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 by the Provincial Natural Science Foundation Guidance Plan of China (No. 2019-ZD-0294).
References (33)
- et al.
High-precision frequency estimation of real sinusoids with reduced computational complexity using a model-based matched-spectrum approach
Digit. Signal Process.
(2014) - et al.
Phase correction autocorrelation-based frequency estimation method for sinusoidal signal
Signal Process
(2017) - et al.
An exact analysis of Modified Covariance frequency estimation algorithm based on correlation of single-tone
Signal Process
(2012) - et al.
Frequency estimation of real-valued single-tone in colored noise using multiple autocorrelation lags
Signal Process
(2010) - et al.
A closed-form expanded autocorrelation method for frequency estimation of a sinusoid
Signal Process
(2012) - et al.
Two-stage autocorrelation approach for accurate single sinusoidal frequency estimation
Signal Process
(2008) - et al.
Phase correction of discrete Fourier transform coefficients to reduce frequency estimation bias of single tone complex sinusoid
Signal Process
(2014) - et al.
Frequency estimator of sinusoid based on interpolation of three DFT spectral lines
Signal Process
(2018) Fine resolution frequency estimation from three DFT samples: case of windowed data
Signal Process
(2015)- et al.
Frequency estimation by two- or three-point interpolated Fourier algorithms based on cosine windows
Signal Process
(2015)
Impact of harmonics on the interpolated DFT frequency estimator
Mech. Syst. Signal Process.
Frequency estimation of a sinusoidal signal via a three-point interpolated DFT method with high image component interference rejection capability
Digit. Signal Process.
Sine-wave parameter estimation by interpolated DFT method based on new cosine windows with high interference rejection capability
Digit. Signal Process.
Effect of noise and harmonics on sine-wave frequency estimation by interpolated DFT algorithms based on few observed cycles
Signal Process
Multipoint interpolated DFT for sine waves in short records with DC components
Signal Process
Accurate frequency estimation of a noisy sine-wave by means of an interpolated discrete-time Fourier transform algorithm
Measurement
Cited by (9)
Analysis of asymptotic and exact Cramér-Rao lower bounds on the parameters of a noisy periodic signals
2023, Digital Signal Processing: A Review JournalAN APPROACH FOR SINGLE-TONE FREQUENCY ESTIMATION USING DFT INTERPOLATION WITH PARZEN WINDOWING
2023, Kufa Journal of EngineeringHigh-precision frequency estimator by using discrete Fourier transform and asymmetric discrete time Fourier transform samples
2023, Review of Scientific InstrumentsAccurate Frequency Estimator for Real Sinusoid Based on DFT
2022, Electronics (Switzerland)