WorksEngineering
An Accurate Modulation-Assisted Method for Phasor Estimation Under Decaying DC Offsets
Abstract
Decaying DC (DDC) offsets severely compromise the accurate estimation of current phasors immediately following power grid fault occurrence. Consequently, conventional full-cycle discrete Fourier transform (FCDFT) algorithms experience decaying oscillatory magnitude errors, potentially causing distance relays to malfunction. This paper introduces a modulation-assisted technique designed to precisely eliminate the DDC components from fault current signals. By multiplying the fault signal with a high-frequency carrier and applying consecutive sliding Fourier projections, the proposed method effectively isolates the DDC signal while entirely rejecting higher-order harmonic interference. It utilizes a direct algebraic solution derived from a robust forward-difference operation across just three sequential projections to estimate and extract the DDC parameters. Furthermore, the algorithm seamlessly processes both single and multiple DDC components by dynamically tracking them as one equivalent first-order mode. Extensive evaluations are conducted to validate the proposed method accuracy. The proposed algorithm limits the maximum overshoot (MO) to 0.62% (0.59% for ) for the evaluated DDC cases, and to 2.47% (0.95% at 20 dB for ) under noisy conditions.
Introduction
Index Terms— phasor estimation, decaying dc offset, discrete Fourier transform, modulation, relaying, fault current.
Fast and accurate estimation of the fundamental fault-current phasor is a basic requirement of modern power-system protection. Distance, directional, differential, and over-current relays use current phasors to identify the faulted element and issue a secure trip command within a few cycles. The discrete Fourier transform (DFT) is widely employed for this purpose because it has a simple implementation, a fixed computational burden, and exact rejection of integer harmonics under nominal-frequency steady-state conditions. However, its performance deteriorates during the first few cycles following a transmission line fault. Fault currents generally contain harmonics, measurement noise, and one or more exponentially decaying dc (DDC) components. Because a DDC is nonperiodic within the DFT window, it is not orthogonal to the fundamental-frequency basis functions and produces a decaying oscillatory error in the estimated magnitude and phase [1, 2, 3, 4, 5]. This transient error can delay relay convergence and cause relay overreach or underreach.
The DDC observed by a protection relay may have more than one physical origin. The primary component is produced by the transient response of the faulted power network; its initial magnitude depends mainly on the fault-inception angle, whereas its time constant is governed by the post-fault resistance-to-inductance ratio. A second component may be introduced by the transient response of the current transformer (CT). The primary and CT-induced components can have substantially different amplitudes and decay rates, particularly during high-current faults and CT saturation. The problem is also time critical: the DDC may persist beyond the decision interval of a high-speed relay, while waiting for its natural decay defeats the purpose of fast phasor estimation [3], [6], [7]. An effective estimator should therefore suppress both single- and multiple-DDC components without sacrificing harmonic rejection, noise tolerance, or convergence speed.
Existing DDC-mitigation methods can be broadly classified according to whether they preprocess the current waveform, correct the DFT output, or explicitly identify multiple exponential modes. In the first class, the DDC is suppressed before phasor calculation. The digital mimic filter is a representative solution and can substantially reduce the offset before the current is applied to the DFT [8]. Nevertheless, its response depends on the agreement between the selected and actual DDC time constants, which vary with the fault condition. In addition, it may also amplify noise and high-frequency transients. Adaptive mimic filtering reduces the time-constant-mismatch problem but introduces an additional parameter-adaptation stage [9]. Other preprocessing approaches use empirical-mode decomposition, intrinsic time-scale decomposition, recursive wavelets, or complex-frequency filters [10, 11, 12, 13]. Although these methods can attenuate DDC effect before phasor extraction, their decomposition, reconstruction, or auxiliary-filtering requirements are less direct than those of a conventional DFT relay.
The second class modifies the DFT calculation or estimates the leakage introduced by an assumed DDC model. Simplified Fourier-domain corrections use relations among consecutive samples or DFT outputs to compensate for the exponential term [1], [2]. Related modified-DFT formulations combine half- or full-cycle outputs, odd and even sample sets, or selected spectral components to suppress the offset-induced error [14]. Online parameter-estimation methods infer the decay factor and remove its contribution from the measured phasor [15], while analytical and fast-approval formulations seek to shorten the transient interval using compact sample combinations [4], [16], [17]. High-frequency modulation has also been used to move DDC information away from zero frequency and then recover the fundamental phasor through modified DFT operations [6], [18]. These methods can significantly improve the conventional DFT when the adopted signal model is satisfied. The method in [19] segments a one-cycle window into subsets and applies DFT to construct a complex equation system incorporating the fundamental frequency phasor alongside two decaying DC offsets. However, the computational complexity is high for the algorithm.
Window length is a central design variable in these formulations. Subcycle estimators respond early but weaken noise averaging and may forfeit full-cycle harmonic rejection [20]. Weighted least square and adaptive-window estimators provide flexibility at the cost of a larger signal model or parameter-selection mechanism [21, 22, 23, 24]. Conversely, full-cycle integration preserves useful Fourier properties but introduces observation delay. Taylor approximation simplifies the estimation equations, although its truncation error depends on the decay-time range and sampling interval. Frequency-deviation-aware formulations address another error source but require additional estimation stages [25]. These tradeoffs become more consequential when the primary and CT-induced offsets have widely separated decay rates [6], [7], [26].
The explicit treatment of two or more DDC modes has consequently received increasing attention. A modulation-based DFT method estimates the primary and secondary DDC components by combining successive outputs of one-cycle averaging filters and removes their contributions from the conventional DFT result [6]. A differential-equation formulation decomposes the fault current into modal components and permits the moving-window length to be selected according to the required compromise between accuracy and delay [27]. Selective interval integration estimates the system and CT offsets from two properly separated integration intervals without applying a Taylor approximation to the exponential terms [7]. Multiple exponentially decaying components have also been addressed directly in the DFT domain [28], [29]. Half-cycle and hybrid formulations estimate two exponential modes through auxiliary signals, successive integrations, or carrier modulation [30], [31]. These approaches establish that retaining separate decay modes can eliminate the residual oscillations that remain after an equivalent single-DDC approximation. At the same time, their implementations involve different combinations of auxiliary filtering, interval selection, successive-window processing, model assumptions, or numerical parameter extraction. Thus, there remains a need for a concise formulation that preserves the harmonic selectivity of a full-cycle DFT, without requiring prior knowledge of the number of DDC components or iterative nonlinear parameter estimation.
This paper develops a modulation-assisted estimator to meet protective relaying requirements. The measured current is multiplied by a high-frequency carrier signal, which shifts the non-oscillatory DDC component into damped components centered at the selected carrier frequency. A sliding full-cycle projection is then applied to reject standard integer harmonics. The high-frequency modulation removes the contribution of the oscillatory term, allowing the projection sequence to be approximated by a decaying exponential. By shifting these consecutive projections by a small fractional time step, the outputs form a predictable discrete sequence. Unlike conventional estimators that rely on iterative approximations, the proposed sequence is resolved purely algebraically. Applying a simple forward difference operation across consecutive observations entirely removes constant residual integration errors. The resulting difference equation directly yields the precise decay characteristics, allowing the algorithm to dynamically track the DDC component and deliver a fully compensated fundamental phasor estimate continuously. The contributions of this paper are as follows:
A robust high-frequency modulation technique is introduced to decouple DDC offsets from fundamental fault phasors. By translating non-oscillatory components into a higher spectrum, a full-cycle sliding projection precisely isolating pure DDC terms without iterations.
A flexible modulation multiplier selection framework optimizes noise immunity and minimizes estimation transients. This adaptive approach guarantees overall hardware compatibility, seamlessly accommodating both standard relay anti-aliasing filter bandwidth limits and diverse digital distance protection sampling rates.
A direct algebraic forward-difference formulation dynamically extracts equivalent DDC parameters. This robust first-order model accurately approximates the composite transient behaviors, seamlessly scaling to resolve multiple interfering DDC components.
The remainder of this paper is organized as follows. Section II presents the proposed algorithm for eliminating the DDC component via high-frequency modulation, detailing both the time- and frequency-domain derivations. Section III evaluates the performance of the proposed method under various scenarios, including single and double DDC components, measurement noise, varying sampling rates, odd harmonics, and off-nominal frequencies, followed by simulation-based assessments on a test network. Finally, Section IV concludes the paper and summarizes the key capabilities of the method.
Proposed Algorithm
Modulation-Assisted DDC Parameter Extraction
The overall schematic diagram of the proposed method is summarized in the flowchart presented in Figure 1. The proposed method consists of signal modulation, consecutive sliding projections, a forward-difference operation, DDC reconstruction, and finally the fundamental phasor estimation.

Figure 1. Flowchart of the proposed modulation-assisted method for removing DDC of the fault current.
Signal Modulation and Sliding Projection Derivation
Consider a fault current signal containing a single DDC component alongside harmonics:
where and are the initial amplitude and time constant of the DDC, respectively, is the fundamental angular frequency, and and represent the amplitude and phase angle of the -th harmonic component.
Modulating with a harmonic carrier , where is the integer harmonic carrier multiplier, yields:
A sliding-window Fourier projection is applied specifically at the -th harmonic bin over a fundamental cycle , starting at an arbitrary initial time for the -th observation window, where is the sliding time step. Assuming the fault current does not contain a significant harmonic of order , the cross-product terms of integer harmonics integrate precisely to zero over , simplifying the complex projection to:
Applying Euler’s identity, the mixed trigonometric-exponential term expands into a stationary baseband term and a double-frequency oscillatory term:
Substituting (4) into (3) splits the projection into two distinct integrals:
Simplifying both integrals at the boundaries and yields:
where the simplification has been applied.
The significance of the oscillatory phase term relative to the pure decay term is given by the ratio :
For practical power system faults ( [8]), selecting a sufficiently high carrier multiplier () guarantees the condition , making the oscillatory term mathematically negligible (). However, the maximum value of is practically constrained by the digital relay’s hardware configuration. To prevent signal degradation, the translated carrier frequency (, where is the fundamental frequency) must remain strictly below the cut-off frequency () of the relay’s analog anti-aliasing filter (AAF) and strictly within the digital Nyquist limit [32]. Therefore, an optimal modulation index (e.g., for standard commercial relays) is selected to simultaneously eliminate the oscillatory integration error while preserving the integrity of the isolated DDC component within the AAF’s flat passband.
Omitting this negligible component, the complex sliding projection is reduced to a purely decaying sequence without phase oscillations:
where is a constant complex scalar, and the substitution discretizes the temporal decay.
Single DDC
When exactly one physical decaying mode is present, the discrete sequence modeled in (8) follows a first-order linear homogeneous difference equation:
where is the autoregressive coefficient.
While could theoretically be extracted directly from the raw projections, practical fault currents contain low-frequency A/D offsets and unsuppressed spectral leakage. To remove gradual background noise and ensure computational stability, the difference between consecutive points is evaluated as
Input Fault Current (1)
Modulate: (2)
Sliding Projection: over shifted by (8)
Forward Difference: (10)
Extract Equivalent Root & Time Constant: , (22), (24)
Reconstruct DDC & Compensate Waveform
FCDTF
Fundamental Phasor Extraction
The differenced sequence maintains the homogeneous property:
Evaluating this at provides the robust coefficient :
The discrete root is , and the time constant is extracted as:
With known, the initial amplitude is recovered algebraically from the differenced projection components after compensating for the isolated exponential mode.
Two DDC
The fault current may exhibit two DDCs. In this case, the discrete sequence is modeled by a second-order linear homogeneous difference equation:
Similar to the single DDC scenario, applying the difference operator across consecutive equations filters out practical baseline offsets, transforming the system into a robust homogeneous form on the differences:
To determine the unknown autoregressive coefficients , five consecutive sliding windows () are processed to form a fully determined linear system:
Solving this matrix yields and . The discrete roots and are derived from the characteristic equation . The true time constants are subsequently calculated as . The complex scalars and , along with their corresponding initial amplitudes and , are then isolated by fitting the discrete roots back to the differenced projection sequence extract the exponential parameters while preserving baseline-offset rejection.
Generalized Extraction (n-DDC)
The formulation readily scales to resolve an arbitrary integer number of independent decaying modes. The projection sequence observes an -th order linear recurrence relation:
Applying the difference operator to essentially filter out low-frequency components yields a homogeneous relation on the differenced sequence:
Parameterizing distinct modes demands an observation vector of consecutive sliding windows (indices through ). The autoregressive coefficients are determined via the Hankel matrix formulation:
The mode roots are identified as the roots of the -th degree polynomial . Using these extracted roots, an Vandermonde system [33] is constructed from the first differenced sequential projections to jointly estimate the individual amplitude coefficients .
Equivalent Single-Mode Modeling for Multiple DDC
While the previous subsections established the formulations for , , and a general , the number of DDC components need not be known a priori because their combined effect is represented by a dynamically updated equivalent DDC component. Rather than implementing complex multi-order extraction routines, the combined effect of multiple decaying modes can be piecewise approximated by a single, dynamically updating equivalent decaying offset as demonstrated in [2] and [14]. To illustrate this concept, the remainder of this subsection demonstrates the equivalent modeling specifically for two DDCs.
Consider a discrete sliding-window sequence obtained from a fault current containing two distinct DDC components. Under the proposed framework, the complex projection expands into a linear combination of two geometric progressions:
where and are the discrete roots, and the complex constants are defined by .
By treating the equivalent parameters as locally invariant over three consecutive sliding steps , an equivalent dynamically updating DDC component
approximates the combined effect of the two DDC components of , such that:
To evaluate the instantaneous equivalent discrete root , the homogeneous forward difference operator is applied to (20) to eliminate low-frequency integration errors and isolate the decay rates:
By evaluating (21) at two consecutive sliding window steps and , and taking their ratio, the equivalent characteristic root is directly isolated:
Given that the pure decay projection integral renders purely imaginary, we define , where represents the real-valued magnitude coefficient for the -th mode. Substituting this into (22), the imaginary operators precisely cancel, yielding a purely real equivalent root:
The result of (23) confirms the existence of an equivalent characteristic root that provides a local piecewise approximation tracking the true composite trajectory of the two DDC components. The instantaneous equivalent time constant is subsequently calculated as:
A similar algebraic expansion can be performed to demonstrate that this approximation holds for an arbitrary number of DDC components. Therefore, by employing the foundational single DDC homogeneous formulation (), the proposed method can dynamically mitigate the combined effect of multiple DDC components on fundamental phasor estimation without requiring the true order a priori.
Discrete-Time Implementation for Digital Relays
Let be the sampling period, the discrete sample index, and the number of samples per fundamental cycle. The sliding window shifts by samples (), where is an integer sub-cycle shift factor, making the starting sample of the -th window . Modulating the sampled fault current by the -th harmonic carrier yields the sliding projection over one cycle:
For a single DDC component, its sampled form is
where is the single-sample decay factor. Applying the discrete form of Euler’s identity,
the DDC contribution to the projection becomes
The first term in (28) is a finite geometric series :
The second term represents the discrete oscillatory component. Being small for the selected modulation index, it is neglected. The sliding projection is thus approximated by the discrete geometric sequence:
where the projection coefficient and discrete root are respectively defined as:
Although evaluating (30) directly could yield , applying the forward difference provides numerical stability and filters practical projection errors. Within each active observation buffer of length (indexed locally by sub-windows ), evaluating the differences for the first two consecutive steps gives:
Taking their ratio eliminates and yields the discrete decay root , from which the time constant is recovered:
Once is determined, and are obtained algebraically as
Substituting and into (31) gives the estimated DDC amplitude :
Ideally, the dominant decay projection is imaginary, making real. Small complex residuals from finite sampling are suppressed by retaining only the real-valued estimate.
The DDC is reconstructed sample by sample using the estimated initial amplitude and time constant , and is subsequently subtracted from the raw fault current to produce the compensated current :
A conventional FCDFT is then applied to to estimate the fundamental phasor.
For multiple DDC components tracked continuously at an arbitrary sliding step , the equivalent-mode formulation retains this first-order structure. The dynamically varying equivalent root , time constant , and local window amplitude (referenced to the start of the -th window ) are updated as:
where and over the current window. Thus, the relay preserves the computational structure while dynamically tracking multiple decaying offsets without prior knowledge of the number of DDC components.
Performance Evaluation
To validate the proposed modulation-assisted sliding-window method, a series of evaluations is conducted. The algorithm is benchmarked against the standard FCDFT and the two closest methods to the proposed approach, Method [6] and Method [18], as both utilize high-frequency modulation algorithms for DDC mitigation.
The evaluation relies on three established quantitative criteria to measure the transient and steady-state performance of the phasor estimation.
The maximum overshoot (MO) evaluates the worst-case transient spike above the true steady-state magnitude ():
The rise time () measures the tracking speed during the initial fault transition. It represents the time required for the estimated magnitude to transition from 10% to 90% of the fault step change:
Finally, the settling time () identifies when the transient effectively ends. It is defined as the elapsed time following fault inception required for the estimation to permanently enter and remain within a tolerance band around the theoretical magnitude:
The modulation orders , , and are considered to investigate modulation-order sensitivity. The case is included for comparative purposes, although its practical applicability depends on the relay sampling rate and AAF bandwidth.
Evaluation with Single DDC
The DDC component depends on the fault inception angle and the network’s equivalent ratio. This ratio directly dictates the time constant [34]. To evaluate the robustness of the proposed methodology across a wide spectrum of fault conditions, fault currents containing single DDC are generated. The magnitude ratio is varied across , and the time constant is varied across cycles. The proposed method is evaluated at three different modulation orders , which yield nearly identical results for the investigated single-DDC cases.
As observed in Table I and Figure 2, the FCDFT is highly affected by the DDC component. In case of severe faults , the FCDFT generates an overshoot up to 25.32% and a settling time up to 27.86 ms. Method [18] exhibits larger overshoots, reaching 35.68%. While Method [6] reduces the settling time in some cases, it still permits up to a 25.32% overshoot.

Figure 2. Transient magnitude estimation of fault current for a single DDC.
| Test Condition | FCDFT | Method [6] | Method [18] | Proposed Method | |||||||||
| (cyc) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | |
| 0.1 | 0.5 | 2.50 | 10.94 | 13.80 | 0.60 | 9.64 | 12.76 | 1.24 | 9.38 | 12.24 | 0.04 | 11.20 | 14.32 |
| 0.1 | 3 | 0.90 | 11.20 | 14.06 | 0.70 | 9.90 | 13.02 | 0.90 | 9.38 | 12.76 | 0.00 | 11.20 | 14.32 |
| 0.1 | 5 | 0.58 | 11.20 | 14.06 | 0.58 | 9.90 | 13.28 | 0.58 | 9.38 | 14.32 | 0.00 | 11.20 | 14.32 |
| 0.5 | 0.5 | 12.58 | 10.42 | 20.05 | 12.31 | 8.85 | 16.67 | 12.58 | 7.03 | 16.41 | 0.20 | 11.20 | 14.32 |
| 0.5 | 3 | 4.91 | 9.90 | 12.24 | 4.84 | 8.85 | 10.94 | 8.31 | 7.03 | 13.02 | 0.01 | 11.20 | 14.32 |
| 0.5 | 5 | 3.52 | 9.90 | 12.24 | 3.46 | 8.85 | 10.94 | 7.81 | 7.03 | 12.50 | 0.00 | 11.20 | 14.32 |
| 1 | 0.5 | 25.32 | 4.69 | 27.86 | 25.32 | 4.69 | 16.67 | 35.68 | 4.69 | 15.36 | 0.40 | 11.20 | 14.06 |
| 1 | 3 | 18.71 | 3.65 | 27.34 | 18.71 | 3.65 | 16.67 | 34.35 | 3.65 | 15.36 | 0.02 | 11.20 | 14.32 |
| 1 | 5 | 20.62 | 3.65 | 17.97 | 20.62 | 3.65 | 16.67 | 33.85 | 3.65 | 15.36 | 0.01 | 11.20 | 14.32 |
Table 1. Comparative Analysis of Algorithms for Test Signals with a Single DDC Component .
| (%) | (%) | (%) | (%) | (%) | |
| 4 | 3.9757 | 1.7022 | 0.4251 | 0.1732 | 0.1299 |
| 8 | 1.9890 | 1.5864 | 0.3962 | 0.1614 | 0.1210 |
| 16 | 0.9947 | 1.5624 | 0.3902 | 0.1590 | 0.1192 |
Table 2. Modulation-Order Sensitivity for , Cycle
The proposed method provides the smallest MO among the evaluated methods and maintains a settling time of 14.06–14.32 ms for all investigated single-DDC cases. Regardless of the DDC parameters, the proposed method separates the isolated mode, limiting the MO to at most 0.40% (and 0.00%–0.02% for cycles). Its settling time remains tightly bounded within 14.32 ms, indicating that the extraction accuracy is largely insensitive to the DDC time constant.
Although the transient metrics in Table I are nearly identical across valid modulation orders, Table II shows the effect of using higher-resolution error metrics. Increasing reduces the retained oscillatory residual and the resulting estimation errors, without changing the sample-quantized rise and settling times.
Evaluation with two DDC
In transmission networks, secondary currents measured by CTs often contain multiple DDC components. These originate from both the primary line properties and the internal magnetic core of the CT during saturation [35]. A series of test signals including severe DDC conditions are evaluated. Table III details the performance across varied system time constants (), secondary CT time constants (), and DDC amplitude ratios ().
| Test Condition | FCDFT | Method [6] | Method [18] | Proposed Method | ||||||||||||
| (cyc) | (cyc) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | |||
| 1 | 2.5 | 1 | 10 | 4 | 35.20 | 3.13 | 35.94 | 35.20 | 3.13 | 16.67 | 57.58 | 3.13 | 15.36 | 0.38 | 10.94 | 13.28 |
| 8 | 35.20 | 3.13 | 35.94 | 35.20 | 3.13 | 16.67 | 57.58 | 3.13 | 15.36 | 0.37 | 10.94 | 13.28 | ||||
| 16 | 35.20 | 3.13 | 35.94 | 35.20 | 3.13 | 16.67 | 57.58 | 3.13 | 15.36 | 0.37 | 10.94 | 13.28 | ||||
| 0.5 | 2.5 | 1 | 10 | 4 | 10.85 | 4.95 | 19.79 | 10.84 | 4.95 | 16.67 | 19.01 | 4.95 | 15.36 | 0.09 | 11.46 | 13.80 |
| 8 | 10.85 | 4.95 | 19.79 | 10.84 | 4.95 | 16.67 | 19.01 | 4.95 | 15.36 | 0.09 | 11.46 | 13.80 | ||||
| 16 | 10.85 | 4.95 | 19.79 | 10.84 | 4.95 | 16.67 | 19.01 | 4.95 | 15.36 | 0.09 | 11.46 | 13.80 | ||||
| 1 | 5 | 0.5 | 5 | 4 | 26.62 | 3.65 | 27.86 | 26.61 | 3.65 | 16.67 | 45.58 | 3.65 | 15.36 | 0.62 | 11.20 | 13.80 |
| 8 | 26.62 | 3.65 | 27.86 | 26.61 | 3.65 | 16.67 | 45.58 | 3.65 | 15.36 | 0.33 | 11.20 | 13.80 | ||||
| 16 | 26.62 | 3.65 | 27.86 | 26.61 | 3.65 | 16.67 | 45.58 | 3.65 | 15.36 | 0.32 | 11.20 | 13.80 | ||||
| 1 | 2 | 2 | 20 | 4 | 47.91 | 2.86 | 35.94 | 47.91 | 2.86 | 16.67 | 63.14 | 2.86 | 15.36 | 0.59 | 10.68 | 13.02 |
| 8 | 47.91 | 2.86 | 35.94 | 47.91 | 2.86 | 16.67 | 63.14 | 2.86 | 15.36 | 0.59 | 10.68 | 13.02 | ||||
| 16 | 47.91 | 2.86 | 35.94 | 47.91 | 2.86 | 16.67 | 63.14 | 2.86 | 15.36 | 0.59 | 10.68 | 13.02 |
Table III. Comparative evaluations of algorithms for performance under double DC offsets.
Under double-DDC conditions, the FCDFT reaches large overshoots with protracted settling delays, while Method [18] reaches a 63.14% error peak under slow-decay conditions. However, the proposed method adapts to the multi-modal trajectory. As seen in Figure 3, it substantially reduces the compound transients, restricting the overshoot to negligible margins (, and for ) and settling within one fundamental cycle (13.02–13.80 ms).

Figure 3. Current magnitude estimation under two DDC.
Immunity to Measurement Noise
Measurement signals acquired from CT secondaries inherently contain noise, typically exhibiting a signal-to-noise ratio (SNR) between 20 to 40 dB. To evaluate the proposed method under noise condition, Gaussian noise along with a severe fault signal (, cycles), scaling the SNR down to 20 dB, is considered.
As detailed in Table IV, the proposed method provides selective filtering due to the full-cycle trigonometric carrier integral. At 20 dB noise, Method [18] reaches a 42.35% overshoot and an 83.85 ms settling time, while the FCDFT and Method [6] exhibit a 20.64% overshoot. Conversely, the proposed method bounds the maximum error tightly. At 20 dB noise, it restricts overshoot to 0.95% (at ) and at most 1.54% (and across all evaluated noise levels) while settling within one fundamental cycle (14.32–15.10 ms). These results demonstrate that the proposed method maintains accurate phasor estimation under severe noise conditions better than other studied methods.
| Condition | FCDFT | Method [6] | Method [18] | Proposed Method | |||||||||
| (dB) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | |
| 20 | 4 | 20.64 | 3.65 | 15.62 | 20.64 | 3.65 | 17.97 | 42.35 | 3.65 | 83.85 | 0.96 | 11.72 | 15.10 |
| 8 | 20.64 | 3.65 | 15.62 | 20.64 | 3.65 | 17.97 | 42.35 | 3.65 | 83.85 | 0.95 | 11.72 | 14.84 | |
| 16 | 20.64 | 3.65 | 15.62 | 20.64 | 3.65 | 17.97 | 42.35 | 3.65 | 83.85 | 1.54 | 11.72 | 14.58 | |
| 30 | 4 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 19.53 | 0.67 | 11.46 | 14.58 |
| 8 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 19.53 | 0.35 | 11.20 | 14.32 | |
| 16 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 16.41 | 20.63 | 3.65 | 19.53 | 2.47 | 11.46 | 14.32 | |
| 45 | 4 | 20.62 | 3.65 | 17.97 | 20.62 | 3.65 | 16.67 | 30.84 | 3.65 | 15.36 | 0.12 | 11.20 | 14.32 |
| 8 | 20.62 | 3.65 | 17.97 | 20.62 | 3.65 | 16.67 | 30.84 | 3.65 | 15.36 | 0.06 | 11.20 | 14.32 | |
| 16 | 20.62 | 3.65 | 17.97 | 20.62 | 3.65 | 16.67 | 30.84 | 3.65 | 15.36 | 0.52 | 11.20 | 14.32 |
Table IV. Comparative evaluations of algorithms for robustness under different noise levels.
Interference of Harmonics
During severe fault transients or transformer energization, the fault current contains both odd and even harmonics [6]. A composite test signal is generated, superimposing a primary DDC ( cycles) with harmonic combinations. The first case evaluates a heavily distorted signal containing 2nd (20% of the amplitude of the fundamental harmonic) and 4th (10%) harmonics. The second evaluates a spectrum containing 2nd (20%), 3rd (10%), and 4th (10%) harmonics simultaneously.
Under the proposed methodology, orthogonal sinusoids integrate to exactly zero over the fundamental window. By selecting an even carrier parameter (e.g., ), the algorithm rejects the investigated low-frequency harmonic components. As Table V shows, these prominent fault harmonics contribute strictly zero energy to the sliding projection integrals. Thus, the proposed method achieves a negligible 0.02% MO (matching the harmonic-free baseline) with a 14.58 ms settling time and isolates the DDC component regardless of harmonic severity.
| Harmonic Content | FCDFT | Method [6] | Method [18] | Proposed Method | |||||||||
| Distortion | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | MO (%) | (ms) | (ms) | |
| 2nd (20%) & 4th (10%) | 4 | 19.79 | 3.39 | 27.34 | 19.79 | 3.39 | 16.67 | 26.35 | 3.39 | 16.41 | 0.02 | 11.98 | 14.58 |
| 8 | 19.79 | 3.39 | 27.34 | 19.79 | 3.39 | 16.67 | 26.35 | 3.39 | 16.41 | 0.02 | 11.98 | 14.58 | |
| 16 | 19.79 | 3.39 | 27.34 | 19.79 | 3.39 | 16.67 | 26.35 | 3.39 | 16.41 | 0.02 | 11.98 | 14.58 | |
| 2nd (20%), 3rd (10%), 4th (10%) | 4 | 19.79 | 3.12 | 27.34 | 19.79 | 3.12 | 16.67 | 27.17 | 3.12 | 16.41 | 0.02 | 12.24 | 14.58 |
| 8 | 19.79 | 3.12 | 27.34 | 19.79 | 3.12 | 16.67 | 27.17 | 3.12 | 16.41 | 0.02 | 12.24 | 14.58 | |
| 16 | 19.79 | 3.12 | 27.34 | 19.79 | 3.12 | 16.67 | 27.17 | 3.12 | 16.41 | 0.02 | 12.24 | 14.58 |
Table V. Comparative performance under harmonic distortion (, cycles).
Impact of Sampling Rate
To guarantee compatibility with a wide spectrum of commercial digital relays, the proposed method must remain stable across different sampling resolutions. The algorithm is evaluated at standardized sampling rates extending from a sparse 960 Hz ( samples/cycle) up to 7680 Hz () [[36]].
The numerical results in Table VI confirm that the discrete characteristic root extraction remains well conditioned at all practical sampling frequencies, provided the modulation order obeys the Nyquist limit (). When the Nyquist condition is violated, the compensation becomes invalid (e.g., at 960 Hz). However, as long as the Nyquist condition is met (e.g., at 960 Hz, or at 3840 Hz), the proposed method maintains a negligible overshoot (0.01%–0.09%) and fast settling (14.06–14.58 ms) with accurate estimation.
| (Hz) | FCDFT MO (%) | (ms) | (ms) | Method [6] MO (%) | (ms) | (ms) | Method [18] MO (%) | (ms) | (ms) | Proposed Method MO (%) | (ms) | (ms) | |
| 960 () | 4 | 18.77 | 3.12 | 19.79 | 18.77 | 3.12 | 16.67 | 29.27 | 3.12 | 15.62 | 0.09 | 11.46 | 14.58 |
| 8 | – | – | – | – | – | – | – | – | – | – | – | – | |
| 16 | – | – | – | – | – | – | – | – | – | – | – | – | |
| 1920 () | 4 | 18.75 | 3.65 | 19.27 | 18.75 | 3.65 | 16.67 | 32.59 | 3.65 | 14.06 | 0.05 | 10.94 | 14.06 |
| 8 | 18.75 | 3.65 | 19.27 | 18.75 | 3.65 | 16.67 | 32.59 | 3.65 | 14.06 | 0.05 | 10.94 | 14.06 | |
| 16 | – | – | – | – | – | – | – | – | – | – | – | – | |
| 3840 () | 4 | 18.71 | 3.65 | 27.34 | 18.71 | 3.65 | 16.67 | 34.35 | 3.65 | 15.36 | 0.02 | 11.20 | 14.32 |
| 8 | 18.71 | 3.65 | 27.34 | 18.71 | 3.65 | 16.67 | 34.35 | 3.65 | 15.36 | 0.02 | 11.20 | 14.32 | |
| 16 | 18.71 | 3.65 | 27.34 | 18.71 | 3.65 | 16.67 | 34.35 | 3.65 | 15.36 | 0.02 | 11.20 | 14.32 | |
| 7680 () | 4 | 18.69 | 3.65 | 27.21 | 18.69 | 3.65 | 16.67 | 35.12 | 3.65 | 16.02 | 0.01 | 11.20 | 14.32 |
| 8 | 18.69 | 3.65 | 27.21 | 18.69 | 3.65 | 16.67 | 35.12 | 3.65 | 16.02 | 0.01 | 11.20 | 14.32 | |
| 16 | 18.69 | 3.65 | 27.21 | 18.69 | 3.65 | 16.67 | 35.12 | 3.65 | 16.02 | 0.01 | 11.20 | 14.32 |
Table VI. Impact of sampling rate on single DDC estimation (, cyc).
Sensitivity to Off-Nominal Frequency
During grid disturbances, such as multi-mode power swings or large load switching events, the system frequency can deviate from its nominal value. The frequency deviation is usually allowed in the range of 2-3% for a short period [[37]]. To evaluate this, the estimators were tested against static off-nominal frequencies (59.5 Hz and 60.5 Hz in Figure 4 and Figure 5) as well as a dynamic condition with continuous 3% linear frequency fluctuation (see Figure 6).

Figure 4. Impact of a static off-nominal frequency drop (59.5 Hz) on estimation accuracy.

Figure 5. Impact of a static off-nominal frequency rise (60.5 Hz) on estimation accuracy.

Figure 6. Impact of dynamic frequency variation (3% deviation) on estimation accuracy.
As shown in the Figure 4 to Figure 6, the proposed method is not sensitive to normal frequency variations in the grid and is able to reconstruct the fundamental harmonic even better than benchmarked methods.
Computational Complexity
The computational complexity and required number of samples of the evaluated phasor estimation algorithms are summarized in Table VII. All estimators employ an -point DFT for fundamental phasor estimation. While the standard FCDFT requires no DDC compensation, Method [6], Method [18], and the proposed method additionally perform high-frequency modulation-based processing to estimate or compensate the DDC component. Method [6] combines three successive outputs of one-cycle averaging filters applied to the original and modulated signals and therefore requires samples. Method [18] obtains the DDC information from -sample summations of the original and high-frequency-modulated signals and requires only one cycle ( samples).
| Algorithm | DDC extraction operation | DFT | Additional nonlinear ops. | Required number of samples | Complexity |
| FCDFT | – | -point | – | ||
| Method [6] | Two -sample averaging filters, three outputs | -point | Square root | ||
| Method [18] | Two -sample sums (original and modulated) | -point | – | ||
| Proposed Algorithm | Three modulated -sample sliding projections | -point | Log/exp |
Table VII. Comparison of the computation burden and number of samples required for phasor estimation algorithms.
-sample summations of the original and high-frequency-modulated signals and requires only one cycle ( samples). The proposed method uses three modulation-assisted sliding projections separated by samples, spanning an observation buffer of samples for parameter extraction (e.g., 80 samples or 1.25 fundamental cycles for and ). Once extracted, the DDC parameters are used to reconstruct and subtract the decaying offset across the active -sample DFT window, enabling the compensated phasor magnitude to enter the 5% settling band within one fundamental cycle ( ms). Furthermore, because parameter updates occur every samples and reuse previously computed sliding projections, only one new projection is evaluated per sub-cycle shift. Although the parameter extraction and time-domain reconstruction involve logarithmic and exponential operations (whereas Method [6] employs an algebraic square-root calculation), all three DDC-compensated algorithms retain linear computational complexity because their dominant operations consist of finite-length summations, modulation, and DFT processing. Thus, the additional computational burden of the proposed method remains compatible with real-time implementation in digital protective relays [38].
Simulation-Based Analysis
The dynamic performance of the phasor-estimation methods is coded and evaluated using a Python-based model of a 230 kV, 60 Hz, 100 km transmission line supplied from two equivalent sources. The source impedances are derived from short-circuit capacities of 8000 MVA and 8500 MVA with an ratio of 6 (see Figure 7 for more details). The transmission line is represented using its positive- and zero-sequence series parameters. kV and MVA are selected for per unit conversion. A solid single-phase-to-ground (A-G) fault with zero fault resistance is applied at ms.

Figure 7. Single-line diagram of the tested grid adapted from [27]
Two complementary evaluations are performed. First, a fault at 50% of the line is used to assess the transient magnitude-estimation performance of the different methods. Second, faults at 77% and 85% of the line are used to investigate their effect on the apparent-impedance trajectories of the distance relay. The relay is located at the Source-1 terminal, and its Zone 1 Mho characteristic is configured to cover 80% of the protected line. Thus, the 77% fault represents a boundary-adjacent internal fault, whereas the 85% fault represents a nearby external fault.
The 50% fault is used to quantify the transient tracking performance in terms of , rise time, and settling time. Table VIII summarizes the corresponding results for the FCDFT, Method [6], Method [18], and the proposed method with modulation orders , 8, and 16.
Table VIII. Transient performance for the A-G fault at 50% of the line.
As shown in Figure 8 and Table VIII, the conventional FCDFT is strongly affected by the DDC component and exhibits the largest transient deviation, with a of 16.041% and a settling time of 37.760 ms. Method [6] reduces the to 2.016% and provides the shortest rise time of 9.375 ms, with a settling time of 15.625 ms. Method [18] further limits the overshoot to 0.004% and provides the shortest settling time of 14.583 ms. The proposed method provides the smallest of 0.001% for all three modulation orders. Its settling time is 16.146 ms, which is slightly less than one fundamental cycle at 60 Hz. The identical results obtained for , 8, and 16 indicate that, for the considered fault waveform, the modulation order has negligible influence on the transient-magnitude estimation performance.

Figure 8. Estimated fundamental-current magnitude for the A-G fault at 50% of the protected line.
The influence of phasor-estimation errors on distance-relay operation is examined separately using faults close to the Zone 1 boundary. The 77% fault lies inside the 80% Zone 1 reach and is therefore used to assess dependability near the protection boundary. In contrast, the 85% fault lies immediately outside Zone 1 and is used to evaluate relay response against transient overreach.
For the 77% internal fault, the theoretical fault impedance lies close to the Zone 1 characteristic, making the relay decision particularly sensitive to transient phasor-estimation errors. As illustrated in Figure 9, the uncompensated FCDFT

Figure 9. Apparent-impedance trajectories for the internal A-G fault at 77% of the protected line, close to the Zone 1 Mho boundary.
exhibits a relatively large transient loop around the expected fault impedance before convergence. Method [6] also produces noticeable transient excursions, whereas Method [18] approaches the expected operating point with reduced local fluctuation. In comparison, the proposed method produces the most confined trajectories, with the results for , 8, and 16 closely overlapping around the expected 77% reach point. This behavior is particularly important for a boundary-adjacent internal fault, where excessive transient displacement can delay or destabilize the in-zone decision.
The 85% fault provides a complementary reliability test because its theoretical apparent impedance lies outside Zone 1. As shown in Figure 10, transient errors in the estimated current phasor displace the apparent-impedance trajectories during the initial fault period. The FCDFT exhibits a large transient loop extending toward and into the Zone 1 characteristic, indicating susceptibility to transient overreach. Method [6] and Method [18] reduce the local oscillation around the final operating point, although noticeable transient excursions remain. In contrast, the proposed method maintains tightly confined trajectories around the expected 85% reach point for all investigated modulation orders and remains outside the Zone 1 characteristic. This reduced transient displacement decreases the risk of transient overreach and improves Zone 1 discrimination for the investigated external fault.

Figure 10. Apparent-impedance trajectories for the external A-G fault at 85% of the protected line, outside the Zone 1 Mho boundary.
Conclusion
This paper proposed a modulation-assisted algorithm for fault current phasor estimation that effectively eliminates DDC offsets causing severe spectral leakage in conventional DFT. By modulating the signal with a high-order harmonic carrier, the algorithm isolates DDC offsets and utilizes a forward difference operator to dynamically track multiple DDC components as a single equivalent decaying signal. Extensive evaluations validated the algorithm’s accuracy, demonstrating a MO strictly bounded below 0.62% under complex double-DDC conditions and settling within one fundamental cycle (under 13.80 ms). Furthermore, the method maintained robust performance under severe 20 dB noise, limiting the MO to 2.47%. Ultimately, this approach guarantees fast, secure distance relay operations during complex Zone 1 grid faults.
Acknowledgments
The authors acknowledge the use of AI tools for English writing assistance, grammar checking, and mathematical derivation verification in the preparation of this manuscript.
References
References
- [1]Y. Guo, M. Kezunovic, and D. Chen, “Simplified algorithms for removal of the effect of exponentially decaying DC-offset on the fourier algorithm,” IEEE Transactions on Power Delivery, vol. 18, no. 3, pp. 711–717, 2003.DOI
- [2]S. H. Kang, D.-G. Lee, S.-R. Nam, P. A. Crossley, and Y.-C. Kang, “Fourier transform-based modified phasor estimation method immune to the effect of the DC offsets,” IEEE Transactions on Power Delivery, vol. 24, no. 3, pp. 1104–1111, 2009.
- [3]S. Mohammadi, A. Mahmoudi, S. Kahourzade, A. Yazdani, and G. M. Shafullah, “Decaying DC offset current mitigation in phasor estimation applications: A review,” Energies, vol. 15, no. 14, p. 5260, 2022.DOI
- [4]S. Mohammadi, N. Rezaei, A. Mahmoudi et al., “A novel analytical method for dc offset mitigation enhancing dft phasor estimation,” Electric Power Systems Research, vol. 209, p. 108036, 2022.DOI
- [5]S. Mohammadi and W. Su, “Symbolicphasor: Power system phasor estimation via deep symbolic regression,” arXiv preprint arXiv:2608.08552, 2026.
- [6]S. Afrandideh and F. Haghjoo, “A DFT-based phasor estimation method robust to primary and secondary decaying DC components,” Electric Power Systems Research, vol. 208, p. 107907, 2022.
- [7]S. Jamali and H. Zamani, “Removal of system and ct decaying dc component in fault signal using selective interval integration,” IEEE Transactions on Power Delivery, vol. 37, no. 2, pp. 905–912, 2022.DOI
- [8]G. Benmouyal, “Removal of DC-offset in current waveforms using digital mimetic filtering,” IEEE Transactions on Power Delivery, vol. 10, no. 2, pp. 621–630, 1995.DOI
- [9]C.-S. Yu, “A discrete fourier transform-based adaptive mimic phasor estimator for distance relaying applications,” IEEE Transactions on Power delivery, vol. 21, no. 4, pp. 1836–1846, 2006.DOI
- [10]Z. Jiang, S. Xiao, and P. Liu, “A modified empirical mode decomposition filtering-based adaptive phasor estimation algorithm for removal of exponentially decaying dc offset,” IEEE transactions on power delivery, vol. 29, no. 3, pp. 1326–1334, 2014.
- [11]M. Pazoki, “A new dc-offset removal method for distance-relaying application using intrinsic time-scale decomposition,” IEEE Transactions on Power Delivery, vol. 33, no. 2, pp. 971–980, 2017.DOI
- [12]J. Ren and M. Kezunovic, “Elimination of DC offset in accurate phasor estimation using recursive wavelet transform,” in 2009 IEEE Bucharest PowerTech, 2009, pp. 1–5.
- [13]B. Jafarpisheh, S. M. Madani, and F. Parvaresh, “Phasor estimation algorithm based on complex frequency filters for digital relaying,” IEEE Transactions on Instrumentation and Measurement, vol. 67, no. 3, pp. 582–592, 2018.DOI
- [14]K. M. Silva and F. A. Nascimento, “Modified dft-based phasor estimation algorithms for numerical relay applications,” IEEE transactions on power delivery, vol. 33, no. 3, pp. 1165–1173, 2017.
- [15]K. N. A. Al-Tallq, G. A. Putrus, and X. Li, “Online algorithm for removal of decaying DC-offset from phasor estimates,” Electric Power Systems Research, vol. 81, no. 12, pp. 2209–2218, 2011.
- [16]A. A. Abdoos, M. K. Ghataki, and J. Sadeh, “Accurate and fast DC offset removal method for digital relaying schemes,” IET Generation, Transmission & Distribution, vol. 10, no. 8, pp. 1877–1885, 2016.DOI
- [17]H. Hajizadeh, S.-A. Ahmadi, and M. Sanaye-Pasand, “An analytical fast decaying dc mitigation method for digital relaying applications,” IEEE Transactions on Power Delivery, vol. 36, no. 6, pp. 3529–3537, 2020.DOI
- [18]B. Jafarpisheh, S. M. Madani, and S. Mohammadh Shahrestani, “A new dft-based phasor estimation algorithm using high-frequency modulation,” IEEE Transactions on Power Delivery, vol. 32, no. 6, pp. 2416–2423, 2017.DOI
- [19]Z. Liu, Y. Zou, H. Zhao, J. Han, and Y. Liu, “An analytical phasor estimation method for transient currents considering double decaying dc offsets,” IET Generation, Transmission & Distribution, vol. 20, no. 1, p. e70309, 2026.
- [20]B. R. Kumar, A. Mohapatra, and S. Chakrabarti, “A novel sub-cycle-based method for estimation of decaying dc component and fundamental phasor,” IEEE Transactions on Instrumentation and Measurement, vol. 70, pp. 1–10, 2021.DOI
- [21]P. Jafarian and M. Sanaye-Pasand, “Weighted least error squares based variable window phasor estimator for distance relaying application,” IET generation, transmission & distribution, vol. 5, no. 3, pp. 298–306, 2011.DOI
- [22]J. K. H. Hwang, “Least error squares phasor estimation with identification of a decaying dc component,” IET Generation, Transmission & Distribution, vol. 12, no. 7, pp. 1486–1492, 2018.DOI
- [23]A. Mahari, M. Sanaye-Pasand, and S. M. Hashemi, “Adaptive phasor estimation algorithm to enhance numerical distance protection,” IET Generation, Transmission & Distribution, vol. 11, no. 5, pp. 1170–1178, 2017.DOI
- [24]M. Sachdev and M. Nagpal, “A recursive least error squares algorithm for power system relaying and measurement applications,” IEEE Transactions on Power Delivery, vol. 6, no. 3, pp. 1008–1015, 1991.DOI
- [25]M. Tajdinian, M. Z. Jahromi, K. Mohseni, and S. M. Kouhshari, “An analytical approach for removal of decaying dc component considering frequency deviation,” Electric Power Systems Research, vol. 130, pp. 208–219, 2016.
- [26]P. D. Acherlkar and B. K. Panigrahi, “Assessment of dc offset in fault current signal for accurate phasor estimation considering current transformer response,” IET Science, Measurement & Technology, vol. 13, no. 3, pp. 403–408, 2019.
- [27]A. Haji-Mohammadi and M. Sanaye-Pasand, “Novel method for decaying dc component removal based on solving the differential equations,” IEEE Transactions on Power Delivery, 2024.
- [28]M. Parpaei, H. Askariyan Abyaneh, and F. Razavi, “An innovative method to eliminate multiple exponentially decaying dc components based on the discrete fourier transform in the numerical distance relay,” IET Generation, Transmission & Distribution, vol. 16, no. 18, pp. 3731–3743, 2022.DOI
- [29]J. K. Hwang, C. K. Song, and M. G. Jeong, “Dft-based phasor estimation for removal of the effect of multiple dc components,” IEEE Transactions on Power Delivery, vol. 33, no. 6, pp. 2901–2909, 2018.DOI
- [30]M. Tajdinian, A. R. Seifi, and M. Allahbakhshi, “Half-cycle method for exponentially dc components elimination applicable in phasor estimation,” IET Science, Measurement & Technology, vol. 11, no. 8, pp. 1032–1042, 2017.DOI
- [31]M. Allahbakhshi, M. Tajdinian, A. R. Seifi, M. Zareian Jahromi, and A. Bagheri, “Hybrid approach for immunisation of dft-based phasor estimation method against decaying dc components,” IET Science, Measurement & Technology, vol. 13, no. 2, pp. 238–246, 2019.DOI
- [32]S. M. Brahma, P. L. De Leon, and R. G. Kavasseri, “Investigating the option of removing the antialiasing filter from digital relays,” IEEE Transactions on Power Delivery, vol. 24, no. 4, pp. 1864–1868, 2009.DOI
- [33]Å. Björck and V. Pereyra, “Solution of vandermonde systems of equations,” Mathematics of Computation, vol. 24, no. 112, pp. 893–903, 1970.DOI
- [34]H. Yu, Z. Jin, H. Zhang, and V. Terzija, “A phasor estimation algorithm robust to decaying dc component,” IEEE Transactions on Power Delivery, vol. 37, no. 2, pp. 860–870, 2021.DOI
- [35]H. Dashti, M. Sanaye-Pasand, and M. Davarpanah, “Fast and reliable ct saturation detection using a combined method,” IEEE Transactions on Power Delivery, vol. 24, no. 3, pp. 1037–1044, 2009.
- [36]S. A. Saleh, E. Okzop, A. Al-Drura, T. Hill, J. Meng, and M. E. Valdes, “On the assessment of sampling rate impacts on responses of digital protective relays,” IEEE Transactions on Industry Applications, vol. 59, no. 5, pp. 5420–5431, 2023.
- [37]C. Luo, H. G. Far, H. Banakar, P.-K. Keung, and B.-T. Ooi, “Estimation of wind penetration as limited by frequency deviation,” in 2006 IEEE Power Engineering Society General Meeting. IEEE, 2006, pp. 8–pp.
- [38]A. G. Phadke and J. S. Thorp, “Computer relaying for power systems,” New York, NY, 2009.