Abstract
The convergence rate and the continuous tracking precision are two main problems of the existing adaptive notch filter (ANF) for frequency tracking. To solve the problems, the frequency is detected by interpolation FFT at first, which aims to overcome the convergence rate of the ANF. Then, referring to the idea of negative feedback, an evaluation factor is designed to monitor the ANF parameters and realize continuously high frequency tracking accuracy. According to the principle, a novel adaptive frequency estimation algorithm based on interpolation FFT and improved ANF is put forward. Its basic idea, specific measures and implementation steps are described in detail. The proposed algorithm obtains a fast estimation of the signal frequency, higher accuracy and better universality qualities. Simulation results verified the superiority and validity of the proposed algorithm when compared with original algorithms.
Introduction
Frequency estimation is a fundamental task in signal processing, particularly in power systems and fault diagnosis. Adaptive Notch Filters (ANFs) are widely used due to their ability to tune to a signal's frequency dynamically. However, traditional ANFs suffer from a critical trade-off: a wide notch bandwidth allows for faster convergence but reduces tracking precision due to noise sensitivity, whereas a narrow bandwidth improves precision but slows down initial locking and risks losing track during sudden frequency shifts.
Proposed Methodology
The proposed algorithm addresses these limitations through a two-phase hybrid approach:
Phase 1: Rapid Initialization via Interpolation FFT
To eliminate the slow convergence phase typical of ANFs, the system uses an Interpolation FFT for a "warm start." Standard FFTs are limited by frequency resolution (the "picket-fence" effect). By analyzing the two highest amplitude peaks in the spectrum and applying a correction factor ($\delta$), the algorithm achieves a high-accuracy initial frequency estimate. This estimate initializes the ANF, placing it immediately near the true frequency.
Phase 2: Long-term Tracking with an Improved ANF
Once initialized, the ANF maintains tracking. To ensure robustness, an evaluation factor $h(n)$ based on the Least Mean Squares (LMS) algorithm is introduced. This factor acts as a "health monitor," calculating the correlation between the input and estimated signals. If $h(n)$ falls below a threshold $T_h$, indicating a loss of lock (e.g., due to a frequency jump), the system automatically adjusts the notch bandwidth to recapture the signal.
The logic flow of this integrated system is depicted in the flowchart below, highlighting the decision-making process involving the evaluation factor and parameter adjustment.
Results and Discussion
The algorithm's performance was evaluated using simulated sinusoidal signals across three different ANF architectures: Lattice ANF (L-ANF), Simplified Lattice ANF (SL-ANF), and SMM-ANF.
Quantitative results demonstrate that the proposed enhancement is "universal," significantly improving the accuracy of all tested architectures. The Mean Square Error (MSE) was reduced by approximately one order of magnitude. For instance, the L-ANF's MSE dropped from $3.72 \times 10^{-7}$ to $1.66 \times 10^{-8}$.
| ANF Type | Original MSE | Proposed MSE |
|---|---|---|
| L-ANF | $3.72 \times 10^{-7}$ | $1.66 \times 10^{-8}$ |
| SL-ANF | $3.57 \times 10^{-7}$ | $1.65 \times 10^{-8}$ |
| SMM-ANF | $3.19 \times 10^{-7}$ | $1.64 \times 10^{-8}$ |
Conclusion
By combining the rapid initial locking of Interpolation FFT with the robust, feedback-monitored tracking of an improved ANF, this algorithm effectively breaks the traditional convergence-precision trade-off. While future work is needed to address multi-tonal signals and optimize computational overhead for real-time use, the current results show a significant leap in frequency estimation performance and reliability.
Summary updated