Showing papers on "Fundamental frequency published in 2021"

An Impedance Model-Based Multiparameter Identification Method of PMSM for Both Offline and Online Conditions

Abstract: Existing online motor parameter identification methods mostly depend on the fundamental frequency voltage equations, which lead to the unsatisfactory identification effect at low current and low speed operation. This article proposes a parameter identification method based on the high frequency (HF) equivalent impedance model of permanent magnet synchronous motor with HF signal injection at both the dq -axes. This method identifies the resistance and the dq -axis inductances offline and online, along with the flux linkage online. In order to improve the identification accuracy, the parameter sensitivity analysis-based algorithm is proposed to detect the resistance and the flux linkage. Meanwhile, the inverter nonlinearities and the HF influence on parameter identification are compensated effectively. In order not to affect the normal operation of the motor drive, the selection of the amplitude, and the frequency of the injected signal is investigated. The proposed method is validated on a 2.2-kW motor and confirmed by finite element analysis. The experimental results show the good identification effect in different operation conditions.

Frequency-dependent damped vibrations of multifunctional foam plates sandwiched and integrated by composite faces

Abstract: This paper investigates damped vibrational behavior of a lightweight sandwich plate subjected to a periodic load within a limited time. The lightweight sandwich structure includes a thick polymeric porous core with either functionally graded or uniformly distributions of voids which is sandwiched by two thin layers of laminate composites. To investigate the effect of void distribution properly, the same void volume fraction has been considered while different types of core have been analyzed. Using the first-order shear deformation theory of plates, the governing equations for the free and forced vibrations have been developed. By involving structural damping, these equations which are able to treat thin to moderately thick plates have been solved by developing a computationally cost-effective finite element approach. An extensive sensitivity analysis has been performed to examine the effects of fiber orientation in composite layers, void’s volume and dispersion in core, and geometrical dimensions on the vibrational behavior of such porous composite sandwich plates (PCSPs). The results show that the use of foam in PCSPs considerably reduces the amplitude of vibrations and improves the fundamental frequency. Furthermore, it was found that the use of [45, −45]2 composite layers offers PCSPs with the highest natural frequency and the lowest amplitude of vibrations.

An Improved Hybrid Prefiltered Open-Loop Algorithm for Three-Phase Grid Synchronization

Abstract: In this article, a robust three-phase grid synchronization technique has been proposed for rapid detection of fundamental frequency, phase, and amplitude. The widely accepted phase locked-loop (PLL) algorithms possess complex architectures and require tedious tuning process for attaining a good stability margin. In order to surpass the shortfalls of PLL algorithms, a computationally efficient, stable, and open-loop scheme has been reported in this article. A novel two consecutive samples based frequency estimator is developed for fast detection of the fundamental frequency. Moreover, an efficient hybrid prefiltering approach is implemented based on the demodulation of the grid voltage signal. Additionally, the combination of a delayed signal cancellation operator and a band-pass filter allowed rapid rejection of dc-offset and harmonics, respectively. In the event of a grid voltage imbalance, the instantaneous symmetrical component method is a rescuer for the rejection of the fundamental negative sequence component without any delay. Subsequently, overall transient response time of the scheme is observed to be improved. On the other hand, the fundamental positive sequence component facilitates the estimation of amplitude and phase angle information. Importantly, the dynamic performance of the proposed scheme has been experimentally validated in presence of various grid disturbances.

Topology optimization of vibrating structures with frequency band constraints

Abstract: Engineering structures usually operate in some specific frequency bands. An effective way to avoid resonance is to shift the structure’s natural frequencies out of these frequency bands. However, in the optimization procedure, which frequency orders will fall into these bands are not known a priori. This makes it difficult to use the existing frequency constraint formulations, which require prescribed orders. For solving this issue, a novel formulation of the frequency band constraint based on a modified Heaviside function is proposed in this paper. The new formulation is continuous and differentiable; thus, the sensitivity of the constraint function can be derived and used in a gradient-based optimization method. Topology optimization for maximizing the structural fundamental frequency while circumventing the natural frequencies located in the working frequency bands is studied. For eliminating the frequently happened numerical problems in the natural frequency topology optimization process, including mode switching, checkerboard phenomena, and gray elements, the “bound formulation” and “robust formulation” are applied. Three numerical examples, including 2D and 3D problems, are solved by the proposed method. Frequency band gaps of the optimized results are obtained by considering the frequency band constraints, which validates the effectiveness of the developed method.

Harmonic and Interharmonic Phasor Estimation Using Matrix Pencil Method for Phasor Measurement Units

Abstract: This paper proposes a novel approach to estimate the fundamental frequency, harmonic and interharmonic phasors in power networks. The proposed method is based on the Matrix Pencil Method (MPM) and Principle Component Analysis (PCA). It estimates parameters of the Prony model that represents the measured signal under noisy conditions. The frequency decomposition feature of the MPM allows concurrent estimation of all frequency components (including all harmonics) of the signal without an additional computational cost. The performance of the proposed method is evaluated with respect to all the requirements defined in the IEEE Std. C37.118 for both M- and P-class phasor measurement units (PMUs). It is shown that the method is compliant with the operating conditions of both PMU classes and at the same time, is able to provide accurate harmonic phasor estimation.

Enhancement of the performance of nonlinear vibration energy harvesters by exploiting secondary resonances in multi-frequency excitations

Abstract: This study is concerned with utilizing secondary resonances in order to harvest energy from low-frequency excitations. Nonlinearities give rise to secondary resonances, which can potentially activate large-amplitude responses when the excitation frequency is a fraction of the fundamental frequency of the system. Such resonances offer an untapped and unique opportunity for harvesting vibratory energy from excitation sources with low-frequency components. This issue has propelled the current study. Based on multi-frequency excitation, we develop a novel theoretical framework for a piezomagnetoelastic energy harvester to enhance its performance. The proposed scheme is implemented in both monostable and bistable piezomagnetoelastic under low-frequency excitations. It is shown throughout the paper that when the excitation frequencies are certain fractions of the system's fundamental frequency, the combination and simultaneous resonance activate large-amplitude responses. Another advantage of the scheme is that the energy could be harvested from low-frequency ambient vibrations, which is a considerable concern in this field of study. Different responses of the system, such as low-amplitude and high-amplitude limit-cycle oscillations and chaotic motions, are studied through perturbation theory and numerical techniques. Various numerical tools, including phase portrait, Poincare section, and Lyapunov exponent, are used to explore complex dynamical behavior of the system. The performance of the harvester is also compared in different regions. Numeral simulations clearly confirm that the proposed framework dramatically enhances the performance of the energy harvester.

A Third-Order MAF Based QT1-PLL That is Robust Against Harmonically Distorted Grid Voltage With Frequency Deviation

Abstract: The quasi-type-l phase-locked loop (QT1-PLL) is a grid synchronization technique that has become very popular in recent years thanks to its attractive performance such as easy implementation, fast dynamic response, and good accuracy in steady-state operation. However, it is still vulnerable to operation under harmonically distorted grid voltages with frequency drift. This paper proposes a novel QT1-PLL based synchronization algorithm that makes an appropriate combination of two filters’ types: an in-loop third-order moving average filter (MAF) with a reduced window width, and a simplified second-order fast delayed signal cancellation (FDSC) based prefiltering stage. The proposed PLL is named third-order MAF based QT1-PLL (TQT1-PLL). Though both TQT1-PLL's filters do not need any adaptive algorithm, it is able to reject non-triplen odd-harmonics and the fundamental frequency negative sequence (FFNS) even under grid frequency drift. Its correct operation is confirmed through numerical simulations and real-time implementation on a digital signal processor (DSP). Moreover, the obtained results confirm its ability to reduce the ripple in the estimated frequency and phase under distorted grid voltages and off-nominal frequency operation. Authors show also through an analytical development that the topology of the proposed TQT1-PLL can be extended to enable the rejection of the DC-offset.

Fundamental frequency optimization of variable stiffness composite skew plates

Abstract: In this study, natural frequencies and vibrational mode shapes of variable stiffness composite skewed plates are optimized applying a genetic algorithm. The variable stiffness behavior is obtained by altering the fiber angles continuously according to two selected curvilinear fiber path functions in the composite laminates. Fundamental frequency and related mode shapes of the plates are optimized for two different fiber path functions using the structural model obtained based on the virtual work principle. A three-layer composite skewed plate with four types of boundary conditions and different plate geometries is considered as case study in this research. Diverse sweptback angles as well as different aspect ratios are considered as various plate geometries. The present study aims to calculate the best fiber path with maximized fundamental frequency or in-plane strengths for a composite skewed plate. The generalized differential quadrature method of solution is employed to solve the governing equations of motion. Moreover, the linear kinematic strain assumptions are used, and the first-order shear deformation theory is employed to generalize the formulation for the case of moderately thick plates including transverse shear effects. Numerical results demonstrate the effect of the fiber angles, boundary conditions, and diverse geometries on the natural frequencies of the composite plate. The optimal fiber angles of each layer are presented for the above cases in free vibration analysis. It is verified that the application of optimized curvilinear fibers instead of the traditional straight fibers introduces a higher degree of flexibility, which can be used to adjust frequencies and mode shapes.

Stabilization of a Harmonic Mode-Locking by Shifting the Carrier Frequency

Abstract: Harmonically mode-locked soliton fiber lasers have been intensively investigated in recent years due to their wide range of applications. Drawback of these lasers is relatively large timing jitter which is significantly higher than the timing jitter in lasers operating at fundamental frequency. We report the stabilizing frequency shift effect in harmonic mode-locking ring soliton fiber laser that is studied theoretically and numerically. It is known that a harmonic mode-locking regime in a fiber laser occurs due to interpulse repulsion, which leads to a uniform distribution of pulses in the cavity. We demonstrate that the frequency shift contributes to an increase in the hardness of interpulse interactions and can led to stabilization of the periodic arrangement of pulses. The experiment carried out confirms the theoretical predictions and the results of numerical simulation.

Impedance Characteristic Analysis and Reshaping Method of DFIG System Based on DPC Without PLL

Abstract: The phase locked loop (PLL) will introduce the negative resistance near the fundamental frequency, which deteriorates the stability of doubly fed induction generators (DFIG) system under the inductive weak grid. This article develops the impedance model of the DFIG system based on direct power control without PLL, and analyzes the frequency coupling characteristic of established impedance model. The phase margin around the fundamental frequency is enhanced after cancelling the PLL. However, there will be a high-frequency resonance issue related to the frequency coupling characteristic. Based on the simplified high-frequency impedance model and the Bode diagram of equivalent single-input single-output impedance, it can be found that the DFIG system is more prone to have stability issues when the degree of frequency coupling is enhanced. An impedance reshaping control strategy is proposed to reduce the degree of frequency coupling at high frequency and suppress the high-frequency resonance. The experimental results validate the availability of the proposed reshaping control strategy.

Improved Fundamental Frequency Estimator for Three-Phase Application

Abstract: This letter reports a digital signal processing based frequency estimation technique that relies on the storage of fundamental three-phase grid voltage samples. Unlike phase-locked loop algorithms, the proposed algorithm does not require the phase error information when estimating the grid frequency. Moreover, only four consecutive samples of the fundamental orthogonal components are required in order to estimate the deviation in the grid frequency. The proposed scheme is facilitated with an inherent nonadaptive comb filtering behavior in order to reject the dc offset and harmonics while attaining a good immunity to voltage sag and the fundamental negative sequence component. Additionally, the off-nominal harmonic rejection capability of the proposed frequency estimator is enhanced by using a postfiltering method. The experimental validations reported in this study demonstrate that the proposed frequency estimator is suitable for three-phase grid applications.

Design and modeling of an improved bridge-type compliant mechanism with its application for hydraulic piezo-valves

Abstract: Bridge-type compliant mechanisms have been frequently utilized as the micro displacement amplifier for a variety of precision manipulation applications. However, the inertial movement of internal actuators (e.g. piezo-stacks) limits the system’s dynamic bandwidth in some traditional design. This paper presents an improved bridge-type compliant mechanism with double output ports that can generate homodromous bi-motions actuated by only one group of piezo-stacks. The inertial motion of piezo-stacks is avoided and the dynamic bandwidth is enhanced. The two-port dynamic stiffness model is established to straightforwardly capture its kinetostatic and dynamic characteristics from the perspective of input and output ports. The displacement amplification ratio, input stiffness, fundamental frequency and dynamic response spectrum of the improved bridge-type compliant mechanism are curved against key geometric parameters, then the optimal performance can be confirmed. Of particular interest for the mechanism application is applying it to develop a new type of piezoelectric two-stage flow control valve with relatively fast dynamic response and large flow rate. The inner leakage and oil contamination is effectively overcome in contrast to traditional nozzle-flapper servovalves. The presented piezoelectric flow control valve is fabricated and experimentally measured with the step response time of 8.5 ms, frequency bandwidth of 120 Hz, and stroke of ±0.8 mm (corresponding to the flow rate of 180 L/min at the supply pressure of 210 bar).

Frequency-Adaptive Virtual Variable Sampling-Based Selective Harmonic Repetitive Control of Power Inverters

Abstract: Due to the $n$ -pulse commutation, the power harmonic distortions caused by power inverters usually concentrate on particular $(nk\pm m)$ -order harmonic frequencies. The conventional repetitive control (RC) uses an identical gain to equally compensate for the distortions at all harmonic frequencies. This leads to slow dynamics as it fails to optimize the convergence rate of the RC at dominant harmonic frequencies. The selective harmonic RC (SHRC) can efficiently suppress dominant power harmonic distortions. However, the ratio of the sampling rate of the SHRC to the fundamental frequency must be an integer. This severely limits the use and lowers the performance of the SHRC with a fixed sampling rate in the presence of frequency variations of interested harmonics. To address this problem, this article proposes a frequency-adaptive virtual variable sampling-based SHRC (FA-VVS-SHRC) scheme that is immune to the issue of the fractional ratio of the sampling rate to the interested harmonics. The proposed FA-VVS-SHRC scheme can provide flexible fractional phase-led compensation to achieve accurate power harmonic control in the presence of frequency variations. Moreover, it is a low-cost and easy-to-implement solution that does not require hardware modifications. Experiments on a three-phase power inverter and comparison studies are presented to verify its effectiveness.

Resonance in Atomic-Scale Sliding Friction

Abstract: Friction represents a major energy dissipation mode, yet the atomistic mechanism of how friction converts mechanical motion into heat remains elusive. It has been suggested that excess phonons are mainly excited at the washboard frequency, the fundamental frequency at which relative motion excites the interface atoms, and the subsequent thermalization of these nonequilibrium phonons completes the energy dissipation process. Through combined atomic force microscopy measurements and atomistic modeling, here we show that the nonlinear interactions between a sliding tip and the substrate can generate excess phonons at not only the washboard frequency but also its harmonics. These nonequilibrium phonons can induce resonant vibration of the tip and lead to multiple peaks in the friction force as the tip sliding velocity ramps up. These observations disclose previously unrecognized energy dissipation channels associated with tip vibration and provide insights into engineering friction force through adjusting the resonant frequency of the tip-substrate system.

Readout Distance Enhancement of Battery-Free Harmonic Transponder

Abstract: This work proposes to enlarge the maximum readout distance of battery-free harmonic transponders through a careful Schottky diode selection and a consideration of low-temperature operations. Diode SMS7621 is identified to deliver lower conversion loss (CL) of harmonic transponders based on a diode selection guide. Moreover, an analytical method for studying temperature effects on the transponder CL performance is introduced and derived with satisfactory accuracy. A diplexer inserted in the harmonic transponder design can help reduce the antenna number. To mitigate matching difficulties at both fundamental and second-harmonic frequencies, a third-order diplexer is developed to enhance its bandwidth performance. Experimental verification has shown that low-temperature operations can effectively reduce the CL of the SMS7630-based harmonic transponder, thus increasing its readout distance. Specifically, its maximum readout distance at −40 °C has increased by more than 10%, from 6.4 to 7.1 m, compared with that at +40 °C. A complete harmonic transponder based on SMS7621, with dimensions of 85 mm $\times45$ mm has reached a maximum readout distance of 8 m when the fundamental frequency is 3.5 GHz.

Mamyshev Oscillator With a Widely Tunable Repetition Rate

Abstract: Controlling an ultrafast laser oscillator repetition rate is crucial for the emerging laser burst micromachining. The efficiency of a material surface treatment process as well as the quality of live tissue ablation depends, to a considerable extent, on the number of laser pulses in a burst window. In the following article, we present an all-PM fiber harmonically mode-locked Mamyshev ring oscillator capable of producing laser pulses with a tunable period. The oscillator can be operated at any frequency, being a multiple of the fundamental repetition rate up to 305 MHz at 19th harmonic. We found that the operation frequency is limited by the available pump power only. What is worth noting is that the energy of the emitted pulses is independent of the operating frequency. The essential laser parameters are excellent: an amplitude noise of 0.1% and fundamental frequency suppression levels of 70 dB at every repetition rate. Additionally, the mechanism of harmonic mode-locking is discussed in detail and is found to be caused by gain depletion and recovery effect.

Precessional spin-torque dynamics in biaxial antiferromagnets

Abstract: The N\'eel order of an antiferromagnet subject to a spin torque can undergo precession in a circular orbit about any chosen axis. To orient and stabilize the motion against the effects of magnetic anisotropy, the spin polarization should have components in-plane and normal to the plane of the orbit, where the latter must exceed a threshold. For biaxial antiferromagnets, the precessional motion is described by the equation for a damped-driven pendulum, which has hysteresis as a function of the spin current with a critical value where the period diverges. The fundamental frequency of the motion varies inversely with the damping and as ${({x}^{p}\ensuremath{-}1)}^{1/p}$, with the drive-to-criticality ratio $x$ and the parameter $pg2$. An approximate closed-form result for the threshold spin current is presented, which depends on the minimum cutoff frequency the orbit can support. Precession about the hard axis has zero cutoff frequency and the lowest threshold, while the easy axis has the highest cutoff. A device setup is proposed for electrical control and detection of the dynamics, which is promising to demonstrate a tunable terahertz nano-oscillator.

DeepF0: End-To-End Fundamental Frequency Estimation for Music and Speech Signals

Abstract: We propose a novel pitch estimation technique called DeepF0, which leverages the available annotated data to directly learns from the raw audio in a data-driven manner. f 0 estimation is important in various speech processing and music information retrieval applications. Existing deep learning models for pitch estimations have relatively limited learning capabilities due to their shallow receptive field. The proposed model addresses this issue by extending the receptive field of a network by introducing the dilated convolutional blocks into the network. The dilation factor increases the network receptive field exponentially without increasing the parameters of the model exponentially. To make the training process more efficient and faster, DeepF0 is augmented with residual blocks with residual connections. Our empirical evaluation demonstrates that the proposed model outperforms the baselines in terms of raw pitch accuracy and raw chroma accuracy even using 77.4% fewer network parameters. We also show that our model can capture reasonably well pitch estimation even under the various levels of accompaniment noise.

Vibration Investigation of Spoke-Type PM Machine With Asymmetric Rotor Considering Modulation Effect of Stator Teeth

Abstract: In this article, we analyze the vibration of the spoke-type permanent-magnet (PM) machine with an asymmetric rotor considering the modulation effect of the stator teeth. First, the semianalytical expression of the radial pressure is derived to quickly estimate the radial pressure characteristic. The equivalent spatial order of the radial pressure is determined by considering the modulation effect. Second, a structural finite-element (FE) model is set up. The modal test validates the accurateness of the structural FE model and modal analysis. Third, the vibration is predicted by an FE method and verified by the experimental results. The results show that the radial pressure harmonics become richer, and the fundamental frequency and lowest nonzero order of the radial pressure are both reduced due to the asymmetric magnetic barrier. Furthermore, the zero-order radial pressure is no longer the dominant vibration source in this integer-slot PM machine with an asymmetric rotor. Some high-order radial pressures that can be modulated as equivalent low-order harmonics excite the low-order and large vibration. In terms of vibration and noise, the equivalent lowest nonzero-order harmonic should be fully considered in the performance optimization for the asymmetric rotor design.

Teager Energy Operator for Fast Estimation of Three-Phase Grid Frequency

Abstract: This article proposes a fundamental frequency estimation technique for three-phase voltage systems under unbalanced and harmonic conditions. It is based on the Clarke transformation, recursive discrete Fourier transform, and Teager energy operator. The proposed technique is computationally efficient and relatively simple to implement, thus suitable for low-cost applications. It does not require real-time evaluation of the computationally demanding inverse trigonometric functions. The proposed method does not suffer from the accumulation error and is also not affected when the voltage signal is sampled at zero or close to zero. It is immune to various voltage harmonics and can produce quick transient response with a time of less than 75% of one fundamental period. When compared with the competing techniques based on phase-locked loops and three consecutive samples, the proposed method can produce faster response under dynamic conditions. The advantages of the proposed technique are confirmed by simulated and experimental results in the laboratory.

A Novel Frequency Estimation Method based on Complex Bandpass Filters for P-class PMUs with Short Reporting Latency

Abstract: The measurement data of P-class phasor measurement units (PMUs) is used for protection and control applications. The reporting latency is critical for these applications. However, the data window of fundamental frequency estimation is commonly longer than that of phasor estimation, which introduces additional latency. In this paper, a novel frequency estimation method that uses sampling values rather than phase angles is proposed for P-class PMUs to reduce their reporting latency. This method uses two complex bandpass filters with different gains to extract the two positive fundamental components of the power signals. A one-to-one correspondence between the frequency and the ratio of the two positive fundamental components is established. By polynomial fitting this relationship, the frequency under static and dynamic conditions can be accurately estimated. In addition, to ensure the accuracy, the two applied complex bandpass filters are designed according to standard requirements, which can effectively suppress the negative fundamental component and harmonics. The simulation and experimental test results show that the proposed method achieves a frequency measurement accuracy that is 10 times higher than the standard requirements with a low computational burden and a data window of two cycles, indicating that the proposed method is suitable for P-class PMUs.

Harmonic dissipative soliton resonance in an Yb doped fiber laser with 110 W all fiber polarization-maintaining amplifier

Abstract: The formation of harmonic dissipative soliton resonance (DSR) is experimentally investigated in an Yb-doped fiber laser operating in large normal dispersion regime. The DSR is observed at a fundamental frequency of 221 KHz, which delivered square-shaped with of 18.7–98.5 ns and single pulse energy of 0.25 μJ at 1056 nm. By adjusting the PCs, the DSR square pulses could be split into their harmonic mode-locked (HML) counterparts. These experimental results provide a new viewpoint for the further understanding of the mechanism and characteristics of HML square pulses. Based on a two-stage all-fiber PM master oscillator power amplifier (MOPA) system seeded by this oscillator, the pulses are amplified up to 110 W output power with the polarization extinction ratio (PER) of 18.7 dB at different frequencies ranging from 221KHz to 663 KHz and a record pulse energy of 0.51 mJ at the frequency of 221 KHz is achieved.

Spectral Analysis and Sideband Harmonic Cancellation of Six-Step Operation With Low Carrier–Fundamental Frequency Ratio for High-Speed Brushless DC Drives

Abstract: High-speed brushless dc (BLDC) drives usually operate with low carrier–fundamental frequency ratio, which causes abundant sideband harmonic current components. Some sideband harmonics will be induced in low-frequency range, leading to low-frequency current oscillations and low-frequency torque ripples. To minimize these undesired sideband harmonics, a novel pulsewidth modulation commutation pattern, i.e., carrier-synchronized commutation (CSC), is proposed in this article. For comparison, two conventional commutation patterns, including regular-sampled commutation (RSC), and natural-sampled commutation (NSC), are analyzed. To reveal the characteristics of sideband harmonics caused by RSC and NSC patterns, an extended geometric wall model is introduced, which can be used for the Fourier decomposition of complex vectors. Theoretical, simulation, and experimental results demonstrate the conventional RSC and NSC patterns will produce additional sideband harmonic currents and torque ripples, while the proposed CSC pattern can considerably suppress these adverse influences and meanwhile presents high performance and robustness.

Optomechanical Synchronization across Multi-Octaves Frequency Spans

Abstract: Experimental exploration of synchronization in scalable oscillator micro systems has unfolded a deeper understanding of networks, collective phenomena, and signal processing. Cavity optomechanical devices have played an important role in this scenario, with the perspective of bridging optical and radio frequencies through nonlinear classical and quantum synchronization concepts. In its simplest form, synchronization occurs when an oscillator is entrained by a signal with frequency nearby the oscillator's tone, and becomes increasingly challenging as their frequency detuning increases. Here, we experimentally demonstrate entrainment of a silicon-nitride optomechanical oscillator driven up to the fourth harmonic of its 32 MHz fundamental frequency. Exploring this effect, we also experimentally demonstrate for the first time a purely optomechanical RF frequency divider, where we performed frequency division up to a 4:1 ratio, i.e., from 128 MHz to 32 MHz. Further developments could harness these effects towards frequency synthesizers, phase-sensitive amplification and nonlinear sensing.

Quasi-Periodic Parallel WaveGAN: A Non-Autoregressive Raw Waveform Generative Model With Pitch-Dependent Dilated Convolution Neural Network

Abstract: In this paper, we propose a quasi-periodic parallel WaveGAN (QPPWG) waveform generative model, which applies a quasi-periodic (QP) structure to a parallel WaveGAN (PWG) model using pitch-dependent dilated convolution networks (PDCNNs). PWG is a small-footprint GAN-based raw waveform generative model, whose generation time is much faster than real time because of its compact model and non-autoregressive (non-AR) and non-causal mechanisms. Although PWG achieves high-fidelity speech generation, the generic and simple network architecture lacks pitch controllability for an unseen auxiliary fundamental frequency ( $F_{0}$ ) feature such as a scaled $F_{0}$ . To improve the pitch controllability and speech modeling capability, we apply a QP structure with PDCNNs to PWG, which introduces pitch information to the network by dynamically changing the network architecture corresponding to the auxiliary $F_{0}$ feature. Both objective and subjective experimental results show that QPPWG outperforms PWG when the auxiliary $F_{0}$ feature is scaled. Moreover, analyses of the intermediate outputs of QPPWG also show better tractability and interpretability of QPPWG, which respectively models spectral and excitation-like signals using the cascaded fixed and adaptive blocks of the QP structure.