Search

In this paper, we investigated a speech augmentation based unsupervised learning approach for keyword spotting (KWS) task. KWS is a useful speech application, yet also heavily depends on the labeled data. We designed a CNN-Attention architecture to conduct the KWS task. CNN layers focus on the local acoustic features, and attention layers model the long-time dependency. To improve the robustness of KWS model, we also proposed an unsupervised learning method. The unsupervised loss is based on the similarity between the original and augmented speech features, as well as the audio reconstructing information. Two speech augmentation methods are explored in the unsupervised learning: speed and intensity. The experiments on Google Speech Commands V2 Dataset demonstrated that our CNN-Attention model has competitive results. Moreover, the augmentation based unsupervised learning could further improve the classification accuracy of KWS task. In our experiments, with augmentation based unsupervised learning, our KWS model achieves better performance than other unsupervised methods, such as CPC, APC, and MPC., Comment: accepted by WCCI 2022

Aiming at the missile Defense system of systems process, based on complex network, firstly, the Application of missile Defense system-of-systems is abstracted for Topologically, and the supernetwork Modeling of the application of missile Defense system-of-systems is built, which is composed of shard subnet, information subnet, Engagement subnet configures and sensor subnet configures. Secondly, the static interaction mechanism of models is analyzed deeply, and which nodes interaction is described mathematically. Thirdly, The dynamic interaction mechanism of models is analyzed based on the timing sequence of operation, Extracted and the characteristic parameters which reflect the network evolution characteristics between subnet and internal subnet are extracted. An example is illustrated at last. The simulation results show The overall characteristic value of the network changes in phases, such as network Density, while, the information flow changes regularly with phases, As well as the role and status of each node, which is a helpful exploration for the following optimization problems.

This paper introduces the construction and application of the electro-thermal co-simulation platform. Based on the finite element method, we first established a device-level simulation model of IGBT and diode with dynamic characteristics; at the same time, we extracted the corresponding network model through CFD simulation calculation of the transient operating conditions of the IGBT device; then, using 1200V 600A IGBT as an example to build a complete electro-thermal co-simulation platform; finally, the simulation data of the electro-thermal co-simulation platform and the test data of the developed 1200V 600A device are tested and verified under the same working conditions.

This paper analyzes in detail the failure of MCL1204 air compressor, an important equipment in a refinery. Based on the mathematical modeling of the gas leakage of the bulkhead, the influence of the factor on the leakage is analyzed theoretically. With the help of laser alignment, the key parts of the unit were measured, the main cause of the fault was found and the fault was successfully repaired.

The island inverter under dual-sequence control method controls the negative sequence voltage by introducing a negative sequence branch to ensure the symmetry of the output voltage. In this paper, the small-signal wide-band sequence impedance model of the island inverter under dual-sequence control is established by harmonic linearization method. Then, the wide-band frequency-sweeping simulation of this control method is carried out in MATLAB. The sweep result is consistent with the established sequence impedance model, which proves the correctness of the established sequence impedance. The analysis of the sequence impedance model shows: the fundamental cause the double-sequence control method can suppress the negative sequence voltage is that the dual-sequence control greatly reduces the negative sequence impedance of the island inverter at the fundamental frequency (50Hz), thereby suppressing the negative sequence voltage drop generated by the negative sequence current of the load on the negative sequence impedance of the inverter. This impedance model can be applied later to the stability analysis of the AC microgrid.

With the rapid development of electric vehicles (EV s), the lack of the charging infrastructure has become the bottleneck of the development of EVs. Aiming at the problem of the low charging efficiency and power capacity of AC network, the DC distribution system has been introduced for a settlement. However, a large amount of EVs charging would inevitably result in the DC neutral point voltage imbalance, especially using bipolar structure. In this paper, the main topology structures and neutral grounding modes of DC distribution network is researched and analyzed. Based on the structural feature of DC bipolar distribution network, the influencing factors of DC neutral point voltage, such as network parameter and load distribution, is studied. Then, aiming at the problems of the voltage unbalance, a new coordinated control method is designed to suppress the voltage fluctuation. Finally, a study case of DC distribution network is simulated and analyzed to verify the proposed method, and the results show that the unbalanced voltage could be depressed effectively.

This paper proposes a method of correcting the nonlinear parasitic capacitor of the input pair of comparator in successive approximations analog-to-digital converters (SAR ADCs). The correction method is proposed for the conventional binary-weighted capacitor array topology used in most of high resolution and high speed SAR ADCs. The effects of dynamic capacitor mismatch are first analyzed and then two feasible correction schemes are proposed to mitigate the impact of the nonlinear parasitic capacitor of the comparator. To verify the effectiveness of the proposed method, we designed a SAR ADC in a CMOS 40 nm process and characterized the design by intensive post-simulations. With the proposed correction schemes, the SFDR and SNDR of the SAR ADC increase about 7 and 4 dB, respectively, the differential nonlinearity and integral nonlinearity after calibration are improved from 1.00 and 3.81 to 0.67 LSB/0.57 LSB and 1.46 LSB/0.77 LSB, respectively.