Search

Automatic and accurate onset time picking of the acoustic emission (AE) signals is of great significance for the location of acoustic events. However, the onset times are difficult to be picked accurately due to the presence of noise in signal records. To overcome this problem, an improved method including three consecutive processes is proposed in this study. First, the windowing Lempel-Ziv (WLZ) complexity method is applied to determine an approximate onset time. Then, the time series selected from the start time to the vicinity of the approximate onset time is coarsened at different scales with the application of multi-scale (MS) theory. Finally, Akaike information criterion (AIC) method is utilized to pick up the accurate onset time. Feasibility and performance of the proposed improved method are tested and compared with the other automatic onset time picking methods, such as AIC, short and long time average ratio (STA/LTA), modified energy ratio (MER), AIC with low pass filter and STA/LTA with low pass filter, using the same broken load experiment and 100 rock fracture signals with various signal-to-noise ratios (SNRs). The results show that the accuracy of automatic picks provided by the improved method are far beyond the other methods mentioned. Meanwhile, both the accuracy stability and SNR adaptability of the improved method for rock fracture signals are better than other methods, which proves that the improved method has the excellent performance to pick up onset times of AE signals.

Anomalous trajectory detection plays a significant role in fraud detection and adverse events monitoring for ride-hailing services. The spatial and temporal dynamics of road networks and the sparsity of trajectories make anomalous trajectory detection a challenging task. Most existing methods are based on density and isolation approaches, which ignore geographical information. Motivated by these challenges and shortcomings, we propose a novel method, which considers geospatial constraints of the trajectories and avoids sparsity issues. In our method, the geographical information and topological constraints of trajectories are embedded into structured vector space. Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN) are used to model common trajectory features. Our method could identify anomalous trajectories and determine which parts are responsible for anomalies by using these features. Experiments on two real-world datasets have been conducted, and results demonstrate the effectiveness and feasibility of the proposed method.

In this paper, an improved automatic method for picking arrival times of acoustic emission (AE) signals, has been proposed to overcome the interference of spike noise and channel crosstalk. First, median filter is applied to eliminate spike noise presented in signals. Second, the preliminary arrival time is obtained by the AIC picker. Then a judgment whether the signal is disturbed by channel crosstalk is conducted. If the signal is triggered normally, the preliminary arrival time is regarded as the final arrival time of the AE signal. If the signal is triggered in advance, the signal series is selected from the preliminary arrival time to the time corresponding to the maximum of absolute values of signal series. At last the AIC picker is used again to pick the arrival time of selected series and the second AIC arrival time is interpreted as the final arrival time. To validate the performance of the improved automatic method, AIC picker, AIC picker applied after the median filter (MF-AIC) are also used simultaneously to pick the arrival times of AE signals and perform event location. The results show that the improved automatic method has better waveform adaptability and location accuracies than other two methods.

Understanding passenger behaviors is of great importance in intelligent transportation and infrastructure planning. However, the passenger trajectories are actually complex temporal data, which consist of rich spatial and temporal information. What's more, the observed passenger trajectories may be a mixture of different types of passengers with various travelling purposes. These difficulties make the prediction of passenger behaviors a challenging work. To address these problems, this paper improves the existing passenger behavior prediction methods from the following two aspects: 1) Encoding the travelling sequence with personalized semantic sensing, and 2) constructing multi-pattern prediction models to capture multiple travelling purposes and dynamics. Along this line, this paper provides a novel passenger behavior prediction model, namely, Semantic and multi-Pattern Long Short-Term Memory (SP-LSTM) model. Particularly, 1) a translation unit is designed, which is able to encode an observed travelling sequence into a structured sequence with consideration of individual travelling semantics; 2) a multi-pattern learning schematic is proposed, which first identifies the travelling patterns of passengers and then handles different patterns with different learning modules; 3) a unified learning framework is provided to integrate the semantic sensing module and multi-pattern learning module together, and present the final prediction results. To evaluate the proposed method, this paper conducts experiments on real-world passenger travelling data. Results demonstrate the superiority of SP-LSTM over both classical and the state-of-the-art methods.

In order to gain insights into the potential and behavior of training-based MIMO systems, the relationship of joint decoding scheme and separate decoding scheme is considered, and the equivalence between these two decoding schemes is proved. For the considered joint decoding scheme, receiver decodes out data by joint processing of received signals of both training symbols and data symbols in the maximum likelihood (ML) sense. We refer it as the joint ML-decoder. By contrast, in the considered separate decoding scheme, receiver first estimates channels in the minimum mean-square-errors (MMSE) sense, and then, based on the estimated channel information and the received signals of data symbols, the receiver decodes out data in the ML sense. We refer it as the separate MMSE-ML decoder. Notice that this separate MMSE-ML decoder is different from the decoder appeared in the most of existing works, where a kind of mismatched ML decoding is used after the phase of channel estimation. Although the above-mentioned decoding schemes have different decoding procedures, we prove that the joint ML-decoder and the separate MMSE-ML decoder are equivalent, while they outperform the mismatched ML decoder. With this equivalence, it is implied that the MMSE channel estimator is optimal when the overall system performance is considered, and the separate MMSE-ML decoding scheme can achieve the same performance with that of joint ML decoding scheme. Furthermore, this equivalence also provides us another way to analyze the system performance of the considered separate or joint decoding scheme. Namely, the results obtained from the considered joint decoding scheme can be directly applied to the considered separate decoding scheme, and vice versa.

For a massive multiple-input multiple-output system in the multi-cell scenarios, it generally suffers from pilot contamination due to the fact that the adjacent cells may use the same pilot sequences. In order to reduce the impact of pilot contamination on the system performance, a novel iterative channel estimation method is proposed. Specifically, in the proposed method, besides the pilot sequences, the data symbols decoded with high reliability are selected to enhance the channel estimation. Furthermore, the selection criteria are derived to determine which data symbols are reliably decoded. The simulation results show that the proposed method can effectively improve the channel estimation accuracy and the error performance of the considered systems.