Yang

Polynomial Chirplet Transform With Application to Instantaneous Frequency Estimation

In this paper, a new time-frequency analysis method known as the polynomial chirplet transform (PCT) is developed by extending the conventional chirplet transform (CT). By using a polynomial function instead of the linear chirp kernel in the CT, the PCT can produce a time-frequency distribution with excellent concentration for a wide range of signals with a continuous instantaneous frequency (IF). In addition, an effective IF estimation algorithm is proposed based on the PCT, and the effectiveness of this algorithm is validated by applying it to estimate the IF of a signal with a nonlinear chirp component and seriously contaminated by a Gaussian noise and a vibration signal collected from a rotor test rig.

General Parameterized Time-Frequency Transform

Interest in parameterized time-frequency analysis for non-stationary signal processing is increasing steadily. An important advantage of such analysis is to provide highly concentrated time-frequency representation with signal-dependent resolution. In this paper, a general scheme, named as general parameterized time-frequency transform (GPTF transform), is proposed for carrying out parameterized time-frequency analysis. The GPTF transform is defined by applying generalized kernel based rotation operator and shift operator. It provides the availability of a single generalized time-frequency transform for applications on signals of different natures. Furthermore, by replacing kernel function, it facilitates the implementation of various parameterized time - frequency transforms from the same standpoint. The desirable properties and the dual definition in the frequency domain of GPTF transform are also described in this paper. One of the advantages of the GPTF transform is that the generalized kernel can be customized to characterize the time - frequency signature of non-stationary signal. As different kernel formulation has bias toward the signal to be analyzed, a proper kernel is vital to the GPTF. Thus, several potential kernels are provided and discussed in this paper to develop the desired parameterized time - frequency transforms. In real applications, it is desired to identify proper kernel with respect to the considered signal. This motivates us to propose an effective method to identify the kernel for the GPTF.

Component Extraction for Non-Stationary Multi-Component Signal Using Parameterized De-chirping and Band-Pass Filter

In most applications, component extraction is important when components of non-stationary multi-component signal are key features to be monitored and analyzed. Existing methods are either sensitive to noise or forced to select a proper time-frequency representation for the considered signal. In this paper, we present a novel component extraction method for non-stationary multi-component signal. The proposed method combines parameterized de-chirping and band-pass filter to obtain components of multi-component signal, which avoids dealing with time-frequency representation of the signal and works well under heavy noise. In addition, it is able to analyze the multi-component signal whose components have intersected instantaneous frequency trajectories. Simulation results show that the proposed method is promising in analyzing complicated multi-component signals. Moreover, it works effective in a high noise environment in terms of improving the output signal-to-noise rate for the interested component.

Application of Parameterized Time-Frequency Analysis on Multicomponent Frequency Modulated Signals

Parameterized time-frequency (TF) transforms, with signal-dependent kernel parameters, have been proposed to analyze multicomponent frequency modulated (FM) signals. Usually, the kernel parameters are estimated through recursive approximation of TF representation (TFR) ridge when instantaneous frequency models of the components have the same parameter settings. However, it will be inapplicable if the components have the different FM sources. In this paper, we introduce a novel method that enables the parameterized TF transform to generate the well-concentrated TFR for both the monocomponent signal and a wide class of multicomponent FM signals, whose components are modulated by either the same or the different sources. The proposed method contains two aspects: 1) estimating kernel parameters based on spectrum concentration index and 2) separating components and assembling the parameterized TFRs of the separated components. An advantage of the proposed method is that it avoids the dependence of the TFR while estimating the parameters. Moreover, it is effective at low signal-to-noise rate. The validity and practical utility of the proposed method are demonstrated by both the simulated and real signals. The results show that it outperforms the traditional TF methods in providing the TFR of the improved concentration for various multicomponent FM signals.

Time-Varying Frequency-Modulated Component Extraction Based on Parameterized Demodulation and Singular Value Decomposition

To analyze the valuable frequency component for time-varying frequency-modulated (FM) signals, component extraction is necessary in most applications. Considering the advantage of parameterized demodulation (PD) in transforming FM signals to be stationary, a novel component extraction method based on PD and singular value decomposition (PD-SVD) for both monocomponent and multicomponent signals is proposed. By extending the idea of PD, the time-varying term of the continuous phase function for the interested FM component can be removed, thus resulting in a highly self-correlated component with constant phase. Then, the extraction of the target component from noise or other components can be realized by SVD. Compared with the existing methods, the proposed algorithm is able to analyze the multicomponent signal with crossed instantaneous frequency trajectories and effectively improve the signal-to-noise ratio of the extracted component. The effectiveness of the proposed method is demonstrated by applying it on several numerical signals and the radial vibration signal of a hydroturbine rotor, indicating the potential of analyzing many practical FM signals.

Intrinsic chirp component decomposition by using Fourier Series representation

We introduce a general model to characterize multi-component chirp signals.The model represents instantaneous frequencies and instantaneous amplitudes of the intrinsic chirp components as Fourier series.We propose a novel chirp component decomposition method based on the model.The method is able to decompose multi-component chirp signals with intersecting IFs. In this paper, we consider the decomposition problem for multi-component chirp signals (MCCSs). We develop a general model to characterize MCCSs, where instantaneous frequencies (IFs) and instantaneous amplitudes (IAs) of the intrinsic chirp components (ICCs) are modeled as Fourier series. The decomposition problem thus boils down to identifying the developed model. The IF estimation is addressed using the framework of the general parameterized time-frequency transform and then the signal can be easily reconstructed by solving a linear system. For the practical implementation of our method, we present a two-step algorithm, which is initiated by an iterative scheme to achieve preliminary separation of the ICCs, followed by a joint-refinement step to get high-resolution ICC reconstructions. Our method acts as a time-varying band-pass filter and can even separate ICCs that cross in the time-frequency domain. The method is applied to analyze several simulated and real signals, which indicates its usefulness in various applications.

Chirplet Path Fusion for the Analysis of Time-Varying Frequency-Modulated Signals

We propose a novel method, called chirplet path fusion, to analyze nonstationary signals with time-varying frequencies. As opposed to many existing methods that assume the signal persists throughout the whole time, the proposed method can be applied in cases where the signals are only present in short time frames. A demodulation-operator-based method is introduced to estimate the locally matched chirplet atoms which can form a chirplet path in the time-frequency domain. According to the density of the multiple chirplet paths obtained under different scales, the effective time support and the instantaneous frequency (IF) of the target signal can be estimated. Our method is effective in analyzing the nonstationary signal with discontinuous IF curve and works well under heavy noise. Several simulated signals and the vibration signal of a rotor test rig are considered to verify the effectiveness of the proposed method.

Machine learning for classifying tuberculosis drug-resistance from DNA sequencing data

Motivation: Correct and rapid determination of Mycobacterium tuberculosis (MTB) resistance against available tuberculosis (TB) drugs is essential for the control and management of TB. Conventional molecular diagnostic test assumes that the presence of any well‐studied single nucleotide polymorphisms is sufficient to cause resistance, which yields low sensitivity for resistance classification. Summary: Given the availability of DNA sequencing data from MTB, we developed machine learning models for a cohort of 1839 UK bacterial isolates to classify MTB resistance against eight anti‐TB drugs (isoniazid, rifampicin, ethambutol, pyrazinamide, ciprofloxacin, moxifloxacin, ofloxacin, streptomycin) and to classify multi‐drug resistance. Results: Compared to previous rules‐based approach, the sensitivities from the best‐performing models increased by 2‐4% for isoniazid, rifampicin and ethambutol to 97% (P < 0.01), respectively; for ciprofloxacin and multi‐drug resistant TB, they increased to 96%. For moxifloxacin and ofloxacin, sensitivities increased by 12 and 15% from 83 and 81% based on existing known resistance alleles to 95% and 96% (P < 0.01), respectively. Particularly, our models improved sensitivities compared to the previous rules‐based approach by 15 and 24% to 84 and 87% for pyrazinamide and streptomycin (P < 0.01), respectively. The best‐performing models increase the area‐under‐the‐ROC curve by 10% for pyrazinamide and streptomycin (P < 0.01), and 4‐8% for other drugs (P < 0.01). Availability and implementation: The details of source code are provided at http://www.robots.ox.ac.uk/˜davidc/code.php. Contact: david.clifton@eng.ox.ac.uk Supplementary information: Supplementary data are available at Bioinformatics online.

Risk prediction for cardiovascular disease using ECG data in the China kadoorie biobank

We set out to use machine learning techniques to analyse ECG data to improve risk evaluation of cardiovascular disease in a very large cohort study of the Chinese population. We performed this investigation by (i) detecting “abnormality” using 3 one-class classification methods, and (ii) predicting probabilities of “normality”, arrhythmia, ischemia, and hypertrophy using a multiclass approach. For one-class classification, we considered 5 possible definitions for “normality” and used 10 automatically-extracted ECG features along with 4 blood pressure features. The one-class approach was able to identify abnormality with area-under-curve (AUC) 0.83, and with 75.6% accuracy. For four-class classification, we used 86 features in total, with 72 additional features extracted from the ECG. Accuracy for this four-class classifier reached 75.1%. The methods demonstrated proof-of-principle that cardiac abnormality can be detected using machine learning in a large cohort study.

Prediction of Type III Secreted Effectors Based on Word Embeddings for Protein Sequences

The type III secreted effectors (T3SEs) are virulence proteins that play an important role in the pathogenesis of Gram-negative bacteria. They are injected into the host cells by the pathogens, interfere with the immune system of the host cells, and help the growth and reproduction of the pathogens. It is a very challenging task to identify T3SEs because of the high diversity of their sequences and the lack of defined secretion signals. Moreover, their working mechanisms have not been fully understood yet. In order to speed up the recognition of T3SEs and the studies of type III secretion systems, computational tools for the prediction of T3SEs are in great demand. In this study, we regard the protein sequences as a special language. Inspired by the word2vec model in natural language processing, we convert the sequences into word embedding vectors in a similar manner with a specific segmentation strategy for protein sequences. And then we construct the T3SE predictor based on the new sequence feature representation. We conduct experiments on both mono-species data and multi-species data. The experimental results show that the new feature representation model has a competitive performance and can work together with the traditional features to enhance the identification of T3SEs.