A log-linearized Gaussian mixture network and its application toEEG pattern classification

Proposes a new probabilistic neural network (NN) that can estimate the a-posteriori probability for a pattern classification problem. The structure of the proposed network is based on a statistical model composed by a mixture of log-linearized Gaussian components. However, the forward calculation and the backward learning rule can be defined in the same manner as the error backpropagation NN. In this paper, the proposed network is applied to the electroencephalogram (EEG) pattern classification problem. In the experiments described, two types of a photic stimulation, which are caused by eye opening/closing and artificial light, are used to collect the data to be classified. It is shown that the EEG signals can be classified successfully and that the classification rates change depending on the amount of training data and the dimension of the feature vectors     

Performance evaluation of FSK and CPFSK optical communicationsystems: a stable and accurate method

The modified moments method for evaluating the performance of coherent optical FSK and CPFSK systems is presented. Since the classical procedure becomes ill-conditioned as the order of the moments increases, we consider the construction of Gaussian quadrature rules (GQR) from the modified moments. The analysis accounts for the influences of IF bandwidth, transmitter and local oscillator laser phase noise, postdetection filters, and additive Gaussian noise. It is found that the proposed approach is a highly reliable and efficient method for calculating the error probability. A comparison with results obtained from the Gaussian quadrature rule, Gaussian approximation method, and analytical approximation formulas shows that this technique is very accurate. Analytical expressions are derived for FSK and CPFSK receivers which include polarization and phase diversity techniques. The use of numerical programming to avoid many unnecessary computations is discussed. This evaluation method can be used to account for the effects of crosstalk in multichannel systems and the influence of error-control codes     

Performance of energy detector in the presence of noise uncertainty in cognitive radio networks

Energy detector is simple in structure and easy to implement. Therefore, it is a promising candidate for spectrum sensing in cognitive radio networks. However, its detection performance is typically challenged by the noise uncertainty. Thus, the detection performance of energy detector in the presence of noise uncertainty needs to be evaluated. In this paper, we derive the decision rules for the energy detector in the presence of noise uncertainty by employing a widely used model. Firstly, we derive the decision rule for unknown deterministic signal when the noise power is uncertain. Second, we derive the decision rule for random Gaussian distributed signal when there is noise uncertainty. Then, we analyze the detection performance of the energy detector in the presence of noise uncertainty for both unknown deterministic signal and random Gaussian distributed signal. Both theoretical analyses and simulation results show that in the presence of noise uncertainty, our derived decision rules provide precise sensing thresholds for the energy detector. Furthermore, compared with the conventional decision rule obtained by overestimating the noise power, our decision rules provide performance gains in terms of signal to noise ratio.     

基于特征值分解的频移键控信号检测与识别

提出一种基于协方差矩阵特征值分布特性的频移键控(FSK)信号检测方法以及基于信源数估计技术的FSK识别方法.根据噪声协方差矩阵特征值的直线分布特征,利用直线拟合度构造了信号检测判定准则.给出了常用的信息论(AIG)、最短描述长度(MDL)及盖氏圆盘算法(GDE)信源数判定准则的定义,并根据特征值分布特征提出了一种最大斜率准则(MSR).仿真表明,MSR准则对于FSK识别更有效.同时,给出了检测及识别参数的设置方法.仿真结果证明了检测及识别方法的有效性.文章将阵列信号处理中的方法应用于单通道信号的分析中,为FSK信号的检测与识别提供了一种新的思路.     

Novel procedure to determine statistical functions of impulsive noise

One of the problems found when measuring impulsive noise is to distinguish this kind of noise from Gaussian noise. Usually, a threshold level is used to make the difference. The problem is that a high threshold level will miss low amplitude pulses, while a low level will include Gaussian noise samples as being impulsive noise. In this paper, results of a novel radio UHF impulsive noise measurement procedure are presented. This work exhibits the peculiarity that data was taken in both horizontal and vertical polarizations simultaneously. One polarization is used to determine the presence of impulsive noise so analysis can be performed in the other polarization considering this circumstance. Measurements were made in four different locations on UHF TV channels around 800 MHz with a bandwidth of 10 MHz, demodulating the in-phase and quadrature phase components in each polarization. When environments are constituted by varied and scattered impulsive sources, horizontal and vertical polarizations show similar statistical behavior. However, horizontal and vertical emissions may be different when a single source is individually measured. The conditional amplitude probability density function (pdf) and conditional amplitude cumulative distribution function (cdf) found for the amplitude in all the locations lead to the conclusion that the variable that best fits the results is a lognormal one.     

Speaker Adaptation of Hybrid NN/HMM Model for Speech Recognition Based on Singular Value Decomposition

Recently several speaker adaptation methods have been proposed for deep neural network (DNN) in many large vocabulary continuous speech recognition (LVCSR) tasks. However, only a few methods rely on tuning the connection weights in trained DNNs directly to optimize system performance since it is very prone to over-fitting especially when some class labels are missing in the adaptation data. In this paper, we propose a new speaker adaptation method for the hybrid NN/HMM speech recognition model based on singular value decomposition (SVD). We apply SVD on the weight matrices in trained DNNs and then tune rectangular diagonal matrices with the adaptation data. This alleviates the over-fitting problem via updating the weight matrices slightly by only modifying the singular values. We evaluate the proposed adaptation method in two standard speech recognition tasks, namely TIMIT phone recognition and large vocabulary speech recognition in the Switchboard task. Experimental results have shown that it is effective to adapt large DNN models using only a small amount of adaptation data. For example, recognition results in the Switchboard task have shown that the proposed SVD-based adaptation method may achieve up to 3-6 % relative error reduction using only a few dozens of adaptation utterances per speaker.     

基于模糊神经网络的目标识别

结合模糊推理和神经网络两种方法的优点,从网络的结构、工作过程、学习算法等方面,探讨了一种基于模糊神经网络(FNN)的目标识别方法。通过仿真结果证明,此方法确实可行。     

基于模糊分类器及多层前馈神经网络混合结构的说话人辨认

提出了基于模糊超椭球聚类算法的说话人辨认新方法。该算法首先将某一类的训练数据分成若干子类, 对每一子类在其中心周围定义具有超椭球区域的模糊规则。实验表明,该系统可以较快的聚类速度取得与HMM 相当的识别效果。进一步的研究表明,基于模糊超椭球聚类算法的说话人辨认系统与传统的基于HMM的识别方法存在一个共同的缺点,即抗噪性能较差。为此,通过引入多层前馈神经网络(MLFNN)与模糊超椭球分类器构成混合模型,使系统的识别性能和抗噪能力显著提高。     

Point-based rigid-body registration using an unscented Kalman filter

We present and validate a novel registration algorithm mapping two data sets, generated from a rigid object, in the presence of Gaussian noise. The proposed method is based on the Unscented Kalman Filter (UKF) algorithm that is generally employed for analyzing nonlinear systems corrupted by additive Gaussian noise. First, we employ our proposed registration algorithm to fit two randomly generated data sets in the presence of isotropic Gaussian noise, when the corresponding points between the two data sets are assumed to be known. Then, we extend the registration method to the case where the data (with known correspondences) is stimulated by anisotropic Gaussian noise. The new registration method not only reliably converges to the correct registration solution, but it also estimates the variance, as a confidence measure, for each of the estimated registration transformation parameters. Furthermore, we employ the proposed registration algorithm for rigid-body, point-based registration where corresponding points between two registering data sets are unknown. The algorithm is tested on point data sets which are garnered from a pelvic cadaver and a scaphoid bone phantom by means of computed tomography (CT) and tracked free-hand ultrasound imaging. The collected 3-D points in the ultrasound frame are registered to the 3-D meshes in the CT frame by using the proposed and the standard Iterative Closest Points (ICP) registration algorithms. Experimental results demonstrate that our proposed method significantly outperforms the ICP registration algorithm in the presence of additive Gaussian noise. It is also shown that the proposed registration algorithm is more robust than the ICP registration algorithm in terms of outliers in data sets and initial misalignment between the two data sets.     

Cooperative automatic modulation recognition in cognitive radio

In this article, a new effective method of cooperative modulation recognition (CMR) is proposed to recognize different modulation types of primary user for cognitive radio receivers. In the cognitive radio (CR) system, two CR users respectively send their feature parameters to the cooperative recognition center, which is composed of back propagation neural network (BPNN). With two users' cooperation and the application of an error back propagation learning algorithm with momentum, the center improves the performance of modulation recognition, especially when one of the CR users' signal-to-noise ratio (SNR) is low. To measure the performance of the proposed method, simulations are carried out to classify different types of modulated signals corrupted by additive white Gaussian noise (AWGN). The simulation results show that this cooperation algorithm has a better recognition performance than those without cooperation.     

一种融合流形学习和深度学习的时变信道自动调制识别技术

自动调制识别在军事领域和民用领域都发挥了巨大作用。现有的大多数研究都是基于高斯白噪声信道,但是时变信道下的自动调制识别才更符合实际并且具有挑战性。该文针对时变信道提出了一种融合流形学习和深度学习的自动调制识别方法,第一次将格拉斯曼流形引入到信号的特征提取,通过将信号星座图建模到格拉斯曼流形上完成特征提取。分类网络由基于流形学习和深度学习的两部分组成,流形数据先经过流形学习网络进行降维,然后映射到平滑子空间,最后通过简单的卷积神经网络完成分类。实验结果表明,与传统的卷积神经网络相比该文所提出的方案具有良好的性能,同时为自动调制识别提供了新的解决思路。      

雷达组网目标检测性能的概率分析

雷达组网能有效提高目标的检测性能。在目标瑞利起伏的模型下，噪声服从零均值高斯分布时，对比了秩K规则和最大信噪比规则这两种融合方法，对目标检测性能的影响。选取六部雷达进行数据融合，推导分析了在不同的虚警概率条件下，采用两种融合规则对雷达检测概率的提高程度，并指出在检测概率方面最大信噪比规则融合方法明显优于秩K规则融合方法，但雷达之间传输的数据量增大。     

Frame Synchronization for Gaussian Channels

The problem of locating a periodically inserted frame synchronization pattern in random data for aM-ary digital communication system operating over the additive white Gaussian noise channel is considered. The optimum maximum-likelihood decision rule, high signal-to-noise approximate maximum likelihood decision rule, and ordinary correlation decision rule for frame synchronization are derived for both coherent and noncoherent phase demodulation. A general lower bound on synchronization probability is derived for the coherent correlation rule. Monte Carlo computer simulations of all three decision rules, along with evaluations of the lower bound for the coherent correlation rule, were performed for the coherent MPSK, coherent, and noncoherentMary orthogonal, and 16 QAM signaling schemes. These results show that in each case the high signal-to-noise maximum-likelihood rules have a performance nearly equal to that of the maximum-likelihood rules over a wide range of practically interesting signal-to-noise ratios (SNR's). These high SNR decision rules also provide significant performance improvement over the simple correlation rules. Moreover, they are much simpler to implement than the maximum-likelihood decision rules and, in fact, are no more complex than the correlation rules.     

Detection, classification, and superposition resolution of actionpotentials in multiunit single-channel recordings by an on-linereal-time neural network

Determination of single-unit spike trains from multiunit recordings obtained during extracellular recording has been the focus of many studies over the last two decades. In multiunit recordings, superpositions can occur with high frequency if the firing rates of the neurons are high or correlated, making superposition resolution imperative for accurate spike train determination. In this work, a connectionist neural network (NN) was applied to the spike sorting challenge. A novel training scheme was developed which enabled the NN to resolve some superpositions using single-channel recordings. Simulated multiunit spike trains were constructed from templates and noise segments that were extracted from real extracellular recordings. The simulations were used to determine the performances of the NN and a simple matched template filter (MTF), which was used as a basis for comparison. The network performed as well as the MTF in identifying nonoverlapping spikes, and was significantly better in resolving superpositions and rejecting noise. An on-line, real-time implementation of the NN discriminator, using a high-speed digital signal processor mounted inside an IBM-PC, is now in use in six laboratories     

A theory of adaptive filtering

This paper considers the adaptive signal extraction problem for time-discrete data when only very general a priori assumptions regarding the distributions of signal and noise are possible. Specifically, it is assumed that the noise is white, additive, and signal independent with mean zero and unknown variance and that the signal is band-limited. No stationarity assumptions are required. After a procedure is found under these conditions, the mean-square-error is derived asymptotically under narrower conditions-stationary Gaussian data with mean zero. Finally, a method of estimating the error variance from the data (without knowing the signal directly) is found.     

Support Vector Machine and Neural Network Classification of Metallic Objects Using Coefficients of the Spheroidal MQS Response Modes

Two different supervised learning algorithms, support vector machine (SVM) and neural networks (NN), are applied in classifying metallic objects according to size using the expansion coefficients of their magneto-quasistatic response in the spheroidal coordinate system. The classified objects include homogeneous spheroids and composite metallic assemblages meant to resemble unexploded ordnance. An analytical model is used to generate the necessary training data for each learning method. SVM and NN are shown to be successful in classifying three different types of objects on the basis of size. They are capable of fast classification, making them suitable for real-time application. Furthermore, both methods are robust and have a good tolerance of 20-dB SNR additive Gaussian noise. SVM shows promise in dealing with noise due to uncertainty in the object's position and orientation.     

Neural network impedance force control of robot manipulator

The performance of an impedance controller for robot force tracking is affected by the uncertainties in both the robot dynamic model and environment stiffness. The purpose of this paper is to improve the controller robustness by applying the neural network (NN) technique to compensate for the uncertainties in the robot model. NN control techniques are applied to two impedance control methods: torque-based and position-based impedance control, which are distinguished by the way of the impedance functions being implemented. A novel error signal is proposed for the NN training. In addition, a trajectory modification algorithm is developed to determine the reference trajectory when the environment stiffness is unknown. The robustness analysis of this algorithm to force sensor noise and inaccurate environment position measurement is also presented. The performances of the two NN impedance control schemes are compared by computer simulations. Simulation results based on a three-degrees-of-freedom robot show that highly robust position/force tracking can be achieved in the presence of large uncertainties and force sensor noise     

基于邻域统计特性的概率神经网络及其在自动目标识别中的应用

在概率神经的一种改进模型－ＦＤＯ网络的基础上，提出在设计网络收敛域时进一步考虑每一像素点周围８邻域的影响，对网络的作用函数加以修，使改进后的网络具有稳定性好且收敛速度快的优点。通过实验对改进前后网络的识别性能加以比较，证明改进后的网络特别适用于噪声图像的识别。     

基于免疫克隆高斯过程隐变量模型的SAR目标特征提取与识别

作为一种非线性维数约减算法,高斯过程隐变量模型(Gaussian process latent variable model,GPLVM)由于其适合处理小样本、高维数据,因而在模式识别、计算机视觉等领域得到了广泛应用.基于此,提出一种基于改进GPLVM的SAR图像目标特征提取及自动识别方法,其中利用改进的GPLVM进行特征提取,高斯过程分类进行目标识别.传统GPLVM使用共轭梯度法对似然函数进行优化,为避免梯度估值易受噪声干扰、步长对算法影响严重等缺点,提出基于免疫克隆选择算法的GPLVM,利用其具有快速收敛到全局最优的特性提高算法性能.实验结果表明,该算法不仅降低了特征维数,且提高了识别精度,从而验证了算法用于SAR图像目标识别的有效性.     

Optimum Frame Synchronization

This paper considers the optimum method for locating a sync word periodically imbedded in binary data and received over the additive white Gaussian noise channel. It is shown that the optimum rule is to select the location that maximizes the sum of the correlation and a correction term. Simulations are reported that show approximately a 3-dB improvement at interesting signal-to-noise ratios compared to a pure correlation rule. Extensions are given to the "phase-shift keyed (PSK) sync" case where the detector output has a binary ambiguity and to the case of Gaussian data.