• Journal of Institute of Control, Robotics and Systems
• /
• v.9 no.11
• /
• pp.917-924
• /
• 2003

Recurrent neural networks have been successfully applied to communications channel equalization. Major disadvantages of gradient-based learning algorithms commonly employed to train recurrent neural networks are slow convergence rates and long training sequences required for satisfactory performance. In a high-speed communications system, fast convergence speed and short training symbols are essential. We propose decision feedback equalizers using a recurrent neural network trained with Kalman filtering algorithms. The main features of the proposed recurrent neural equalizers, utilizing extended Kalman filter (EKF) and unscented Kalman filter (UKF), are fast convergence rates and good performance using relatively short training symbols. Experimental results for two time-varying channels are presented to evaluate the performance of the proposed approaches over a conventional recurrent neural equalizer.

• Proceedings of the KIEE Conference
• /
• 1998.07b
• /
• pp.552-554
• /
• 1998

In target tracking problems the fixed gain Kalman filter is primarily used to predict a target state vector. This filter, however, has a poor precision for maneuvering targets while it has a good performance for non-maneuvering targets. To overcome the problem this paper proposes the system which estimates the acceleration with neural networks using the input estimation technique. The ability to efficiently fuse information of different forms is one of the major capabilities of trained multi-layer neural networks. The primary motivation for employing neural networks in these applications comes from the efficiency with which more features can be utilized as inputs for estimating target maneuvers. The parallel processing capability of a properly trained neural network can permit fast processing of features to yield correct acceleration estimates. The features used as inputs can be extracted from the combinations of innovation data and heading changes, and for this we set the two dimensional model. The properly trained neural network system outputs the acceleration estimates and compensates for the primary Kalman filter. Finally the proposed system shows the optimum performance.

• International Journal of Naval Architecture and Ocean Engineering
• /
• v.12 no.1
• /
• pp.314-324
• /
• 2020

Due to the nonlinearity and environmental uncertainties, the design of the ship's steering controller is a long-term challenge. The purpose of this study is to design an intelligent autopilot based on Extended Kalman Filter (EKF) trained Radial Basis Function Neural Network (RBFNN) control algorithm. The newly developed free running model scaled surface vessel was employed to execute the motion control experiments. After describing the design of the EKF trained RBFNN autopilot, the performances of the proposed control system were investigated by conducting experiments using the physical model on lake and simulations using the corresponding mathematical model. The results demonstrate that the developed control system is feasible to be used for the ship's motion control in the presences of environmental disturbances. Moreover, in comparison with the Back-Propagation (BP) neural networks and Proportional-Derivative (PD) based control methods, the EKF RBFNN based control method shows better performance regarding course keeping and trajectory tracking.

• Proceedings of the KIEE Conference
• /
• 2007.04a
• /
• pp.3-5
• /
• 2007

This paper presents decision feedback equalizers using a recurrent neural network trained algorithm using extended Kalman filter(EKF) and sigma-point Kalman filter(SPKF). EKF is propagated, analytically through the first-order linearization of the nonlinear system. This can introduce large errors in the true posterior mean and covariance of the Gaussian random variable. The SPKF addresses this problem by using a deterministic sampling approach. The features of the proposed recurrent neural equalizer And we investigate the bit error rate(BER) between EKF and SPKF.

• 제어로봇시스템학회:학술대회논문집
• /
• 1997.10a
• /
• pp.524-527
• /
• 1997

Conventional target tracking algorithms based on the linear estimation techniques perform quite efficiently when the target motion does not involve maneuvers. Target maneuvers involving short term accelerations, however, cause a bias in the measurement sequence. Accurate compensation for the bias requires processing more samples of which adds to the computational complexity. The primary motivation for employing a neural network for this task comes from the efficiency with which more features can be as inputs for bias compensation. A system architecture that efficiently integrates the fusion capabilities of a trained multilayer neural net with the tracking performance of a Kalman filter is described. The parallel processing capability of a properly trained neural network can permit fast processing of features to yield correct acceleration estimates and hence can take the burden off the primary Kalman filter which still provides the target position and velocity estimates.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2018.11a
• /
• pp.133-134
• /
• 2018

MPEG-NNR (Compressed Representation of Neural Networks) aims to define a compressed and interoperable representation of trained neural networks. In this paper, a compressed representation of NN and its evaluation performance along with use cases of image/video compression in MPEG-NNR are presented. In the compression of NN, a CNN to replace the in-loop filter in VVC (Versatile Video Coding) intra coding is compressed by applying uniform quantization to reduce the trained weights, and the compressed CNN is evaluated in terms of compression ratio and coding efficiency compared to the original CNN. Evaluation results show that CNN could be compressed to about quarter with negligible coding loss by applying simple quantization to the trained weights.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.15 no.5
• /
• pp.552-559
• /
• 2005

Recurrent neural networks(RNNs) trained with gradient based such as real time recurrent learning(RTRL) has a drawback of slor convergence rate. This algorithm also needs the derivative calculation which is not trivialized in error back propagation process. In this paper a derivative free Kalman filter, so called the unscented Kalman filter(UKF), for training a fully connected RNN is presented in a state space formulation of the system. A derivative free Kalman filler learning algorithm makes the RNN have fast convergence speed and good tracking performance without the derivative computation. Through experiments of nonlinear channel equalization, performance of the RNNs with a derivative free Kalman filter teaming algorithm is evaluated.

• 제어로봇시스템학회:학술대회논문집
• /
• 2003.10a
• /
• pp.2498-2501
• /
• 2003

In this paper, we present a wavelet neural network (WNN) approach to the solution of the path tracking problem for mobile robots that possess complexity, nonlinearity and noise. First, we discuss a WNN based control system where the control signals are directly obtained by minimizing the difference between the reference track and the pose of a mobile robot. This compact network structure is helpful to determine the number of hidden nodes and the initial value of weights. Then, the data with various noises provided by odometric and external sensors are here fused together by means of an Extended Kalman Filter (EKF) approach for the pose estimation problem of mobile robots. This control process is a dynamic on-line process that uses the wavelet neural network trained via the gradient-descent method with estimates from EKF. Finally, we verify the effectiveness and feasibility of the proposed control system through simulations.

• The Transactions of the Korean Institute of Electrical Engineers
• /
• v.43 no.4
• /
• pp.553-563
• /
• 1994

This paper presents a filtering method using neural networks to extract fundamental frequency components of the transient wave signals in power systems. Based on the ability of multilayer feedforward neural networks to approximate any continuous function, a neural networks mapping filter is proposed for the protective distance relaying systems to extract the effective components efficiently. A characteristic feature of this mapping filter is composed of the multilayer perceptron neural networks which are trained by using random signals and those are mapped to the DFT filtering computational structure by GDR(Generalized Delta Rule). The advantage of this approach is demonstrated by the random waves and the fault transient wave signals of EMTP(electromagnetic transients program) in power systems fault conditions. The proposed method is compared with the conventional method and the simulation results show the efficiency of the neural networks.

• Journal of Institute of Control, Robotics and Systems
• /
• v.22 no.9
• /
• pp.773-777
• /
• 2016

In this paper, a novel multi-object tracking method to track an unknown number of objects is proposed. To handle multiple object states and uncertain observations efficiently, a probability hypothesis density (PHD) filter is adopted and modified. The PHD filter is capable of reducing false positives, managing object appearances and disappearances, and estimating the multiple object trajectories in a unified framework. Although the PHD filter is robust in cluttered environments, it is vulnerable to false negatives. For this reason, we propose to exploit local observations in an RFS of the observation model. Each local observation is generated by using an online trained object detector. The main purpose of the local observation is to deal with false negatives in the PHD filtering procedure. The experimental results demonstrated that the proposed method robustly tracked multiple objects under practical situations.