Multi-Band CNN With Band-Dependent Kernels and Amalgamated Cross Entropy Loss for Motor Imagery Classification.

In this paper, we present a novel MI classification method based on multi-band convolutional neural network (CNN) with band-dependent kernel sizes, named MBK-CNN, to improve classification performance, by resolving the subject dependency issue of the widely used CNN-based approaches due to the kernel size optimization problem. The proposed structure exploits the frequency diversity of the EEG signals and simultaneously resolves the subject dependent kernel size issue. EEG signal is decomposed into overlapping multi-band and passed through multiple CNNs (termed 'branch-CNNs') with different kernel sizes to generate frequency dependent features, which are combined by a simple weighted sum. In contrast to the existing works where single-band multi-branch CNNs with different kernel sizes are used to resolve the subject dependency issue, a unique kernel size per frequency band is used. To prevent possible overfitting induced by a weighted sum, each branch-CNN is additionally trained by tentative cross entropy loss while overall network is optimized by the end-to-end cross entropy loss, which is named amalgamated cross entropy loss. In addition, we further propose multi-band CNN with enhanced spatial diversity, named MBK-LR-CNN, by replacing each branch-CNN with several sub branch-CNNs applied for channel subsets (termed 'local region') to improve the classification performance. We evaluated the performance of the proposed methods, MBK-CNN and MBK-LR-CNN, on publicly available datasets, BCI Competition IV dataset 2a and High Gamma Dataset. The experimental results confirm the performance improvement of the proposed methods compared to the currently existing MI classification methods.

A cascade of preconditioned conjugate gradient networks for accelerated magnetic resonance imaging.

BACKGROUND AND OBJECTIVE: Recent unfolding based compressed sensing magnetic resonance imaging (CS-MRI) methods only reinterpret conventional CS-MRI optimization algorithms and, consequently, inherit the weaknesses of the alternating optimization strategy. In order to avoid the structural complexity of the alternating optimization strategy and achieve better reconstruction performance, we propose to directly optimize the ℓ1 regularized convex optimization problem using a deep learning approach. METHOD: In order to achieve direct optimization, a system of equations solving the ℓ1 regularized optimization problem is constructed from the optimality conditions of a novel primal-dual form proposed for the effective training of the sparsifying transform. The optimal solution is obtained by a cascade of unfolding networks of the preconditioned conjugate gradient (PCG) algorithm trained to minimize the mean element-wise absolute difference (ℓ1 loss) between the terminal output and ground truth image in an end-to-end manner. The performance of the proposed method was compared with that of U-Net, PD-Net, ISTA-Net+, and the recently proposed projection-based cascaded U-Net, using single-coil knee MR images of the fastMRI dataset. RESULTS: In our experiment, the proposed network outperformed existing unfolding-based networks and the complex version of U-Net in several subsampling scenarios. In particular, when using the random Cartesian subsampling mask with 25 % sampling rate, the proposed model outperformed PD-Net by 0.76 dB, ISTA-Net+ by 0.43 dB, and U-Net by 1.21 dB on the positron density without suppression (PD) dataset in term of peak signal to noise ratio. In comparison with the projection-based cascade U-Net, the proposed algorithm achieved approximately the same performance when the sampling rate was 25% with only 1.62% number of network parameters at the cost of a longer reconstruction time (approximately twice). CONCLUSION: A cascade of unfolding networks of the PCG algorithm was proposed to directly optimize the ℓ1 regularized CS-MRI optimization problem. The proposed network achieved improved reconstruction performance compared with U-Net, PD-Net, and ISTA-Net+, and achieved approximately the same performance as the projection-based cascaded U-Net while using significantly fewer network parameters.

Design Method for a Wideband Non-Uniformly Spaced Linear Array Using the Modified Reinforcement Learning Algorithm.

In this paper, we present a design method for a wideband non-uniformly spaced linear array (NUSLA), with both symmetric and asymmetric geometries, using the modified reinforcement learning algorithm (MORELA). We designed a cost function that provided freedom to the beam pattern by setting limits only on the beam width (BW) and side-lobe level (SLL) in order to satisfy the desired BW and SLL in the wide band. We added the scan angle condition to the cost function to design the scanned beam pattern, as the ability to scan a beam in the desired direction is important in various applications. In order to prevent possible pointing angle errors for asymmetric NUSLA, we employed a penalty function to ensure the peak at the desired direction. Modified reinforcement learning algorithm (MORELA), which is a reinforcement learning-based algorithm used to determine a global optimum of the cost function, is applied to optimize the spacing and weights of the NUSLA by minimizing the proposed cost function. The performance of the proposed scheme was verified by comparing it with that of existing heuristic optimization algorithms via computer simulations.

Multi-Target Localization Based on Unidentified Multiple RSS/AOA Measurements in Wireless Sensor Networks.

All existing hybrid target localization algorithms using received signal strength (RSS) and angle of arrival (AOA) measurements in wireless sensor networks, to the best of our knowledge, assume a single target such that even in the presence of multiple targets, the target localization problem is translated to multiple single-target localization problems by assuming that multiple measurements in a node are identified with their originated targets. Herein, we first consider the problem of multi-target localization when each anchor node contains multiple RSS and AOA measurement sets of unidentified origin. We propose a computationally efficient method to cluster RSS/AOA measurement sets that originate from the same target and apply the existing single-target linear hybrid localization algorithm to estimate multiple target positions. The complexity analysis of the proposed algorithm is presented, and its performance under various noise environments is analyzed via simulations.

Target Localization with Unknown Transmit Power and Path-Loss Exponent Using a Kalman Filter in WSNs.

We present a novel hybrid localization algorithm for wireless sensor networks in the absence of knowledge regarding the transmit power and path-loss exponent. Transmit power and the path-loss exponent are critical parameters for target localization algorithms in wireless sensor networks, which help extract target position information from the received signal strength. In the absence of information on transmit power and path-loss exponent, it is critical to estimate them for reliable deployment of conventional target localization algorithms. In this paper, we propose a simultaneous estimation of transmit power and path-loss exponent based on Kalman filter. The unknown transmit power and path-loss exponent are estimated using a Kalman filter with the tentatively estimated target position based solely on angle information. Subsequently, the target position is refined using a hybrid method incorporating received signal strength measurements based on the estimated transmit power and path-loss exponent. Our proposed algorithm accurately estimates transmit power and path-loss exponent and yields almost the same target position accuracy as the simulation results confirm, as the hybrid target localization algorithms with known transmit power and path-loss exponent. Simulation results confirm the proposed algorithm achieves 99.7% accuracy of the target localization performance with known transmit power and path-loss exponent, even in the presence of severe received signal strength measurement noise.

Hybrid RSS/AOA Localization using Approximated Weighted Least Square in Wireless Sensor Networks.

We present a target localization method using an approximated error covariance matrix based weighted least squares (WLS) solution, which integrates received signal strength (RSS) and angle of arrival (AOA) data for wireless sensor networks. We approximated linear WLS errors via second-order Taylor approximation, and further approximated the error covariance matrix using a least-squares solution and the variance in measurement noise over the sensor nodes. The algorithm does not require any prior knowledge of the true target position or noise variance. Simulations validated the superior performance of our new method.

Selective Feature Generation Method Based on Time Domain Parameters and Correlation Coefficients for Filter-Bank-CSP BCI Systems.

This paper presents a novel motor imagery (MI) classification algorithm using filter-bank common spatial pattern (FBCSP) features based on MI-relevant channel selection. In contrast to existing channel selection methods based on global CSP features, the proposed algorithm utilizes the Fisher ratio of time domain parameters (TDPs) and correlation coefficients: the channel with the highest Fisher ratio of TDPs, named principle channel, is selected and a supporting channel set for the principle channel that consists of highly correlated channels to the principle channel is generated. The proposed algorithm using the FBCSP features generated from the supporting channel set for the principle channel significantly improved the classification performance. The performance of the proposed method was evaluated using BCI Competition III Dataset IVa (18 channels) and BCI Competition IV Dataset I (59 channels).

Target Localization Using Double-Sided Bistatic Range Measurements in Distributed MIMO Radar Systems.

We develop a novel approach improving existing target localization algorithms for distributed multiple-input multiple-output (MIMO) radars based on bistatic range measurements (BRMs). In the proposed algorithms, we estimate the target position with auxiliary parameters consisting of both the target-transmitter distances and the target-receiver distances (hence, "double-sided") in contrast to the existing BRM methods. Furthermore, we apply the double-sided approach to multistage BRM methods. Performance improvements were demonstrated via simulations and a limited theoretical analysis was attempted for the ideal two-dimensional case.

Frequency-Optimized Local Region Common Spatial Pattern Approach for Motor Imagery Classification.

This paper presents a novel feature extraction approach for motor imagery classification overcoming the weakness of conventional common spatial pattern (CSP) methods, especially for small sample settings. We consider local CSPs generated from individual channels and their neighbors (termed "local regions") rather than a global CSP generated from all channels. The novelty is to select a few good local regions using interquartile range (IQR) or an "above the mean" rule based on variance ratio dispersion score (VRDS) and inter-class feature distance (ICFD); instead of computationally expensive cross-validation method. Furthermore, we develop frequency optimization using filter banks by extending the VRDS and ICFD to frequency-optimized local CSPs. The proposed methods are tested on three publicly available brain-computer interface (BCI) datasets: BCI competition III dataset IVa, BCI competition IV dataset I, and BCI competition IV dataset IIb. The proposed method exhibits substantially improved classification accuracy compared to recent related motor imagery (MI) classification methods.

Complex artifact suppression using vestigial sideband filter in Fourier-domain optical coherence tomography.

In order to achieve computationally efficient mirror image rejection during the off-pivot, full-range approach in spectral-domain optical coherence tomography, we used a vestigial sideband (VSB) filter in place of a Hilbert transform. The appropriate choice of the VSB filter parameters enabled almost complete removal of one sideband with much reduced computational load. To determine the optimal filter parameters, we acquired images of the infrared card and analyzed the mirror suppression ratio of the card surface. Comparison between images obtained using the two filters revealed that the computational load is reduced by 52.4±0.17% when using the VSB filter as it requires a much shorter truncation length. Finally, we present the anterior segment images of a human volunteer's eye processed using the VSB filter.

Transmitter IQ mismatch compensation in coherent optical OFDM systems using pilot signals.

Transmitter in-phase/quadrature (IQ) mismatch in coherent optical orthogonal frequency division multiplexing (CO-OFDM) systems is difficult to mitigate at the receiver using conventional time domain methods such as the Gram-Schmidt orthogonalization procedure, particularly in the presence of channel distortion. In this paper, we present a scheme that mitigates both transmitter IQ mismatch and channel distortion. We propose a pilot structure to estimate both channel and IQ mismatch, and develop a minimum mean square error compensation method. Numerical results show that the proposed method is effective in reducing transmitter IQ mismatch for a CO-OFDM system.