Nuha Abdul Sahib Alwan

In this work, deep learning (DL) is incorporated in the form of deep neural networks (DNN) into the design of a class of neuroadaptive control systems, namely model reference adaptive inverse control (MRAIC) systems. As inverse control is essentially feedforward control, it is much simpler to design and analyze than most current control methods especially when considering the control of nonlinear plants. Using the filtered-ε adaptation method with deep adaptive controller, it is demonstrated that the nonlinear plant output tracks the reference model output in the all-deep MRAIC system much more efficiently than the MRAIC system with a shallow adaptive controller. First- and second-order reference models have been experimented with. The proposed deep MRAIC system is also robust to plant parameter change

The problem of frequency estimation of a single sinusoid observed in colored noise is addressed. Our estimator is based on the operation of the sinusoidal digital phase-locked loop (SDPLL) which carries the frequency information in its phase error after the noisy sinusoid has been acquired by the SDPLL. We show by computer simulations that this frequency estimator beats the Cramer-Rao bound (CRB) on the frequency error variance for moderate and high SNRs when the colored noise has a general low-pass filtered (LPF) characteristic, thereby outperforming, in terms of frequency error variance, several existing techniques some of which are, in addition, computationally demanding. Moreover, the present approach generalizes on existing work that addresses different methods of sinusoid frequency estimation involving specific colored noise models such as the moving average (MA) noise model. An insightful theoretical analysis is presented to support the practical findings.

An adaptive deep neural network is used in an inverse system identification setting to approximate the inverse of a nonlinear plant with the aim of constituting the plant controller by copying to the latter the weights and architecture of the converging deep neural network. This deep learning (DL) approach to the adaptive inverse control (AIC) problem is shown to outperform the adaptive filtering techniques and algorithms normally used in adaptive control, especially when in nonlinear plants. The deeper the controller, the better the inverse function approximation, provided that the nonlinear plant has an inverse and that this inverse can be approximated. Simulation results prove the feasibility of this DL-based adaptive inverse control scheme. The DL-based AIC system is robust to nonlinear plant parameter changes in that the plant output reassumes the value of the reference signal considerably faster than with the adaptive filter counterpart of the deep neural network. The settling and rise times of the step response are shown to improve in the DL-based AIC system.

An adaptive deep neural network is used in an inverse system identification setting to approximate the inverse of a nonlinear plant with the aim of constituting the plant controller by copying to the latter the weights and architecture of the converging deep neural network. This deep learning (DL) approach to the adaptive inverse control (AIC) problem is shown to outperform adaptive filtering techniques and algorithms normally used in adaptive control, especially when the plant is nonlinear. The deeper the controller the better the inverse function approximation, provided that the nonlinear plant have an inverse and that this inverse can be approximated. Simulation results prove the feasibility of this DL-based adaptive inverse control scheme. The DL-based AIC system is robust to parameter change of the nonlinear plant in that, under such change, the plant output reassumes the value of the reference signal considerably faster than with the adaptive filter counterpart of the deep neural network. Settling times and rise times of the step response are shown to improve in the DL-based AIC system.

This work combines compressive sensing and short word-length techniques to achieve localization and target tracking in wireless sensor networks with energy-efficient communication between the network anchors and the fusion center. Gradient descent localization is performed using time-of-arrival (TOA) data which are indicative of the distance between anchors and the target thereby achieving range-based localization. The short word-length techniques considered are delta modulation and sigma-delta modulation. The energy efficiency is due to the reduction of the data volume transmitted from anchors to the fusion center by employing any of the two delta modulation variants with compressive sensing techniques. Delta modulation allows the transmission of one bit per TOA sample. The communication energy efficiency is increased by RⱮ, R ≥ 1, where R is the sample reduction ratio of compressive sensing, and Ɱ is the number of bits originally present in a TOA-sample word. It is found that the localization system involving sigma-delta modulation has a superior performance to that using delta-modulation or pure compressive sampling alone, in terms of both energy efficiency and localization error in the presence of TOA measurement noise and transmission noise, owing to the noise shaping property of sigma-delta modulation.

This work combines compressive sensing and short word-length techniques to achieve localization and target tracking in wireless sensor networks with energy-efficient communication between the network anchors and the fusion center. Gradient descent localization is performed using time-of-arrival (TOA) data which are indicative of the distance between anchors and the target thereby achieving range-based localization. The short word-length techniques considered are delta modulation and sigma-delta modulation. The energy efficiency is due to the reduction of the data volume transmitted from anchors to the fusion center by employing any of the two delta modulation variants with compressive sensing techniques. Delta modulation allows the transmission of one bit per TOA sample. The communication energy efficiency is increased by RⱮ, R≥1, where R is the sample reduction ratio of compressive sensing and Ɱ is the number of bits originally present in a TOA-sample word. It is found that the localization system involving sigma-delta modulation has a superior performance to that using delta-modulation or pure compressive sampling alone, in terms of both energy efficiency and localization error in the presence of TOA measurement noise, owing to the noise shaping property of sigma-delta modulation.

Frequency estimation of a single sinusoid in colored noise has received a considerable amount of attention in the research community. Taking into account the recent emergence and advances in compressive covariance sensing (CCS), the aim of this work is to combine the two disciplines by studying the effects of compressed measurements of a single sinusoid in moving-average colored noise on its frequency estimation accuracy. CCS techniques can recover the second-order statistics of the original uncompressed signal from the compressed measurements, thereby enabling correlation-based frequency estimation of single tones in colored noise using higher order lags. Acceptable accuracy is achieved for moderate compression ratios and for a sufficiently large number of available compressed signal samples. It is expected that the proposed method would be advantageous in applications involving resource-limited systems such as wireless sensor networks.

Frequency estimation of a single sinusoid in colored noise has received a considerable amount of attention in the research community. Taking into account the recent emergence and advances in compressive covariance sensing (CCS), the aim of this work is to combine the two disciplines by studying the effects of compressed measurements of a single sinusoid in moving-average (MA) colored noise on its frequency estimation accuracy. CCS techniques can recover the second-order statistics of the original uncompressed signal from the compressed measurements, thereby enabling correlation-based frequency estimation of single tones in colored noise using higher-order lags. Acceptable accuracy is achieved for moderate compression ratios and for a sufficiently large number of available compressed signal samples. It is expected that the proposed method would be advantageous in applications involving resource-limited systems such as wireless sensor networks.

Training data for a deep learning (DL) neural network (NN) controller are obtained from the input and output signals of a conventional digital controller that is designed to provide the suitable control signal to a specified plant within a feedback digital control system. It is found that if the DL controller is sufficiently deep (four hidden layers), it can outperform the conventional controller in terms of settling time of the system output transient response to a unit-step reference signal. That is, the DL controller introduces a damping effect. Moreover, it does not need to be retrained to operate with a reference signal of different magnitude, or under system parameter change. Such properties make the DL control more attractive for applications that may undergo parameter variation, such as sensor networks. The promising results of robustness against parameter changes are calling for future research in the direction of robust DL control.

In this work, an estimate of the power spectrum of a real-valued wide-sense stationary autoregressive signal is computed from sub-Nyquist or compressed measurements in additive white Gaussian noise. The problem is formulated using the concepts of compressive covariance sensing and Blackman-Tukey nonparametric spectrum estimation. Only the second-order statistics of the original signal, rather than the signal itself, need to be recovered from the compressed signal. This is achieved by solving the resulting overdetermined system of equations by application of least squares, thereby circumventing the need for applying the complicated ℓ 1 -minimization otherwise required for the reconstruction of the original signal. Moreover, the signal need not be spectrally sparse. A study of the performance of the power spectral estimator is conducted taking into account the properties of the different bases of the covariance subspace needed for compressive covariance sensing, as well as different linear sparse rulers by which compression is achieved. A method is proposed to benefit from the possible computational efficiency resulting from the use of the Fourier basis of the covariance subspace without considerably affecting the spectrum estimation performance.