The Sum-Line Extrapolative Algorithm and Its Application to Statistical Classification Problems

This paper describes an investigation of the capability of a two-level adaptive linear threshold element (LTE) system to perform speaker discriminations. The study also includes an investigation of discriminating a speaker from an unknown population. The problem has been confined to the verification of an utterance as that of an expected informant. The environment of the experiments is discussed, and the experimental system is described. At the first level LTE, four different kinds of training have been developed for effective transformation and data reduction. At the second-level LTE, different training conditions and different decision processes are investigated and evaluated. Over 90% accuracy is obtained in separating a known speaker from impostors.

Active power filter requires accurate harmonic current identification to compensate harmonics in power system distribution. The purpose of this paper is to present a comparative study of two techniques for harmonics currents identification based on P-Q theory and Adaline(ADAptiveLINear Element) neural networks. The harmonic current can be identified from powers or currents. The first method is based on the instantaneous powers taking advantage from relationship between load currents and power transferred from the supply source to the loads. The second method concerns the artificial neural networks based on the LMS (least mean square) algorithm. This approach adjusts the weights by iteration and provides more flexibility to perform the compensation. The developed architecture is validated by computer simulation proving its effectiveness, capability and robustness.

Classical linear neural network architectures, such as the optimal linear associative memory (OLAM) Kohonen and Ruohonen (IEEE Trans Comp 22(7):701---702, 1973) and the adaptive linear element (Ada...

This paper presents improvement of a harmonics extraction algorithm, known as the fundamental active current (FAC) adaptive linear element (ADALINE) neural network with the integration of photovoltaic (PV) to shunt active power filters (SAPFs) as active current source. Active PV injection in SAPFs should reduce dependency on grid supply current to supply the system. In addition, with a better and faster harmonics extraction algorithm, the SAPF should perform well, especially under dynamic PV and load conditions. The role of the actual injection current from SAPF after connecting PVs will be evaluated, and the better effect of using FAC ADALINE will be confirmed. The proposed SAPF was simulated and evaluated in MATLAB/Simulink first. Then, an experimental laboratory prototype was also developed to be tested with a PV simulator (CHROMA 62100H-600S), and the algorithm was implemented using a TMS320F28335 Digital Signal Processor (DSP). From simulation and experimental results, significant improvements in terms of total harmonic distortion (THD), time response and reduction of source power from grid have successfully been verified and achieved.

Harmonics and interharmonics may introduce operational problems in electrical and electronic equipment. Therefore, monitoring harmonics/interharmonics for improving the power quality is of importance for both electric utilities and their customers. In this paper, a cascaded two-stage adaptive linear element (ADALINE) structure for both harmonics and interharmonics measurement is proposed. In addition, a simple laboratory setup implemented by MATLAB and the dedicated hardware for measuring power signals is built to verify the performance of proposed method. Results are compared with those obtained by short-time Fourier transform and two other conventional ADALINE-based methods. It shows that the proposed method is with a better accuracy, even if the power frequency deviation and interharmonic components are present in the measured signal. The proposed method also can be adopted for harmonic/interharmonic compensation devices in real time.

In this paper, we derive and analyze a Bayesian detector for a multiple access ultra-wideband (UWB) system, which only requires the knowledge of the signature code of the user of interest. The receiver structure is equivalent to a single user matched filter (MF) receiver followed by an adaptive threshold. Since the multiple access interference (MAI) is not Gaussian, this detector outperforms the conventional MF detector at the expense of the additional complexity necessary for calculating the adaptive threshold. In the two-user case, as the AWGN tends to zero, the probability of error of the Bayesian detector goes to zero for any finite-length time-hopping codes. This is in contrast to the MF receiver, which requires infinitely long codes to achieve the same result. Finally, we propose a linear MMSE adaptive receiver to mitigate MAI with low complexity.

The combination of multi-input multi-output (MIMO) millimeter-wave radar sensor and synthetic aperture radar (SAR) makes the implementation of 3-D near-field millimeter-wave imaging cheaper and more practical. In this paper, a 3-D SAR imaging system based on the millimeter-wave sensor is constructed, through experiment contrast and analyzed based on the theory of matched filtering (MF) imaging algorithm and based on the theory of compressed sensing (CS) in the nearfield millimeter-wave imaging algorithm of 3-D imaging results of SAR system, the pros and cons of two kinds of algorithms are verified in theory, the imaging algorithm based on CS theory imaging effect is better than conventional MF imaging algorithm of imaging effect is concluded, along with various real imaging results.

Least Mean Square (LMS) algorithm is undoubtedly the most resorted to algorithm in diverse fields of engineering. Due to its simplicity it has been applied to solve numerous problems including side lobe reduction in matched filters, adaptive equalization, system identification, adaptive noise cancellation etc. In this paper we present a simple architecture for the implementation of a variant of Block LMS algorithm where the weight updation and error calculation are both calculated block wise. The algorithm performs considerably well with a slight trade off in the learning curve time and misadjustment, both of which can be adjusted by varying the step size depending on the requirement. The architecture can be further modified to perform the variants of LMS algorithm such as sign-sign, signerror and sign-data algorithms. The performance of the Simplified BLMS and LMS algorithms are compared in MATLAB simulations and the hardware outputs from the FPGA are verified with the simulations.

We revisit the problem of searching for gravitational waves from inspiralling compact binaries in Gaussian coloured noise. For binaries with quasicircular orbits and non-precessing component spins, considering dominant mode emission only, if the intrinsic parameters of the binary are known then the optimal statistic for a single detector is the well-known two-phase matched filter. However, the matched filter signal-to-noise ratio is /not/ in general an optimal statistic for an astrophysical population of signals, since their distribution over the intrinsic parameters will almost certainly not mirror that of noise events, which is determined by the (Fisher) information metric. Instead, the optimal statistic for a given astrophysical distribution will be the Bayes factor, which we approximate using the output of a standard template matched filter search. We then quantify the possible improvement in number of signals detected for various populations of non-spinning binaries: for a distribution of signals uniformly distributed in volume and with component masses distributed uniformly over the range $1\leq m_{1,2}/M_\odot\leq 24$, $(m_1+m_2) /M_\odot\leq 25$ at fixed expected SNR, we find $\gtrsim 20\%$ more signals at a false alarm threshold of $10^{-6}\,$Hz in a single detector. The method may easily be generalized to binaries with non-precessing spins.

Linear threshold elements (LTEs) are the basic processing elements in artificial neural networks. An LTE computes a function that is a sign of a weighted sum of the input variables. The weights are arbitrary integers; actually they can be very big integers-exponential in the number of input variables. However, in practice, it is very difficult to implement big weights. So the natural question that one can ask is whether there is an efficient way to simulate a network of LTEs with big weights by a network of LTEs with small weights. We prove the following results: 1) every LTE with big weights can be simulated by a depth-3, polynomial size network of LTEs with small weights, 2) every depth-d polynomial size network of LTEs with big weights can be simulated by a depth-(2d+1), polynomial size network of LTEs with small weights. To prove these results, we use tools from harmonic analysis of Boolean functions. Our technique is quite general, it provides insights to some other problems. For example, we were able to improve the best known results on the depth of a network of threshold elements that computes the COMPARISON, ADDITION and PRODUCT of two n-bits numbers, and the MAXIMUM and the SORTING of n n-bit numbers.

This paper presents the results of an experimental investigation about target detecting with passive sonar in Persian Gulf. Detecting propagated sounds in the water is one of the basic challenges of the researchers in sonar field. This challenge will be complex in shallow water (like Persian Gulf) and noise less vessels. Generally, in passive sonar, the targets are detected by sonar equation (with constant threshold) that increases the detection error in shallow water. The purpose of this study is proposed a new method for detecting targets in passive sonars using adaptive threshold. In this method, target signal (sound) is processed in time and frequency domain. For classifying, Bayesian classification is used and posterior distribution is estimated by Maximum Likelihood Estimation algorithm. Finally, target was detected by combining the detection points in both domains using Least Mean Square (LMS) adaptive filter. Results of this paper has showed that the proposed method has improved true detection rate by about 24% when compared other the best detection method.

The paper proposes a new noncoherent detection algorithm, based on the maximum-likelihood (ML) criterion, for orthogonal multicode (OM) systems in Nakagami fading channels. Simulation results show that the new ML algorithm can provide at least 0.7 dB performance gain over the conventional matched-filter (MF) algorithm in uncoded systems. Finally, a simplified version of the ML algorithm with significant complexity reduction and negligible performance degradation is also presented.

Cubic boron nitride (cBN) synthesis has gained lot of interest during the past decade as it offers outstanding physical and chemical properties like high hardness, high wear resistance, and chemical inertness. Despite of their excellent properties, every application of cBN is hindered by high compressive stresses and poor adhesion. The cost of equipment is also high in almost all the techniques used so far. This thesis deals with the synthesis of cubic phase of boron nitride on Si (100) wafers using electron beam evaporator, a low cost equipment that is capable of depositing films with reduced stresses. Using this process, need of ion beam employed in ion beam assisted processes can be eliminated thus reducing the surface damage and enhancing the film adhesion. Four sets of samples have been deposited by varying substrate temperature and the deposition time. scanning electron microscopy (SEM), energy dispersive spectroscopy (EDS), X-ray photoelectron spectroscopy (XPS), and Fourier transform infrared spectroscopy (FTIR) techniques have been used to determine the structure and composition of the films deposited. X-ray diffraction (XRD) was performed on one of the samples to determine the thickness of the film deposited for the given deposition rate. Several samples showed dendrites being formed as a stage of film formation. It was found that deposition at substrate temperature of 400oC and for a period of one hour yielded high quality cubic boron nitride films.

There is a wide interest in the segmentation of retinal vessel structure from fundus images of preterm infants. This has lead to focus on the processing of dark or low resolution images acquired with portable medical cameras. The concept of signal matched filter (MF) has been used to recognize the retinal vascular structure and the matched filter and first-order derivative of Gaussian (MFFDOG) method was introduced to get a binarized vascular network. In this paper we present a new variation of MF for segmenting retinal vessels network. We apply the original MF to distinguish the vessels network in a gray scale image. Then we obtain the vascular network as a binary image using an adaptive thresholding based on our denominated hard kernel (MFHK). We have also generalized the application of matched filter with any filter function and any domain. We assessed both methods MFFDOG and MFHK in a set of fifty images of children with retinopathy of prematurity (ROP).

An approximate degrees of freedom solution to the multivariate Behrens-Fisher problem* Ying Yao Ying Yao Bell Telephone Laboratories, Inc., Holmdel, New Jersey Search for other works by this author on: Oxford Academic Google Scholar Biometrika, Volume 52, Issue 1-2, June 1965, Pages 139–148, https://doi.org/10.1093/biomet/52.1-2.139 Published: 01 June 1965