Recent Patents on Signal Processing (Discontinued) - Current Issue

This paper reviews the digital signal processing (DSP) methods used in coherent fiber optic communication systems. With the advances in high speed DSP, linear impairments such as chromatic dispersion (CD) and polarization mode dispersion (PMD) can be compensated for using fixed/adaptive equalizers. It is also possible to compensate for the interplay between dispersion and nonlinearity by digital back propagation (DBP), in which the virtual fibers whose signs of dispersion, loss and nonlinear coefficients are opposite of those of the transmission fiber, are realized in the digital domain by numerically solving the nonlinear Schrdinger equation (NLSE). DSP equalization enhances the transmission performance and error-free reach of coherent fiber optic systems significantly.

When liquid steel is tapped from a basic oxygen furnace (BOF), it is essential to minimize the quantity of slag carry-over, because the high level slag may result in the oxidation and the phosphorus reversion during the secondary metallurgy process. The system with the infrared thermal camera is widely used to detect the slag. However, two problems still exist. The first problem is the tracking of tapping stream. The second problem is to automatically discriminate between the slag and the steel according to their different infrared emissivities. In this paper, a system used to inspect the slag content is designed. A method that can quickly identify and track the tapping stream is proposed to only measure slag from the area identified as the stream. This reduces any errors caused by background heat sources in the field of view. A method to discriminate between the slag and steel is also proposed. Experimental results prove that the quantity of slag carry-over is decreased with using the system.

An image processing patent technique for the improvement of entropic compression with digital images has been presented. “Twodimension wavelet transform” (2D-DWT), “two-dimension discrete cosine transform” (2D-DCT) and multi-resolution perfect reconstruction structures, with quincunx filter and grid are used for image processing and simulations using MATLAB, up to 2-level of analysis and synthesis. The possibility of suppressing the high and middle frequencies components in order to increase compression rates was considered. As a result, higher rates of entropic compression were obtained with the application of 2D-DCT after 2D-DWT in the sub-images generated from the multiresolution perfect reconstruction structures. High quality reconstructed images were obtained, even when high and middle frequencies components were suppressed.

The “Internet-of-Things" (IoT) paradigm has been extensively focused in the last years, since it covers a wide scope of applications, that range from spam detection in e-mail datasets to on-body sensors. However, this data flooding, also known as “big data" phenomenon, has required high levels of security, since private and important data are now surfing over our heads. In this paper, we introduced a recently developed pattern recognition technique called Optimum-Path Forest (OPF) for the task of data mining in IoT-oriented applications, such as spam detection in e-mail- and web-content, as well as intrusion detection in computer networks. An extensive comparison against fourteen classification techniques can drive us a picture about the effectiveness and efficiency of OPF classifier in that context.

The monitoring of the power level during nuclear reactors operation is done by instruments, which measure the neutron flux. The calibration of this power is carried out by thermal means. Its accurate knowledge is critical to the fuel burnup calculations, and mainly to electrical power. Nuclear fuel burning is proportional to the thermal power released by the reactor and is extremely relevant for the calculation of uranium-235 mass burned, profile knowledge of fission products, the value of the radioactive decay heat, the activity and radiotoxicity. The objective of this paper is to present two methods for real time monitoring of nuclear reactor power using thermal processes. These methods have been developed and experimentally validated in the TRIGA nuclear research reactor at Nuclear Technology Development Center – CDTN (Belo Horizonte, Brazil), and are the focus of the paper described here. One uses the temperature difference between an instrumented fuel element and the pool water cooling below the reactor core. This method is innovative and the most reliable way of on-line power monitoring of the CDTN TRIGA reactor. For this methodology, a patent was deposited at the Brazilian Institute of Industrial Property (INPI) in 2012. The other method consists in real-time monitoring of the steady-state energy balance of the primary and secondary reactor cooling loops. This process is now the standard methodology for power calibration used in the CDTN TRIGA reactor. It also presents the uncertainty analysis for these thermal power measure procedures.

The output of an N-point FFT (Fast Fourier Transform) is zero at odd positions if its inputs at odd positions with N/2 distance are equal and this property is exploited in order to reduce the power consumed by the FFT circuitry when operating on sparse data. An OFDM (Orthogonal Frequency Division Multiplexing) transceiver with such a configurable FFT is described, where undersampling is applied in order to reduce the ADC (Analog Digital Conversion) power and the buffering requirements by up to 25% or even 37.5% if additional FFT properties are exploited. As will be shown in this paper, up to 3/8 of the FFT butterfly operations can be deactivated in this case without any effect on the error caused by the undersampling procedure. Deactivating a higher number of FFT operations is also possible with a small error overhead. The reconstruction of the sparse images captured by a surveillance camera as well as X-ray images is also used as case studies demonstrating the use of the proposed method in image transfer applications over OFDM infrastructures. The Normalized Mean Square Error (NMSE) in these case studies is as low as 0.01 if 12.5% of the FFT input samples are omitted by the undersampling process.