Signal Separation Method Research Articles

Signal Separation Method for Tower Mode Shape Measurement

Vibration shapes of the fundamental modes of eight-legged, K-braced, offshore towers undergo significant changes when important braces are damaged. These changes, measurable at deck-level, provide valuable clues for detecting damage. Because of the small difference between vibration frequencies of the two fundamental translational modes in these towers, the mode shapes are difficult to measure. A simple method is presented which solves this difficulty. The method uses transient acceleration response records that can be produced by different excitation procedures, and no knowledge of the input that causes the excitation is necessary. The finite Fourier transform of such records is examined, and its important properties are pointed out for low damping. Based on these properties, procedures are set forth by which mode shape data can be retrieved from closely spaced peaks in the spectra. The retrieval is done by separating these peaks from each other. Applications of the method in a numerical experiment, and in a physical experiment on a 1/50th scale offshore platform model, are presented. Application of the method to full-scale towers is discussed.

Analogue adaptive hybrid for digital transmission on subscriber lines

Duplex transmission of digital signals on the single-pair subscriber line is possible if the signals in the two directions are sufficiently separated. An analogue adaptive hybrid is one method of signal separation. We discuss here the theory of analogue adaptive hybrids. We present a model for the system and the adaptive hybrid, and show the result of simulation. The parameters of the adaptive hybrid converge fast to the correct values if no signal is present from the far end.

Optical methods for separation of signals from a periodic noise background

Optical methods for separation of signals from a periodic noise background

Method and apparatus for control signal separation to regain synchronization between a visual image projector and an audio program

First Page

Pattern classification of renal electrolyte cycles in primary aldosteronism

The cycles of urinary Na, K, and Cl in six controls and in four patients with aldosterone-secreting adenoma were studied over several days of constant intake and controlled posture ( R = 24-hr recumbency; S d = erect position 0700–1900 hr). The logarithmically transformed data are analyzed by Fourier series, analysis of variance and power spectrum. Signal analysis indicates that the Fourier mean and coefficients of the 24-hr harmonic from S d are the pattern components of choice for training a Pattern Classifier. The classifier is a committee machine composed of a decision-maker for each electrolyte. The decision-makers are based upon the Ho-Kashyap algorithm for linear separability; the classification of the individual as normal or primary aldosteronism patient is given by the committee by a majority decision. The machine proved reliable and highly efficient (error rate <5%) when tested by the data from twelve normal subjects. Of two additional patients with clinical primary aldosteronism, one (adrenal adenoma at laparotomy) was correctly classified; the second one (not operated upon yet) was placed in the “no decision” region. Pattern classification, being deterministic, requires no assumption of data distribution; it may also achieve statistical significance with a sufficiently large number of training sets. Its first successful application to renal electrolyte cycles in primary aldosteronism may be expanded and upgraded by including other diseases and other methods of signal analysis and separation.