Data processing: artificial intelligence – Neural network
Reexamination Certificate
2000-10-04
2004-04-06
Starks, Jr., Wilbert L. (Department: 2121)
Data processing: artificial intelligence
Neural network
C706S022000, C706S025000
Reexamination Certificate
active
06718316
ABSTRACT:
STATEMENT OF THE GOVERNMENT INTEREST
The invention described herein may be manufactured and used by or for the Government of the United States of America for Governmental purposes without the payment of any royalties thereon or therefore.
BACKGROUND OF THE INVENTION
(1) Field of the Invention
The present invention relates generally to signal processing and, more specifically, to a neural network trained to determine when an input deviates from pure noise characteristics.
(2) Description of the Prior Art
Prior art signal processors attempt to detect the presence of an object by filtering out background noise and applying detection techniques. These detectors try to identify whether a signal is embedded in background noise by comparing, for example, the received waveform with a model of the signal to see if there is any correlation. One disadvantage with these techniques is that the transmitted signal may become distorted because the amplitude, phase and frequency characteristics of the transmitted signal are adversely affected as the signal propagates through the medium. Hence, detection performance decreases. Such signal distortion may occur in an environment where a sinusoidal pulse impacts an object, traverses multiple paths and combines in an unfavorable manner at the receiver array. In an underwater acoustic environment, for instance, an adverse multipath effect occurs when multiple reflected signals propagate through the ocean after a transmitted signal has impacted an underwater object like another vehicle. Multipath effects are also present in most types of radio and wireless communications resulting in reduced detectability.
Artificial neural networks (ANN) are commonly referred to as neural networks or neural nets. Neural networks may typically be comprised of many very simple processors, commonly referred to as units or neurons, each normally having an allocated amount of local memory. The units may typically be connected by unidirectional communication channels or connections, which may carry numeric as opposed to symbolic data. The units operate only on their local data and on the inputs they receive via the connections. An artificial neural network is a processing device, either software or actual hardware, whose design was inspired by the design and functioning of neural networks such as biological nervous systems and components thereof. Most neural networks have some sort of training rule whereby the weights of connections may be adjusted on the basis of presented patterns. Neural networks learn from examples, just like children learn to recognize dogs from examples of dogs, and exhibit some structural capability for generalization. The term “neural net” should logically, but in common usage never does, also include biological neural networks, whose elementary structures are far more complicated than the mathematical models used for ANNs.
The patents discussed below describe use of a neural network to act as a detector wherein an attempt is made to recognize a signal pattern within noise.
U.S. Pat. No. 5,402,520, issued Mar. 28, 1995, to B. Schnitts, discloses an apparatus for retrieving signal embedded in noise and analyzing the signals. The apparatus includes an input device for receiving input signals having noise. At least one filter retrieves data signals embedded in the input signals. At least one adaptive pattern recognition filter generates coefficients of a polynomial expansion representing the pattern of the filtered data signals. A storage device stores the coefficients generated. It is determined when an event has occurred, the event being located at any position within the data signals. An adaptive autoregressive moving average pattern recognition filter generates coefficients of a polynomial expansion representing an enhanced pattern of filtered data signals. At least one weighting filter compares the stored patterns with the enhanced pattern of data signals. The neural network is trained to recognize and predict signal patterns within noise as discussed above, e.g., stock price patterns, rather than to recognize noise itself.
U.S. Pat. No. 5,778,152, issued Jul. 7, 1998, to Oki et al., discloses a neural network designed to recognize a particular character. The network is supplied with initial tap weights for a first hidden node, which are an image of the character to be recognized. The additive inverse of this set of weights is used as the tap weights for a second hidden node. A third node, if used is initialized with random noise. The network is then trained with back propagation. The neural network is trained to recognize signal patterns within noise, e.g., letters, rather than to recognize noise itself.
The above patents do not address the value or approach of recognizing noise itself. For certain types of waveforms, particularly those which may or may not contain a signal embedded in noise, this type of information is especially useful for efficient detection. Consequently, it would be desirable to provide a neural network trained to detect noise and programmed to indicate if any non-noise anomalies are present. Those skilled in the art will appreciate the present invention that addresses the above and other needs and problems.
SUMMARY OF THE INVENTION
Accordingly, it is an object of the present invention to provide an improved signal detector.
It is yet another object of the present invention to provide a means for determining the presence or absence of a non-noise component within noise.
These and other objects, features, and advantages of the present invention will become apparent from the drawings, the descriptions given herein, and the appended claims.
In accordance with the present invention, a method is provided for determining the presence or absence of a non-noise anomaly within noise by processing a received waveform including steps such as producing a plurality of samples of the received waveform and applying the plurality of samples to one or more initial neural networks. Each of the one or more initial neural networks may be trained to recognize noise. The initial neural networks produce one or more respective outputs related to the presence or absence of the non-noise anomaly. Another step includes analyzing the one or more respective outputs of the one or more initial neural networks to determine if the non-noise anomaly is present in the received waveform. The step of analyzing may further comprise applying the one or more outputs to a decision making circuit for determining if a non-noise anomaly is present in the received waveform.
The step of producing a plurality of samples may further comprise dividing the received waveform into one or more windows whereupon the received waveform within each of the one or more windows is sampled and applied to a respective one of the one or more initial neural networks. The one or more windows may be incremented so as to slide relative to the received waveform with each increment such that the windows are incremented until all of the received waveform is sampled. Another step may include storing the respective outputs from the one or more initial neural networks in a database.
In one example, the initial neural networks are trained to recognize Gaussian noise. The step of analyzing may include calculating standard deviations related to the respective outputs.
The anomaly recognition system of the present invention comprises a plurality of initial neural networks, wherein each of the plurality of initial neural networks may be programmed for recognizing noise. The plurality of initial neural networks may produce a respective plurality of outputs related to the presence or absence of a non-noise anomaly. A decision making aid is preferably provided for receiving and evaluating the plurality of outputs from the neural networks. The decision making aid may be programmed to determine if a non-noise element is present or not after analyzing the plurality of outputs. The system may further comprise a plurality of sampling members for providing a plurality of samples of the received waveform for
Booker Kelvin
Kasischke James M.
Nasser Jean-Paul A.
Oglo Michael F.
Starks, Jr. Wilbert L.
LandOfFree
Neural network noise anomaly recognition system and method does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Neural network noise anomaly recognition system and method, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Neural network noise anomaly recognition system and method will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3244065