Boots – shoes – and leggings
Patent
1994-04-11
1995-08-15
Gordon, Paul P.
Boots, shoes, and leggings
364151, 395 23, G05B 1304, G06F 1518
Patent
active
054425433
DESCRIPTION:
BRIEF SUMMARY
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates generally to a filter architecture using a neural network, and to a method and an apparatus for implementing the architecture.
2. Description of the Related Art
Neural networks are already used in a wide range of applications of adaptive non-linear filters. In the case of most of these applications, the number of input variables of the network is small and the amount of noise on the training signal is relatively low. If these conditions do not prevail, it is frequently difficult to train a network, particularly if rapid adaptation is required.
The transfer characteristic of a general, discrete linear filter can be described by the formula ##EQU1## In this case, g(n) designates the output variable of the linear filter at time n, f(n-i) designates the input variable of the filter at time n-i, and the functions k and r designate the response functions of the discrete linear filter. A so-called FIR filter (filter with a finite pulse response) is provided for the case in which the function r disappears at all times i. The output function of the filter is in this case a linear superposition of instantaneous and preceding input signals and preceding output signals.
This filter structure can be generalized to form a non-linear filter architecture, which is given by the formula input and output signals. The non-linear function N is in this case approximated by a neural network, which is the case for the filter architecture proposed by Lapedes and Farber (A. Lapedes, R. Farber, "How neural nets work." In Neural information processing systems, ed. D. Z. Anderson, pages 442-456, New York, American Institute of Physics, 1988) Waibel (A. Waibel, "Modular construction of time-delay neural networks for speech recognition", Neural Computation, Vol. 1, pages 39-46, 1989) has described a neural network in the case of which only the input signals of preceding times are supplied to the input variables.
SUMMARY OF THE INVENTION
In the context of this patent application, it is intended to consider systems in the case of which the properties of the input functions influence the properties of the output function of the system over long periods. Using recursive networks, such systems can be modelled with long-lasting response functions. As an alternative to this, the length of the response function M could be selected to be sufficiently large. This necessitates a large number of input variables for the neural network used, and the training of the network is in consequence inefficient.
The present invention is therefore based on the object of specifying a non-linear filter architecture using which long-lasting filter response functions can be implemented, which filter architecture is at the same time insensitive to noise in the training signal and whose transfer characteristics can be rapidly matched to a specific object (rapid adaptation).
This and other objects and advantages are achieved according to the invention using a non-linear filter architecture having an input signal f(n-i) to the non-linear filter, which input signal is associated with time n-i and is possibly multi-dimensional, a time signal for the time i and a parameter vector p(n-i) associated with time n-i are linked, at time i, to the input nodes of a neural network;
b) the output signal g(n) of the non-linear filter at time n results from summation of M+1 output signals, which are associated with the times n, . . . , n-M, of the neural network, in accordance with the formula ##EQU2## where N designates the output function of the neural network. An adaptation method for a non-linear filter architecture is provided, wherein each weighting w of the neural network is varied in such a manner that a predetermined error E, which represents the value of the deviation, determined over a predetermined time interval, of the output signals g(n) of the filter from the required output signals gm(n), is minimized by varying the weightings of the neural network. A circuit arrangement for imple
REFERENCES:
patent: 4937872 (1990-06-01), Hopfield et al.
patent: 5253329 (1993-10-01), Villarreal et al.
patent: 5313407 (1994-05-01), Tiernan et al.
"A Nonlinear Digital Filter using Multi-Layered Neural Network" by Arakawa, et al., IEEE International Conference on Communications ICC '90 Including Supercomm Technical Sessions, Apr. 15-19, 1990, pp. 424-428.
"Adaptive Processing with Neural Network Controlled Resonator-Banks" by Sztipanovits, IEEE Transactions on Circuits and Systems, vol. 37, No. 11, Nov. 1990, pp. 1436-1440.
"Neural Information Processing Systems" by Lapedes, American Institute of Physics 1988, pp. 442-457.
"Modular Construction of Time-Delay Neural Networks for Speech Recognition" by Waibel, Neural Computation, vol. 1, 1989, pp. 39-46.
Gordon Paul P.
Siemens Aktiengesellschaft
LandOfFree
Neural filter architecture for overcoming noise interference in 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 filter architecture for overcoming noise interference in , we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Neural filter architecture for overcoming noise interference in will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2187365