Method for fuzzifying input signals present at the inputs of a f

Details Method for fuzzifying input signals present at the inputs of a f Method for fuzzifying input signals present at the inputs of a f

1996-10-28

1998-06-09

MacDonald, Allen R.

395 61, 395900, G06G 700

Patent

active

057648557

DESCRIPTION:

BRIEF SUMMARY
BACKGROUND OF THE INVENTION

1. Field of the Invention
The present invention relates to a fuzzy logic processor which contains a fuzzifier which calculates membership values of the input membership functions hit from the digital input signal.
2. Description of the Related Art
The structure and mode of operation of fuzzy logic processors are known (see, for example, the publication by H. Eichfeld, T. Kunemund, M. Klimke "An 8b Fuzzy Coprocessor for Fuzzy Control" ISSCC 93, San Francisco, 24 to 26 Feb. 1993, pages 180, 181, 286). Such a fuzzy logic processor contains a fuzzy circuit, also termed a fuzzifier, whose task consists in using input membership functions stored in a memory--also termed a knowledge base memory (KBM)--to determine the membership values for a digital input signal which is to be fuzzified. This firstly requires a determination of which input membership functions (termed membership function below) are hit by the input signal. Subsequently, the membership values assigned to the input signal can then be fetched from the memory.
It has been customary to date to store the membership functions in a point-by-point fashion. Assuming that the input signal is resolved with 8 bits, the membership functions were stored in a pointwise fashion with 256 points. This has advantages. The calculation of the membership values is performed by means of memory access (a sampling method). Any desired shape of the function is permitted. However, the disadvantage resides in that the required storage space is an exponential function of the resolution of the input signals, with the result that in the case of a resolution with 12 bits, for example, the sampling method leads to large memory requirements.
When searching for other possibilities of storing the membership functions, in which the memory requirements turn out to be smaller, it must always be borne in mind that the outlay on calculating time for calculating the membership values must not become excessively large.


SUMMARY OF THE INVENTION

The problem on which the invention is based thus consists in specifying a fuzzification method in which the memory requirements for storing the membership functions do not become excessively large and the calculating time for calculating the membership values nevertheless remains within acceptable limits. The problem consists, furthermore, in specifying a circuit arrangement for carrying out the method.
The problem is solved by means of a method for storing input membership functions and for fuzzifying digital input signals present at the inputs of a fuzzy processor by using these input membership functions, information containing the shape of the respective membership function, wherein ae=number of the bits, of each input is divided into elementary intervals, elementary interval the respective values of restart points, at which the slope of the membership function changes inside the elementary interval or the membership function intersects the boundaries of the elementary interval, and contains the slopes of the membership function which start from the restart points and extend in the elementary interval, performed in accordance with the division condition that each membership function inside an elementary interval, has precisely one restart point, the elementary interval are hit by the input signal, each membership function hit. A circuit arrangement for carrying out the method carries out the fuzzification of input signals using a latch memory connected to a bus to the memory for storing the first memory word, read out from the memory, of the sector for the membership functions, type number from the first memory word, multiplexer and a comparator circuit to which at the second input the number of membership functions per elementary interval is fed by the latch memory, the values xs, ys, p, q, according to the formula:
The memory requirements for storing the membership values can be reduced by virtue of the fact that shape information is stored which records the shape of the membership function. This shape informatio

REFERENCES:
patent: 5410633 (1995-04-01), Ota et al.
patent: 5566274 (1996-10-01), Ishida et al.
patent: 5615303 (1997-03-01), Abruzzese et al.
Eichfeld et al., "Architecture of a CMOS Fuzzy Logic Controller with Optimized Memory Organisation and Operator Design", Proceedings of the International Conference on Fuzzy Systems, 8 Mar. 1992, San Diego, pp. 1317-1323.
Eichfeld et al., "An 8b Fuzzy Coprocessor for Fuzzy Control", 1993 IEEE International Solid-State Circuits Conference, San Francisco, pp. 180, 181, 286.
Thaler, "Neurogenerator Vereinfacht Fuzzy-Entwicklung", Design & Elektronik-Zukunftstechnologien, pp. 10-14, Sep. 1993.

Affiliated with

Also associated with

Rating

Method for fuzzifying input signals present at the inputs of a f does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Method for fuzzifying input signals present at the inputs of a f, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method for fuzzifying input signals present at the inputs of a f will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-2211697