Signal discrimination device using neural network

• US 5,287,430 A
• Filed: 04/14/1992
• Issued: 02/15/1994
• Est. Priority Date: 04/15/1991
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Denki Kabushiki Kaisha (Mitsubishi Electric Corporation)
• Original Assignee: Mirsubishi Denki Kabushiki Kaisha
• Inventors: Okada, Takamitsu, Iwamoto, Masafumi, Kondoh, Michimasa, Fujisaka, Takahiko
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Shapiro, Stuart
• Application Number:
  US07/868,460
• Time in Patent Office: 672 Days
• Field of Search: 395/22, 395/20, 395/23, 382/14, 382/15
• US Class Current: 706/20
• CPC Class Codes: G01S 7/417 involving the use of neural networks
• CPC Class Codes: G06V 30/194 References adjustable by an adaptive method, e.g. learning

Abstract

A signal discrimination device using a neural network for discriminating input signals such as radar reception signals includes an adaptive code generator means for generating codes for representing the discrimination categories. The distances between the codes for closely related categories are smaller than the distances between the codes for remotely related categories. During the learning stage, the neural network is trained to output the codes for respective inputs. The discrimination result judgment means determines the categories by comparing the outputs of the neural network and the codes for the respective categories.

Claims

1. A signal discrimination device using a neural network for classifying input data signals into a plurality of categories, said signal discrimination device comprising:

input means for supplying input data signals;

parameter supply means for supplying distinct values of parameters for characterizing respective categories into which said input data signals are classified;

adaptive code generator means coupled to said parameter supply means to receive the distinct values of parameters supplied from said parameter supply means, and for generating codes for representing said respective categories on the basis of said values of said parameters supplied from said parameter supply means, wherein distances between said codes correspond to degrees of affinity between said respective categories;

neural network means coupled to said input means and responsive to said codes from said adaptive code generator means for outputting discrimination signals in response to said input data signals, wherein said discrimination signals correspond to said codes for representing said respective categories; and

discrimination result judgment means coupled to said neural network means and responsive to said codes for respective categories from said adaptive code generator, for determining said categories corresponding to respective input data signals by comparing said discrimination signals with said codes for respective categories.

2. A signal discrimination device as claimed in claim 1, wherein a code length L of said codes for representing respective categories is equal to a number M of said parameters;

said parameters are numbered from 1 to M;

said categories are numbered from 1 to N, N being the number of said categories; and

said adaptive code generator means includes;

means for determining a minimum P_j1 and a maximum P_j2 of jth parameter P_j, for respective values of j from 1 to M; and

means responsive to said means for determining a minimum P_j1 and a maximum P_j2 for determining code C_i for an ith category by;

space="preserve" listing-type="equation">C.sub.ij =(P.sub.j -P.sub.j1)/(P.sub.j2 -P.sub.j1)

where i ranges from 1 through N and j ranges from 1 through L.

3. A signal discrimination device as claimed in claim 1, wherein a code length L of said codes for representing respective categories is equal to a number M of said parameters;

said parameters are numbered from 1 to M;

said categories are numbered from 1 to N, N being the number of said categories; and

said adaptive code generator means includes;

means for determining from the distinct values of parameters supplied from said parameter supply means a minimum P_j1, a maximum P_j2, and a minimum separation d_j of a jth one of said parameters, P_j, for respective values of j from 1 to M;

means responsive to said means for determining a minimum for determining a code length L_j for said jth parameter by;

space="preserve" listing-type="equation">L.sub.j =INT((P.sub.j2 -P.sub.j1)/d.sub.j)

where j ranges from 1 to M and a function INT() represents rounding fractions up;

means responsive to said means for determining a code length for setting a total code length L equal to a sum of L_j for respective values of j from 1 to M;

means responsive to said means for determining a minimum for determining a partial code c_j for said jth parameter P_j for an ith category by;

##EQU14## where j and k range from 1 to M and from 1 to L_j, respectively, and c_jk represents a kth component of said partial code c_j ; and

means responsive to said means for determining a partial code for determining a total code C_i for representing said ith category by;

space="preserve" listing-type="equation">C.sub.i =c.sub.1 ⊕

c.sub.2 ⊕

- - - ⊕

c.sub.m

where i ranges from 1 to N and a symbol ⊕

represents a concatenation operator.

4. A signal discrimination device as claimed in claim 1, further comprising:

learning input data generator means coupled to said parameter supply means for generating input learning data signals corresponding to said distinct values of said parameters supplied from said parameter supply means said neural network being responsive to said input learning data signals;

wherein weights of respective nodes of said neural network means are adjusted to output said codes for respective categories in response to said input learning data signals supplied from said learning input data generator means.

5. A signal discrimination device as claimed in claim 1, further comprising:

learning input data selector means, coupled to said neural network, said input means and said parameter supply means, for selecting between input learning signals which are obtained from said input data signals based on said distinct values of parameters and said input data signals supplied from said input means to provide a selection to the neural network;

wherein weights of respective nodes of said neural network means are adjusted to output said codes for respective categories in response to said input learning signals supplied from said learning input data selector means.

6. A signal discrimination device using a neural network for classifying input data signals into a plurality of categories, said signal discrimination device comprising:

input means for supplying input data signals;

parameter supply means for supplying distinct values of parameters for characterizing respective categories into which said input data signals are classified;

adaptive code generator means coupled to said parameter supply means to receive the distinct values of parameters supplied from said parameter supply means, and for generating codes for representing said respective categories on the basis of said values of said parameters supplied from said parameter supply means, wherein distances between said codes correspond to degrees of affinity between said respective categories;

neural network means coupled to said input means for outputting discrimination signals in response to said input data signals, wherein said discrimination signals correspond to said codes for representing said respective categories;

discrimination result judgment means coupled to said neural network means for determining said categories corresponding to respective input data signals by comparing said discrimination signals with said codes for respective categories; and

wherein a code length L of said codes for representing respective categories is equal to a number M of said parameters;

said parameters are numbered from 1 to M;

said categories are numbered from 1 to N, N being the number of said categories; and

said adaptive code generator means includes;

means for determining a minimum P_j1 and a maximum P_j2 of jth parameter P_j, for respective values of j from 1 to M; and

means responsive to said means for determining a minimum P_j1 and a maximum P_j2 for determining code C_i for an ith category by;

space="preserve" listing-type="equation">C.sub.ij =(P.sub.j -P.sub.j1)/(P.sub.j2 -P.sub.j1)

where i ranges from 1 through N and j ranges from 1 through L.

7. A signal discrimination device using a neural network for classifying input data signals into a plurality of categories, said signal discrimination device comprising:

input means for supplying input data signals;

parameter supply means for supplying distinct values of parameters for characterizing respective categories into which said input data signals are classified;

adaptive code generator means coupled to said parameter supply means to receive the distinct values of parameters supplied from said parameter supply means, and for generating codes for representing said respective categories on the basis of said values of said parameters supplied from said parameter supply means, wherein distances between said codes correspond to degrees of affinity between said respective categories;

neural network means coupled to said input means for outputting discrimination signals in response to said input data signals, wherein said discrimination signals correspond to said codes for representing said respective categories;

discrimination result judgment means coupled to said neural network means for determining said categories corresponding to respective input data signals by comparing said discrimination signals with said codes for respective categories; and

wherein a code length L of said codes for representing respective categories is equal to a number M of said parameters;

said parameters are numbered from 1 to M;

said categories are numbered from 1 to N, N being the number of said categories; and

said adaptive code generator means includes;

means for determining from the distinct values of parameters supplied from said parameter supply means a minimum P_j1, a maximum P_j2, and a minimum separation d_j of a jth one of said parameters, P_j, for respective values of j from 1 to M;

means responsive to said means for determining a minimum for determining a code length L_j for said jth parameter by;

space="preserve" listing-type="equation">L.sub.j =INT((P.sub.j2 -P.sub.j1)/d.sub.j)

where j ranges from 1 to M and a function INT() represents rounding fractions up;

means responsive to said means for determining a code length for setting a total code length L equal to a sum of L_j for respective values of j from 1 to M;

means responsive to said means for determining a minimum for determining a partial code c_j for said jth parameter P_j for an ith category by;

##EQU15## where j and k range from 1 to M and from 1 to L_j, respectively, and c_jk represents a kth component of said partial code c_j ; and

means responsive to said means for determining a partial code for determining a total code C_i for representing said ith category by;

space="preserve" listing-type="equation">C.sub.i =c.sub.1 ⊕

c.sub.2 ⊕

- - - ⊕

c.sub.m

where i ranges from 1 to N and a symbol ⊕

represents a concatenation operator. X