Signal processing apparatus having at least one neural network having pulse density signals as inputs and outputs

• US 5,259,064 A
• Filed: 01/24/1992
• Issued: 11/02/1993
• Est. Priority Date: 01/25/1991
• Status: Expired due to Fees

• Current Assignee: Ricoh Company Limited
• Original Assignee: Ricoh Company Limited
• Inventors: Kitaguchi, Takashi, Eguchi, Hirotoshi, Furuta, Toshiyuki
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Hafiz, Tariq R.
• Application Number:
  US07/826,906
• Time in Patent Office: 648 Days
• Field of Search: 395/23, 395/22, 395/27, 395/61, 395/24, 364/DIG. 3, 364/900 MS File, 364/DIG. 1, 364/200 MS File
• US Class Current: 706/23
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A signal processing apparatus for controlling an object includes an input unit, a neural network, an output unit, a teaching unit, and an error signal generator for generating a teaching signal that makes the neural network learn in real time. An error signal generator generates an error signal from the teaching signal and information contained in the network output signal. The error signal controls the neural network so that the control output signal has correct control information with respect to the output signal from the controlled object.

Claims

1. A signal processing apparatus for controlling a controlled object having an output signal, comprising:

input means for converting said output signal of the controlled object into a network input signal, said network input signal being a pulse density signal having a pulse density defined by a number of first values and second values within a predetermined time, the first values and the second values being arranged at random, and the first and second values respectively corresponding to high and low binary signal levels;

a first neural network receiving the network input signal and outputting a network output signal which is also a pulse density signal;

output means, coupled to said first neural network, for converting said network output signal into a control output signal applied to the controlled object;

teaching means for generating a teaching signal which makes the first neutral network learn in real-time; and

error signal generating means, coupled to said first neutral network and said teaching means, for generating an error signal from the teaching signal and information contained in said network output signal, said error signal controlling the first neural network so that the control output signal has correct control information with respect to the output signal from the controlled object.

2. A signal processing apparatus as claimed in claim 1, wherein said teaching means comprises means for manually generating said teaching signal.

3. A signal processing apparatus as claimed in claim 1, wherein said teaching means. comprises means for automatically generating said teaching signal from the information contained in the network output signal and information contained in the output signal from the controlled object.

4. A signal processing apparatus as claimed in claim 1, wherein said teaching means comprises:

delay means for generating a first delayed version of the information contained in the output signal from the controlled object and for generating a second delayed version of the information contained in the network output signal;

a second neural network, coupled to said delay means, for receiving the information contained in the output signal without delay and said first and second delayed versions of the information and for generating a predicted output signal of the controlled object;

comparator means, coupled to said second neural network, for generating an instruction signal from the information contained in the output signal from the controlled object and the predicted output signal, said instruction signal controlling whether or not said first neural network should be made to learn, said instruction signal controlling said first neural network to learn only when the output signal from the controlled object has a predetermined relationship with the predicted output signal; and

converter means for generating said error signal from the information contained in the network output signal.

5. A signal processing apparatus as claimed in claim 4, wherein said predetermined relationship which causes said instruction signal to be generated which makes said first neural network learn occurs only when a value indicated by the information contained in the output signal from the controlled object is closer to a target value than a value indicated by the information contained in the predicted output signal from the first neutral network.

6. A signal processing apparatus as claimed in claim 4, further comprising selector means for outputting the first and second delayed versions to said second neural network when the second neural network is made to learn and for directly outputting to the second neural network the information contained in the output signal from the controlled object and the information contained in the network output signal when the second neural network is not made to learn.

7. A signal processing apparatus as claimed in claim 1, wherein said teaching means comprises:

delay means for generating a first delayed version of the information contained in the output signal from the controlled object and for generating a second delayed version of the information contained in the network output signal;

a second neural network, coupled to said delay means, for receiving the information contained in the output signal without delay and said first and second delayed versions of the information;

a third neural network for directly receiving the information contained in the output signal from the controlled object and information contained in the control output signal and for generating a predicted output signal of the controlled object;

transfer means, coupled to said second and third neural networks, for transferring parameter values of the second neural network to the third neural network so that parameter values of the third neural network are updated in real time;

comparator means, coupled to said third neural network, for generating an instruction signal from the information contained in the output signal from the controlled object and the predicted output signal, said instruction signal showing whether or not said first neural network should be made to learn; and

converter means for generating said error signal from the information contained in the network output signal.