Cellular neural network

• US 5,140,670 A
• Filed: 10/05/1989
• Issued: 08/18/1992
• Est. Priority Date: 10/05/1989
• Status: Expired due to Term

• Current Assignee: Regents of the University of California (University of California)
• Original Assignee: Regents of the University of California (University of California)
• Inventors: Yang, Lin, Chua, Leon O.
• Primary Examiner(s): Fleming, Michael R.
• Assistant Examiner(s): Downs, Robert W.
• Application Number:
  US07/417,728
• Time in Patent Office: 1,048 Days
• Field of Search: 364/513, 364/900, 395/24, 395/22
• US Class Current: 706/29
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/065 Analogue means

Abstract

A novel class of information-processing systems called a cellular neural network is discussed. Like a neural network, it is a large-scale nonlinear analog circuit which processes signals in real time. Like cellular automata, it is made of a massive aggregate of regularly spaced circuit clones, called cells, which communicate with each other directly only through its nearest neighbors. Each cell is made of a linear capacitor, a nonlinear voltage-controlled current source, and a few resistive linear circuit elements. Cellular neural networks share the best features of both worlds; its continuous time feature allows real-time signal processing found within the digital domain and its local interconnection feature makes it tailor made for VLSI implementation. Cellular neural networks are uniquely suited for high-speed parallel signal processing.

Claims

1. A cellular neural network comprising:

a plurality of identical analog cells arranged in a multi-dimensional lattice array of said plurality of cells, the cells closest to any single cell in each of the multiple dimensions being neighbor cells that form a first layer of neighbor cells that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell, all of the cells separated from any single cell by one other cell in any of the multiple dimensions form a second layer of neighbor cells that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell, and the layers of neighbor cells expanding concentrically from said second layer of neighbor cells in the multiple dimensions around said any single cell forming additional layers of neighbor cells to the full extent of said lattice array that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell,

each of said cells interacting non-linearly and continuously in time with selected layers of said neighbor cells and at least said first layer of each of said cell'"'"'s nearest neighbor cells with independently selected parameters for each of cell of said layers depending on the relative position of each cell in said layer to said any single cell,

wherein each cell includes;

non-linear feed-forward means coupled to each cell in the selected layers of said each cell'"'"'s neighbor cells for influencing the state of each cell in said selected layers of neighbor cells,

non-linear feedback means coupled to each cell in the selected layers of each cell'"'"'s neighbor cells for influencing the state of each cell in said selected layers of said neighbor cells, and

non-linear self feedback means for influencing the state of itself.

2. The cellular neural network of claim 1 wherein said array of cells is an n-dimensional array wherein n is an integer number of at least 2.

3. The cellular neural network of claim 2 wherein n equals 2 with the array of cells having i rows and j columns, i and j being independent integers of at least 2.

4. The cellular neural network of claim 1 wherein there are m levels of neighbors wherein m is an integer number of at least 1.

5. The cellular neural network of claim 1 wherein said feed-forward means of each cell is controlled by an independent signal source.

6. The cellular neural network of claim 1 wherein said feedback and self feedback means of each cell are each controlled by the state of its cell.

7. The cellular neural network of claim 1 wherein:

said feed forward means of each of said any single cell includes multiple output voltage controlled current source means for generating fixed feed forward signals to the cells in each layer of neighbor cells depending on relative position of the neighbor cell to said any single cell, and

said feedback and self feedback means of each of said any single cell includes multiple output voltage controlled current source means for generating fixed feedback signals to the cells in each layer of neighbor cells depending on relative position of the neighbor cell to said any single cell and the same self feedback signal to said any single cell.

8. The cellular neural network of claim 1 wherein each cell interacts with a selected number of contiguous layers of the cell'"'"'s nearest neighbor cells.

17. The pattern recognition system of claim 7 further including multiplex means coupled to each cell in each cellular neural network for extraction of the output signal from the cells consecutively a row or column at a time to minimize the number of external terminals required by the system.

9. A pattern recognition system comprising:

a plurality of cellular neural networks each functioning concurrently with each other to substantially simultaneously identify a different particular portion of the pattern of interest and to signal if that portion of the pattern was detected from an input signal that is simultaneously applied to each of said plurality of cellular neural networks, and

decision means coupled to receive said signal from each of said plurality of cellular neural networks for determining if each element of the pattern of interest is present,

each of said cellular neural networks includes;

a plurality of identical analog cells arranged in a multi-dimensional lattice array of said plurality of cells, the cells closest to any single cell in each of the multiple dimensions being neighbor cells that form a first layer of neighbor cells that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell, all of the cells separated from any single cell by one other cell in any of the multiple dimensions form a second layer of neighbor cells that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell, and the layers of neighbor cells expanding concentrically from said second layer of neighbor cells in the multiple dimensions around said any single cell forming additional layers of neighbor cells to the full extent of said lattice array that are interconnected with said any single cell in the same way depending on their relative position as a neighbor cell to said any single cell,

each of said cells interacting non-linearly and continuously in time with selected layers of said neighbor cells and at least said first layer of each of said cell'"'"'s nearest neighbor cells with independently selected parameters for each of cell of said layers depending on the relative position of each cell in said layer to said any single cell,

wherein each cell includes;

non-linear feed-forward means coupled to each cell in the selected layers of said each cell'"'"'s neighbor cells of influencing the state of each cell in said selected layers of neighbor cells,

non-linear feedback means coupled to each cell in the selected layers of each cell'"'"'s neighbor cells for influencing the state of each cell in said selected layers of said neighbor cells, and

non-linear self feedback means for influencing the state of itself.

10. The pattern recognition system of claim 9 wherein each cell interacts with a selected number of contiguous layers of the cell'"'"'s nearest neighbor cells.

11. The pattern recognition system of claim 9 wherein said feed-forward means of each cell is controlled by an independent signal source.

12. The pattern recognition system of claim 9 wherein said feedback and self feedback means of each cell are each controlled by the state of its cell.

13. The pattern recognition system of claim 9 wherein:

said feed forward means of each of said any single cell includes multiple output voltage controlled current source means for generating fixed feed forward signals to the cells in each layer of neighbor cells depending on relative position of the neighbor cell to said any single cell, and

said feedback and self feedback means of each of said any single cell includes multiple output voltage controlled current source means for generating fixed feedback signals to the cells in each layer of neighbor cells depending on relative position of the neighbor cell to said any single cell and the same self feedback signal to said any single cell.

14. The pattern recognition system of claim 9 wherein each cell within each of the cellular neural networks is disposed to receive a corresponding single pixel of a display that is the subject of the desired pattern recognition.

15. The pattern recognition system of claim 14 further includes a plurality of optical detector means with one such means connected to each cell of each cellular neural network for being optically exposed to the corresponding pixel of the subject pattern recognition display for entry of that pixel into the cell.

16. The pattern recognition system of claim 9 further including multiplex means coupled to each cell in each cellular neural network for entry of the input signal into the cells consecutively a row or column at a time to minimize the number of external terminals required by the system.

18. The pattern recognition system of claim 9 further including multiplex means coupled to each cell in each cellular neural network for entry of the input signal into the cells consecutively a row or column at a time prior to the processing of the information, for extraction of the output signal from the cells consecutively a row or column at a time after the processing of the signal, and for minimizing of the number of external terminals required by the system.