GB2191869A - Optical processor - Google Patents

Description

SPECIFICATION Wavelength multiplexing optical processor The present invention relates to an optical processor for multiplexing wavelengths. A recently developed new generation of optical processors has enabled algebraic operations such as matrix-matrix or matrix-vector products to be performed, as well as more specific functions such as those which relate to the processing of images and the convolution of images, the conversion of incoherent images into coherent images, and the conversion of wavelengths. The optical processor consists essentially of three parts: the input transducer, the processor unit, and the output transducer. The input and output transducers perform electronic/optical transformations or inversions of the information which is to be processed. The processing of the information takes place in a unit for optical modulation functions referred to below as a spatial light modulator. Previously disclosed are spatial light modulators with an acousto-optical effect utilizing a Bragg cell made up of one or more channels. An optical processor is formed by adding to this type of modulator an input transducer and an output transducer which perform the necessary electronic/optical conversions. The application of mathematical operations directly to numerically coded numbers demands the same number of circuits as there are elementary signals, and a practical limit due to the size of the modulator is soon imposed. Another type of architecture permits a systolic processor to be designed on the basis of two spatial light modulators used as an input transducer and a directional module, for example a separator cube which transmits to a common outlet the output signals from one or other of the modulators modulated by its opposite number. This type of processor is also restricted in size by the maximum possible dimensions for the input modulators. An object of the present invention is not to impose any limits of a practical nature. According to the invention, with the above object in view, an optical processor is characterized in that it consists of several groups of principal input transistors which perform the wavelength multiplexing 1, 2, An and are allocated to each number to be multiplied, each of which principal input transducers represents a principal operative circuit, said processor also consisting of several groups of principal output transducers which perform wavelength demultiplexing after the operative processing, each of which output transducers represents the terminal point of an elementary operative circuit and supplies the result of an elementary operation,said principal input and output transducers being interconnected on the operative circuits via a multi-track spatial light modulator in accordance with the operating points situated in zones of modifiable optical characteristics under the effect of electrical commands originating from the numbers to be multiplied. The optical processor in accordance with the invention exhibits numerous advantages: # very low operating energy; # high speed; # the high degree of parallelism permits the dimensions to be reduced; # the processor permits a large number of algebraic mathematical functions to be performed in matrical calculations, as well as more complex operations by application of the method of iterative algorithms; # multiplicity of applications and functions: -a rapid, multi-track optical switch; -the analysis of images and the processing of images by Fourrie transform and by Laplace transform; # shape recognition; # enhancement of contrasts; # applications in robotics: information processing in real time. The technical characteristics and other advantages of the invention are set out in the following description, which is presented as a non-restrictive example, with reference to the accompanying drawings, in which: Figure 1 is the general block diagram of an optical processor in accordance with the invention; Figure 2 is the detailed block diagram of a principal input transducer; Figure 3 is the functional diagram of a spatial light modulator; Figure 4 is the detailed block diagram of a principal output transducer; Figure 5 is a simplified view in longitudinal section of a type of multiplexer or demultiplexer of the micro-engraved system type known as "blaze"; Figures 6 to 8 are diagrams which illustrate one or other variant of the spatial light modulator: Figure 6 represents an acousto-optical modulator; Figure 7 represents an electro-optical modulator;Figure 8 represents a magneto-optical modulator. Figure 9 is the explanatory diagram of the multiplication of two numerically coded numbers on a single circuit; Figure 10 is the general diagram for a version with multiplication of the analogue values in the case of two matrices with three lines and three columns. The following is a description of the optical processor in general terms by association of the functions which are executed by each basic module when applied to the multiplication of two matrices with three lines and three columns, each module being liable to be replaced by an equivalent module with regard to the means or the technology. Generally speaking, and with reference to Fig. 1, the optical processor 1 in accordance with the invention consists of several principal input transducers 2, 3, 4 relative to each principal operative circuit, for example three in the example chosen here, a multi-track spatial light modulator 5, and several principal output transducers, for example nine, 6, 7, 8, 9, 10, 11, 12, 13, 14, in the example chosen here of the multiplication of a matrix with three lines and three columns a 11, a 12, ... aij, a33 by a matrix of identical size with three lines and three columns b 11, b 12, ... bij, b33. In conformity with the principal objective of the invention, the number of lines and columns of the matrices to be multiplied is, of course, limited only by considerations of space of a practical nature, although this still permits a number of circuits largely superior to that which is currently possible to be achieved. This limitation on the modulator thus does not constitute a constraining factor because of the considerable capability which results from the great integration density. Described below in greater detail are the elementary functions which go to make up the principal input and output transducers and the principal functions of the spatial light modulator, with reference to Figs. 2 to 4. Fig. 2 represents the functional details of a principal input transducer. It consists in the traditional manner of an analogue/numerical converter 15 which executes the numerical coding in mixed binary form of the numbers which are present at its input, a shift register 16 which operates at the rhythm of a central clock H, a module for transposition into wavelengths 1, 2, ... i̇, ... An peculiar to each bit, for example a unit 17 of light-emitting diodes, for example laser diodes such as 18, connected individually by means of optical fibres such as 19 and together making up a bundle 20 at the input to a wavelength multiplexer 21.This multiplexer consists of a system 22, for example a microengraved system of the type known as "blaze", the output from which, in the form of a single optical fibre 23, is connected to the spatial light modulator 5 on as many identical circuits as there are columns bij in the matrix to be multiplied by means of an optical separator 24. A principal circuit must be allocated to each line aij of the first matrix, which circuit terminates in as many operating points 25 in the spatial light modulator 5. These operating points 25 are connected to as many principal output transducers as there are operative circuits, for example to nine references from 6 to 14, such as those represented in Fig. 1, and in which are performed the inverse demultiplexing operations and the operations involving the numerical/analogue conversion of the numbers. Each operating point 25 at the input to the modulator is connected to the output from a shift register 26 for the modulator controlled by the central clock H at a high frequency of, for example, 1 GHz. The register 26 receives the mixed binary expression of the number to be multiplied and applies each binary element to the operating point concerned of the spatial modulator at the rhythm of the central clock H. Similarly, the information present at each operating point 25 is transmitted sequentially to the operating point adjacent to each impulse of the clock. More precisely, each output transducer consists of a demultiplexer 27, for example of the same type as those used for multiplexing, that is to say with a system of the type known as "blazé", from which are emitted the various individualized wavelengths attributed to each binary element of the resulting number from the operation performed on the operative circuit concerned. The various wavelengths are transposed into electronic signals by a detection unit 28 equipped with photodiodes 29 followed by an amplifier/integrator or by detection by means of a photo-sensitive device with charge transfer known as CCD (charge coupled device) or its equivalent particular to each demultiplexing circuit, said unit serving as an optical electronic transducer. The result, expressed in mixed binary form, is then converted into simple binary form and, if appropriate, is then introduced into a numerical/analogue converter 30 enabling it to be expressed in decimal form. Described below is an example of a multiplexer or a demultiplexer 21 or 27 with a system of the type known as "blaze", that is to say a micro-engraved optical system arranged, as shown in Fig. 5, at an inclined angle with a central passageway 31 for the source of light originating from the bundle 19 of optical fibres after transposition into wavelengths. The beam of light belonging to each fibre crossing the "blaze" system 22 is reflected by the opposite wall equipped with a concave or parabolic mirror 32 on the plane of the system 22 carrying the micro-engraving, which deflects it back onto the mirror 32, which in turn reflects it back towards a single outlet point at which the extremity of the single optical fibre 23 is present. Conversely, in the case of a demultiplexer, this fibre 23 constitutes the input and the bundle of fibres 19 the output. In a more advanced and powerful embodiment it will be possible to replace the "blaze" system by a holographic system with high diffraction efficiency, the characteristics and the nature of which will produce a distinct improvement, both in wavelength selectivity and diffraction. The various equivalent forms of the spatial light modulator 5 are represented diagrammatically in Figs. 6, 7 and 8. The basic embodiment is an acousto-optical modulator 33, for example a Bragg cell, in which each operating point 25 corresponds to an elementary zone 34 in which the index is varied under the effect of an electrical modulation wave transformed into an acoustical wave by an appropriate transducer. The beam will then be caused to deviate, or not, as the case may be, towards the optical fibre as a function of the presence or the absence of the modulation wave at the input, as represented in Fig. 6. The spatial modulator 5 may also be executed in the form of electro-optical technology. In this type of modulator an electrical field modifies the optical properties of a crystal in such a way as to blank off the crystal to the beam towards the output fibre (Fig. 7). The modulator may also be of the kind known as magneto-optical, that is to say involving the variation of the polarization of the light, as represented in Fig. 8. This modulator consists of a polarization filter or a polarizer 35 crossed by beams originating from the multiplexers. The polarized light crosses a matrix 36 of magneto-optical valves made up of elementary valves such as 37 particular to each circuit. Each elementary zone is controlled by a shift register which denotes the presence or the absence of a voltage in order to cause the plane of polarization to rotate at a constant value, or not to rotate, as the case may be. The light then crosses a second polarizing filter known as an analyser 38, which, depending on the change in the rotation of the plane-of polarization, will be transparent or opaque to the beam. The analyser 38 behaves in the manner of an opaque element to all light which has been caused to rotate to a sufficient degree f on the plane of polarization. An equivalent to the modulators already described above has been achieved in this way. The assemblies and the sub-assemblies of the opto-processor in accordance with the invention are interconnected by means of optical fibres. It is conceivable to execute the connections between the various elements by means of traditional optical components, that is to say lenses or their equivalent. The following is a description, with reference to Fig. 9, of an example of the numerical multiplication of two numbers. This type of multiplication takes place simultaneously on each circuit during the operation of multiplying one matrix by another. Multiplication is produced by the convolution method by using a mixed binary representation. The two numbers Ai and Bj represented in mixed binary form are injected into the two shift registers 16 and 26 particular to the multiplexing and to the modulator. On each pulse of the clock H the two numbers Ai and Bj are shifted respectively towards the top and towards the left. The result of the multiplication of the contents of the multiplexing register 16 by the bit on the extreme left of the register 26 particular to the modulator is added to the preceding product at the level of the detectors/integrators. The function of the spatial light modulator 5 is to connect the beam of light to the ouput fibre, if this bit is equal to one, or, if this bit is equal to zero, not to cause the beam to deviate or not to allow the beam to pass towards the output fibre. A single wavelength is associated with each bit of the mixed binary expression for the number to be multiplied. The multiplexer permits all the wavelengths to be transmitted in a single fibre. A considerable reduction is made in this way in the number of operating points 25 on the modulator, and thus in its dimensions, which is the principal advantage of the invention. As may be appreciated from Fig. 9, this reduction produced by multiplexing is such that it is possible to perform the multiplication of two numbers at a single operating point 25 on the modulator 5, irrespective of the number of bits used for the representation of these two numbers, that is to say irrespective of the size of these numbers. The demultiplexer then separates the wavelengths, and the representation of the numbers as electronic signals can be processed and converted with a view to its directly usable expression in decimal form. In this particular embodiment multiplexing is performed directly on the numbers in each column of the first matrix Aij, and multiplication is performed simultaneously on three operative circuits by the numbers on the same line of the second matrix Bij. Thus, for each pulse of the clock H, a wavelength having an intensity which is proportional to the value of the number is associated simultaneously on each principal operative circuit with each of the numbers in the same column of the first matrix Aij. The wavelengths representing the value of each number in each column are multiplexed and arrive- at each operating point 25 before being multiplied sequentially by the numbers on each of the three lines of the matrix Bij. The results are demultiplexed, that is to say are individualized into partial products converted into electronic information, and are then stored. Thus, for the example of matrical calculation examined here, all that exists is a single lightemitting unit 17 consisting simply of three light-emitting diodes 18, for example laser di- odes, connected to a single multiplexer 21 and continuing across the separator 24 towards the spatial wavelength modulator 5 on three operative circuits which simply receive at the three operating points 25 only the electrical control signals originating from each line of the second matrix Bij on each pulse of the clock H.The products corresponding to the columns of the matrical result Cij are extracted on the three operative circuits by separate demultiplexers 27 in order to find on the three elementary operative circuits particular to each principal operative circuit the numbers cij for the matrical result Cij according to the columns of this matrix after transposition into electronic signals or information in detectorintegrators such as 39, each consisting of three elementary detectors 40, for example photo-diodes or charge coupled devices. It will be appreciated that the limitation of the dimensions of the modulator makes it possible for reasons of a technical nature to process matrices of a very high order and, as a consequence, imposes no restrictions on the execution of voluminous and complex operations. The invention described above is subject to modifications, variants and variations within the competence of a person skilled in the art, yet without departing from the scope of the protection conferred.

Claims (11)

1. An optical processor for performing algebraic operations, in particular matrical calculations, characterized in that it consists of several groups of principal input transducers which perform wavelength multiplexing 1, 2, An and are allocated to each number or expression for each number to be multiplied, each of which principal input transducers represents a principal operative circuit, said processor also consisting of several groups of principal output transducers which perform wavelength demultiplexing following the operative processing, each of which output transducers represents the terminal point of an operative circuit and supplies the result of an operation, said principal input and output transducers being interconnected on the operative circuits via a multi-track spatial light modulator (5) in accordance with the operating points (25) situated in zones of modifiable optical characteristics under the effect of electrical commands generated from the numbers to be multiplied.

2. An optical processor according to Claim 1 for the performance of numerical algebraic operations, characterized in that each wavelength 1, 2, An is allocated to a bit of the binary expression for each number to be multiplied, in that each of the principal input transducers represents a principal operative circuit, in that the processor also consists of several groups of principal output transducers which perform wavelength demultiplexing following the operative processing, in that each output transducer represents the terminal point of an elementary operative circuit and supplies the result of an elementary operation, and in that the principal input and output transducers are interconnected on the operative circuits via a multi-track spatial light modulator (5) in accordance with the operating points (25) situated in zones of modifiable optical characteristics under the effect of electrical commands originating from shift registers (26) each of which receives the mixed binary expression of the number to be multiplied.

3. A processor according to Claim 2, characterized in that each principal input transducer consists of an analogue/numerical converter (15), a shift register (16), a wavelength transposition module particular to each bit followed by a transmitter unit (17) with specific transmitters (18) for a particular wavelength, and a wavelength multiplexer (21), the single output fibre from which is divided into as many operative circuits as are required by means of an optical separator (24).

4. A processor according to Claim 2, characterized in that each principal output transducer consists of a demultiplexer (27) followed by a detector unit (28) equipped with photodiodes (29) or with a device which is sensitive to the transfer of charges which acts as an opto-electronic transducer, and by a numerical/analogue converter (30).

5. A processor according to Claims 2, 3 or 4, characterized in that the multiplexers (21) or demultiplexers (27) are of the micro-engraved optical system (22) type with a central passageway (31) for the incident and emergent beam, said system being arranged inclined in front of a concave or parabolic mirror (32).

6. A processor according to Claims 1 or 5, characterized in that the micro-engraved optical system is a holographic system.

7. A processor according to Claims 1 or 2, characterized in that the spatial light modulator (5) is an acousto-optical modulator with a Bragg cell.

8. A processor according to Claims 1 or 2, characterized in that the spatial light modulator ' (5) is an electro-optical modulator.

9. A processor according to Claims 1 or 2, characterized in that the spatial light modulator (5) is a magneto-optical modulator consisting of a matrix (36) of magneto-optical valves arranged between two polarization filters (35 and.38), the control field of which matrix produces the rotation of the plane of polarization.

10. A processor according to any preceding Claim, characterized in that the operations are performed on quantities which represent, depending on their level or their intensity, the analogue expression of the numbers to be multiplied, and in that the wavelength multiplexer is unique to all the numbers of the same principal calculation circuit, for example a circuit relative to the numbers on the same row or in the same column of a matrix which is to be multiplied by another matrix.

11. An optical processor constructed, arranged, and adapted for use substantially as hereinbefore described with reference to, and as shown in, the accompanying drawings.