A system for realizing matrix operations of different orders based on monochromatic dispersive media

Abstract

The present invention relates to the field of photon computing network technology, and in particular to a system for realizing matrix operations of different orders based on a monochromatic dispersive medium. The system comprises: a multi-wavelength light source, a modulation processing unit, a micro-ring array unit, a monochromatic dispersive medium unit and an acquisition processing unit connected in sequence. The monochromatic dispersive medium unit is used to perform delay processing on each weighted modulated optical signal, and by controlling the delay time of each weighted modulated optical signal, after summing the optical signal intensities of each delayed weighted modulated optical signal in the same time domain, the calculation result of the convolution calculation performed by the second matrix on the first matrix can be obtained, which greatly improves the calculation efficiency of the matrix multiplication operation. At the same time, the monochromatic dispersive medium is introduced to replace the traditional optical delay line, which is conducive to reducing the size of the optical computing integrated system and reducing the cost.

Description

Technical Field

[0001] The present invention relates to the field of photon computing network technology, and in particular to a system for realizing matrix operations of different orders based on a monochromatic dispersive medium. Background Art

[0002] Optical computing is a technology that uses optical signals to process and calculate data. Compared with traditional electrical computing, the advantages of optical computing are fast transmission and computing speed, large bandwidth, low energy consumption, parallel transmission, and no electromagnetic interference. These characteristics make it more efficient when processing large amounts of data. Especially in the fields of big data, artificial intelligence, and machine learning, optical computing shows great potential. As an important architecture in the field of optical computing, the photonic convolutional neural network (PCNN) has shown its unique advantages. Due to the high complexity of matrix multiplication and addition operations, traditional convolutional neural networks (CNN) take up a lot of computing time and memory resources when performing matrix multiplication and addition operations. Compared with traditional convolutional neural networks (CNN), photonic convolutional neural networks (PCNN) have the characteristics of integrated storage and computing, high speed and low power consumption. Compared with CNN, it has a natural advantage when performing matrix multiplication and addition operations.

[0003] At present, some progress has been made in the photon matrix multiplication and addition operation. In the mainstream photon matrix multiplication and addition operation, in order to avoid aliasing when the modulated optical signal passes through the microring and then passes through the multiplexer (MUX), the usual practice is to add an optical delay line after each microring. In addition, the optical delay line can also achieve the following functions: 1. Signal synchronization and integration: When processing signals from different input sources or after different processing steps, the delay line can help achieve time synchronization to ensure that the signal is integrated and processed within the appropriate time. 2. Realize time convolution operation: In photon convolutional neural networks, convolution operations usually rely on the ability to process data at different time points. By providing time offset, the delay line allows the network to process the same input signal at multiple time points to achieve effective time convolution. 3. Enhance the dynamic range of the network: By introducing the processing capability of the time dimension in the network, the delay line helps increase the network's ability to process complex dynamic signals.

[0004] However, there are also some problems with optical network delay lines, such as: 1. Physical size limitations: The actual photonic delay line may require a longer physical path length, which may increase the size and complexity of the device. 2. Loss problem: During the transmission of optical signals, especially in longer delay lines, the signal may experience loss, affecting the integrity and processing efficiency of the information. 3. Manufacturing cost: The manufacturing and maintenance costs of high-quality photonic delay lines may be relatively high, and once the system's delay line is designed, it can only achieve a specific time delay.

[0005] In addition, in the existing photon matrix multiplication and addition process, the delay time of the optical network delay line for each optical signal is equidistant. After the delay processing, a large amount of transformation and acquisition is required to obtain the final matrix multiplication and addition result. This undoubtedly causes a waste of time in the photon matrix multiplication and addition operation and reduces the computing efficiency. Summary of the invention

[0006] In order to solve the above technical problem or at least partially solve the above technical problem, the present invention provides a system for implementing matrix operations of different orders based on a monochromatic dispersive medium.

[0007] The present invention provides a system for realizing matrix operations of different orders based on a monochromatic dispersive medium, the system comprising:

[0008] A multi-wavelength light source, used to generate a plurality of initial optical signals with different wavelengths;

[0009] A modulation processing unit, used for converting the first matrix into a first matrix electrical signal based on a time sequence, and modulating the first matrix electrical signal into each initial optical signal to form a multi-channel modulated optical signal;

[0010] The micro-ring array unit includes a plurality of micro-rings arranged in sequence, the weight of each micro-ring is determined by the value of each element in the second matrix, and each micro-ring is coupled to a modulated optical signal to load different weights on each modulated optical signal to obtain a plurality of weighted modulated optical signals;

[0011] A monochromatic dispersive medium unit includes a plurality of monochromatic dispersive media corresponding to a plurality of microrings arranged in sequence, each monochromatic dispersive medium receives a weighted modulated optical signal output by the microring corresponding thereto, and performs delay processing on each weighted modulated optical signal to obtain a delayed weighted modulated optical signal, wherein the delay time length of each monochromatic dispersive medium is determined according to the number of arrangement bits thereof, the order of the first matrix and the order of the second matrix;

[0012] The acquisition and processing unit is used to integrate the delayed weighted modulated optical signals into the same optical path to obtain a multiplexed optical signal, and convert the multiplexed optical signal into a multiplexed electrical signal, and extract the data information of the third matrix from the multiplexed electrical signal, wherein the third matrix is ​​the calculation result of the convolution calculation performed by the second matrix on the first matrix.

[0013] Furthermore, the weight of each micro-ring is determined by the value of each element of the second matrix, including:

[0014] Determine the target matrix position in the second matrix corresponding to the target microring according to the number of permutation bits of the target microring:

[0015] k=(i-1)×n+j

[0016] Wherein, k is the number of permutation bits of the target micro-ring, n is the order of the second matrix, i is the number of rows of the target position of the second matrix, and j is the number of columns of the target position of the second matrix;

[0017] Select the value W of the element at the target position of the second matrix ij As the weight of the target microring.

[0018] Furthermore, the delay time t of the target monochromatic dispersive medium corresponding to the target microring is k Satisfies the following calculation formula:

[0019] t k =[(ni)×m+nj]×Δτ

[0020] Wherein, m is the order of the first matrix, and Δτ is the time length used by the modulation processing unit to modulate one bit of signal.

[0021] Furthermore, the monochromatic dispersive medium is a fiber Bragg grating based on lithium niobate material, and the monochromatic dispersive medium unit further includes:

[0022] The electric control unit is used to control the delay time of the fiber Bragg grating by adjusting the voltage acting on the fiber Bragg grating; wherein,

[0023] The delay time of the fiber Bragg grating and the voltage U acting on the fiber Bragg grating satisfy the following calculation formula:

[0024]

[0025]

[0026] Where △n is the change in the refractive index of the fiber Bragg grating, L is the length of the fiber Bragg grating, C is the transmission rate of light on the optical fiber, n effis the initial refractive index of the fiber Bragg grating, r is the electro-optic coefficient of the fiber Bragg grating, and d is the electrode spacing acting on the fiber Bragg grating.

[0027] Furthermore, the acquisition processing unit includes:

[0028] The result collection unit is used to respectively s Starting from the time instant, sequentially collecting P-bit sampled electrical signals from the multiplexed electrical signal as the s-th row data of the third matrix;

[0029] T s The moment satisfies the following calculation formula:

[0030] T S =[(n-1)(m+1)+(s-1)m]·Δτ

[0031] P satisfies the following calculation formula:

[0032] P=m-n+1

[0033] Where s=1,2,...,P.

[0034] Furthermore, the monochromatic dispersive medium unit further comprises:

[0035] The temperature control unit is used to control the temperature of each monochromatic dispersive medium in the monochromatic dispersive medium unit so as to keep the temperature of each monochromatic dispersive medium stable.

[0036] Furthermore, the acquisition processing unit also includes:

[0037] A multiplexer, used for receiving each delayed weighted modulated optical signal, combining each delayed weighted modulated optical signal into one multiplexed optical signal, so as to sum the optical signal strengths of different delayed weighted modulated optical signals in the same time domain;

[0038] a photodetector for converting the multiplexed optical signal into an original analog multiplexed electrical signal;

[0039] an analog-to-digital converter, used for converting an initial analog multiplexed electrical signal into an initial digital multiplexed electrical signal;

[0040] A digital signal processor, used for performing distortion compensation on each initial digital multiplexed electrical signal to obtain a digital multiplexed electrical signal;

[0041] The digital-to-analog converter is used to convert the digital multiplexed electrical signal into a multiplexed electrical signal for executing the convolution calculation result.

[0042] Furthermore, the digital signal processor comprises:

[0043] A Fourier transform module, used for performing Fourier transform on the initial digital multiplexed electrical signal, so as to convert the initial digital multiplexed electrical signal based on the time domain into a frequency domain electrical signal based on the frequency domain;

[0044] The phase compensation filter is used to perform phase compensation on the frequency domain electrical signal to obtain a compensated frequency domain electrical signal. The transfer function of the phase compensation satisfies the following calculation formula:

[0045]

[0046] Where f is the offset frequency, β2 is the second-order dispersion coefficient of the system, and β3 is the third-order dispersion coefficient of the system;

[0047] The inverse transform module is used to perform inverse Fourier transform on the compensated frequency domain electrical signal to obtain a digital multiplexed electrical signal based on the time domain.

[0048] Furthermore, the modulation unit comprises:

[0049] A wavelength division multiplexer, used for receiving initial optical signals and multiplexing the initial optical signals onto the same optical signal transmission medium to obtain a wavelength division multiplexed optical signal;

[0050] A modulator, configured to receive a first matrix electrical signal and load the first matrix electrical signal into the wavelength division multiplexed optical signal to form a multiplexed modulated optical signal;

[0051] The demultiplexer is used to receive the multiplexed modulated optical signal and demultiplex the multiplexed modulated optical signal into a plurality of modulated optical signals with different wavelengths according to different wavelengths.

[0052] Furthermore, the modulation processing unit also includes:

[0053] The input matrix processing unit is used to flatten the first matrix, and convert the values ​​of each element of the first matrix into corresponding electrical signals in time sequence to form a first matrix electrical signal, and input the first matrix electrical signal into the modulator.

[0054] The system provided by the present invention realizes matrix operations of different orders based on monochromatic dispersive media, and uses monochromatic dispersive media units to perform delay processing on each weighted modulated optical signal, so that after summing the optical signal intensities of each delayed weighted modulated optical signal in the same time domain, the calculation result of the convolution calculation performed by the second matrix on the first matrix can be obtained, which greatly improves the calculation efficiency of the matrix multiplication operation. At the same time, the introduction of monochromatic dispersive media to replace traditional optical delay lines is conducive to reducing the size of the optical computing integrated system and reducing costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0056] Figure 1 This is a schematic diagram of the structure of a system for implementing matrix operations of different orders based on a monochromatic dispersive medium according to an embodiment of the present invention;

[0057] Figure 2 It is a schematic diagram of the structure of a system for realizing matrix operations of different orders based on a monochromatic dispersive medium according to another embodiment of the present invention;

[0058] Figure 3 Schematic diagram of a convolution calculation process according to an embodiment of the present invention;

[0059] Figure 4 The figure is a schematic diagram of a process of modulating and delaying an optical signal according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0061] Before introducing the system of the present invention for realizing matrix operations of different orders based on a monochromatic dispersive medium, the convolution calculation principle of the present invention is first introduced in detail. Figure 1 FIG. 4 shows a process of performing convolution calculation on matrices of different orders according to a specific embodiment of the present invention, wherein Figure 1 (a) is the first matrix of order 4, Figure 1 (b) is a second matrix of order 2. The second matrix is ​​used as a convolution kernel to perform convolution calculation on the first matrix. The convolution calculation process is to slide the second matrix in the first matrix in sequence. The convolution calculation result is as follows: Figure 1 (c) The third matrix shown.

[0062] In the photon convolution calculation process of the embodiment of the present invention, it is necessary to first flatten the first matrix, that is, to connect the rows of the first matrix in sequence to form a one-dimensional vector as shown in 1(d), and the second matrix is ​​processed in the same way as shown in Figure 1 As shown in (e), the one-dimensional vectors of the first matrix are multiplied by each element in the second matrix to obtain Figure 1(f) Array calculation results shown.

[0063] By observation Figure 1 The array calculation results and the actual convolution calculation results shown in (f) can be obtained as follows Figure 1 (g) shows the operation rule, when the first matrix is ​​a 4-order matrix and the second matrix is ​​a 2-order matrix, the first row of the array is padded with one more zero than the second row, the second row is padded with 3 more zeros than the third row, and the third row is padded with one more zero than the fourth row. After the zero padding operation, you only need to add the data in the same column of the array to obtain an element in the convolution calculation result.

[0064] Further summarizing the above rules, we can see that for the convolution calculation result when the first matrix is ​​an m-order matrix and the second matrix is ​​an n-order matrix, the array starts from the first row to the Nth (N=n 2 ) rows, the previous row is padded with one more zero than the leftmost row of the next row, and the nth row is padded with m-n+1 more zeros than the leftmost row of the n+1th row. For example, when the first matrix is ​​a 3x3 matrix and the second matrix is ​​a 2x2 matrix, at this time m=3, n=2, when performing the zero-padding operation, the first row is padded with one more zero than the second row on the left; and the second row (n=2) is padded with m-n+1=2 more zeros than the third row (n+1=3); the zero-padding process can be expressed by the following formula:

[0065] τ k =[(ni)×m+nj] (1)

[0066] where τ k It represents the number of zero padding in the k-th row, k=(i-1)×n+j, which corresponds to the i-th row and j-th column of the second matrix.

[0067] Based on the above convolution calculation principle, the embodiment of the present invention proposes the following Figure 2 The structural diagram of the system based on monochromatic dispersive medium to realize matrix operations of different orders is shown below. Figure 2 The system provided by the present invention for realizing matrix operations of different orders based on monochromatic dispersive media is introduced in detail.

[0068] The system comprises a multi-wavelength light source (not shown in the drawings), a modulation processing unit 1, a micro-ring array unit 2, a monochromatic dispersion medium unit 3 and a collection processing unit 4 connected in sequence, wherein:

[0069] A multi-wavelength light source, used to generate a plurality of initial optical signals with different wavelengths;

[0070] A modulation processing unit 1, used for converting the first matrix into a first matrix electrical signal based on a time sequence, and modulating the first matrix electrical signal into each initial optical signal to form a multi-channel modulated optical signal;

[0071] The micro-ring array unit 2 includes a plurality of micro-rings arranged in sequence, the weight of each micro-ring is determined by the value of each element in the second matrix, and each micro-ring is coupled to a modulated optical signal to load different weights on each modulated optical signal to obtain a plurality of weighted modulated optical signals;

[0072] The monochromatic dispersive medium unit 3 comprises a plurality of monochromatic dispersive media 31 corresponding to a plurality of microrings arranged in sequence, each monochromatic dispersive medium 31 receives a weighted modulated optical signal output by the microring corresponding thereto, and performs delay processing on each weighted modulated optical signal to obtain a delayed weighted modulated optical signal, wherein the delay time length of each monochromatic dispersive medium 31 is determined according to the number of arrangement bits thereof, the order of the first matrix and the order of the second matrix;

[0073] The acquisition and processing unit 4 is used to integrate each delayed weighted modulated optical signal into the same optical path to obtain a multiplexed optical signal, and convert the multiplexed optical signal into a multiplexed electrical signal, and extract the data information of the third matrix from the multiplexed electrical signal. The third matrix is ​​the calculation result of the convolution calculation of the second matrix on the first matrix.

[0074] Furthermore, as attached Figure 2 The wavelengths shown are λ1, λ2, ..., λ N The light source is the initial light signal, wherein, in order to correspond to the order n of the second matrix, the number N of the initial light signals generally satisfies N=n 2 .

[0075] Furthermore, the second matrix of the embodiment of the present invention is expressed as:

[0076]

[0077] It is understandable that the initial optical signals are different in the frequency domain and have the same optical signal strength in the time domain. In a specific embodiment of the present invention, a distributed feedback laser (DFB) is used to generate a multi-wavelength light source through a microcavity optical frequency comb.

[0078] Furthermore, the modulation processing unit 1 is used to receive a first matrix signal, and load the first matrix signal into each of the initial optical signals to form a plurality of modulated optical signals.

[0079] Furthermore, the first matrix can be expressed as:

[0080]

[0081] Wherein, m is the order of the first matrix.

[0082] It should be noted that before modulating each initial optical signal based on the first matrix, the first matrix needs to be flattened to obtain a first matrix signal. The flattening process is to connect the first digits of each row of the first matrix in sequence to form a one-dimensional vector group, and select an electrical signal of corresponding intensity according to the numerical value of each element in the one-dimensional vector group to form the first matrix signal.

[0083] Therefore, the system of the embodiment of the present invention also includes an input matrix processing unit, which is used to flatten the first matrix, and convert the values ​​of each element of the first matrix into corresponding electrical signals in time sequence to form a first matrix electrical signal, and input the first matrix electrical signal into the modulator.

[0084] Furthermore, in order to simplify the description process, the optical signal strength of the initial optical signal is set to 1 in the embodiment of the present invention, and the optical signal strength of the modulated optical signal in the time domain can be expressed as:

[0085] [A 11 , A 12 , ..., A xy , ..., A m(m-1) , A mm ](4)

[0086] Among them, A xy is the optical signal intensity of the zth signal in the modulated optical signal, z=(x-1)×m+y, which is the optical signal intensity corresponding to the value of the xth row and yth column of the first matrix.

[0087] Further, the microring array unit 2 includes a microring array 21, each microring in the microring array 21 has a one-to-one correspondence with each element in the second matrix, and the weight of each microring is determined by the value of each element of the second matrix, including:

[0088] Determine the target matrix position in the second matrix corresponding to the target microring according to the number of permutation bits of the target microring:

[0089] k=(i-1)×n+j

[0090] Wherein, k is the number of permutation bits of the target micro-ring, n is the order of the second matrix, i is the number of rows of the target position of the second matrix, and j is the number of columns of the target position of the second matrix;

[0091] Select the value W of the element at the target position of the second matrix ij As the weight of the target microring. Specifically, the weight of each microring in the microring array 21 can be expressed as:

[0092] [W 11 , W 12 , ..., W ij ,...,Wn(n-1) , W nn ](5)

[0093] Furthermore, the optical signal intensity of the kth weighted modulated optical signal output by the kth micro-ring in the time domain is expressed as:

[0094]

[0095] Among them, W k It can also be expressed as W ij , B kz Represents the optical signal intensity of the z-th signal in the k-th weighted modulated optical signal in the time domain.

[0096] Furthermore, it can be known from the above embodiments that the delay time length t of the target monochromatic dispersive medium 31 corresponding to the target microring is k Satisfies the following calculation formula:

[0097] t k =τ k ×Δτ

[0098] =[(ni)×m+nj]×Δτ (7)

[0099] Among them, Δτ is the time length used by the modulation processing unit 1 to modulate a bit signal, τ k is the number of zeros padded in the kth row;

[0100] Further, refer to Figure 1 The convolution calculation process of the present invention, the acquisition processing unit 4 of the embodiment of the present invention includes a result acquisition unit (not shown in the drawings), which is used to respectively s Starting from the time instant, sequentially collecting P-bit sampled electrical signals from the multiplexed electrical signal as the s-th row data of the third matrix;

[0101] T s The moment satisfies the following calculation formula:

[0102] T S =[(n-1)(m+1)+(s-1)m]·Δτ (8)

[0103] P satisfies the following calculation formula:

[0104] P=m-n+1 (9)

[0105] Where s=1,2,...,P.

[0106] It should be noted that P is the order of the third matrix. For the first row of the third matrix, after the delay time of the first row is reached, the multiplexed electrical signal can be collected. At this time, each sampled electrical signal collected can correspond to a parameter of the third matrix. After collecting P-bit sampled electrical signals, all the data of the first row of the third matrix can be obtained. It can be understood that at this time, the delay time of the last row is 0, that is, the delay time of the Nth monochromatic dispersion medium 31 is 0.

[0107] Through observation and theoretical analysis, it can be known that after the delay time of the first row is reached, and after the time interval of m-bit sampling electrical signals, the second row data of the third matrix can be collected. Similarly, by sequentially collecting P-bit sampling electrical signals, all the data of the second row of the third matrix can be obtained.

[0108] From the above analysis, we can see that the sampling time T of the third matrix s Satisfies the following calculation formula:

[0109] T S =[τ k +(s-1)m]·Δτ (10)

[0110] By combining formula (1) and formula (10), we can obtain the calculation result of the sampling time as shown in formula (8).

[0111] Further, Figure 3 A system for realizing matrix operations of different orders based on a monochromatic dispersive medium according to another embodiment of the present invention is shown. Figure 3 It can be seen that the modulation processing unit 1 provided in the embodiment of the present invention specifically also includes a wavelength division multiplexer 11 (represented as WDM in the figure), a modulator 12 (represented as MZM in the figure) and a demultiplexer 13 (represented as DEMUX in the figure) connected in sequence.

[0112] The wavelength division multiplexer 11 is used to receive the initial optical signal and multiplex the initial optical signals onto the same optical signal transmission medium to obtain a wavelength division multiplexed optical signal; the modulator 12 is used to receive the first matrix signal corresponding to the first matrix and load the first matrix signal into the wavelength division multiplexed optical signal to form a multiplexed modulated optical signal; the demultiplexer 13 is used to receive the multiplexed modulated optical signal and demultiplex the multiplexed modulated optical signal into N modulated optical signals with different wavelengths according to different wavelengths.

[0113] In a specific embodiment of the present invention, the wavelength division multiplexer 11 multiplexes each initial optical signal onto the same optical fiber for transmission, and multiplexes multiple initial optical signals into a single channel to form a wavelength division multiplexed optical signal, which can save the number of transmission channels and is beneficial to the synchronous modulation of each initial optical signal by the modulator 12.

[0114] Furthermore, in a specific embodiment of the present invention, the modulator 12 is a Mach-Zehnder Modulator (MZM), and an electrical signal with input matrix information (i.e., a first matrix signal) is loaded on the MZM to modulate the wavelength division multiplexed optical signal so that the optical signal intensity of each bit of the multiplexed modulated optical signal in the time domain corresponds to the value of the first matrix. Then, the optical signals of different wavelengths in the multiplexed modulated optical signal are separated by the demultiplexer 13 to obtain modulated optical signals with different wavelengths.

[0115] Further, refer to Figure 3 It can be seen that the microring array unit 2 of the embodiment of the present invention further includes a weight control unit 22. The weight of each microring can be controlled by the weight control unit 22. The unit includes a weight library. Different voltages acting on the microrings correspond to different weights.

[0116] Furthermore, the monochromatic dispersive medium unit 3 of the embodiment of the present invention further includes an electric control unit 32 and a temperature control unit 33. The electric control unit 32 can achieve different delays by applying different voltages to each monochromatic dispersive medium 31 and changing the refractive index and dispersion coefficient of the medium, so that it can be dynamically adjusted according to the delay requirements of the convolution operation of different-order matrices. The temperature control unit 33 is used to control the temperature of the monochromatic dispersive medium unit 3 so that the monochromatic dispersive medium unit 3 maintains stable performance, that is, the temperature of the monochromatic dispersive medium 31 can be kept stable through the feedback control system, so that it has stable optical properties.

[0117] The application of dispersive media in optical computing mainly relies on its ability to affect the phase and group velocity of light passing through the medium. In optical computing networks, traditional optical delay lines are often used to adjust the transmission time and phase of optical signals to achieve precise time control and signal synchronization. The introduction of dispersive media can provide a more flexible and efficient way to achieve these functions.

[0118] The following are several specific implementation methods:

[0119] 1. Dispersive fiber delay line: Dispersive fiber uses its unique wavelength-dependent dispersion characteristics to control the propagation speed of light pulses. For example, positive dispersion fiber (such as standard single-mode fiber) will cause light pulses of different wavelengths to propagate at different speeds at a specific wavelength, thereby achieving time delay adjustment. This method is suitable for occasions where dynamic delay adjustment is required.

[0120] 2. Photonic crystal structure: Photonic crystals have a periodic dielectric constant distribution, and photonic bandgap structures with specific dispersion characteristics can be designed. Such structures can be used to precisely control the propagation path and speed of light waves and achieve local delay of optical signals. The design of photonic crystals can be very flexible and support complex optical path design and integration.

[0121] 3. Dispersive metamaterials: Metamaterials are materials that have been engineered to have physical properties that do not exist in nature through artificial structures. In optics, metamaterials can be designed to have unusual dispersion properties, such as negative group velocity dispersion. This can be used to realize inverse delay lines, that is, to achieve long optical delays over very short physical distances, which is particularly important for high-density optical integration.

[0122] 4. MEMS-based adjustable dispersion delay line: Micro-Electro-Mechanical System (MEMS) technology can be used to manufacture dynamically adjustable optical components. For example, by adjusting the micro-mirrors or refractive index in the MEMS structure, the path length and speed of the light passing through can be changed, thereby realizing an adjustable delay line. This technology supports fast adjustment and high-precision control.

[0123] The selection and application of these technologies will depend on the specific requirements of the optical computing system, such as the magnitude of delay, bandwidth, and complexity of system integration. In practical applications, factors such as the linear and nonlinear optical properties of the medium, temperature stability, and optical loss may also need to be considered.

[0124] In a specific embodiment of the present invention, the monochromatic dispersive medium 31 is a fiber Bragg grating based on lithium niobate material, and the electric control unit 32 of the embodiment of the present invention can change the delay time of the fiber Bragg grating by changing the voltage acting on the fiber Bragg grating.

[0125] Specifically, the delay time of the fiber Bragg grating and the voltage U acting on the fiber Bragg grating satisfy the following calculation formula:

[0126]

[0127] Where Δn is the change in the refractive index of the fiber Bragg grating, L is the length of the fiber Bragg grating, C is the transmission rate of light on the optical fiber, and n eff is the initial refractive index of the fiber Bragg grating, r is the electro-optic coefficient of the fiber Bragg grating, and d is the electrode spacing acting on the fiber Bragg grating. That is, the voltage applied to each monochromatic dispersive medium 31 can be determined according to the time length Δτ used by the modulation processing unit 1 to modulate a bit signal.

[0128] In a specific embodiment of the present invention, the modulation signal rate on the Mach-Zehnder modulator 12 is 25Gbit / s, that is, the time required to modulate 1bit of data is 40ps. In order to flexibly realize multi-order matrix operations, an adjustable voltage of 0-50V is used in the electronic control unit 32 of the monochromatic dispersive medium 31, and the optical fiber length L is 10mm. In the present invention, for every increase of 5V in the voltage U acting on the fiber Bragg grating, the light delay through the fiber Bragg grating increases by -40ps, and the range of the 0-50V delay time is 0-400ps.

[0129] Furthermore, the electro-optic coefficient of the lithium niobate fiber Bragg grating is 30 pm / V, the initial refractive index is n = 2.2, the operating wavelength is 1550 nm, and the speed of light is 3×10 8 m / s, the propagation speed of light in the optical fiber is C = 1.36 × 10 8 Substituting the above parameters into formula (11) and formula (12), it can be concluded that to achieve a delay of 40ps, Δn = 0.8, and then substituting this value into formula (12) it can be obtained that when the spacing between the electrodes is 1nm and Δn = 0.8, U is approximately equal to 5V.

[0130] Furthermore, for Figure 1 In the specific embodiment, the first matrix is ​​4×4, the second matrix is ​​2×2, the first row after convolution is not padded with zero on the right, that is, the time advance is 0, the second row is padded with 1 0 on the right, the time advance is 1 bit, 40ps, the third row is padded with 3 0s on the right, the time advance is 3 bits, 120ps, the fourth row is padded with 4 0s on the right, the time advance is 160ps. Based on the above data, Figure 1 By simply adding a 0V voltage to the grating of the first channel of the monochromatic dispersion medium 31, a 5V voltage to the second channel, a 15V voltage to the third channel, and a 20V voltage to the fourth channel, the shift zero-filling operation of the convolution operation with an input matrix of order 3 can be implemented.

[0131] It can be seen from the above specific embodiments that the present invention uses a controllable monochromatic dispersive medium 31 to flexibly adjust the delay time by electrical control, thereby realizing matrix convolution operations of different orders on a fixed optical structure, greatly improving the flexibility of the system.

[0132] Furthermore, the integrated processing unit provided in the embodiment of the present invention also includes a multiplexer 41 (represented as MUX in the drawing), a photodetector 42, an analog-to-digital converter 43 (represented as ADC in the drawing), a digital signal processor 44 (represented as DSP in the drawing) and a digital-to-analog converter 45 (represented as DAC in the drawing) connected in sequence.

[0133] Among them, the multiplexer 41 is used to receive each delayed weighted modulated optical signal, and merge each delayed weighted modulated optical signal into a multiplexed optical signal to sum the optical signal strengths of different delayed weighted modulated optical signals in the same time domain; the photodetector 42 is used to convert the multiplexed optical signal into an initial analog multiplexed electrical signal; the analog-to-digital converter 43 is used to convert the initial analog multiplexed electrical signal into an initial digital multiplexed electrical signal; the digital signal processor 44 is used to perform distortion compensation on each initial digital multiplexed electrical signal to obtain a digital multiplexed electrical signal; the digital-to-analog converter 45 is used to convert the digital multiplexed electrical signal into a multiplexed electrical signal for executing the convolution calculation result.

[0134] In optical computing networks, wavelength division multiplexing (WDM) technology can significantly increase the transmission capacity of the network, but it may also introduce distortions such as dispersion, nonlinear effects, signal crossover, etc. The digital signal processor 44 (DSP) plays a key role in this case, which is used to compensate for these distortions and ensure signal quality. The present invention uses DSP technology to perform distortion compensation, so that the obtained calculation results are more accurate and reliable.

[0135] After the optical signal passes through the photodetector 42, it is sampled and quantized into a digital signal by the ADC analog-to-digital conversion unit. In a specific embodiment of the present invention, the second-order dispersion coefficient of the channel of the entire optical computing system is approximately β2 = -20 (ps) 2 / km, the third-order dispersion coefficient is about β3=0.1(ps) 3 / km, the phase response caused by dispersion is The phase compensation filter is designed based on the phase response caused by dispersion. The transfer function of the compensation filter should be the negative value of the original phase response, that is:

[0136]

[0137] Wherein f is the offset frequency, i.e., the difference between the frequency of light and the frequency corresponding to the wavelength of 1550nm. In order to design and implement the phase compensation filter in hardware, the present invention adopts the DSP chip of the ADSP series of Analog Devices, so in a specific embodiment of the present invention, the digital signal processor 44 includes:

[0138] A Fourier transform module, used for performing Fourier transform on the initial digital multiplexed electrical signal, so as to convert the initial digital multiplexed electrical signal based on the time domain into a frequency domain electrical signal based on the frequency domain;

[0139] The phase compensation filter is used to perform phase compensation on the frequency domain electrical signal to obtain a compensated frequency domain electrical signal. The transfer function of the phase compensation satisfies the following calculation formula:

[0140]

[0141] Where f is the offset frequency, β2 is the second-order dispersion coefficient of the system, and β3 is the third-order dispersion coefficient of the system;

[0142] The inverse transform module is used to perform inverse Fourier transform on the compensated frequency domain electrical signal to obtain a digital multiplexed electrical signal based on the time domain.

[0143] The system for implementing matrix operations of different orders based on a monochromatic dispersive medium provided in an embodiment of the present invention uses a monochromatic dispersive medium unit 3 to perform delay processing on each weighted modulated optical signal, so that the convolution calculation result can be executed after the optical signal intensity of each delayed weighted modulated optical signal in the same time domain is summed up, which greatly improves the computational efficiency of the photon convolution calculation. At the same time, the introduction of a monochromatic dispersive medium to replace the traditional optical delay line is conducive to reducing the size of the optical computing integrated system and reducing the cost.

[0144] In addition, those skilled in the art will appreciate that, although some embodiments herein include certain features included in other embodiments but not other features, the combination of features of different embodiments is meant to be within the scope of the present invention and form different embodiments. For example, any one of the claimed embodiments may be used in any combination.

[0145] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A system for realizing matrix operations of different orders based on monochromatic dispersive media, characterized in that: The system comprises: A multi-wavelength light source, used to generate a plurality of initial optical signals with different wavelengths; A modulation processing unit, used for converting the first matrix into a first matrix electrical signal based on a time sequence, and modulating the first matrix electrical signal into each initial optical signal to form a multi-channel modulated optical signal; The microring array unit includes a plurality of microrings arranged in sequence, the weight of each microring is determined by the value of each element in the second matrix, and each microring is coupled to a modulated optical signal to load different weights on each modulated optical signal to obtain a plurality of weighted modulated optical signals; wherein, The weight of each micro-ring is determined by the value of each element of the second matrix, including: Determine the target matrix position in the second matrix corresponding to the target microring according to the number of permutation bits of the target microring: k=(i-1)×n+j Wherein, k is the number of permutation bits of the target micro-ring, n is the order of the second matrix, i is the number of rows of the target position of the second matrix, and j is the number of columns of the target position of the second matrix; Select the value W of the element at the target position of the second matrix ij as the weight of the target micro-ring; A monochromatic dispersive medium unit includes a plurality of monochromatic dispersive media corresponding to a plurality of microrings arranged in sequence, each monochromatic dispersive medium receives a weighted modulated optical signal output by the microring corresponding thereto, and performs delay processing on each weighted modulated optical signal to obtain a delayed weighted modulated optical signal, wherein the delay time length of each monochromatic dispersive medium is determined according to the number of arrangement bits thereof, the order of the first matrix and the order of the second matrix; The acquisition and processing unit is used to integrate the delayed weighted modulated optical signals into the same optical path to obtain a multiplexed optical signal, and convert the multiplexed optical signal into a multiplexed electrical signal, and extract the data information of the third matrix from the multiplexed electrical signal, wherein the third matrix is ​​the calculation result of the convolution calculation performed by the second matrix on the first matrix.

2. The system according to claim 1, characterized in that The delay time t of the target monochromatic dispersive medium corresponding to the target microring k Satisfies the following calculation formula: t k =[(n-i)×m+n-j]×Δτ Wherein, m is the order of the first matrix, and Δτ is the time length used by the modulation processing unit to modulate one bit of signal.

3. The system according to claim 2, characterized in that The monochromatic dispersive medium is a fiber Bragg grating based on lithium niobate material, and the monochromatic dispersive medium unit further includes: The electric control unit is used to control the delay time of the fiber Bragg grating by adjusting the voltage acting on the fiber Bragg grating; wherein, The delay time of the fiber Bragg grating and the voltage U acting on the fiber Bragg grating satisfy the following calculation formula: Where △n is the change in the refractive index of the fiber Bragg grating, L is the length of the fiber Bragg grating, C is the transmission rate of light on the optical fiber, n eff is the initial refractive index of the fiber Bragg grating, r is the electro-optic coefficient of the fiber Bragg grating, and d is the electrode spacing acting on the fiber Bragg grating.

4. The system according to claim 2, characterized in that The acquisition and processing unit includes: The result collection unit is used to respectively s Starting from the time instant, sequentially collecting P-bit sampled electrical signals from the multiplexed electrical signal as the s-th row data of the third matrix; T s The moment satisfies the following calculation formula: T S =[(n-1)(m+1)+(s-1)m]·Δτ P satisfies the following calculation formula: P=m-n+1 Where s=1,2,...,P.

5. The system according to claim 1, characterized in that The monochromatic dispersive medium unit further comprises: The temperature control unit is used to control the temperature of each monochromatic dispersive medium in the monochromatic dispersive medium unit so as to keep the temperature of each monochromatic dispersive medium stable.

6. The system according to claim 1, characterized in that The acquisition processing unit also includes: A multiplexer, used for receiving each delayed weighted modulated optical signal, combining each delayed weighted modulated optical signal into one multiplexed optical signal, so as to sum the optical signal strengths of different delayed weighted modulated optical signals in the same time domain; a photodetector for converting the multiplexed optical signal into an original analog multiplexed electrical signal; an analog-to-digital converter, used for converting an initial analog multiplexed electrical signal into an initial digital multiplexed electrical signal; A digital signal processor, used for performing distortion compensation on each initial digital multiplexed electrical signal to obtain a digital multiplexed electrical signal; The digital-to-analog converter is used to convert the digital multiplexed electrical signal into a multiplexed electrical signal for executing the convolution calculation result.

7. The system according to claim 6, characterized in that The digital signal processor includes: A Fourier transform module, used for performing Fourier transform on the initial digital multiplexed electrical signal, so as to convert the initial digital multiplexed electrical signal based on the time domain into a frequency domain electrical signal based on the frequency domain; The phase compensation filter is used to perform phase compensation on the frequency domain electrical signal to obtain a compensated frequency domain electrical signal. The transfer function of the phase compensation satisfies the following calculation formula: Where f is the offset frequency, β2 is the second-order dispersion coefficient of the system, and β3 is the third-order dispersion coefficient of the system; The inverse transform module is used to perform inverse Fourier transform on the compensated frequency domain electrical signal to obtain a digital multiplexed electrical signal based on the time domain.

8. The system according to claim 1, characterized in that The modulation unit includes: A wavelength division multiplexer, used for receiving initial optical signals and multiplexing the initial optical signals onto the same optical signal transmission medium to obtain a wavelength division multiplexed optical signal; A modulator, configured to receive a first matrix electrical signal and load the first matrix electrical signal into the wavelength division multiplexed optical signal to form a multiplexed modulated optical signal; The demultiplexer is used to receive the multiplexed modulated optical signal and demultiplex the multiplexed modulated optical signal into a plurality of modulated optical signals with different wavelengths according to different wavelengths.

9. The system according to claim 8, characterized in that The modulation processing unit also includes: The input matrix processing unit is used to flatten the first matrix, and convert the values ​​of each element of the first matrix into corresponding electrical signals in time sequence to form a first matrix electrical signal, and input the first matrix electrical signal into the modulator.

Images