Memristor measurement matrix determination method and device, equipment, storage medium and product

Abstract

The application provides a memristor measurement matrix determination method, device, equipment, storage medium and product. The method comprises the following steps: determining the maximum value of the candidate measurement matrix bit width according to the conductance deviation characteristics of the target memristor and the target compression ratio; obtaining the target bit width interval based on the least significant bit width corresponding to the target memristor and the maximum value; generating the candidate measurement matrix corresponding to each bit width in the target bit width interval based on the statistical distribution characteristics of the original measurement matrix; performing signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each candidate measurement matrix to obtain the signal reconstruction accuracy corresponding to each candidate measurement matrix; and taking the candidate measurement matrix with the highest signal reconstruction accuracy as the measurement matrix of the memristor array corresponding to the target memristor, so as to enable the memristor array to perform compressive sensing. The measurement matrix used can meet the actual situation of the operation of the target memristor, and overcome the negative influence of the device characteristics on the reconstruction accuracy of compressive sensing.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, device, storage medium and product for determining a memristor measurement matrix. Background Technology

[0002] With the rapid development of the Internet of Things (IoT) and edge computing, edge devices face the pressure of storing and transmitting massive amounts of data during signal acquisition and processing. Compressed sensing technology can significantly reduce the amount of data by compressing signals at the sampling end. Furthermore, the in-memory computing architecture based on memristor cross arrays can efficiently complete matrix operations in compressed sensing by leveraging its analog computing characteristics. This breaks through the energy efficiency and latency bottlenecks of traditional digital processors, providing a hardware implementation path for low-power, high-concurrency signal acquisition at the edge.

[0003] When compressed sensing algorithms are deployed on memristor hardware, the theory relies on a high-precision, continuously distributed measurement matrix to ensure signal reconstruction quality. However, the inherent programming errors and conductance drift of memristor devices cause the physical weights mapped onto the array to deviate significantly from their theoretical values. This introduces significant hardware noise into the compressed measurement and signal reconstruction stages, ultimately degrading the quality of the recovered signal. Existing solutions struggle to achieve efficient signal acquisition in resource-constrained edge scenarios and cannot overcome the negative impact of device characteristics on the accuracy of compressed sensing reconstruction. Summary of the Invention

[0004] This application provides a method, apparatus, device, storage medium, and product for determining a memristor measurement matrix, aiming to solve the technical problem in the related art of being unable to overcome the negative impact of device characteristics on the accuracy of compressed sensing reconstruction.

[0005] In a first aspect, this application provides a method for determining a memristor measurement matrix, the method comprising: Based on the conductance deviation characteristics of the target memristor and the target compression ratio, determine the maximum bit width of the candidate measurement matrix; Based on the minimum effective bit width and the maximum value corresponding to the target memristor, the target bit width range is obtained; Based on the statistical distribution characteristics of the original measurement matrix, candidate measurement matrices corresponding to each bit width in the target bit width range are generated; Based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices, signal reconstruction simulation is performed to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices; The candidate measurement matrix with the highest signal reconstruction accuracy is used as the measurement matrix of the memristor array corresponding to the target memristor, for use in the memristor array to perform compressed sensing.

[0006] In some possible implementations, generating candidate measurement matrices corresponding to each bit width in the target bit width interval based on the statistical distribution characteristics of the original measurement matrix includes: For each bit width within the target bit width range, perform the following steps: Calculate the mean and standard deviation of the original measurement matrix to obtain the statistical distribution characteristics of the original measurement matrix; The element values ​​of each element in the original measurement matrix are mapped to corresponding discrete values ​​to obtain the first matrix. The discrete values ​​include different fixed values, and the number of fixed values ​​corresponds to the bit width. Based on the statistical distribution characteristics, the first matrix is ​​transformed to obtain the candidate measurement matrix corresponding to the bit width.

[0007] In some possible implementations, transforming the first matrix based on the statistical distribution characteristics to obtain the candidate measurement matrix corresponding to the bit width includes: Based on the standard deviation of the original measurement matrix, the first matrix is ​​scaled to obtain a second matrix, the second matrix having the same standard deviation as the original measurement matrix; Based on the mean of the original measurement matrix, the second matrix is ​​shifted to obtain a candidate measurement matrix corresponding to the bit width. The candidate measurement matrix has the same standard deviation and mean as the original measurement matrix.

[0008] In some possible implementations, scaling the first matrix according to the standard deviation of the original measurement matrix to obtain the second matrix includes: Calculate the standard deviation of the first matrix, and determine the ratio of the standard deviation of the original measurement matrix to the standard deviation of the first matrix; Multiply each element of the first matrix by the ratio to obtain the second matrix.

[0009] In some possible implementations, the step of shifting the second matrix according to the mean of the original measurement matrix to obtain the candidate measurement matrix corresponding to the bit width includes: Calculate the mean of the second matrix, and determine the difference between the mean of the original measurement matrix and the mean of the second matrix; By adding the difference to each element in the second matrix, a candidate measurement matrix corresponding to the bit width is obtained.

[0010] In some possible implementations, the step of performing signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices includes: Based on the preset noise simulation model corresponding to the target memristor, the corresponding noise is injected into each of the candidate measurement matrices to obtain the corresponding noise matrices. Based on each of the noise matrices, the data is compressed and recovered using the preset noise simulation model to determine the signal reconstruction accuracy corresponding to each of the candidate measurement matrices.

[0011] In some possible implementations, determining the maximum bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio includes: Based on the conductance deviation characteristics of the target memristor, determine the average value of the relative deviation parameters of the target memristor in each conductance state; The maximum value of the candidate measurement matrix bit width is determined by multiplying the target compression ratio, the average value, and the preset threshold parameter.

[0012] Secondly, this application provides a signal processing method based on memristors, the method comprising: Acquire the raw signals collected by the sensor; The original signal is converted into an electrical signal; The electrical signal is input to a memristor array, and the compressed signal is obtained by parallel calculation of each memristor in the memristor array corresponding to the element value of the measurement matrix; the measurement matrix is ​​determined according to the method described above.

[0013] Thirdly, this application provides a memristor measurement matrix determination apparatus, the apparatus comprising: The determination module is used to determine the maximum value of the bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio; The determining module is also used to obtain the target bit width range based on the lowest effective bit width corresponding to the target memristor and the maximum value; The generation module is used to generate candidate measurement matrices for each bit width in the target bit width range based on the statistical distribution characteristics of the original measurement matrix. The simulation module is used to perform signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices, so as to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices; The execution module is used to take the candidate measurement matrix with the highest reconstruction accuracy of the signal as the measurement matrix of the memristor array corresponding to the target memristor, so as to perform compressed sensing on the memristor array.

[0014] Fourthly, this application provides a memristor measurement matrix determination device, the device comprising: a processor, and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the memristor measurement matrix determination method as described above.

[0015] Fifthly, this application provides a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the memristor measurement matrix determination method described above.

[0016] Sixthly, this application provides a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the memristor measurement matrix determination method as described above.

[0017] The memristor measurement matrix determination method, apparatus, device, storage medium, and product provided in this application determine the maximum value of the candidate measurement matrix bit width based on the conductance deviation characteristics and target compression ratio of the target memristor, thereby determining the search range of the bit width. This avoids redundant searches for high bit widths and improves optimization efficiency. Secondly, within this range, a corresponding candidate measurement matrix is ​​generated for each candidate bit width. Based on a preset device noise simulation model, end-to-end signal reconstruction simulation is performed on each candidate measurement matrix to simulate the system performance under hardware characteristics, evaluate the reconstruction accuracy of each matrix, and select the candidate measurement matrix with the highest signal reconstruction accuracy in the simulation as the final deployed measurement matrix. The scheme determines a reasonable target bit width range based on the conductance deviation characteristics of the target memristor, thereby generating multiple candidate measurement matrices corresponding to the bit width. Then, based on the preset noise simulation model corresponding to the target memristor, simulation tests are performed on the multiple candidate measurement matrices to determine the matrix with the best signal reconstruction accuracy among the multiple candidate measurement matrices as the measurement matrix of the target memristor. The corresponding measurement matrix is ​​determined based on the actual situation of the target memristor, so that the measurement matrix used can conform to the actual situation of the target memristor running the compressed sensing algorithm, and overcome the negative impact of the device characteristics of the memristor on the reconstruction accuracy of compressed sensing. Attached Figure Description

[0018] This application can be better understood from the following description of specific embodiments in conjunction with the accompanying drawings, wherein: Other features, objects, and advantages of this application will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings, wherein the same or similar reference numerals denote the same or similar features.

[0019] Figure 1 This is a flowchart of a memristor measurement matrix determination method provided in one embodiment of this application; Figure 2 This is a flowchart of a memristor measurement matrix determination method provided in another embodiment of this application; Figure 3 This is a flowchart of a method for determining a memristor measurement matrix provided in another embodiment of this application; Figure 4 This is a schematic flowchart of a signal processing method based on memristors provided in one embodiment of this application; Figure 5 This is an overall flowchart of the memristor measurement matrix determination method of this application; Figure 6 This is a schematic diagram of the structure of a memristor measurement matrix determination device provided in one embodiment of this application; Figure 7 This is a schematic diagram of the hardware structure of the memristor measurement matrix determination device provided in the embodiments of this application. Detailed Implementation

[0020] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0021] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0022] With the rapid development of the Internet of Things (IoT) and edge computing, edge devices face the pressure of massive data collection and transmission. Compressed Sensing (CS), as an efficient signal acquisition paradigm, projects high-dimensional signals into low-dimensional observation vectors through a measurement matrix, achieving significant data compression at the sampling end, thereby greatly reducing storage and transmission bandwidth requirements. In terms of specific hardware implementation, a memristor cross array completes the compressed sensing calculation by mapping the weights of the measurement matrix to the conductance values ​​of the memristors, achieving efficient signal compression at the edge.

[0023] While compressed sensing schemes based on memristor cross arrays offer energy efficiency advantages, they are prone to mismatches between algorithm and device characteristics when deploying measurement matrices in practice. Specifically, compressed sensing theory typically relies on measurement matrices that satisfy constrained equidistant properties, with element values ​​usually represented by high-precision floating-point numbers and continuously distributed. However, programming errors, conductance drift, and inter-device variability in memristor devices cause significant deviations between the actual stored physical weights and the ideal weights when the continuously distributed high-precision matrix is ​​mapped onto the memristor array. This deviation results in the compressed signal calculated by the memristor chip containing substantial hardware noise, and noise is also introduced during the signal reconstruction process by the memristor chip, leading to a low signal-to-noise ratio in the final recovered signal.

[0024] To address the problems of the prior art, embodiments of this application provide a method, apparatus, device, storage medium, and product for determining a memristor measurement matrix. The method for determining a memristor measurement matrix provided in this application embodiment will be described first below.

[0025] Figure 1 A flowchart illustrating a method for determining a memristor measurement matrix according to an embodiment of this application is shown. Figure 1 As shown, the method includes the following steps: S101 to S105.

[0026] S101: Determine the maximum bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio.

[0027] S102: Based on the minimum effective bit width and maximum value corresponding to the target memristor, obtain the target bit width range.

[0028] S103: Based on the statistical distribution characteristics of the original measurement matrix, generate candidate measurement matrices corresponding to each bit width in the target bit width interval.

[0029] S104: Based on the preset noise simulation model corresponding to the target memristor and each candidate measurement matrix, perform signal reconstruction simulation to obtain the signal reconstruction accuracy corresponding to each candidate measurement matrix.

[0030] S105: The candidate measurement matrix with the highest signal reconstruction accuracy is used as the measurement matrix of the memristor array corresponding to the target memristor, so as to perform compressed sensing on the memristor array.

[0031] In the specific implementation of S101, measured conductance data of the target memristor under stable operating conditions are acquired. Statistical analysis of this data is performed to extract statistical characteristics of its conductance deviation, such as calculating its standard deviation or fitting a probability distribution model of the deviation. Subsequently, combined with a preset target compression ratio (the ratio of the original signal length to the number of compressed measurements), a mathematical model is established based on the constraint relationship between measurement matrix accuracy and reconstruction error in compressed sensing theory. This model quantizes the conductance deviation as equivalent noise interference to the measured values ​​and uses the maximum information loss allowed by the target compression ratio as a constraint. Through numerical calculation, the maximum quantization error allowed by the measurement matrix elements under the initial reconstruction requirements is solved, and then this maximum quantization error is converted into the corresponding digital bit width.

[0032] In order to accurately determine the target bit width range, in some implementations, S101 may include the following steps: S1011 to S1012.

[0033] S1011: Based on the conductance deviation characteristics of the target memristor, determine the average value of the relative deviation parameters of the target memristor in each conductance state.

[0034] S1012: Determine the maximum value of the bit width of the candidate measurement matrix based on the product of the target compression ratio, the average value, and the preset threshold parameter.

[0035] In the specific implementation of S1011, multiple repeated read operations are performed under various preset conductance states of the memristor array to measure and record the stable conductance value in each conductance state. Then, for each independent conductance state, its relative deviation parameter is calculated, and the coefficient of variation is used to characterize the volatility of that conductance state. Specifically, the standard deviation of all measurements under a single conductance state is calculated, and then this standard deviation is divided by the arithmetic mean of all measurements under that conductance state to obtain the coefficient of variation corresponding to that conductance state. This calculation is performed for all conductance states to obtain the corresponding coefficient of variation values. Finally, the arithmetic mean of the coefficients of variation for all conductance states is calculated, which is the average value of the relative deviation parameter of the target memristor in each conductance state.

[0036] In the specific implementation of S1012, the target compression ratio, the average value of the relative deviation parameter calculated in the previous step, and the preset threshold parameter are multiplied to obtain the maximum value of the candidate measurement matrix bit width.

[0037] The above-described embodiments of this application determine the average value of the relative deviation parameter of the target memristor in each conductance state by using the conductance deviation characteristics of the target memristor. Then, based on the product of the target compression ratio, the average value, and the preset threshold parameter, the maximum value of the candidate measurement matrix bit width is determined, thereby accurately determining the target bit width range.

[0038] In the specific implementation of S102, the determination of the minimum effective bit width depends on the calibration of the physical characteristics of the target memristor array. Specifically, under an ideal simulation environment that ignores other noise sources, the numerical bit width of the measurement matrix is ​​gradually reduced during compressed sensing reconstruction testing. The bit width value corresponding to the point where the reconstruction accuracy begins to decrease significantly is defined as the minimum effective bit width that the memristor array can effectively respond to and utilize. Then, the maximum bit width calculated in the first step and this minimum effective bit width are used as the two ends of an interval to form a continuous set of integer bit widths, namely the target bit width interval.

[0039] In the specific implementation of S103, the original floating-point measurement matrix that meets the requirements of compressed sensing theory is obtained, and the statistical distribution of all its matrix elements is analyzed to determine its value range, mean, and variance. Then, for each specific bit width value within the target bit width range, uniform quantization is performed. Based on the number of integers that the bit width can represent, the continuous value range of the original matrix elements is uniformly divided into corresponding quantization intervals. Each original floating-point element value is mapped to its corresponding quantization interval, and the center value or boundary value corresponding to that interval is used as its quantized value, thereby generating a candidate measurement matrix with fixed-point elements. This process is repeated for all bit widths to generate a series of candidate matrix sets with different numerical precisions.

[0040] To determine the optimization matrix for each bit width separately, in some implementations, reference is made to... Figure 2 S103 may include the following steps: S1031 to S1033.

[0041] For each bit width within the target bit width range, perform the following steps: S1031: Calculate the mean and standard deviation of the original measurement matrix to obtain the statistical distribution characteristics of the original measurement matrix.

[0042] S1032: Map the element values ​​of each element in the original measurement matrix to their corresponding discrete values ​​to obtain the first matrix. The discrete values ​​include different fixed values, and the number of fixed values ​​corresponds to the bit width.

[0043] S1033: Transform the first matrix based on statistical distribution characteristics to obtain the candidate measurement matrix corresponding to the bit width.

[0044] In the specific implementation of S1031, the original measurement matrix is ​​obtained as a reference. This matrix is ​​typically a floating-point matrix randomly generated according to a specific probability distribution, and its dimension is determined by the number of compressed sensing measurements and the length of the original signal. Then, statistics are performed: each element in the original measurement matrix is ​​traversed, the values ​​of all elements are summed, and this sum is divided by the total number of elements in the matrix to calculate the arithmetic mean of all elements in the original measurement matrix. Subsequently, the standard deviation of these elements is calculated. The mean and standard deviation constitute the core statistical features describing the central location and dispersion of the matrix's numerical distribution.

[0045] In the specific implementation of S1032, for the current target bit width, the total number of uniformly distributed discrete values ​​that the bit width can represent is determined. Each floating-point element value in the original measurement matrix is ​​traversed, and based on its relative position within the range defined by the minimum and maximum values ​​of all elements in the original matrix, it is linearly normalized to an integer corresponding to the discrete value index. After performing this operation on all elements in the original matrix, a "first matrix" is obtained where all elements are composed of discrete values.

[0046] In the specific implementation of S1033, the first matrix after discrete quantization is standardized to restore or approximate the statistical distribution characteristics of the original measurement matrix, thereby satisfying the normative requirements of compressed sensing theory for the measurement matrix. Specifically, based on the mean and standard deviation of the original measurement matrix calculated in S1031, and the mean and standard deviation of the first matrix obtained in S1032, a linear transformation is performed on each element value of the first matrix. Specifically, each element value is subtracted from the mean of the first matrix itself to center it to zero mean; then, the centered value is divided by the standard deviation of the first matrix itself to scale it to unit standard deviation; finally, this standardized value is multiplied by the standard deviation of the original matrix and added to the mean of the original matrix to generate a candidate measurement matrix. This transformation ensures that the generated candidate measurement matrix has a mean and standard deviation consistent with the original measurement matrix in mathematical expectation, thus maintaining the necessary statistical distribution characteristics while preserving the quantization characteristics caused by the bit width limitation.

[0047] The above-described implementation method of this application calculates the mean and standard deviation of the original measurement matrix to obtain the statistical distribution characteristics of the original measurement matrix. The element values ​​of each element in the original measurement matrix are mapped to corresponding discrete values ​​to obtain a first matrix. The discrete values ​​include different fixed values, and the number of fixed values ​​corresponds to the bit width. Then, the first matrix is ​​transformed based on the statistical distribution characteristics to obtain the candidate measurement matrix corresponding to the bit width, thereby determining the optimization matrix corresponding to each bit width.

[0048] To obtain a usable optimization matrix, in some implementations, reference is made to... Figure 3 S1033 may include the following steps: S301 to S302.

[0049] S301: Based on the standard deviation of the original measurement matrix, the first matrix is ​​scaled to obtain the second matrix, which has the same standard deviation as the original measurement matrix.

[0050] S302: Based on the mean of the original measurement matrix, the second matrix is ​​shifted to obtain the candidate measurement matrix corresponding to the bit width. The standard deviation and mean of the candidate measurement matrix are the same as those of the original measurement matrix.

[0051] In the specific implementation of S301, the statistical standard deviation of all elements in the first matrix is ​​calculated. This value characterizes the dispersion of element values ​​within the first matrix. Simultaneously, the standard deviation of the original measurement matrix, analyzed in previous steps, is obtained. Next, the standard deviation of the original measurement matrix is ​​compared with the standard deviation of the first matrix, and the proportional relationship between the two is calculated. This proportion serves as the scaling factor for subsequent operations. Then, based on this scaling factor, a holistic numerical scaling process is performed on the first matrix. Specifically, each independent element in the first matrix is ​​traversed, and its current value is multiplied by the calculated scaling factor to generate a new value. After performing this multiplication operation on all elements in the matrix, a new matrix, called the second matrix, is obtained. Through this series of proportionally uniform scaling operations, the discrete distribution range of values ​​within the second matrix is ​​precisely adjusted, ensuring that its standard deviation remains consistent with the standard deviation of the original measurement matrix.

[0052] In the specific implementation of S302, the arithmetic mean of all elements in the second matrix is ​​calculated, reflecting the central position of the numerical distribution of the second matrix. Simultaneously, the mean of the original measurement matrix stored in the previous step is retrieved and retrieved. Next, the mean of the original measurement matrix is ​​subtracted from the mean of the second matrix; the difference is the required translation amount. Then, based on this translation amount, a global numerical translation process is performed on the second matrix. Specifically, each independent element in the second matrix is ​​traversed, and its current value is added to the calculated translation amount to generate a final adjusted value. After performing this addition operation on all elements in the matrix, the resulting new matrix is ​​the candidate measurement matrix corresponding to the current specific positioning width. Through this translation operation, the center of the numerical distribution of the candidate measurement matrix is ​​calibrated to the same position as the mean of the original measurement matrix. Since the previous scaling operation has ensured the consistency of the standard deviations of the two, the resulting candidate measurement matrix after this shift has the same statistical mean and standard deviation as the original measurement matrix. Thus, under the premise of satisfying the specific quantization bit width constraint, the statistical distribution characteristics of the original matrix are maintained to the greatest extent.

[0053] In the above implementation of this application, the first matrix is ​​scaled down using the standard deviation of the original measurement matrix to obtain a second matrix. The second matrix has the same standard deviation as the original measurement matrix. Then, the second matrix is ​​shifted according to the mean of the original measurement matrix to obtain a candidate measurement matrix corresponding to the bit width. The candidate measurement matrix has the same standard deviation and mean as the original measurement matrix, thereby obtaining an usable optimized matrix.

[0054] In order to accurately scale, in some embodiments, S301 may include the following steps: S3011 to S3012.

[0055] S3011: Calculate the standard deviation of the first matrix and determine the ratio of the standard deviation of the original measurement matrix to the standard deviation of the first matrix.

[0056] S3012: Multiply each element in the first matrix by the ratio to obtain the second matrix.

[0057] In the specific implementation of S3011, all numerical elements in the first matrix are accessed and read, and the arithmetic mean of these elements is calculated. Then, the square of the difference between each element value and the arithmetic mean is calculated to obtain a set of squared deviation values. Next, all squared deviation values ​​are summed, and this sum is divided by the total number of elements to obtain the variance. Finally, the square root of this variance is performed to obtain the standard deviation of the first matrix, which quantitatively describes the dispersion of the elements within the first matrix. Simultaneously, the standard deviation of the original measurement matrix, calculated and stored in the previous steps, is retrieved. The standard deviation of the original measurement matrix is ​​used as the dividend, and the standard deviation of the first matrix is ​​used as the divisor; a division operation is performed, and the quotient is the precise ratio between the two.

[0058] In the specific implementation of S3012, the first matrix is ​​traversed, and each independent numerical element in the first matrix is ​​accessed in turn. The current value of the element is used as the multiplicand, and the ratio mentioned above is used as the multiplier to perform a multiplication operation. This multiplication operation is repeated for each element in the matrix. After all elements have been calculated, the resulting series of new values ​​are recombine into a new matrix according to the original row and column structure of the first matrix. This matrix is ​​the second matrix.

[0059] The above-described implementation method of this application calculates the standard deviation of the first matrix, determines the ratio of the standard deviation of the original measurement matrix to the standard deviation of the first matrix, and then multiplies each element in the first matrix by the ratio to obtain the second matrix. This allows for accurate scaling to obtain the second matrix.

[0060] In order to perform translation accurately, in some embodiments, S302 may include the following steps: S3021 to S3022.

[0061] S3021: Calculate the mean of the second matrix and determine the difference between the mean of the original measurement matrix and the mean of the second matrix.

[0062] S3022: Add the difference to each element in the second matrix to obtain the candidate measurement matrix corresponding to the bit width.

[0063] In the specific implementation of S3021, all numerical elements in the second matrix are accessed and read, and these elements are summed one by one to obtain the total value of all elements. Then, this total value is divided by the total number of elements in the second matrix, and the quotient is the arithmetic mean of the second matrix, which represents the center position of the numerical distribution of the second matrix. Simultaneously, the arithmetic mean of the original measurement matrix stored in the previous process is called and read. Next, a difference operation is performed to determine the translation amount. The mean of the original measurement matrix is ​​used as the minuend, and the mean of the second matrix just calculated is used as the subtrahend; the result is the difference between the two.

[0064] In the specific implementation of S3022, the precise difference calculated in S3021 is obtained. Next, the second matrix is ​​traversed, sequentially accessing each independent numerical element. The current value of that element is used as the addend, and the aforementioned difference is used as the addend, to perform an addition operation, thereby calculating a new value after final adjustment. This addition operation is repeated for each element in the second matrix. After all elements have been calculated, the resulting series of new values ​​are recombine into a new matrix according to the original row and column structure of the second matrix. This matrix is ​​the candidate measurement matrix corresponding to the current specific binary bit width.

[0065] In the above implementation of this application, the difference between the mean of the original measurement matrix and the mean of the second matrix is ​​determined by calculating the mean of the second matrix. Then, the difference is added to each element in the second matrix to obtain the candidate measurement matrix corresponding to the bit width. The matrix is ​​then accurately shifted to obtain the optimized matrix.

[0066] In the specific implementation of S104, a pre-established noise simulation model capable of simulating the non-ideal characteristics of the target memristor array is constructed or invoked. This model, in the form of function or circuit simulation, takes the ideal target conductance value as input and outputs the actual measured value after superimposing comprehensive effects such as conductance deviation noise and thermal noise from the read / write circuit. Subsequently, for each candidate measurement matrix, multiple sets of signals from the standard test signal library are selected as the original signals; the candidate matrix is ​​used to perform a simulated "compressed measurement" of the original signals to obtain an ideal measurement vector; this ideal vector is input into a preset noise simulation model to obtain a simulated, noisy actual measurement vector; finally, a specified reconstruction algorithm, such as the orthogonal matching pursuit algorithm, is used to reconstruct the signal from the noisy measurement vector. The signal reconstruction accuracy of the candidate matrix in this simulation environment is quantified by calculating the error index between the reconstructed signal and the original signal.

[0067] In order to perform accurate simulation, in some implementations, S104 may include the following steps: S1041 to S1042.

[0068] S1041: Based on the preset noise simulation model corresponding to the target memristor, inject the corresponding noise into each candidate measurement matrix to obtain the corresponding noise matrix.

[0069] S1042: Based on each noise matrix, the data is compressed and recovered using a preset noise simulation model to determine the signal reconstruction accuracy corresponding to each candidate measurement matrix.

[0070] In the specific implementation of S1041, a pre-established simulation model of non-ideal characteristics for the target memristor is invoked. This model encapsulates mathematical or statistical descriptions of various physical effects such as conductance drift, write operation fluctuations, and thermal noise in the read circuit. Subsequently, for each matrix in the candidate measurement matrix set, noise injection is performed. Each numerical element in the candidate measurement matrix is ​​traversed, and the ideal conductance value corresponding to that element is passed as an input parameter to the pre-defined noise simulation model. This model, based on its internally defined noise generation mechanism, calculates a noise value or noise offset associated with that input value. Then, this calculated noise offset is algebraically added to the original value of the corresponding element in the candidate measurement matrix, generating a new element value superimposed with simulated noise. After performing the noise offset calculation and superposition operation on all elements in the candidate measurement matrix, a new matrix with the same dimensions as the original candidate matrix, but where each element value already contains simulated hardware non-ideal noise, is obtained; this new matrix is ​​called the noise matrix.

[0071] In the specific implementation of S1042, one or more sets of known original test signals are selected from a standard signal library. Then, for each noise matrix obtained, a simulation test is performed. Specifically, the noise matrix is ​​used as the measurement matrix in compressed sensing. Simulated compressed sampling calculations are performed on each set of original test signals, i.e., the product of the original signal vector and the noise matrix is ​​calculated to obtain a theoretical compressed measurement value vector. To more realistically simulate the complete noise impact of the hardware link, this measurement value vector can be input again into the aforementioned preset noise simulation model, applying additional noise that may exist in subsequent stages such as analog-to-digital conversion, forming the final simulated measurement value vector. Then, a predetermined signal reconstruction algorithm is used to process the simulated measurement value vector to recover the original signal, obtaining a reconstructed signal vector. Finally, the reconstructed signal vector is compared with the original test signal vector, and the difference between the two is quantified by calculating a preset accuracy evaluation index. This index value characterizes the signal reconstruction accuracy of the currently used candidate measurement matrix in this simulation environment. Repeating this complete simulation process for each noise matrix and its corresponding candidate measurement matrix allows each candidate measurement matrix to be assigned a quantified reconstruction accuracy value.

[0072] The above-described implementation method of this application involves injecting corresponding noise into each candidate measurement matrix based on a preset noise simulation model corresponding to the target memristor to obtain corresponding noise matrices. Then, based on each noise matrix, the preset noise simulation model is used to compress and recover the data to determine the signal reconstruction accuracy corresponding to each candidate measurement matrix, thereby accurately performing the simulation.

[0073] In the specific implementation of S105, the signal reconstruction accuracy values ​​corresponding to each candidate measurement matrix obtained in step S104 are compared. By sorting or finding the maximum value, the candidate measurement matrix that optimizes the reconstruction accuracy index is determined. The bit width specifications of this matrix and all its specific fixed-point values ​​are finally determined as the measurement matrix parameters adapted to the current target memristor array.

[0074] The memristor measurement matrix determination method provided in this application determines the maximum value of the candidate measurement matrix bit width based on the conductance deviation characteristics and target compression ratio of the target memristor, thereby determining the search range for the bit width. This avoids redundant searches for high bit widths and improves optimization efficiency. Secondly, within this range, a corresponding candidate measurement matrix is ​​generated for each candidate bit width. Based on a preset device noise simulation model, end-to-end signal reconstruction simulation is performed on each candidate measurement matrix to simulate system performance under hardware characteristics, evaluate the reconstruction accuracy of each matrix, and select the candidate measurement matrix with the highest signal reconstruction accuracy in the simulation as the final deployed measurement matrix. The scheme determines a reasonable target bit width range based on the conductance deviation characteristics of the target memristor, thereby generating multiple candidate measurement matrices corresponding to different bit widths. Then, based on a preset noise simulation model corresponding to the target memristor, simulation tests are performed on multiple candidate measurement matrices to determine the matrix with the best signal reconstruction accuracy among the multiple candidate measurement matrices as the measurement matrix for the target memristor. By determining the corresponding measurement matrix based on the actual situation of the target memristor, the used measurement matrix can conform to the actual situation of the target memristor running the compressed sensing algorithm, overcoming the negative impact of the memristor's device characteristics on the compressed sensing reconstruction accuracy.

[0075] Figure 4 A schematic flowchart of a memristor-based signal processing method according to an embodiment of this application is shown. Figure 4 As shown, the method includes the following steps: S401 to S403.

[0076] S401: Acquire the raw signal collected by the sensor.

[0077] S402: Converts the original signal into an electrical signal.

[0078] S403: The electrical signal is input to the memristor array. Parallel calculations are performed by each memristor in the array corresponding to the element values ​​of the measurement matrix to obtain the compressed signal. The measurement matrix is ​​determined according to the method described above.

[0079] In the specific implementation of S401, the physical signal collected by the target sensor is digitized according to the preset sampling frequency and quantization accuracy to form a set of discrete digital sequences representing the original physical information. This digital sequence is the original signal to be processed.

[0080] In the specific implementation of S402, the original signal is converted into an electrical signal proportionally, which can be a current input signal or a voltage input signal.

[0081] In the specific implementation of S403, multiple electrical signals corresponding to each component of the original signal are applied in parallel to the corresponding input ports of the memristor array via pre-arranged wires. The conduction state of each memristor cell in the memristor array has been pre-programmed according to the corresponding element values ​​of the finally selected target measurement matrix. When the electrical signals representing the original signal are input simultaneously, the electrical signals are processed at the intersection structure inside the array with the measurement matrix element values ​​represented by the conductance values ​​of each memristor. This is equivalent to performing a matrix multiplication between the original signal vector and the measurement matrix, resulting in a compressed signal.

[0082] The memristor-based signal processing method provided in this application acquires the raw signal collected by the sensor, converts the raw signal into an electrical signal, and then inputs the electrical signal into a memristor array. The memristors in the memristor array, corresponding to the element values ​​of the measurement matrix, perform parallel calculations to obtain a compressed signal. This results in a suitable compressed signal.

[0083] In some implementations, to reduce the impact of hardware noise on signal acquisition, the matrix weights can be quantized to reduce dependence on unstable intermediate conductance states. However, direct quantization alters the statistical distribution of the matrix weights, thereby violating the finite isometry property upon which compressed sensing theory relies. Therefore, a linear transformation can be used to align the mean and standard deviation of the matrices before and after quantization, thus maintaining the finite isometry property. In the specific execution flow, the system uses the input original measurement matrix... Compression ratio of compressed sensing tasks Relative deviation parameters of each conductance state and threshold parameters The empirical upper limit for quantization bits is calculated as follows: For example, compression ratio Average relative deviation of each conductance state , Experience points Calculated .

[0084] Based on the range of values ​​for the quantization bits, for Internally, based on the constrained quantization strategy, an optimized measurement matrix for the current bit width is generated. Subsequently, the verification module calls the input memristor non-ideal characteristic parameters to verify the current... At most under bit quantization Each conductance state is injected into a corresponding noise simulation model to calculate and record the signal reconstruction accuracy under the influence of noise. After traversing all legal bit widths within the interval, a global optimization is performed, and the bit width corresponding to the highest reconstruction accuracy is output as the optimal quantization bit width. and output the corresponding measurement matrix. As the final deployment solution, this process ensures that the optimized matrix not only meets the mathematical and statistical requirements of compressed sensing, but also adapts to the non-ideal characteristics of memristors.

[0085] Specifically, in the process of quantization strategy, the matrix weights are quantized to... After that, scaling and translation are used to make the processed matrix... Compared with the matrix before quantization Maintain the same mean and standard deviation. Optimized matrix. The Euclidean length of the vector is preserved in the statistical expectation sense. Therefore, the generated low-bit-width measurement matrix... While effectively suppressing the influence of hardware non-ideal characteristics, it still maintains a high-quality measurement matrix that meets the requirements of compressed sensing theory. For details, please refer to... Figure 5 .

[0086] The embodiments described in this application significantly overcome the limitations of traditional schemes in deploying compressed sensing algorithms on non-ideal devices through joint optimization of physical mechanisms and mathematical statistics. First, this invention, through a constrained quantization strategy, maps the measurement matrix weights to more stable conductance states with stronger noise immunity in the memristor array at the physical level, greatly reducing the interference of non-ideal device characteristics on computational accuracy. Simultaneously, at the mathematical level, aligning the mean and variance ensures that the low-bit matrix still satisfies the finite equidistant property of compressed sensing, thereby simplifying hardware deployment while guaranteeing theoretical signal recovery capability. Second, this invention introduces a hardware-aware closed-loop verification process, automatically searching for the optimal quantization bit width based on the noise model of a specific process node without adding additional hardware circuit overhead such as time or spatial replication. This method achieves the best balance between reconstruction fidelity and noise robustness, and experiments have shown that it can significantly improve signal reconstruction quality under the same noise environment, exhibiting good cost-effectiveness and versatility.

[0087] Based on the memristor measurement matrix determination method provided in the above embodiments, this application also provides specific implementation methods of the memristor measurement matrix determination device. Please refer to the following embodiments.

[0088] First see Figure 6 The memristor measurement matrix determination device 600 provided in this application embodiment includes the following modules: The determination module 601 is used to determine the maximum value of the bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio.

[0089] The determination module 601 is also used to obtain the target bit width range based on the minimum effective bit width and the maximum value corresponding to the target memristor.

[0090] The generation module 602 is used to generate candidate measurement matrices corresponding to each bit width in the target bit width range based on the statistical distribution characteristics of the original measurement matrix.

[0091] The simulation module 603 is used to perform signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each candidate measurement matrix, so as to obtain the signal reconstruction accuracy corresponding to each candidate measurement matrix.

[0092] The execution module 604 is used to select the candidate measurement matrix with the highest signal reconstruction accuracy as the measurement matrix of the memristor array corresponding to the target memristor, so as to perform compressed sensing on the memristor array.

[0093] As one implementation of this application, the generation module 602 includes: The calculation unit is used to calculate the mean and standard deviation of the original measurement matrix to obtain the statistical distribution characteristics of the original measurement matrix.

[0094] The mapping unit is used to map the element values ​​of each element in the original measurement matrix to corresponding discrete values ​​to obtain the first matrix. The discrete values ​​include different fixed values, and the number of fixed values ​​corresponds to the bit width.

[0095] The transformation unit is used to transform the first matrix based on statistical distribution characteristics to obtain the candidate measurement matrix corresponding to the bit width.

[0096] As one implementation of this application, the transformation unit includes: The scaling sub-unit is used to scale the first matrix according to the standard deviation of the original measurement matrix to obtain the second matrix, which has the same standard deviation as the original measurement matrix.

[0097] The translation sub-unit is used to translate the second matrix according to the mean of the original measurement matrix to obtain the candidate measurement matrix corresponding to the bit width. The standard deviation and mean of the candidate measurement matrix are the same as those of the original measurement matrix.

[0098] As one implementation of this application, the scaling subunit includes: The calculation subunit is used to calculate the standard deviation of the first matrix and determine the ratio of the standard deviation of the original measurement matrix to the standard deviation of the first matrix.

[0099] Determine the sub-units used to multiply each element in the first matrix by a ratio to obtain the second matrix.

[0100] As one implementation of this application, the translation sub-unit includes: The calculation sub-unit is used to calculate the mean of the second matrix and determine the difference between the mean of the original measurement matrix and the mean of the second matrix.

[0101] The sub-unit is determined to add the difference to each element in the second matrix to obtain the candidate measurement matrix corresponding to the bit width.

[0102] As one implementation of this application, the simulation module 603 includes: The injection unit is used to inject corresponding noise into each candidate measurement matrix based on the preset noise simulation model corresponding to the target memristor, so as to obtain the corresponding noise matrix.

[0103] The simulation unit is used to compress and recover data based on each noise matrix using a preset noise simulation model, and to determine the signal reconstruction accuracy corresponding to each candidate measurement matrix.

[0104] As one implementation of this application, module 601 includes: The determination unit is used to determine the average value of the relative deviation parameter of the target memristor in each conductance state based on the conductance deviation characteristics of the target memristor.

[0105] The determining unit is also used to determine the maximum value of the bit width of the candidate measurement matrix based on the product of the target compression ratio, the average value, and the preset threshold parameter.

[0106] Each module in the memristor measurement matrix determination device provided in this application embodiment can implement each step in the above-described memristor measurement matrix determination method and achieve the corresponding effect. For the sake of brevity, it will not be described in detail here.

[0107] Figure 7 A schematic diagram of the hardware structure for determining the memristor measurement matrix provided in an embodiment of this application is shown.

[0108] The memristor measurement matrix determination device may include a processor 701 and a memory 702 storing computer program instructions.

[0109] Specifically, the processor 701 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0110] Memory 702 may include mass storage for data or instructions. For example, and not limitingly, memory 702 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 702 may include removable or non-removable (or fixed) media. Where appropriate, memory 702 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 702 is non-volatile solid-state memory.

[0111] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the memristor measurement matrix determination method according to any embodiment of this disclosure.

[0112] The processor 701 reads and executes computer program instructions stored in the memory 702 to implement any of the memristor measurement matrix determination methods in the above embodiments.

[0113] In one example, the memristor measurement matrix determination device may also include a communication interface 703 and a bus 710. Wherein, as Figure 7 As shown, the processor 701, memory 702, and communication interface 703 are connected through bus 710 and complete communication with each other.

[0114] The communication interface 703 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0115] Bus 710 includes hardware, software, or both, that couples components of an online data traffic metering device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 710 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, any suitable bus or interconnect is contemplated herein.

[0116] Furthermore, in conjunction with the memristor measurement matrix determination method in the above embodiments, this application embodiment can provide a computer storage medium for implementation. The computer storage medium stores computer program instructions; when these computer program instructions are executed by a processor, they implement any of the memristor measurement matrix determination methods in the above embodiments.

[0117] This application also provides a computer program product, including a computer program that, when executed, implements any of the memristor measurement matrix determination methods described in the above embodiments.

[0118] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0119] The functional blocks shown in the above-described structural diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0120] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0121] The aspects of this disclosure have been described above with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block in the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowchart illustrations and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowchart illustrations, and combinations of blocks in the block diagrams and / or flowchart illustrations, can also be implemented by an FPGA performing the specified functions or actions, or can be implemented by a combination of an FPGA and computer instructions.

[0122] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for determining a memristor measurement matrix, characterized in that, The method includes: Based on the conductance deviation characteristics of the target memristor and the target compression ratio, determine the maximum bit width of the candidate measurement matrix; Based on the minimum effective bit width and the maximum value corresponding to the target memristor, the target bit width range is obtained; Based on the statistical distribution characteristics of the original measurement matrix, candidate measurement matrices corresponding to each bit width in the target bit width range are generated; Based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices, signal reconstruction simulation is performed to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices; The candidate measurement matrix with the highest signal reconstruction accuracy is used as the measurement matrix of the memristor array corresponding to the target memristor, for use in the memristor array to perform compressed sensing.

2. The memristor measurement matrix determination method according to claim 1, characterized in that, The process of generating candidate measurement matrices for each bit width within the target bit width range based on the statistical distribution characteristics of the original measurement matrix includes: For each bit width within the target bit width range, perform the following steps: Calculate the mean and standard deviation of the original measurement matrix to obtain the statistical distribution characteristics of the original measurement matrix; The element values ​​of each element in the original measurement matrix are mapped to corresponding discrete values ​​to obtain the first matrix. The discrete values ​​include different fixed values, and the number of fixed values ​​corresponds to the bit width. Based on the statistical distribution characteristics, the first matrix is ​​transformed to obtain the candidate measurement matrix corresponding to the bit width.

3. The method for determining the memristor measurement matrix according to claim 2, characterized in that, The transformation of the first matrix based on the statistical distribution characteristics to obtain the candidate measurement matrix corresponding to the bit width includes: Based on the standard deviation of the original measurement matrix, the first matrix is ​​scaled to obtain a second matrix, the second matrix having the same standard deviation as the original measurement matrix; Based on the mean of the original measurement matrix, the second matrix is ​​shifted to obtain a candidate measurement matrix corresponding to the bit width. The candidate measurement matrix has the same standard deviation and mean as the original measurement matrix.

4. The memristor measurement matrix determination method according to claim 3, characterized in that, The step of scaling the first matrix according to the standard deviation of the original measurement matrix to obtain the second matrix includes: Calculate the standard deviation of the first matrix, and determine the ratio of the standard deviation of the original measurement matrix to the standard deviation of the first matrix; Multiply each element of the first matrix by the ratio to obtain the second matrix.

5. The method for determining the memristor measurement matrix according to claim 3, characterized in that, The step of shifting the second matrix according to the mean of the original measurement matrix to obtain the candidate measurement matrix corresponding to the bit width includes: Calculate the mean of the second matrix, and determine the difference between the mean of the original measurement matrix and the mean of the second matrix; By adding the difference to each element in the second matrix, a candidate measurement matrix corresponding to the bit width is obtained.

6. The method for determining the memristor measurement matrix according to claim 1, characterized in that, The step of performing signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices includes: Based on the preset noise simulation model corresponding to the target memristor, the corresponding noise is injected into each of the candidate measurement matrices to obtain the corresponding noise matrices. Based on each of the noise matrices, the data is compressed and recovered using the preset noise simulation model to determine the signal reconstruction accuracy corresponding to each of the candidate measurement matrices.

7. The method for determining the memristor measurement matrix according to any one of claims 1 to 6, characterized in that, The step of determining the maximum bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio includes: Based on the conductance deviation characteristics of the target memristor, determine the average value of the relative deviation parameters of the target memristor in each conductance state; The maximum value of the candidate measurement matrix bit width is determined by multiplying the target compression ratio, the average value, and the preset threshold parameter.

8. A signal processing method based on memristors, characterized in that, The method includes: Acquire the raw signals collected by the sensor; The original signal is converted into an electrical signal; The electrical signal is input to a memristor array, and the compressed signal is obtained by parallel calculation of each memristor in the memristor array corresponding to the element value of the measurement matrix; the measurement matrix is ​​determined by the method according to any one of claims 1 to 7.

9. A memristor measurement matrix determination device, characterized in that, The device includes: The determination module is used to determine the maximum value of the bit width of the candidate measurement matrix based on the conductance deviation characteristics of the target memristor and the target compression ratio; The determining module is also used to obtain the target bit width range based on the lowest effective bit width corresponding to the target memristor and the maximum value; The generation module is used to generate candidate measurement matrices for each bit width in the target bit width range based on the statistical distribution characteristics of the original measurement matrix. The simulation module is used to perform signal reconstruction simulation based on the preset noise simulation model corresponding to the target memristor and each of the candidate measurement matrices, and to obtain the signal reconstruction accuracy corresponding to each of the candidate measurement matrices; The execution module is used to take the candidate measurement matrix with the highest reconstruction accuracy of the signal as the measurement matrix of the memristor array corresponding to the target memristor, so as to perform compressed sensing on the memristor array.

10. A memristor measurement matrix determination device, characterized in that, The device includes: a processor and a memory storing computer program instructions; the processor reads and executes the computer program instructions to implement the memristor measurement matrix determination method as described in any one of claims 1-7.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer program instructions that, when executed by a processor, implement the memristor measurement matrix determination method as described in any one of claims 1-7.

12. A computer program product, characterized in that, When the instructions in the computer program product are executed by the processor of the electronic device, the electronic device performs the functions as described in the claims.

7. The method for determining the memristor measurement matrix as described in any one of the above.

Images