Reservoir calculation method based on rare earth doped nanocrystals

Abstract

The invention provides a reservoir calculation method based on rare earth doped nanocrystals. The reservoir calculation method comprises the following steps: acquiring an original digital image and converting the original digital image into a binary vector set; modulating the binary vector set into a pulse sequence set through time multiplexing coding; inputting the pulse sequence into a reservoir formed by three layers of core-shell-shell structure rare earth doped nanocrystals, and generating a corresponding luminescence response curve set through nonlinear dynamic interaction of the three layers of core-shell-shell structure rare earth doped nanocrystals; extracting features based on the curve set to form a reservoir state vector set; and finally, training and classifying the state vector set through a linear layer to obtain an image classification result. According to the reservoir calculation method based on the rare earth doped nanocrystals, through the synergistic effect of the physical characteristics of the rare earth doped nanocrystals and time multiplexing coding, a single reservoir replaces a traditional complex electronic network, coding, calculation and feature extraction are completed in an optical domain, the algorithm complexity and the training cost are remarkably reduced, and the calculation efficiency is improved. And efficient image classification is realized.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of optical reservoir computing, and particularly relates to a reservoir computing method based on rare earth doped nanocrystals. BACKGROUND

[0002] With the development of Internet of Things and big data applications, there is an urgent demand for intelligent hardware with compact size, low power consumption and edge computing capability in sensing and computing systems. Traditional computing has challenges in energy efficiency and real-time performance. Reservoir computing (RC) has become a frontier in time series data processing due to its simplified training process.

[0003] However, the existing RC scheme still has obvious limitations: the system based on fiber delay loop requires several meters of optical fiber, which is bulky and difficult to realize chip-level integration; the silicon-based micro-ring resonator is highly sensitive to temperature fluctuations and manufacturing tolerances, and is prone to detuning, with poor stability; the free-space scattering structure relies on complex optical alignment, with low system fault tolerance; the hybrid electro-optical scheme partially alleviates the above problems, but introduces additional energy consumption and signal delay, which goes against the original intention of all-optical RC.

[0004] Rare earth doped materials have multi-energy level electronic states and long lifetime memory windows, and can be prepared on a CMOS compatible platform, and are considered as an ideal candidate for building RC systems. However, existing researches mainly focus on the luminescence mechanism and spectral regulation, and have not fully tapped the application potential of rare earth doped materials in the typical RC task of handwritten digital image classification, making it difficult to solve the problems of hardware complexity, high energy consumption and high training cost of traditional RC in this type of task. SUMMARY

[0005] In view of the deficiencies of the prior art, the present application provides a reservoir computing method based on rare earth doped nanocrystals to solve the above problems. The rare earth doped nanocrystal reservoir composed of rare earth doped nanocrystals with a three-layer core-shell-shell structure replaces the complex network in traditional electronic computing through the synergistic effect of the physical properties of rare earth doped nanocrystals and the time multiplexing encoding strategy. The entire process completes encoding, computing and feature extraction in the optical domain, effectively reducing algorithm complexity and training cost, and finally realizing efficient image classification.

[0006] To solve the above technical problems, the present application provides a reservoir computing method based on rare earth doped nanocrystals, comprising the following steps:

[0007] Obtaining an original digital image, and obtaining a binary vector set based on the original digital image;

[0008] Time multiplexing encoding based on the binary vector set to obtain a pulse sequence set;

[0009] inputting the pulse sequence set into a rare earth doped nanocrystal reservoir to make the rare earth doped nanocrystal reservoir perform reservoir computing based on the pulse sequence set, to obtain a light emission response curve set;

[0010] The rare earth doped nanocrystal reservoir is composed of rare earth doped nanocrystals with a three-layer core-shell-shell structure. A plurality of pulse sequences in the pulse sequence set are sequentially input into the rare earth doped nanocrystal reservoir to make the pulse sequence set interact with the nonlinear dynamics of the rare earth doped nanocrystal, to obtain a plurality of light emission response curves corresponding to the pulse sequences. The light emission response curve set is obtained based on the plurality of light emission response curves.

[0011] Based on the light emission response curve set, feature extraction is performed to obtain a reservoir state vector set.

[0012] The reservoir state vector set is input into a linear layer to make the linear layer perform training classification based on the reservoir state vector set, to obtain an image classification result.

[0013] In the above scheme, the binary vector set representing the original digital image information is expanded in the time dimension and modulated into a pulse sequence set through time multiplexing coding, so as to load the digital information into the optical domain. Subsequently, the pulse sequence set is input into a rare earth doped nanocrystal reservoir composed of rare earth doped nanocrystals with a three-layer core-shell-shell structure. Through the nonlinear dynamics interaction between the pulse sequence set and the rare earth doped nanocrystal, a plurality of light emission response curves with unique transient and sustained responses are obtained to form a light emission response curve set, thereby completing high-dimensional nonlinear feature transformation through a single physical material. Finally, the reservoir state vector set is extracted from the light emission response curve set, and the linear layer performs training classification based on the reservoir state vector set to obtain an image classification result.

[0014] The above scheme replaces the traditional complex electronic network with a rare earth doped nanocrystal reservoir with a three-layer core-shell-shell structure, greatly simplifying the hardware structure. Through time multiplexing coding, the binary vector set is multiplexed on the time axis, so that a single physical reservoir virtually creates a sufficient high-dimensional state space to carry out complex computing tasks, achieving efficient processing of time sequence information without increasing hardware complexity. The two cooperate to form a reservoir computing scheme with simple hardware, convenient training, and excellent energy efficiency. The entire process is completed in the optical domain for encoding, computing, and feature extraction, effectively reducing algorithm complexity and training cost, and finally achieving efficient image classification.

[0015] Further, the obtaining of the original digital image and the obtaining of the binary vector set based on the original digital image comprise:

[0016] Obtaining an original digital image;

[0017] binarize the original digital image to obtain a binarized image;

[0018] divide the binarized image into a plurality of non-overlapping image blocks;

[0019] arrange the plurality of non-overlapping image blocks in a preset order to obtain a binary vector set.

[0020] In the above scheme, the original digital image is obtained and binarized to convert continuous gray scale information into discrete binary images. Then, the binarized image is divided into a plurality of non-overlapping image blocks, and finally these blocks are arranged and combined in a preset order to form a structured binary vector set. By image blocking and binarization, the spatial distribution information of the original image is converted into a regular binary data organization form, which not only preserves the spatial structure characteristics of the original digital image, but also realizes dimension reduction and regularization of data through blocking processing, which helps to improve the compatibility and efficiency of subsequent optical processing links.

[0021] Further, the binary vector set is time-multiplexed coded to obtain a pulse sequence set; comprising:

[0022] sequentially time-divide each bit in the binary vector set to obtain a plurality of corresponding time units;

[0023] For each bit in the binary vector set, map it according to the corresponding time unit to obtain a corresponding time unit pulse.

[0024] Obtain a pulse sequence set based on a plurality of time unit pulses.

[0025] In the above scheme, the binary vector set is converted into a pulse sequence through time-multiplexed coding. Specifically, each binary bit is mapped into an independent time unit, and the parallel digital information is converted into a serial time sequence optical signal by generating corresponding pulses in different time units. By expanding the data in the time dimension, a single physical reservoir can virtually have multiple processing nodes through time sequence response, which expands the high-dimensional state space while keeping the hardware structure simple, avoids complex optical device arrays, and effectively solves the contradiction between hardware complexity and computing capacity.

[0026] Further, in the above scheme, for each bit in the binary vector set, mapping it according to the corresponding time unit to obtain a corresponding time unit pulse; comprising:

[0027] For each bit in the binary vector set, if the bit value is a first preset value, map it according to the corresponding time unit to obtain a time unit pulse with a preset amplitude and a preset pulse width.

[0028] In the above scheme, for each bit in the binary vector set, if the bit value is the first preset value, an optical pulse with a uniform amplitude and pulse width is generated in the corresponding time unit, thereby converting the digital logic value into a standardized time unit pulse. By presetting the amplitude and pulse width, consistent optical excitation conditions are ensured for each valid bit, eliminating calculation bias introduced by signal differences and providing a stable and reliable nonlinear excitation basis for subsequent rare earth doped nanocrystal reservoirs.

[0029] It should be noted that the first preset value can be "1".

[0030] Further, the mapping of each bit in the binary vector set according to the corresponding time unit to obtain the corresponding time unit pulse; further comprising:

[0031] For each bit in the binary vector set, if the bit value is the second preset value, there is no pulse in the corresponding time unit.

[0032] In the above scheme, by mapping the second preset value in the binary vector to the no-pulse state and combining it with the time unit pulse corresponding to the first preset value, a complete timing encoding scheme is constructed, thereby realizing the conversion of binary information to an optical pulse sequence. With the alternating arrangement of pulses and no pulses, all information features of the original binary data are completely preserved in the time dimension, not only establishing a determined correspondence between digital information and optical excitation, but also reducing the overall energy consumption by introducing the no-pulse state.

[0033] It should be noted that the second preset value can be "0".

[0034] Further, the feature extraction based on the luminescence response curve set to obtain a reservoir state vector set; comprising:

[0035] For each luminescence response curve in the luminescence response curve set, a feature value is sampled at a preset time point to obtain a corresponding scalar feature value;

[0036] Arrange a plurality of scalar feature values in order to obtain a reservoir state vector set.

[0037] In the above scheme, the characteristic sampling is performed on the luminescence response curve at a preset time point, the optical response characteristic value in each time unit is extracted, and all scalar characteristic values are integrated into a reservoir state vector set according to the original time sequence structure. The continuous dynamic response of the rare earth doped nanocrystal is converted into a discrete feature vector, which not only retains the key state information in the nonlinear dynamics of the rare earth doped nanocrystal, but also realizes data dimension reduction and feature enhancement with the aid of time dimension reconstruction. The transient and sustained response characteristics contained in the luminescence dynamics are effectively captured, and the complex simulated optical response is converted into structured data suitable for machine learning processing, providing high-discrimination and high-stability feature input for subsequent processing, thereby fully utilizing the role of the rare earth doped nanocrystal reservoir in feature sampling.

[0038] It should be noted that the characteristic sampling at the preset time point can be reading the luminescence intensity of the luminescence response curve at the preset time point, and based on the inherent multi-time scale dynamic characteristics of the rare earth doped nanocrystal, the luminescence intensity at a certain time point contains both transient response and historical memory information, and high-dimensional features can be obtained through simple scalar sampling.

[0039] Further, for each luminescence response curve in the luminescence response curve set, characteristic extraction is performed at a preset time point to obtain a corresponding scalar characteristic value; including:

[0040] For each luminescence response curve in the luminescence response curve set, characteristic extraction is performed at a preset time point after the last pulse of the corresponding pulse sequence ends to obtain a corresponding scalar characteristic value.

[0041] The scalar characteristic values are arranged in order to obtain a reservoir state vector set.

[0042] In the above scheme, the binary vector set is converted into a pulse sequence through time multiplexing encoding, and the nonlinear dynamic characteristics of the rare earth doped nanocrystal are used to map each pulse sequence into a unique luminescence response curve, and then the scalar characteristic value is extracted at a preset time point after each pulse sequence ends, and is arranged in order to construct a state vector set. By combining a single rare earth doped nanocrystal reservoir with a time multiplexing mechanism, a high-dimensional state space is virtually created without complex hardware conditions, the transient and sustained features in the dynamic response are fully captured through timed sampling, the optical nonlinear response is efficiently converted into structured data suitable for linear classification, thereby simplifying the hardware structure, reducing the data processing complexity and training cost, and realizing high-energy efficient image classification.

[0043] The application also provides a rare earth doped nanocrystal reservoir computing system, comprising:

[0044] An image acquisition module is configured to acquire an original digital image and obtain a binary vector set based on the original digital image.

[0045] a pulse coding module, configured to perform time-multiplexing coding based on the binary vector set to obtain a pulse sequence set;

[0046] a reservoir computing module, configured to input the pulse sequence set into a rare earth doped nanocrystal reservoir to enable the rare earth doped nanocrystal reservoir to perform reservoir computing based on the pulse sequence to obtain a luminescence response curve set;

[0047] a feature extraction module, configured to perform feature extraction based on the luminescence response curve set to obtain a reservoir state vector set;

[0048] a training classification module, configured to input the reservoir state vector set into a linear layer to enable the linear layer to perform training classification based on the reservoir state vector set to obtain an image classification result.

[0049] In the scheme, the pulse coding module expands the binary vector set representing the original digital image information in the time dimension and modulates it into a pulse sequence set through time-multiplexing coding, thereby loading the digital information into the optical domain. Subsequently, the reservoir computing module inputs the pulse sequence set into a rare earth doped nanocrystal reservoir composed of a three-layer core-shell-shell structure of rare earth doped nanocrystals, and obtains a plurality of luminescence response curves with unique transient and sustained responses through the nonlinear dynamic interaction between the pulse sequence set and the rare earth doped nanocrystals, thereby forming a luminescence response curve set and completing high-dimensional nonlinear feature transformation through a single physical material. Finally, the feature extraction module extracts a reservoir state vector set from the luminescence response curve set, and the training classification module performs training classification based on the reservoir state vector set through a linear layer to obtain an image classification result.

[0050] The scheme replaces the traditional complex electronic network with a three-layer core-shell-shell structure of rare earth doped nanocrystal reservoirs, greatly simplifying the hardware structure. The pulse coding module multiplexes the binary vector set in the time axis through time-multiplexing coding, enabling a single physical reservoir to virtually create sufficient high-dimensional state space to carry out complex computing tasks without increasing hardware complexity, thereby achieving efficient processing of time sequence information. The two modules work together to form a reservoir computing scheme with simple hardware, convenient training, and excellent energy efficiency. The entire process is completed in the optical domain to encode, compute, and extract features, effectively reducing algorithm complexity and training cost, and ultimately achieving efficient image classification.

[0051] Further, the image acquisition module is configured to acquire an original digital image and obtain a binary vector set based on the original digital image, and includes:

[0052] acquiring an original digital image;

[0053] performing binaryzation processing on the original digital image to obtain a binary image;

[0054] divide the binary image into a plurality of non-overlapping image blocks;

[0055] arrange the plurality of non-overlapping image blocks in a preset order to obtain a binary vector set.

[0056] In the above scheme, the image acquisition module converts continuous gray scale information into discrete binary images by acquiring an original digital image and performing binary processing, and then divides the binary image into a plurality of non-overlapping image blocks, and finally arranges and combines these blocks in a preset order to form a structured binary vector set. By image blocking and binary processing, the spatial distribution information of the original image is converted into a regular binary data organization form, which not only retains the spatial structure characteristics of the original digital image, but also realizes dimension reduction and regularization of data through blocking processing, which helps to improve the compatibility and efficiency of subsequent optical processing links.

[0057] Further, the pulse encoding module is configured to perform time multiplexing encoding based on the binary vector set to obtain a pulse sequence set, and the pulse encoding module comprises:

[0058] time-dividing each bit in the binary vector set to obtain a plurality of corresponding time units;

[0059] For each bit in the binary vector set, mapping according to the corresponding time unit to obtain a corresponding time unit pulse;

[0060] obtaining a pulse sequence set based on a plurality of time unit pulses.

[0061] In the above scheme, the pulse encoding module converts the binary vector set into a pulse sequence through time multiplexing encoding. Specifically, each binary bit is mapped to an independent time unit, and the parallel digital information is converted into a serial time sequence optical signal by generating corresponding pulses in different time units. By expanding the data in the time dimension, a single physical reservoir can virtually have multiple processing nodes through time sequence response, which expands the high-dimensional state space while maintaining the simplicity of the hardware structure, avoids complex optical device arrays, and effectively solves the contradiction between hardware complexity and computing capacity. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A rare earth doped nanocrystal based reservoir computing method flowchart is provided for an embodiment of the present application;

[0063] Figure 2 A rare earth doped nanocrystal morphology characterization diagram is provided for an embodiment of the present application;

[0064] Figure 3 A reservoir computing scheme based on rare earth doped nanocrystal luminescence kinetics and an energy level transition schematic diagram are provided for an embodiment of the present application.

[0065] Figure 4 A time multiplexing and memory mechanism schematic diagram of a reservoir computing method based on rare earth doped nanocrystals is provided for an embodiment of the present application.

[0066] Figure 5 A rare earth doped nanocrystal reservoir computing architecture schematic diagram of a reservoir computing method based on rare earth doped nanocrystals is provided for an embodiment of the present application.

[0067] Figure 6 A response and classification performance evaluation diagram of a single node to 16 kinds of 4-bit combinations of a reservoir computing method based on rare earth doped nanocrystals is provided for an embodiment of the present application.

[0068] Figure 7 A reservoir computing system architecture schematic diagram based on rare earth doped nanocrystals is provided for an embodiment of the present application.

[0069] Figure 8 A reservoir computing system device based on rare earth doped nanocrystals and a pulse response characterization schematic diagram are provided for an embodiment of the present application. DETAILED DESCRIPTION

[0070] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application.

[0071] Please refer to Figure 1 The present embodiment provides a reservoir computing method based on rare earth doped nanocrystals, comprising the following steps:

[0072] An original digital image is obtained, and a binary vector set is obtained based on the original digital image;

[0073] Time multiplexing coding is performed based on the binary vector set to obtain a pulse sequence set;

[0074] The pulse sequence set is input to a rare earth doped nanocrystal reservoir, so that the rare earth doped nanocrystal reservoir performs reservoir computing based on the pulse sequence set to obtain a luminescence response curve set;

[0075] The rare earth doped nanocrystal reservoir is composed of rare earth doped nanocrystals with a three-layer core-shell-shell structure; a plurality of pulse sequences in the pulse sequence set are sequentially input to the rare earth doped nanocrystal reservoir, so that the pulse sequence set and the nonlinear dynamics of the rare earth doped nanocrystals interact, obtaining a plurality of luminescence response curves corresponding to the pulse sequences, and obtaining the luminescence response curve set based on the plurality of luminescence response curves;

[0076] Feature extraction is performed based on the luminescence response curve set, obtaining a reservoir state vector set;

[0077] The reservoir state vector set is input to the linear layer, so that the linear layer performs training classification based on the reservoir state vector set, obtaining an image classification result.

[0078] Please refer to Figure 2 , in an embodiment, a rare earth doped nanocrystal based reservoir computing method is provided, and the rare earth doped nanocrystal reservoir used in the method is a NaYF4 nanocrystal with a three-layer core-shell-shell structure. Specifically, the preparation method includes: synthesizing a NaYF4 core (size about 14 nm) containing 10 mol% Gd 3+ by co-precipitation method, then epitaxially growing a NaYF4:Tm shell layer (thickness about 5 nm) containing 95 mol% Yb 3+ and 5 mol% Tm 3+ on the core, and finally coating an inert NaYF4 passivation shell layer (thickness about 6 nm), and the obtained nanocrystals are dispersed in cyclohexane to form a suspension.

[0079] It should be noted that, Figure 2 (a) shows the transmission electron microscope image of NaYF4:Gd@NaYF4:Tm@NaYF4 nanocrystals, showing that the particle size distribution is uniform, and the average size is about 25 nm. The suspension is spin-coated on a glass sheet, and a uniform nanocrystal film can be formed on the surface of the glass sheet (see Figure 1 (b)).

[0080] Please refer to Figure 3 , Figure 3 (a) shows the schematic design of the reservoir implementation scheme based on the luminescence dynamics of rare earth ions (Re 3+ ). In Re 3+ doped luminescent materials, fast non-radiative transitions between adjacent energy levels result in abundant cross relaxation processes (see Figure 3(b)). The interaction between radiative transitions and non-radiative cross-relaxation processes establishes a highly nonlinear relationship between the material luminescence and the incident optical excitation. In addition, the long-lived intermediate states commonly found in Re3+doped materials provide an intrinsic temporal memory capability for continuous input signals. These features collectively facilitate the construction of rare-earth doped nanocrystal reservoirs.

[0081] In the present embodiment, the reservoir computation method based on rare-earth doped nanocrystals includes: encoding an input signal into a pulse sequence to irradiate the nanocrystal thin film; under 980 nm pumping, the energy is captured by the absorbing ions (such as Yb 3+ ) and transferred to the luminescent ions (such as Tm 3+ , Gd 3+ , etc.) through energy transfer, thereby forming an excitation state level system with different lifetime characteristics; the processes of radiative and non-radiative transitions, cross-relaxation and energy transfer jointly act to produce residual excitation states decaying with time, constituting the "temporal memory" of the reservoir; by sampling the time sequence response and using a linear layer for training and mapping, the computation function is realized.

[0082] It should be noted that the material system can be optimized according to actual application requirements: by adjusting the doping ratio and shell thickness to control the proportion and time constant of different lifetime states; by surface passivation to reduce non-radiative recombination processes; or by introducing local field enhancement structures to improve nonlinear response and suppress noise. The width and interval of the input pulse can also be designed to match different time scales, thereby optimizing the time coupling effect between virtual nodes.

[0083] It should be further noted that, Figure 3 (a) represents the process of signal from input to output, where the pump wavelength is 980 nm and the emission wavelength is 451 nm. "ETU" and "CR" represent energy transfer upconversion and cross-relaxation processes, respectively. Figure 3 (b) demonstrates the multi-level structure and energy transfer path in Re 3+ doped systems, including absorbing ions (such as Yb 3+ ), emitting / active ions (such as Tm 3+ , Gd 3+ ) and base lattice ions (Y 3+ ); each horizontal line represents a different energy level; arrows represent optical absorption, energy transfer, radiative transition and non-radiative processes. Figure 3 The transfer path of excitation energy from absorption to emission and its time characteristics are described.

[0084] In the present embodiment, by time-multiplexed encoding, the binary vector set representing the original digital image information is unfolded in the time dimension, modulated into a pulse sequence set, thereby loading the digital information into the optical domain. Subsequently, the pulse sequence set is input into a rare-earth doped nanocrystal reservoir composed of three-layer core-shell-shell structure of rare-earth doped nanocrystals, through the nonlinear dynamics interaction of the pulse sequence set with the rare-earth doped nanocrystals, a number of light-emitting response curves with unique transient and sustained response are obtained, forming a light-emitting response curve set, thereby completing the high-dimensional nonlinear feature transformation through a single physical material. Finally, the reservoir state vector set is extracted from the light-emitting response curve set, and the image classification result is obtained by training classification based on the reservoir state vector set through the linear layer.

[0085] Please refer to Figure 4 In an embodiment, under time-multiplexed encoding, each pulse corresponds to a virtual node. The intermediate state or residual excitation excited by the previous pulse decays with a certain time constant, and this residual excitation will change the initial state of the rare-earth doped nanocrystal reservoir when the subsequent pulse is injected, affecting its response characteristics, thereby forming a coupling relationship between virtual nodes that decays over time, realizing the internal state correlation and information fusion required for reservoir computing. When the interval between consecutive light pulses is shorter than the Re 3+ excited state lifetime, Re 3+ The input information of the previous node can be stored as a hidden state h(t), which together with the new input data x(t) determines the output response y(t). The persistent decay of the previous node affects the interconnection between adjacent virtual nodes, and this memory effect is the key to achieving excellent reservoir computing performance. Based on the material of Re 3+ , it has strong memory effect and multi-scale dynamic characteristics, so that the rare-earth doped nanocrystal reservoir can construct recursive coupling between virtual nodes through time-multiplexing, and realize state memory and internal connection without additional feedback loop.

[0086] Need to be specified, Figure 4A time multiplexing and memory mechanism schematic diagram of a reservoir computing method based on rare earth doped nanocrystals is provided for an embodiment of the present application, which is used to visually display the logic of input signal timing, reservoir state memory and output mapping in the embodiment. Input (input layer): signal input end of reservoir computing, used to carry the pulse sequence signal processed by time multiplexing coding, convert the original digital image into a binary vector set, expand into a serial time sequence optical signal through the time dimension, and provide an excitation source for the rare earth doped nanocrystal reservoir. Hidden state (hidden state layer): corresponding to the internal dynamic state of the rare earth doped nanocrystal reservoir in the embodiment, the state change directly depends on the physical properties of the rare earth doped nanocrystal. Output (output layer): signal output end of reservoir computing, used to receive dynamic information transmitted by the hidden state layer, and output the calculation result through linear mapping. i (t) (i = 1, 2, …, N) represents the pulse signal in the i-th time unit, corresponding to the optical mapping of the i-th bit in the binary vector set; t is a time variable; N is the number of input pulse sequences, that is, the number of input nodes virtually output by a single physical reservoir through the time multiplexing mechanism, and the value of N is determined by the total number of bits in the binary vector set.h j (t) (j = 1, 2, …, K) represents the j-th dynamic state of the rare earth doped nanocrystal reservoir at time t; K is the dimension of the internal virtual state of the reservoir, and the value is determined by the multi-time scale dynamic characteristics of the rare earth doped nanocrystal, without the need for additional hardware expansion.y m (t) (m = 1, 2, …, M) represents the output signal after linear transformation of the hidden state h(t); M is the dimension of the output signal, corresponding to the number of categories of the final task; the value of y(t) is determined by the product of h(t) and the linear layer weight matrix, directly reflecting the effective features of the reservoir after nonlinear processing, providing input for subsequent training and classification.

[0087] The embodiment replaces the traditional complex electronic network with a three-layer core-shell-shell structure of the rare earth doped nanocrystal reservoir, greatly simplifying the hardware structure. The binary vector set is multiplexed on the time axis through time multiplexing coding, so that a single physical reservoir virtually outputs sufficient high-dimensional state space to carry out complex computing tasks, realizing efficient processing of time sequence information without increasing the hardware complexity. The two cooperate to form a reservoir computing scheme with simple hardware, convenient training and excellent energy efficiency. The entire process completes coding, computing and feature extraction in the optical domain, effectively reduces the algorithm complexity and training cost, and finally realizes efficient image classification.

[0088] Further, the obtaining of the original digital image and the obtaining of the binary vector set based on the original digital image comprise:

[0089] The original digital image is obtained.

[0090] binarize the original digital image to obtain a binarized image;

[0091] divide the binarized image into a plurality of non-overlapping image blocks;

[0092] arrange the plurality of non-overlapping image blocks in a preset order to obtain a binary vector set.

[0093] In this embodiment, by obtaining an original digital image and binarizing it, the continuous gray scale information is converted into a discrete binary image, and then the binarized image is divided into a plurality of non-overlapping image blocks, and finally these blocks are arranged and combined in a preset order to form a structured binary vector set. By image blocking and binarization, the spatial distribution information of the original image is converted into a regular binary data organization form, which not only retains the spatial structure characteristics of the original digital image, but also realizes the dimensionality reduction and regularization of the data through blocking processing, which helps to improve the compatibility and efficiency of the subsequent optical processing link.

[0094] Further, the binary vector set is time-multiplexed coded to obtain a pulse sequence set; comprising:

[0095] time-dividing each bit in the binary vector set in turn to obtain a plurality of corresponding time units;

[0096] For each bit in the binary vector set, mapping according to the corresponding time unit to obtain a corresponding time unit pulse;

[0097] obtaining a pulse sequence set based on a plurality of time unit pulses.

[0098] In this embodiment, the binary vector set is converted into a pulse sequence through time-multiplexed coding, specifically, each binary bit is mapped into an independent time unit, and the parallel digital information is converted into a serial time sequence optical signal by generating corresponding pulses in different time units. By expanding the data in the time dimension, a single physical reservoir can virtually have multiple processing nodes through time sequence response, while maintaining the simplicity of the hardware structure, realizing the expansion of the high-dimensional state space, avoiding complex optical device arrays, and effectively solving the contradiction between hardware complexity and computing capacity.

[0099] Further, in the binary vector set, each bit is mapped according to the corresponding time unit to obtain a corresponding time unit pulse; comprising:

[0100] For each bit in the binary vector set, if the bit value is a first preset value, a time unit pulse with a preset amplitude and a preset pulse width is generated in the corresponding time unit.

[0101] In this embodiment, for each bit in the binary vector set, if the bit value is a first preset value, an optical pulse with a uniform amplitude and a pulse width is generated in the corresponding time unit, thereby converting the digital logic value into a standardized time unit pulse. By presetting the amplitude and the pulse width, consistent optical excitation conditions are ensured for each valid bit, eliminating calculation bias introduced by signal differences and providing a stable and reliable nonlinear excitation basis for subsequent rare earth doped nanocrystal reservoirs.

[0102] It should be noted that the first preset value can be "1".

[0103] Further, the mapping of each bit in the binary vector set according to the corresponding time unit to obtain the corresponding time unit pulse further includes:

[0104] For each bit in the binary vector set, if the bit value is a second preset value, there is no pulse in the corresponding time unit.

[0105] In this embodiment, by mapping the second preset value in the binary vector to a non-pulse state and combining it with the time unit pulse corresponding to the first preset value, a complete timing encoding scheme is constructed, thereby realizing the conversion of binary information to an optical pulse sequence. With the alternating arrangement of pulses and non-pulses, all information features of the original binary data are completely preserved in the time dimension, not only establishing a determined correspondence between digital information and optical excitation, but also reducing the overall energy consumption by introducing a non-pulse state.

[0106] It should be noted that the second preset value can be "0".

[0107] Further, the feature extraction based on the set of luminescence response curves to obtain the reservoir state vector set includes:

[0108] For each luminescence response curve in the set of luminescence response curves, a feature value is sampled at a preset time point to obtain a corresponding scalar feature value.

[0109] A number of scalar feature values are arranged in order to obtain a reservoir state vector set.

[0110] In this embodiment, by sampling the luminescence response curve at a preset time point, the optical response characteristic value in each time unit is extracted, and all scalar characteristic values are integrated into a reservoir state vector set according to the original time sequence structure. The continuous dynamic response of the rare earth doped nanocrystal is converted into a discrete feature vector, which not only retains the key state information in the nonlinear dynamics of the rare earth doped nanocrystal, but also realizes data dimension reduction and feature enhancement with the help of time dimension reconstruction. The transient and sustained response characteristics contained in the luminescence dynamics are effectively captured, and the complex simulated optical response is converted into structured data suitable for machine learning processing, providing high-discrimination and high-stability feature input for subsequent processing, thereby fully utilizing the role of rare earth doped nanocrystal reservoir in feature sampling.

[0111] It should be noted that the feature sampling at the preset time point can be reading the luminescence intensity of the luminescence response curve at the preset time point. Based on the inherent multi-time scale dynamic characteristics of the rare earth doped nanocrystal, the luminescence intensity at a certain time point contains both transient response and historical memory information, and high-dimensional features can be obtained through simple scalar sampling.

[0112] Further, for each luminescence response curve in the luminescence response curve set, a scalar characteristic value is extracted at a preset time point to obtain a corresponding scalar characteristic value.

[0113] For each luminescence response curve in the luminescence response curve set, a scalar characteristic value is extracted at a preset time point after the last pulse of the corresponding pulse sequence ends to obtain a corresponding scalar characteristic value.

[0114] The scalar characteristic values are arranged in order to obtain a reservoir state vector set.

[0115] In this embodiment, the binary vector set is converted into a pulse sequence through time multiplexing encoding, and the nonlinear dynamic characteristics of the rare earth doped nanocrystal are used to map each pulse sequence into a unique luminescence response curve. Then, at a preset time point after each pulse sequence ends, a scalar characteristic value is extracted to construct a state vector set in order. By combining a single rare earth doped nanocrystal reservoir with a time multiplexing mechanism, a high-dimensional state space is virtually created without complex hardware conditions. By timing sampling, transient and sustained features in the dynamic response are fully captured, and the optical nonlinear response is efficiently converted into structured data suitable for linear classification, thereby simplifying the hardware structure, reducing the data processing complexity and training cost, and realizing high-energy efficient image classification.

[0116] See Figure 5 In an embodiment, a rare earth doped nanocrystal reservoir computing method is provided, including the following steps:

[0117] First, the original 28×28 pixel grayscale image is acquired and binarized according to a set threshold. Pixel values ​​greater than the threshold are set to "1" and otherwise to "0". Then, the image is divided into non-overlapping 2×2 pixel blocks, resulting in 196 image blocks. The four pixels in each block are arranged in a preset order to form a 4-bit binary vector.

[0118] Time-multiplexed pulse coding is performed on each binary vector: bit "1" is mapped to a pump light pulse with an intensity of 45mW / cm2 and a pulse width of 25μs, while bit "0" does not output a pulse, thereby converting spatial image information into a temporal optical pulse sequence;

[0119] The pulse sequence is input into a reservoir composed of rare earth-doped nanocrystals with a three-layer core-shell-shell structure. The nonlinear luminescence dynamics of the nanocrystals are used to respond to the pulse. Each block corresponds to a luminescence response curve, and all response curves constitute a luminescence response curve set.

[0120] The luminescence intensity at a specific time point is extracted from each response curve as a feature to construct a state vector; specifically, sampling is performed at the end of the response curve corresponding to each block to obtain a 196-dimensional state vector x(t), which is used as the input of the linear layer;

[0121] The state vector is input into a linear classifier for training and classification, and the final image classification result is output.

[0122] It should be noted that 45mW / cm² is selected in this embodiment. 2 The pump intensity and 25 μs pulse width are based on the characterization results of the nonlinear characteristics of multi-level transitions and the pulse promotion (PPF) effect of rare-earth nanocrystals. This parameter combination can generate a stable optical response under single-pulse excitation, while achieving a balance between temporal resolution and signal-to-noise ratio, making it suitable for image classification tasks. Through temporal multiplexing coding, the spatial structure information of the image is mapped to adjacent pulse sequences in the temporal dimension. By utilizing the multi-scale relaxation dynamics of rare-earth materials, including transient response, pulse promotion (PPF) enhancement effect, and long-lifetime decay tail—capturing memory effects at different time scales—high-dimensional virtual nodes can be constructed and nonlinear feature transformations can be achieved in a single physical reservoir.

[0123] It needs to be further explained that, Figure 5The Input layer, Reservoir and Output layer shown in the middle correspond to the input layer, reservoir (hidden layer) and output layer in the neural network architecture. The input is a 28x28 pixel MNIST handwritten digit image, each pixel is binarized; in the input layer, the image is divided into 196 2x2 pixel blocks, each pixel block is converted into a 4-bit binary number. The reservoir maps each 4-bit binary input to a scalar response value and passes it to the output layer; the output layer maps the features to one of the ten classes from 0 to 9 through linear regression, completing the classification task.

[0124] Referring to Figure 6 , Figure 6 (a) shows different response trajectories generated by the reservoir for all 16 possible 4-bit input combinations. After training the linear layer, this embodiment achieves a classification accuracy of 90.7% on a test set containing 10,000 images, and the corresponding confusion matrix is shown in Figure 6 (b) to characterize the classification prediction effect.

[0125] It should be noted that, Figure 6 (a) shows the process of mapping 4-bit binary input to one-dimensional scalar response x(t) by the reservoir, and the response curves under all 16 input combinations are plotted. Figure 6 (b) is the confusion matrix corresponding to the classification accuracy, which intuitively reflects the prediction performance of the model on different classes. As shown in Figure 6 (c) shows that after the pump light stops, the rare earth nanocrystal still maintains a decay luminescence of about 40μs, and the classification accuracy remains stable within this time window. The abscissa represents the delay of the sampling time relative to the end time of the last input pulse, and the larger the abscissa value, the farther the sampling time of x(t) is from the input completion time. This characteristic shows that this embodiment has good robustness to timing errors and signal jitter, and is suitable for practical application scenarios with timing disturbance or intermittent input.

[0126] Referring to Figure 7 , the embodiment also provides a reservoir computing system based on rare earth doped nanocrystals, comprising:

[0127] An image acquisition module is configured to acquire an original digital image and obtain a set of binary vectors based on the original digital image;

[0128] A pulse encoding module is configured to perform time multiplexing encoding based on the set of binary vectors to obtain a set of pulse sequences;

[0129] A reservoir computing module is configured to input the set of pulse sequences to a rare earth doped nanocrystal reservoir, so that the rare earth doped nanocrystal reservoir performs reservoir computing based on the pulse sequences to obtain a set of luminescence response curves.

[0130] a feature extraction module configured to perform feature extraction based on the set of luminescence response curves to obtain a set of reservoir state vectors;

[0131] a training classification module configured to input the set of reservoir state vectors to a linear layer to cause the linear layer to perform training classification based on the set of reservoir state vectors to obtain an image classification result.

[0132] In the embodiment, the pulse encoding module expands the binary vector set representing the original digital image information in the time dimension by time multiplexing encoding, and modulates the binary vector set into a pulse sequence set, so as to load the digital information into the optical domain. Subsequently, the reservoir computing module inputs the pulse sequence set into the rare earth doped nanocrystal reservoir composed of the rare earth doped nanocrystals with a three-layer core-shell-shell structure, and obtains a plurality of luminescence response curves with unique transient and sustained responses by the nonlinear dynamic interaction between the pulse sequence set and the rare earth doped nanocrystals, to form a set of luminescence response curves, so that the high-dimensional nonlinear feature transformation is completed by a single physical material. Finally, the feature extraction module extracts a set of reservoir state vectors from the set of luminescence response curves, and the training classification module performs training classification based on the set of reservoir state vectors by the linear layer to obtain an image classification result.

[0133] Please refer to Figure 8 In an embodiment, a rare earth doped nanocrystal-based reservoir computing system is provided, which includes an optical pumping unit and a rare earth doped nanocrystal reservoir. The optical pumping unit emits a light beam from a 980 nm continuous laser, modulates the light beam into a rectangular pulse sequence by an electro-optic modulator (EOM), and focuses the pulse sequence onto the rare earth doped nanocrystal film through a 50x objective lens (see Figure 8 (a)). When the pumping intensity is 40 mW / cm 2 and the pulse width is 1700 μs, the emission intensity at 451 nm shows a significant nonlinear increase with the pumping power and the pulse width, and still maintains a decay tail of tens to hundreds of microseconds after the pumping light is cut off (see Figure 8 (b)), which indicates that the rare earth doped nanocrystal reservoir has a multi-time scale memory effect based on the multi-energy level transition of rare earth ions.

[0134] Meanwhile, when the pumping power is fixed at 45 mW / cm 2, under the condition of pulse width 25 μs, two short pulses were input into the rare earth doped nanocrystal reservoir in a variable time interval (Δt), and it was observed that the luminescence peak triggered by the second pulse was higher than that of the first pulse. This phenomenon indicates that the rare earth doped nanocrystal reservoir has both short-time pulse promotion (PPF) and multi-scale memory characteristics of long-time decay tail: the peak enhancement caused by the short pulse interval corresponds to the fast memory channel, and the decay tail lasting tens to hundreds of microseconds after the pump is stopped provides the slow memory retention capability. Based on the above characteristics, the input information is encoded by pulse parameters, and the rare earth doped nanocrystal reservoir can exhibit nonlinear response with intrinsic memory effect, realizing the information processing function of the hierarchical neuron-like biological.

[0135] It needs to be specifically pointed out that, Figure 8 (a) Each element in (a) is: 980 nm pump light source, MMF (multimode optical fiber), AL (aspherical lens), LP (linear polarizer), EOM (electro-optic modulator), lens (focusing lens), PD (photodetector), RFA (radio frequency amplifier), AWG (arbitrary waveform generator), Spectrometer (spectrometer).

[0136] Figure 8 (b) is the luminescence response time curve of the rare earth doped nanocrystal reservoir under single pulse excitation (including experimental measured curve and simulation curve), which is used to verify the long-life memory characteristics of the reservoir; the abscissa Ti is the time variable, the unit is microsecond (μs), which represents the complete time process from the beginning of pulse excitation (left starting point) to the end of decay after pulse cutoff (right end point); the ordinate 451 nm Emission indicates the luminescence intensity of the rare earth doped nanocrystal reservoir at 451 nm wavelength; a.u. is an arbitrary unit (Arbitrary Unit) for quantifying the relative size of the luminescence intensity; the rising section on the left side of the abscissa is the pulse action period, which reflects the transient response characteristics of the rare earth ions rapidly jumping from the ground state to the excited state after excitation, and proves that the rare earth doped nanocrystal reservoir can quickly respond to the input pulse signal. The falling section on the right side of the abscissa is the decay memory period, which indicates that after the pulse is cut off, the luminescence intensity does not disappear immediately, but decays slowly with time, forming a decay tail lasting tens to hundreds of microseconds, which provides a physical basis for the memory of the time sequence information of the reservoir calculation. The trends of the two curves (experimental measurement vs. theoretical simulation) in the figure are highly consistent, verifying the accuracy of the rare earth doped nanocrystal reservoir calculation system provided in the embodiment, and ensuring the reliability of the subsequent reservoir calculation.

[0137] Figure 8(c) is a pulse-promoted fluorescence (PPF) response curve of a rare earth doped nanocrystal reservoir provided in this embodiment under double pulse excitation, used to verify the short-term memory characteristics of the reservoir. PPF is the pulse-promoted fluorescence effect. A1 is the peak intensity of luminescence generated by the first pulse excitation of the reservoir, which is used as a reference intensity to quantify the strength of the PPF effect; A2 is the peak intensity of luminescence generated by the second pulse excitation of the reservoir, and the ratio of A2 to A1 (A2 / A1) directly reflects the degree of enhancement of the PPF effect; Δt is the time interval between the two pulses, with units of microseconds (μs); Figure 8 The upper right corner of (c) is a timing diagram of double pulse input and luminescence response of the reservoir, showing the correspondence between the input pulse sequence and the output luminescence response. The horizontal axis Δt has units of μs, representing the time interval between the two excitation pulses. The vertical axis A2 / A1 represents the ratio of the second pulse luminescence peak to the first pulse luminescence peak, which is unitless. The larger the ratio, the stronger the PPF effect, and the more significant the short-term memory capability of the reservoir.

[0138] This embodiment replaces the traditional complex electronic network with a three-layer core-shell-shell structure of rare earth doped nanocrystal reservoir, greatly simplifying the hardware structure. The pulse encoding module encodes the binary vector set in time domain through time multiplexing, so that a single physical reservoir can virtually have enough high-dimensional state space to carry out complex computing tasks, achieving efficient processing of time sequence information without increasing hardware complexity. The two work together to form a reservoir computing scheme with simple hardware, convenient training, and excellent energy efficiency. The entire process completes encoding, computing, and feature extraction in the optical domain, effectively reducing algorithm complexity and training cost, and ultimately achieving efficient image classification.

[0139] Further, the image acquisition module is configured to acquire an original digital image and acquire a binary vector set based on the original digital image; the method comprises:

[0140] acquiring an original digital image;

[0141] performing binaryzation processing on the original digital image to obtain a binaryzation image;

[0142] dividing the binaryzation image into a plurality of non-overlapping image blocks;

[0143] arranging the plurality of non-overlapping image blocks in a predetermined order to obtain a binary vector set.

[0144] In the embodiment, the image acquisition module acquires an original digital image and performs binaryzation processing on the original digital image to convert continuous gray scale information into discrete binary images, then divides the binaryzation images into a plurality of non-overlapping image blocks, and finally arranges and combines the blocks in a preset order to form a structured binary vector set. The spatial distribution information of the original image is converted into a regular binary data organization form through image blocking and binaryzation, which not only retains the spatial structure characteristics of the original digital image, but also realizes dimension reduction and regularization of data through blocking processing, which helps to improve the compatibility and efficiency of subsequent optical processing links.

[0145] Further, the pulse encoding module is configured to perform time multiplexing encoding based on the binary vector set to obtain a pulse sequence set, and the pulse sequence set comprises:

[0146] time division is performed on each bit in the binary vector set to obtain a plurality of corresponding time units;

[0147] for each bit in the binary vector set, mapping is performed according to the corresponding time unit to obtain a corresponding time unit pulse;

[0148] the pulse sequence set is obtained based on the plurality of time unit pulses.

[0149] In the embodiment, the pulse encoding module converts the binary vector set into a pulse sequence through time multiplexing encoding, specifically, each binary bit is mapped into an independent time unit, and parallel digital information is converted into serial time sequence optical signals by generating corresponding pulses in different time units. By expanding the data in the time dimension, a single physical reservoir can virtually have multiple processing nodes through time sequence response, which realizes the expansion of high-dimensional state space while maintaining the simplicity of the hardware structure, avoids complex optical device arrays, and effectively solves the contradiction between hardware complexity and computing capacity.

[0150] The above is the preferred embodiment of the present application. It should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which are also considered within the scope of protection of the present application.

Claims

1. A reservoir calculation method based on rare-earth-doped nanocrystals, characterized in that, Includes the following steps: Obtain the original digital image, and then obtain a binary vector set based on the original digital image; Based on the binary vector set, time multiplexing coding is performed to obtain a pulse sequence set; The pulse sequence set is input into the rare earth-doped nanocrystal reservoir so that the rare earth-doped nanocrystal reservoir can perform reservoir calculations based on the pulse sequence set to obtain a set of luminescence response curves. The rare earth-doped nanocrystal reservoir is composed of rare earth-doped nanocrystals with a three-layer core-shell-shell structure. Several pulse sequences from the pulse sequence set are sequentially input into the rare earth-doped nanocrystal reservoir so that the pulse sequence set interacts with the nonlinear dynamics of the rare earth-doped nanocrystals to obtain several luminescence response curves corresponding to the pulse sequences. The luminescence response curve set is obtained based on the several luminescence response curves. Feature extraction is performed based on the set of luminescence response curves to obtain a reservoir state vector set; The reservoir state vector set is input into a linear layer so that the linear layer can be trained and classified based on the reservoir state vector set to obtain image classification results.

2. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 1, characterized in that, The step of acquiring the original digital image and acquiring a binary vector set based on the original digital image includes: Acquire the raw digital image; The original digital image is binarized to obtain a binarized image; The binarized image is divided into several non-overlapping image blocks; The binary vector set is obtained by arranging the non-overlapping image blocks in a preset order.

3. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 1, characterized in that, The step of performing time multiplexing coding based on the binary vector set to obtain a pulse sequence set includes: Based on each bit in the binary vector set, time is divided sequentially to obtain several corresponding time units; For each bit in the binary vector set, it is mapped according to the corresponding time unit to obtain the corresponding time unit pulse; A pulse sequence set is obtained based on several time unit pulses.

4. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 3, characterized in that, The step of mapping each bit in the binary vector set according to the corresponding time unit to obtain the corresponding time unit pulse includes: For each bit in the binary vector set, if the bit value is a first preset value, it is mapped according to the corresponding time unit to obtain a time unit pulse with a preset amplitude and preset pulse width.

5. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 4, characterized in that, For each bit in the binary vector set, it is mapped according to the corresponding time unit to obtain the corresponding time unit pulse; Also includes: For each bit in the binary vector set, if the bit value is a second preset value, there is no pulse in the corresponding time unit.

6. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 1, characterized in that, The step of extracting features based on the set of luminescence response curves to obtain a reservoir state vector set includes: For each luminescence response curve in the set of luminescence response curves, feature extraction is performed at a preset time point to obtain the corresponding scalar feature value; Arrange several scalar eigenvalues ​​in order to obtain the reservoir state vector set.

7. The reservoir calculation method based on rare earth-doped nanocrystals according to claim 6, characterized in that, For each luminescence response curve in the set of luminescence response curves, feature extraction is performed at a preset time point to obtain the corresponding scalar feature value; including: For each luminescence response curve in the set of luminescence response curves, feature sampling is performed at a preset time point after the end of the last pulse of the corresponding pulse sequence to obtain the corresponding scalar feature value; Arrange several scalar eigenvalues ​​in order to obtain the reservoir state vector set.

8. A reservoir computing system based on rare-earth-doped nanocrystals, characterized in that, include: The image acquisition module is used to acquire raw digital images and obtain binary vector sets based on the raw digital images; The pulse coding module is used to perform time multiplexing coding based on the binary vector set to obtain a pulse sequence set; The reservoir calculation module is used to input the pulse sequence set into the rare earth-doped nanocrystal reservoir, so that the rare earth-doped nanocrystal reservoir can perform reservoir calculations based on the pulse sequence to obtain a set of luminescence response curves. The feature extraction module is used to extract features based on the set of luminescence response curves to obtain a reservoir state vector set; The training classification module is used to input the reservoir state vector set into the linear layer, so that the linear layer can perform training classification based on the reservoir state vector set to obtain image classification results.

9. A reservoir computing system based on rare-earth-doped nanocrystals according to claim 8, characterized in that, The image acquisition module is used to acquire the original digital image and obtain a binary vector set based on the original digital image; it includes: Acquire the raw digital image; The original digital image is binarized to obtain a binarized image; The binarized image is divided into several non-overlapping image blocks; The binary vector set is obtained by arranging the non-overlapping image blocks in a preset order.

10. A reservoir computing system based on rare-earth-doped nanocrystals according to claim 8, characterized in that, The pulse coding module is used to perform time multiplexing coding based on the binary vector set to obtain a pulse sequence set; including: Based on each bit in the binary vector set, time is divided sequentially to obtain several corresponding time units; For each bit in the binary vector set, it is mapped according to the corresponding time unit to obtain the corresponding time unit pulse; A pulse sequence set is obtained based on several time unit pulses.

Images