Method, system and equipment for generating high-capacity mutually-orthogonal complementary sequence set

Abstract

The invention discloses a method, a system and equipment for generating a high-capacity mutually-orthogonal complementary sequence set, and belongs to the field of data processing, and the method comprises the steps: generating an initial sequence set through the training of a long short-term memory (LSTM) network; remodeling the initial sequence into a symmetric weight matrix through a Hopfield neural network, and performing asynchronous state updating; grouping and cutting the updated sequence set to generate an undetermined sequence set; and evaluating the to-be-determined sequence set based on an MC index, and screening to obtain a final mutually orthogonal complementary sequence set. According to the method, the number theory limitation of traditional algebraic construction is broken through by virtue of the modeling capability of the LSTM on the long-term dependency relationship of the sequence and combining the iterative optimization of the Hopfield neural network on the orthogonal complementary constraint, and the large-scale sequence set meeting the zero cross-correlation and zero autocorrelation side lobe conditions can be quickly generated in a data driving manner.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a method, system and device for generating a large-capacity mutually orthogonal complementary sequence set. Background Art

[0002] A mutually orthogonal complementary sequence set (MOCSS) is a family of sequence sets with strict complementary characteristics, which has important applications in multiple fields such as communication, radar, sonar, and bioinformatics. Its core value lies in simultaneously satisfying complementarity and orthogonality. In code division multiple access (CDMA), non-orthogonal multiple access (NOMA), and 6G communication, a large number of users need to share the same frequency band. Therefore, it is necessary to use mutually orthogonal or low-correlation spreading codes to distinguish users and avoid interference.

[0003] In the frontier exploration of modern communication and radar technologies, etc., the mutually orthogonal complementary sequence set has become a core element to support the breakthrough of system performance due to its unique zero cross-correlation and zero autocorrelation sidelobe characteristics. In the 6G communication scenario, the development of high-frequency spectrum resources, the popularization of large-scale multiple-input multiple-output (MIMO) technology, and the demand of non-orthogonal multiple access (NOMA) systems for high-density user access have all put forward strict requirements for the capacity and generation efficiency of the mutually orthogonal complementary sequence set. Only with a sufficiently large-scale mutually orthogonal complementary sequence set can the ideal isolation of thousands or even tens of thousands of user signals be achieved within a limited spectrum, avoiding signal misjudgment and system capacity bottlenecks caused by multiple access interference (MAI).

[0004] Taking an active phased array radar as an example, it needs to simultaneously track hundreds of dynamic targets and suppress enemy interference in a complex electromagnetic environment, which relies on a large number of orthogonal coding sequences to achieve beamforming and target resolution; the high-resolution imaging of a synthetic aperture radar (SAR) requires long sequences to maintain strict orthogonality in multipath fading to avoid misjudgment of false targets. Thus, it can be seen that the generation of large-capacity MOCSS sequences is very important. With the continuous development of fields such as communication, radar, and sonar, the demand for large-capacity MOCSS is increasing day by day. In communication, whether it is CDMA, NOMA, or 6G, a large-capacity MOCSS is needed to support more user access and improve system performance; in radar and sonar systems, a large-capacity MOCSS can better cope with complex target detection scenarios. Only by realizing the generation of large-capacity MOCSS can its advantages in various fields be fully exerted, meeting the growing technical needs and promoting the further development of related fields.

[0005] This contradiction is particularly prominent in the implementation of technologies: The actual measurement of a 6G test network shows that the MOCSS generated by traditional methods can only support 256 users to access simultaneously, far lower than the theoretical design capacity of tens of thousands; in a certain shipborne radar in a strong clutter environment, due to the insufficient scale of orthogonal sequences, the multi-target recognition accuracy drops sharply from 95% to 68%. It can be seen that breaking through the constraints of algebraic construction and realizing the efficient generation of large-scale and arbitrarily long MOCSS has become the key bottleneck for communication and radar technologies to move towards the new generation.

[0006] Existing MOCSS construction methods mainly include the Hadamard matrix construction method and the generation scheme based on HP-GAN. Among them, the Hadamard matrix construction method includes the direct construction method, the construction method based on Kronecker product, and the recursive construction method, which highly rely on algebraic structures such as finite fields and group theory. The sequence length must meet specific conditions (such as powers of 2), and as the scale increases, the computational complexity increases exponentially. When the actual requirements break through the limitations of traditional algebraic structures (for example, sequences of length 1023 or non-power-of-2 sequences are required), existing methods often fall into the "construction failure" dilemma, resulting in limited user access capabilities of communication systems and decreased radar target detection accuracy.

[0007] The generation scheme of Hp-GAN realizes the approximate generation of MOCSS through the combination of a generative adversarial network and a Hopfield neural network. It introduces the Hopfield neural network to screen the sequence set, iteratively updates the state of the sequence set, and evaluates the sequence set with the metric function MC. The sequence set with an MC value of 0 is selected as MOCSS. Through alternating training of GAN and the Hopfield neural network for iterative optimization, Hp-GAN realizes the generation of MOCSS. However, it needs to go through many rounds of training before it can obtain the sequence set with an MC value of 0 for the first time, which means it needs to go through multiple rounds of training before starting to generate MOCSS. The convergence characteristics of the training process are not good enough, and the feature extraction and expression capabilities are insufficient. Especially when dealing with long sequences and complex structure sequences, its defects are more obvious, ultimately resulting in low quality and search efficiency of the generated sequences.

[0008] Therefore, there is an urgent need for a MOCSS generation technology with faster and more stable training, stronger adaptability in feature learning, and higher sequence quality and search efficiency. Summary of the Invention

[0009] The object of the present invention is to overcome the technical problems existing in the prior art, and provides a method, a system and a device for generating a large-capacity mutually orthogonal complementary sequence set. By means of the modeling ability of the long short-term memory network (LSTM) for the long-term dependence relationship of sequences and in combination with the iterative optimization of the mutually orthogonal complementary constraint by the Hopfield neural network, the number theory limitation of traditional algebraic construction is broken through, and a large-scale sequence set satisfying the conditions of zero cross-correlation and zero sidelobe of auto-correlation can be quickly generated in a data-driven manner.

[0010] The object of the present invention is achieved by the following technical solutions: In a first aspect, a method for generating a large-capacity mutually orthogonal complementary sequence set is provided, including: Training with a long short-term memory network (LSTM) to generate an initial sequence set; Reshaping the initial sequence set into a symmetric weight matrix by a Hopfield neural network and performing asynchronous state update; Grouping and cutting the updated sequence set to generate a to-be-determined sequence set; Evaluating the to-be-determined sequence set based on the MC metric and screening to obtain the final mutually orthogonal complementary sequence set.

[0011] In some embodiments, the long short-term memory network (LSTM) is defined as: where y represents the output, X represents the input tensor, is the LSTM layer operation, is the weight matrix of the LSTM layer, is the weight matrix of the fully connected layer, is the bias vector, represents the hyperbolic tangent activation function.

[0012] In some embodiments, the training with a long short-term memory network (LSTM) to generate an initial sequence set includes: Using the Adam optimizer to minimize the mean square error loss and using the cosine annealing learning rate scheduler to dynamically adjust the learning rate to generate a continuous value sequence; Binarizing the continuous value sequence by applying the sign function.

[0013] In some embodiments, the symmetric weight matrix is: where, represents reshaping the matrix shape, represents a vector of length 1024, and i represents the i-th index in the continuous sequence among.

[0014] In some embodiments, the asynchronous state update includes: Define the update function of the Hopfield neural network and randomly select neurons for update; the update function is as follows: where, represents the state of node i at the next moment t + 1, represents the sign function, which outputs +1 when the input is positive, -1 when the input is negative, and usually outputs 0 or +1 when the input is 0, represents the connection weight from node j to i, represents the state of node j at time t, represents the state of node j at time t + 1, j represents the node number, and I represents the index set. In some embodiments, the MC metric is calculated as follows: where, represents the autocorrelation metric, represents the cross-correlation metric, represents the flip metric.

[0015] In some embodiments, the screening to obtain the final set of mutually positive and complementary sequences includes: If and the sequence does not exist in the historical optimal sequence set S, then include it in the historical optimal sequence set S.

[0016] In a second aspect, a system for generating a large-capacity set of mutually positive and complementary sequences is provided, including: An LSTM sequence generation module for training and generating an initial sequence set using the long short-term memory network LSTM; A Hopfield neural network optimization module for reshaping the initial sequence into a symmetric weight matrix through the Hopfield neural network and performing asynchronous state updates; A sequence grouping and cutting module for grouping and cutting the updated sequence set to generate a set of candidate sequences; A metric evaluation module for evaluating the set of candidate sequences based on the MC metric and screening to obtain the final set of mutually positive and complementary sequences.

[0017] In a third aspect, an electronic device is provided, including a memory and a processor, where a computer instruction executable on the processor is stored on the memory, and when the processor runs the computer instruction, it executes the method for generating a large-capacity set of mutually positive and complementary sequences described in the first aspect.

[0018] It should be further noted that the technical features corresponding to the above embodiments can be combined or replaced with each other without conflict to form a new technical solution.

[0019] Compared with the prior art, the present invention has the following beneficial effects: The present invention uses a long short-term memory network LSTM training to generate an initial sequence set, and uses a Hopfield neural network for optimization. Specifically, the initial sequence set is reshaped into a symmetric weight matrix, and an asynchronous state update is performed. Finally, the pending sequence set is evaluated based on the MC indicator, and the final mutually orthogonal complementary sequence set is screened. With the modeling ability of LSTM for long-term sequence dependencies, combined with the iterative optimization of the Hopfield network for orthogonal complementary constraints, the number theory limitations of traditional algebraic constructions are broken through, and a large-scale sequence set that satisfies the zero mutual correlation and zero autocorrelation sidelobe conditions can be quickly generated in a data-driven manner. At the same time, compared with the Hp-GAN network, the number of generations has increased by about 21%, the convergence speed has increased by about 96%, the round of the first generation of MOCSS has been advanced from 2300 rounds to the first round, the training stability has been significantly improved, the feature learning adaptability is stronger, and the sequence quality and search efficiency are higher. This innovation provides a key technical path for the spectrum efficiency revolution of 6G communications and the upgrade of the anti-interference capability of radar systems, and provides technical support for large-scale MOCSS application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 A schematic flow chart of a method for generating a large-capacity mutually orthogonal complementary sequence set according to the present invention; Figure 2 This is a schematic diagram of the Hp-LSTM network structure of the present invention; Figure 3 Schematic diagram of the training process of Hp-LSTM of the present invention; Figure 4 It is a schematic diagram comparing Hp-LSTM and Hp-GAN of the present invention; Figure 5 It is the generator loss graph of Hp-GAN; Figure 6 It is the discriminator loss diagram of Hp-GAN; Figure 7 This is the loss graph of Hp-LSTM. DETAILED DESCRIPTION

[0021] The technical solution of the present invention is clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the accompanying drawings can be arranged and designed in various configurations. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0022] It should be noted that all the defects existing in the above solutions of the prior art are the results obtained by the inventor through practice and careful research. Therefore, the process of discovering the above problems and the solutions proposed by the embodiments of the present application below for the above problems should be the contributions made by the inventor to the present application during the invention and creation process, and should not be understood as the technical content known to those skilled in the art.

[0023] In view of the technical problems pointed out in the background art, the embodiments provided by the present invention are as follows: Referring to Figure 1 , in an exemplary embodiment, a method for generating a large-capacity mutually positive and complementary sequence set is provided, including: Using a long short-term memory network (LSTM) to train and generate an initial sequence set; Reshaping the initial sequence set into a symmetric weight matrix through a Hopfield neural network and performing asynchronous state updates; Grouping and cutting the updated sequence set to generate a to-be-determined sequence set; Evaluating the to-be-determined sequence set based on the MC metric and screening to obtain the final mutually positive and complementary sequence set.

[0024] The long short-term memory network (Long Short-Term Memory, LSTM) is a special recurrent neural network (RNN) architecture designed to address the vanishing gradient and exploding gradient problems faced by traditional RNNs when processing long sequence data, thereby effectively capturing long-term dependencies in the sequence. The core structure of LSTM consists of a memory cell, an input gate, a forget gate, and an output gate. The gating mechanism of LSTM can adaptively selectively retain or forget historical information, effectively alleviating the defects of traditional RNNs in long sequence modeling. At the same time, compared with the Nash equilibrium that the GAN network needs to achieve, the training difficulty and time cost are greatly reduced. Based on this, the present invention utilizes the modeling ability of LSTM for long-term dependencies in sequences and combines the iterative optimization of the positive and complementary constraints of the Hopfield neural network to achieve a better algorithm for generating a mutually positive and complementary sequence set (MOCSS), abbreviated as Hp-LSTM, and the structure is as Figure 2 shown.

[0025] Specifically, the input of LSTM is a three-dimensional tensor , where, is the number of samples, the time step is set to 1, and the feature dimension is 1024. The long short-term memory network LSTM is defined as:

[0026] where y represents the output, and X represents the input tensor, is the operation of the LSTM layer, is the weight matrix of the LSTM layer, is the weight matrix of the fully connected layer, is the bias vector, represents the hyperbolic tangent activation function.

[0027] Furthermore, the mean squared error (MSE) loss function is adopted in the model training process:

[0028] where L is the loss, represents the batch size, is the true value, is the predicted value, is the square of the norm.

[0029] The Adam optimizer is used with a learning rate of to minimize the loss. Due to the limited binary MOCSS sequence features, the present invention utilizes a cosine annealing learning rate scheduler to dynamically adjust the learning rate, trying to avoid falling into local optima. Finally, a continuous value sequence is generated, and the continuous value sequence is binarized by applying the sign function.

[0030] Furthermore, in the sequence screening stage, after every certain number of training rounds (such as 500 rounds, and since HP-LSTM converges rapidly, the evaluation rounds can be reduced), the trained LSTM is used to predict the input data (such as using the LSTM to predict 200 sequence sets with a length of 1024). Subsequently, it is reshaped into the shape of [200, 32, 32] as the symmetric weight matrix of the Hopfield neural network. The symmetric weight matrix is:

[0031] where, represents reshaping the matrix shape, represents a vector with a length of 1024, and i represents the i-th index / element in the continuous sequence Furthermore, the update function of the Hopfield neural network is defined as:

[0032] where, represents the state of node i at the next moment t + 1, represents the sign function, which outputs +1 when the input is positive, -1 when the input is negative, and usually outputs 0 or +1 when the input is 0, represents the connection weight from node j to i, represents the state of node j at time t, Denote the state of node j at time t + 1, where j represents the node number and I represents the index set. Initialize a binary sequence of length 32 randomly , and perform a certain number of rounds of iterative updates (such as 200 rounds). At the same time, use the evaluation function MC to evaluate the sequence. The MC index is calculated as:

[0033] where, represents the autocorrelation index, represents the cross-correlation index, represents the flip index.

[0034] If and the sequence does not exist in the set S of historical optimal sequences, then include it in the set S of historical optimal sequences. Finally, save the set S of historical optimal sequences with the longest length to generate a MOCSS with a larger capacity.

[0035] In another exemplary embodiment, a specific example of the above method is given, including the following steps: Step 1: Generation of the initial sequence set of Hp-LSTM 1) Given the training data (500 samples, time step 1, feature 1024), every 100 rounds of training, generate 100 consecutive value sequences through the Hp-LSTM model : , where are the parameters of the fully connected layer.

[0036] 2) Apply the sign function to the output:

[0037] Step 2: Construction of the symmetric weight matrix Reshape the 1024-dimensional sequence into a symmetric weight matrix.

[0038] Step 3: Asynchronous state update Randomly select 8 neurons for update, and each time select the index set . Use the above update function of the Hopfield neural network for update.

[0039] Step 4: Sequence grouping and cutting Divide the 1024-dimensional sequence into to form a set of MOCSS undetermined sequences with L = 8, M = 2, and K = 2. Where L is the sequence length, M is the number of tuples, which determines the number of users, and K is the number of group sequences, which determines the number of subcarriers.

[0040] Step 5: MC index calculation 1) Calculate autocorrelation. The autocorrelation function (time delay ) is expressed as:

[0041] where represents the autocorrelation value of the sequence , represents the time delay, represents the index of the sequence element; the autocorrelation values of the remaining sequences are calculated in the same way.

[0042] Thus, its autocorrelation index can be obtained as:

[0043] 2) Calculate cross-correlation. The cross-correlation function (time delay ) is expressed as: , where represents the cross-correlation value of the sequences and , represents the index of the sequence element; and so on to calculate the cross-correlation values of any two sequences; Thus, its cross-correlation index can be obtained as:

[0044] 3) Calculate flip correlation. The flip dot product (time delay ) is expressed as: , where represents the flip correlation value of the sequence , represents the index of the sequence element, and so on to calculate the flip correlation values of the remaining sequences; Thus, its flip index can be obtained as:

[0045] The total MC index is calculated as:

[0046] If the sequence set evaluation calculation yields its , and if this sequence set has never appeared (does not exist in the historical sequence set cluster), then add it to the candidate sequence set cluster.

[0047] Step 6: After reaching the maximum number of training rounds (e.g., 10,000 times), terminate the training and save the candidate sequence set cluster. Finally, 95 new MOCSSs (not present in the historical sequence set) are obtained.

[0048] The following evaluates the effect of the Hp-LSTM of this application.

[0049] Figure 3 The main training process of Hp-LSTM is shown. The abscissa represents the number of training rounds, and the ordinate represents the MC index. For each round of training, 100 sets of undetermined sequences are generated, and the average value ( ) and the minimum value ( ) of the MC for this round are statistically calculated.

[0050] From the figure, we can see that Hp-LSTM reaches 0 for the first time in the 8th round, indicating that the model has found the sequence set of for the first time and has obtained effective learning. After that, it has always maintained the ability to find the sequence set of . At the same time, the average value of its MC begins to gradually decrease and enters a plateau at 40 rounds, and then oscillates slightly between 10 and 20. Thus, it can be seen that Hp-LSTM can quickly extract sequence features in a very short number of training rounds, and at the same time, the average value of MC converges to a lower value extremely quickly, indicating that Hp-LSTM can adapt to the sequence generation task.

[0051] As Figures 5 - 6 shown, the training processes of Hp-LSTM and Hp-GAN are compared. Figure 5 and Figure 6 are the generator loss graph and discriminator loss graph of Hp-GAN respectively. The abscissa represents the number of training rounds, and the ordinate represents the binary cross-entropy loss; Figure 7 is the mean squared error loss image of Hp-LSTM. The abscissa represents the number of training rounds, and the ordinate represents the mean squared error loss. By comparison, it can be found that Hp-LSTM converges very quickly and completes convergence at about 100 rounds, while Hp-GAN basically converges at 3000 rounds, and the generator loss is still oscillating upward, proving that Hp-LSTM is significantly stronger than Hp-GAN in terms of sequence generation ability and feature extraction. In terms of convergence speed, it is much more stable than the Hp-GAN network.

[0052] As Figure 4 shown, by comparing the MC curve graphs, it can be found that Hp-GAN first finds the sequence set of at about 2500 rounds. Hp-LSTMM has the ability to generate MOCSS from the first round, proving that LSTM has more advantages in matrix feature learning than GAN, especially when generating longer and more complex sequences. In addition, the average value of MC of Hp-LSTM has reached the level of Hp-LSTM trained for 6000 rounds at the beginning of training, and its average value of MC is always lower than that of Hp-GAN throughout the entire training process, further confirming the advantages of LSTM.

[0053] Finally, using the same training set, compared with Hp-GAN which found 119 new MOCSS sequence sets after 10,000 rounds of training, the proposed method Hp-LSTM found 144 MOCSS sequence sets in 4 rounds of training.

[0054] In summary, it shows significant performance advantages compared with the Hp-GAN algorithm. From the perspective of the convergence characteristics of the training process, its training process stability and rapidity are significantly better than Hp-GAN. By comparing the MC (sequence metric) curves of the two, it can be seen that Hp-GAN first obtained a sequence set with an MC value of 0 after about 2500 rounds of training, indicating that it began to generate sequences that meet the strict conditions of MOCSS; on the contrary, Hp-LSTM showed the ability to generate MOCSS from the first round of training, which indicates that the LSTM network has stronger adaptability in matrix feature learning. Especially when dealing with long sequences and sequences with complex structures, its feature extraction and expression ability is significantly better than the GAN architecture.

[0055] In terms of sequence quality and search efficiency, the advantages of Hp-LSTM are also prominent. The average value of MC during the training process of Hp-LSTM reached the level of Hp-GAN after 6000 rounds of training at the initial stage of training, fully demonstrating its superiority in sequence optimization efficiency. Under the condition of the same training set, in the original paper of Hp-GAN, only 119 MOCSS sequence sets were found after 10,000 rounds of training, while Hp-LSTM successfully obtained 144 MOCSS sequence sets through 4 independent trainings. Not only the number of training rounds was greatly reduced, but also the number of sequence discoveries increased by about 21%. This result shows that through the deep integration of LSTM and discrete Hopfield neural network, Hp-LSTM has achieved a double breakthrough in the efficiency and quality of sequence search, and will be more adaptable to large-scale communication scenarios in the future, and at the same time provides a more efficient and stable solution for the application of CSS sequence generation and related fields.

[0056] In another exemplary embodiment, based on the same inventive concept as the method embodiment, a generation system for a large-capacity mutually positive and complementary sequence set is provided, including: An LSTM sequence generation module for training and generating an initial sequence set using a long short-term memory network LSTM; A Hopfield neural network optimization module for reshaping the initial sequence set into a symmetric weight matrix through a Hopfield neural network and performing asynchronous state updates; A sequence grouping and cutting module for grouping and cutting the updated sequence set to generate a to-be-determined sequence set; An index evaluation module for evaluating the to-be-determined sequence set based on the MC index and screening to obtain the final mutually positive and complementary sequence set.

[0057] including an MLAG protocol module, a dynamic routing protocol module, and an LACP protocol module; In another exemplary embodiment, based on the same inventive concept as the method embodiment, an electronic device is provided, including a memory and a processor. A computer instruction that can run on the processor is stored on the memory. When the processor runs the computer instruction, it executes a method for generating a large-capacity mutually complementary sequence set provided in the embodiments of the present invention.

[0058] The processor can be a single-core or multi-core central processing unit or a specific integrated circuit, or an integrated circuit configured to implement one or more of the present invention.

[0059] The embodiments of the subject matter and functional operations described in this specification can be implemented in the following: tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or a combination of one or more of them. The embodiments of the subject matter described in this specification can be implemented as one or more computer programs, that is, one or more modules in computer program instructions encoded on a tangible non-transitory program carrier to be executed by a data processing device or to control the operation of the data processing device. Alternatively or additionally, the program instructions can be encoded on an artificially generated propagated signal, such as a machine-generated electrical, optical, or electromagnetic signal, which is generated to encode information and transmit it to a suitable receiver device for execution by the data processing device.

[0060] The processing and logic flows described in this specification can be executed by one or more programmable computers executing one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flows can also be executed by dedicated logic circuits - such as FPGAs (Field Programmable Gate Arrays) or ASICs (Application Specific Integrated Circuits), and the device can also be implemented as a dedicated logic circuit.

[0061] Processors suitable for executing computer programs include, for example, general and / or special purpose microprocessors, or any other type of central processing unit. Generally, the central processing unit will receive instructions and data from read-only memory and / or random access memory. The basic components of a computer include a central processing unit for implementing or executing instructions and one or more memory devices for storing instructions and data. Generally, a computer will also include one or more mass storage devices for storing data, such as magnetic disks, magneto-optical disks, or optical disks, etc., or the computer will be operatively coupled to such a mass storage device to receive data from it or transfer data to it, or both. However, a computer is not necessarily required to have such devices. In addition, a computer may be embedded in another device, such as a mobile phone, a personal digital assistant (PDA), a mobile audio or video player, a game console, a global positioning system (GPS) receiver, or a portable storage device such as a universal serial bus (USB) flash drive, to name just a few examples.

[0062] It should be understood that each block in a flowchart or block diagram can represent a module, a program segment, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, as well as the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0063] The above specific embodiments are detailed descriptions of the present invention. It should not be considered that the specific embodiments of the present invention are limited only to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions and substitutions can still be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for generating a large-capacity mutually orthogonal complementary sequence set, characterized in that, Including: Training with a Long Short-Term Memory (LSTM) network to generate an initial sequence set; Reshaping the initial sequence set into a symmetric weight matrix through a Hopfield neural network and performing asynchronous state updates; Grouping and cutting the updated sequence set to generate a to-be-determined sequence set; Evaluating the to-be-determined sequence set based on the MC metric and screening to obtain a final mutually positive and complementary sequence set.

2. The generating method of a large-capacity mutually positive and complementary sequence set according to claim 1, characterized in that, The Long Short-Term Memory (LSTM) network is defined as: , where y represents the output, X represents the input tensor, is the operation of the LSTM layer, is the weight matrix of the LSTM layer, is the weight matrix of the fully connected layer, is the bias vector, represents the hyperbolic tangent activation function.

3. The generating method of a large-capacity mutually positive interactive complementary sequence set according to claim 2, characterized in that, The step of training with a Long Short-Term Memory (LSTM) network to generate an initial sequence set includes: Using an Adam optimizer to minimize the mean squared error loss and using a cosine annealing learning rate scheduler to dynamically adjust the learning rate to generate a continuous value sequence; Binarizing the continuous value sequence by applying a sign function.

4. A method for generating a large-capacity mutually orthogonal complementary sequence set according to claim 1, characterized in that, The symmetric weight matrix is: , where represents reshaping the matrix shape, represents a vector of length 1024, and i represents the i-th index in the continuous sequence in it.

5. The generating method of a large-capacity mutually positive interleaved complementary sequence set according to claim 4, wherein The asynchronous state update includes: Defining a Hopfield neural network update function and randomly selecting neurons for update; the update function is: , where, represents the state of node i at the next moment t + 1, represents the sign function, which outputs +1 when the input is positive, -1 when the input is negative, and usually outputs 0 or +1 when the input is 0, represents the connection weight from node j to i, represents the state of node j at time t, represents the state of node j at time t + 1, where j represents the node number and I represents the index set.

6. A method for generating a large-capacity mutually positive and complementary sequence set according to claim 1, characterized in that The MC metric is calculated as: , where represents the autocorrelation index, represents the cross-correlation index, represents the flip index.

7. A method for generating a large-capacity mutually positive and complementary sequence set according to claim 6, characterized in that, The step of screening to obtain a final mutually positive and complementary sequence set includes: If and the sequence does not exist in the set S of historical optimal sequences, then incorporate it into the set S of historical optimal sequences.

8. A generation system for a large-capacity mutually positive interactive complementary sequence set, characterized in that, Including: An LSTM sequence generation module for training with a Long Short-Term Memory (LSTM) network to generate an initial sequence set; A Hopfield neural network optimization module for reshaping the initial sequence set into a symmetric weight matrix through a Hopfield neural network and performing asynchronous state updates; A sequence grouping and cutting module for grouping and cutting the updated sequence set to generate a to-be-determined sequence set; A metric evaluation module for evaluating the to-be-determined sequence set based on the MC metric and screening to obtain a final mutually positive and complementary sequence set.

9. An electronic device, comprising a memory and a processor, wherein computer instructions that can run on the processor are stored on the memory, and characterized in that, When the processor runs computer instructions, it executes the method for generating a large-capacity mutually positive and complementary sequence set according to any one of claims 1-7.

Images