An optimal odd complementary sequence pair search method, framework, storage medium and device

Abstract

The application discloses a kind of optimal odd complementary sequence pair search method, framework, storage medium and equipment, belong to the cross field of signal processing and artificial intelligence, which comprises: initialization sequence prefix, search is carried out by the selection, expansion, evaluation, back transmission steps of MCTS;Wherein, evaluation step uses CNN to directly estimate the value of current sequence prefix, avoid the high computing overhead caused by random completion in traditional MCTS;The state, strategy and return information generated in the search process are independently managed and fixed length processing by replay pool, guarantee the stability and generalization ability of CNN training under different length sequence input;And update CNN parameter in closed loop mode, continuously optimize search strategy.The application can efficiently and stably search out optimal odd complementary sequence pair satisfying Krammer lower bound in the field of radar signal design, wireless communication channel estimation and synchronization, etc., significantly improve the efficiency and engineering practicability of sequence search.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of signal processing and artificial intelligence, and in particular to a method, framework, storage medium, and device for searching optimal odd complementary sequence pairs. Background Technology

[0002] Odd complementary sequence pairs (OCSPs) are a class of discrete signal sequences with special mathematical properties. Their aperiodic autocorrelation and cross-correlation functions ideally or nearly ideally sum to zero outside of zero time delay. This property makes them valuable in fields requiring high-precision signal design and processing, such as radar systems, wireless communication, sonar detection, and cryptographic synchronization. For example, in radar systems, using Odd complementary sequence pairs as transmitted waveforms can significantly suppress range sidelobes, improving the accuracy and resolution of target detection. In channel estimation for wireless communication, using them as training sequences can effectively combat inter-symbol interference and improve the accuracy of channel parameter estimation.

[0003] Traditional methods for constructing odd complementary sequence pairs mainly fall into two categories: analytical mathematical construction and exhaustive search methods. Analytical mathematical construction directly generates sequence pairs of a specific length using rigorous mathematical theories (such as generalized Boolean functions and recursive construction). This method is efficient but has significant drawbacks: first, not all optimal odd complementary sequence pairs of all lengths can be constructed using existing mathematical methods, creating a construction blind spot; second, even if a sequence pair can be constructed, it may not be the "optimal" performer (e.g., reaching the Cramer-Rao lower bound), often only satisfying the basic condition of complementarity, failing to achieve optimality in more refined indicators such as estimation error and robustness. Exhaustive search methods, on the other hand, traverse all possible sequence combinations of a given length, selecting the optimal sequence pair that meets the conditions. Theoretically, this method can find the optimal solution, but the computational complexity increases exponentially with the sequence length. For slightly longer sequences, the enormous search space makes the computation time unbearable, resulting in low efficiency and extremely low practical engineering application value.

[0004] To achieve efficient exploration in a vast search space, the Monte Carlo Tree Search (MCTS) method was introduced. Traditional MCTS requires simulation by randomly completing the remaining parts of the sequence during the search process. In sequence search problems, this random completion method has a high degree of randomness, resulting in a high variance in the simulation reward. It requires an extremely large number of simulations to obtain a stable value assessment, leading to excessively high computational costs and poor stability of the search results.

[0005] Currently, deep neural networks (DNNs) have been introduced to assist MCTS in sequence search. However, in specific applications of sequence search, using fully connected DNNs presents the following problems:

[0006] 1. DNN lacks an inductive bias for the local correlation of sequence data, which makes it inefficient in modeling structural inputs such as "sequence prefixes" and redundant parameters. This often results in insufficient focus of policy priors and large variance in value estimation, which in turn makes backpropagation unstable and requires more simulations to converge.

[0007] 2. During training, replay pool samples are prone to mixing in different length scenarios, causing training anomalies or failures, which requires additional padding and resetting methods to avoid.

[0008] In summary, existing technologies always have certain shortcomings in terms of accuracy or efficiency. Even with the introduction of DNNs for optimization, the final results are still affected by inappropriate model structure and flawed training mechanisms. Therefore, there is an urgent need for a new method that can efficiently and stably search for optimal odd complementary sequence pairs in a large-scale sequence space. Summary of the Invention

[0009] The purpose of this invention is to overcome the problems existing in the prior art and to provide an optimal odd complementary sequence pair search method, framework, storage medium and device.

[0010] The objective of this invention is achieved through the following technical solution:

[0011] Firstly, a method for searching for optimal odd complementary sequence pairs is provided, including the following steps:

[0012] S1. Initialize the root node, whose state is the current prefix of the sequence to be generated, and fill the unfilled part with the preset symbol;

[0013] S2. Call the convolutional neural network to predict the prior probability of each action. Starting from the current root node, perform Monte Carlo tree search to recursively select child edges and select actions according to the predictor-tree confidence upper bound algorithm until a leaf node is reached.

[0014] S3. When the leaf node is visited for the first time, perform expansion;

[0015] S4. Call the convolutional neural network to evaluate the sequence prefix state corresponding to the leaf node, and output the value estimate and / or policy prior probability distribution of the state;

[0016] S5. Use the value estimate and / or policy prior probability distribution obtained in step S4 as the simulated reward, back from the current node to the root node, and update the statistics on the path.

[0017] S6. After repeating steps S2 to S5 multiple times for simulation, calculate the action distribution from the number of visits of each action at the root node, and select an action from the action distribution;

[0018] S7. Based on the action selected in step S6, generate the next state and iteratively execute steps S2 to S6 as the new root node until a complete candidate sequence pair is generated;

[0019] S8. Determine whether the complete candidate sequence pair is the optimal odd complementary sequence pair. If so, record and store the sequence pair; otherwise, mark the current candidate sequence pair as invalid.

[0020] In some embodiments, the method further includes a training step:

[0021] The states generated during the search process with different sequence lengths, the policy distribution of the convolutional neural network output, and the final reward are stored in the form of triples in an independent replay buffer corresponding to the sequence length.

[0022] When the number of samples in any of the independent playback buffers reaches the preset batch size, data is sampled from the independent playback buffer to update the convolutional neural network.

[0023] In some embodiments, the independent playback buffer maintains a uniform input dimension by zero-padding or truncation when storing triples.

[0024] In some embodiments, the selection of actions based on the predictor-tree confidence upper bound algorithm includes:

[0025] This is done using the following formula:

[0026]

[0027] in, This indicates the optimal action to choose. Indicates the index of the current action. Indicates the current leaf node, This represents the average reward for each action. Indicates the exploration coefficient. This represents the prior probability of each action. This indicates the number of times each action is accessed. Indicates a temporary action index. This indicates the number of times a temporary action has been accessed.

[0028] In some embodiments, it also includes:

[0029] Before selecting an action at the root node, the prior probability is... Dirichlet noise was injected to increase the exploratory nature of the process.

[0030] In some embodiments, calculating the action distribution from the number of visits to each action at the root node includes:

[0031] Use temperature control strategies to select actions.

[0032] In some embodiments, determining whether the complete candidate sequence pair is an optimal odd complementary sequence pair includes:

[0033] Using the Cramero lower bound as a performance metric for the candidate sequence pair, its reward is defined as:

[0034] For candidate sequence pairs that satisfy the optimal odd complementarity condition and whose relevant information matrix is ​​non-singular, a reward negatively correlated with the lower bound of Cramer-Robb is given; for invalid candidate sequence pairs, a fixed high penalty is given.

[0035] Secondly, a search framework for optimal odd complementary sequence pairs is provided for performing the optimal odd complementary sequence pair search method described in the first aspect, including:

[0036] The Monte Carlo tree search module is used to perform selection, expansion, evaluation, and backpropagation steps sequentially in a large sequence space.

[0037] A convolutional neural network module is used to output policy distribution and value estimate in the evaluation step of the Monte Carlo tree search module;

[0038] An independent replay buffer is used to store the state, strategy, and reward information generated during the Monte Carlo tree search module's search process;

[0039] The network training update module is used to sample training data in batches from an independent replay buffer, update the parameters of the convolutional neural network, and return the updated network to the Monte Carlo tree search module.

[0040] Thirdly, a computer-readable storage medium is provided, the computer-readable storage medium storing a computer program, which, when executed by a processor, implements the optimal odd complementary sequence pair search method described in the first aspect.

[0041] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores computer instructions executable on the processor, and the processor executes an optimal odd complementary sequence pair search method as described in the first aspect when executing the computer instructions.

[0042] It should be further noted that the technical features corresponding to the above-mentioned options and embodiments can be combined or substituted with each other to form new technical solutions without conflict.

[0043] Compared with the prior art, the beneficial effects of the present invention are:

[0044] 1. Dual Improvement in Search Efficiency and Accuracy: This invention creatively combines the global planning capability of Monte Carlo Tree Search with the local feature perception and rapid evaluation capability of Convolutional Neural Networks (CNNs). The value early stopping mechanism implemented through CNNs completely eliminates the computationally expensive and variance-laden random completion simulation step in traditional MCTS, significantly improving the speed of a single simulation. Simultaneously, the high-quality prior probabilities provided by CNNs guide MCTS to focus more on promising search directions, thus enabling the search for better sequence solutions with fewer simulations under the same computational resources, achieving a balance between efficiency and accuracy.

[0045] 2. Model Design Better Fits Sequence Data Structure: Compared to the unstructured fully connected DNNs used in existing technologies, this invention employs a convolutional neural network as the evaluator. The inherent local connectivity and weight sharing characteristics of CNNs make them particularly adept at capturing the correlations (i.e., local structure) between adjacent symbols in sequence data. This inductive bias is highly compatible with sequence prefix evaluation tasks. Therefore, CNNs can more accurately and stably estimate the value of partial sequence states, outputting more focused policy priors, fundamentally reducing the blindness and variance of the search.

[0046] 3. Stable Training Process and Strong Generalization Ability: To address the training anomalies caused by the mixing of sequence samples of different lengths, this invention proposes a strategy of independent management and fixed-length processing of the replay pool. Specifically, it maintains independent replay buffers for different target lengths and normalizes all input states to the same dimension by padding with zeros or truncating them. This design ensures the consistency of data distribution within each batch during CNN training, avoiding training failures caused by inconsistent sample shapes in existing technologies, and greatly improving training stability and convergence speed. Simultaneously, this enables a single CNN model to effectively handle sequence search tasks of various lengths, significantly enhancing its generalization ability.

[0047] 4. Possesses clear engineering practical value: This invention, through a closed-loop optimized system design (search-data collection-training-refined search), enables the algorithm to have self-improvement capabilities. By reasonably setting the simulation budget, temperature parameter attenuation strategy, and regularized training, it can stably produce high-performance optimal odd complementary sequence pairs applicable to radar, communications, and other fields under acceptable computational resource consumption. This solves the engineering problems of unsolvable or non-optimal mathematical construction methods and the infeasibility of exhaustive methods, providing an efficient and reliable automated tool for signal design in related fields.

[0048] 5. Significant Application Effects: Applying the optimal odd complementary sequence pairs obtained by this invention to the aforementioned practical scenarios can bring direct performance gains. For example, in radar systems, it can improve ranging accuracy and resolution; in wireless communication, it can improve the accuracy of channel estimation, thereby reducing the system bit error rate and improving spectral efficiency. The output of this invention has the potential to be directly translated into improving the performance of existing communication and detection systems. Attached Figure Description

[0049] Figure 1 This is a flowchart of an optimal odd complementary sequence pair search method according to the present invention;

[0050] Figure 2 This is a schematic diagram of the search framework of the present invention. Detailed Implementation

[0051] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0052] It should be noted that the defects in the solutions in the prior art are all the results of the inventors' practice and careful research. Therefore, the discovery process of the above problems and the solutions proposed by the embodiments of this application in the following text should be the inventors' contributions to this application in the process of invention and creation, and should not be understood as technical content known to those skilled in the art.

[0053] In view of the technical problems pointed out in the background art, the present invention provides the following embodiments:

[0054] Reference Figure 1 In one exemplary embodiment, an optimal odd complementary sequence pair search method is provided, comprising the following steps:

[0055] S1. Initialize the root node, whose state is the current prefix of the sequence to be generated, and fill the unfilled part with the preset symbol;

[0056] S2. Call the convolutional neural network to predict the prior probability of each action. Starting from the current root node, perform Monte Carlo tree search to recursively select child edges and select actions according to the predictor-tree confidence upper bound algorithm until a leaf node is reached.

[0057] S3. When the leaf node is visited for the first time, perform expansion;

[0058] S4. Call the convolutional neural network to evaluate the sequence prefix state corresponding to the leaf node, and output the value estimate and / or policy prior probability distribution of the state;

[0059] S5. Use the value estimate and / or policy prior probability distribution obtained in step S4 as the simulated reward, back from the current node to the root node, and update the statistics on the path.

[0060] S6. After repeating steps S2 to S5 multiple times for simulation, calculate the action distribution from the number of visits of each action at the root node, and select an action from the action distribution;

[0061] S7. Based on the action selected in step S6, generate the next state and iteratively execute steps S2 to S6 as the new root node until a complete candidate sequence pair is generated;

[0062] S8. Determine whether the complete candidate sequence pair is the optimal odd complementary sequence pair. If so, record and store the sequence pair; otherwise, mark the current candidate sequence pair as invalid.

[0063] The specific implementation methods for each step are described below.

[0064] In step S1, the root node is initialized. The state is the prefix of the current sequence. Unfilled portions are padded with 0s.

[0065] In step S2, the CNN is invoked once to obtain the prior probability of each compound action. Among the actions , q For alphabet size, To fill the step size, starting from the root node, MCTS recursively selects child edges, maximizing them according to the Predictor-Tree Confidence Upper Bound (PUCT) algorithm, until an expandable leaf node (child node) is reached. The action selection formula is as follows:

[0066]

[0067] in, This represents the "optimal action" calculated according to this PUCT formula at the current node. Indicates the index of the current action. Indicates the current leaf node, This represents the average reward for each action. This represents the "exploration coefficient" in the PUCT formula. This represents the prior probability of each action. This indicates the number of times each action is accessed. a′ This represents a "temporary action index" used for summation. Indicates temporary action a′ The number of times it has been accessed by MCTS.

[0068] Specifically, through , , , To control exploration. During the data generation phase, Dirichlet noise is also injected into the root node to prevent an action from being referenced multiple times, thus increasing exploratory power.

[0069]

[0070] in, This indicates that after adding noise, the root node's actions... a The new prior probability, This indicates the root node's actions without added noise. a The original prior probabilities are output by the CNN. Indicates the distribution from Dirichle Dir ( a ) a The sampled vector corresponds to the action. a That one-dimensional component, Indicates the noise mixing coefficient (0 < ε <1, control the weight of "original prior" and "Dirichle noise".

[0071] In step S3, when the new child node Expand upon first access. Index the actions. Decode into a length of The symbol block is concatenated to the prefix generator substate.

[0072] In step S4, when a leaf node is reached, traditional random completion is no longer used. Instead, the CNN is directly called to evaluate the current prefix sequence and obtain the value estimate. This value is immediately returned to the path, thus avoiding the high variance and overhead of random completion.

[0073] In step S5, the simulated reward is propagated back from the current leaf node to the root node, and the statistics on the path are updated:

[0074]

[0075]

[0076] Where z is the reward for one simulation, i.e., the CNN prediction value used in the prefix state. , Represents a node vm Next action a n The average return received Represents a node v m Next action a n The number of times it is accessed (selected) by MCTS.

[0077] In step S6, after m simulations are completed at the root node, the action distribution is calculated based on the number of visits, and a temperature control strategy is used:

[0078]

[0079] in, The temperature parameter is used to adjust the randomness and determinism of action selection. The higher the temperature, the more even the distribution, and the more equal the probability of selecting each action. Conversely, the lower the temperature, the more likely the action will be selected. Indicates at the root node v 0 Action a The probability of being selected. Represents the root node v 0 Next action a Number of visits, This indicates all candidate actions. a′ The number of visits to a single item in the sum of the number of visits.

[0080] In step S7, from the distribution Select an action to generate the next state. It then becomes the new root node and enters the next iteration, repeating steps S1-S6. In the next iteration, a complete candidate sequence pair S is finally generated, and the final reward is obtained, where... N It is the length of the sequence pair.

[0081] In step S8, if the generated candidate sequence pair S satisfies the optimal condition, it will be recorded and stored. Specifically, in this embodiment, the reward function is designed as follows:

[0082] During the search process, the Cramer-Rao lower bound is used as a performance metric. The Cramer-Rao lower bound refers to the minimum achievable variance of any unbiased estimate in parameter estimation problems. When the sequence pairs used to construct the matrix for parameter estimation reach the Cramer-Rao lower bound, this sequence pair is called the optimal sequence pair, and its reward is defined as:

[0083] For candidate sequence pairs that satisfy the optimal odd complementarity condition and whose relevant information matrix is ​​non-singular, a reward negatively correlated with the lower bound of Cramer-Robb is given:

[0084] ;

[0085] Invalid candidate sequence pairs will be subject to a fixed, high penalty (e.g., -10). 6 To prevent the search from getting stuck on infeasible solutions, this design naturally transforms the "minimize CRLB" objective into a "maximize reward" problem under the MCTS framework.

[0086] In step S9, during each complete sequence generation round, the triples are... Store the corresponding length N Independent replay buffers are used, and a uniform input dimension is maintained through zero-padding or truncation. Different sequence lengths maintain corresponding independent replay buffers. When the number of samples reaches the batch size (BATCH_SIZE), a subset of samples is extracted to update the CNN and optimize the objective function.

[0087]

[0088] in, This is a regularization coefficient, used to prevent overfitting and improve the model's generalization ability. This represents the value function estimated by the model, corresponding to the predicted value of the searched leaf nodes. Indicates a reward. The policy network is state-based. s t The probability distribution of the output. This represents all the parameters of the CNN network used for training. This represents the policy distribution calculated from the number of visits to MCTS.

[0089] Therefore, the system in this embodiment reduces the computational complexity of the search process through the value early stopping mechanism, improves the stability and generalization ability of network training through independent management and fixed-length processing, and enhances the efficiency and accuracy of sequence search by combining closed-loop optimization structure, thus possessing good engineering feasibility and application value.

[0090] In another exemplary embodiment, a search framework for optimal odd complementary sequence pairs is provided for performing the aforementioned optimal odd complementary sequence pair search method, such as... Figure 2 As shown, the search framework includes:

[0091] The Monte Carlo tree search module is used to sequentially perform selection, expansion, evaluation, and backpropagation steps in a large sequence space to determine candidate sequence pairs;

[0092] The convolutional neural network module is used to output policy distribution and value estimation in the evaluation step of the Monte Carlo tree search module, so as to achieve early value stopping and avoid the high computational overhead caused by random completion in traditional methods.

[0093] An independent replay buffer is used to store the state, strategy, and reward information generated during the Monte Carlo tree search module's search process. Through independent management and fixed-length processing, the consistency of input under different sequence lengths is ensured.

[0094] The network training update module is used to sample training data in batches from each independent replay buffer, update the parameters of the convolutional neural network, and return the updated network to the Monte Carlo tree search module, thereby realizing closed-loop optimization of search and training.

[0095] In another exemplary embodiment, based on the same inventive concept as the method embodiment, a computer-readable storage medium is provided. This computer-readable storage medium stores a computer program, which, when executed by a processor, implements a method for searching optimal odd complementary sequence pairs provided in this embodiment of the invention. Based on this understanding, the technical solution of this embodiment, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0096] 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. The memory stores computer instructions that can be executed on the processor. When the processor executes the computer instructions, it performs a search method for optimal odd complementary sequence pairs provided in the embodiments of the present invention.

[0097] The processor may be a single-core or multi-core central processing unit or a specific integrated circuit, or one or more integrated circuits configured to implement the present invention.

[0098] The embodiments of the subject matter and functional operation described in this specification can be implemented in: tangibly embodied computer software or firmware, computer hardware including the structures disclosed in this specification and their structural equivalents, or combinations thereof. Embodiments of the subject matter described in this specification can be implemented as one or more computer programs, i.e., one or more modules of computer program instructions encoded on a tangible, non-transitory program carrier for execution by a data processing device or for controlling the operation of a data processing device. Alternatively or additionally, the program instructions may be encoded on artificially generated propagation signals, such as machine-generated electrical, optical, or electromagnetic signals, which are generated to encode information and transmit it to a suitable receiving device for execution by the data processing device.

[0099] The processing and logic flow described in this specification can be executed by one or more programmable computers that execute one or more computer programs to perform corresponding functions by operating on input data and generating output. The processing and logic flow can also be executed by dedicated logic circuitry—such as FPGAs (Field-Programmable Gate Arrays) or ASICs (Application-Specific Integrated Circuits), and the device can also be implemented as dedicated logic circuitry.

[0100] Suitable processors for executing computer programs include, for example, general-purpose and / or special-purpose microprocessors, or any other type of central processing unit. Typically, the central processing unit receives 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. Typically, a computer will also include one or more mass storage devices for storing data, such as disks, magneto-optical disks, or optical disks, or the computer will be operatively coupled to such mass storage devices to receive data from or transfer data to them, or both. However, a computer is not required to have such devices. Furthermore, a computer can 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 a few.

[0101] It should be understood that each block in a flowchart or block diagram can represent a module, segment, or portion of code, which contains one or more executable instructions for implementing the 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 shown in the figures. For example, two consecutive blocks may actually be executed substantially in parallel, or they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram and / or flowchart, and combinations of blocks in block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0102] The above detailed embodiments are a description of the present invention. It should not be considered that the specific embodiments of the present invention are limited to these descriptions. For those skilled in the art, several simple deductions and substitutions can be made without departing from the concept of the present invention, and all of these should be considered to fall within the protection scope of the present invention.

Claims

1. An optimal odd complementary sequence pair search method, applied to signal sequence design in radar systems or wireless communication systems, characterized in that, Includes the following steps: S1. Initialize the root node, whose state is the current prefix of the signal sequence to be generated, and fill the unfilled part with the preset symbol; S2. Call the convolutional neural network to predict the prior probability of each action. Starting from the current root node, perform Monte Carlo tree search to recursively select child edges and select actions according to the predictor-tree confidence upper bound algorithm until a leaf node is reached. S3. When the leaf node is visited for the first time, perform expansion; S4. Call the convolutional neural network to evaluate the prefix state of the signal sequence corresponding to the leaf node, and output the value estimate and / or policy prior probability distribution of the state; S5. Use the value estimate and / or policy prior probability distribution obtained in step S4 as the simulated reward, back from the current node to the root node, and update the statistics on the path. S6. After repeating steps S2 to S5 multiple times for simulation, calculate the action distribution from the number of visits of each action at the root node, and select an action from the action distribution; S7. Based on the action selected in step S6, generate the next signal sequence state and iteratively execute steps S2 to S6 as the new root node until a complete candidate signal sequence pair is generated; S8. Determine whether the complete candidate signal sequence pair is an optimal odd complementary sequence pair. If so, record and store the signal sequence pair and use it as a radar transmission waveform or communication training sequence. Otherwise, mark the current candidate signal sequence pair as invalid. The method also includes a training step: The states generated during the search process with different sequence lengths, the policy distribution of the convolutional neural network output, and the final reward are stored in the form of triples in an independent replay buffer corresponding to the sequence length. When the number of samples in any of the independent playback buffers reaches the preset batch size, data is sampled from the independent playback buffer to update the convolutional neural network.

2. The optimal odd complementary sequence pair search method according to claim 1, characterized in that, The independent playback buffer maintains a uniform input dimension by zero-padding or truncation when storing triples.

3. The optimal odd complementary sequence pair search method according to claim 1, characterized in that, The action selection based on the predictor-tree confidence upper bound algorithm includes: This is done using the following formula: ,in, This indicates the optimal action to choose. Indicates the index of the current action. Indicates the current leaf node, This represents the average reward for each action. Indicates the exploration coefficient. This represents the prior probability of each action. This indicates the number of times each action is accessed. Indicates a temporary action index. This indicates the number of times a temporary action has been accessed.

4. The optimal odd complementary sequence pair search method according to claim 3, characterized in that, Also includes: Before selecting an action at the root node, the prior probability is... Dirichlet noise was injected to increase the exploratory nature of the process.

5. The optimal odd complementary sequence pair search method according to claim 1, characterized in that, The calculation of the action distribution from the number of visits to each action at the root node includes: Use temperature control strategies to select actions.

6. The optimal odd complementary sequence pair search method according to claim 1, characterized in that, The determination of whether the complete candidate sequence pair is an optimal odd complementary sequence pair includes: Using the Cramero lower bound as a performance metric for the candidate signal sequence pair, its reward is defined as: For candidate signal sequence pairs that satisfy the optimal odd complementarity condition and whose related information matrix is ​​non-singular, a reward negatively correlated with the lower bound of Cramer-Rao is given; for invalid candidate signal sequence pairs, a fixed high penalty is given.

7. A search system for optimal odd complementary sequence pairs, used to execute the optimal odd complementary sequence pair search method according to any one of claims 1-6, characterized in that, include: The Monte Carlo tree search module is used to perform selection, expansion, evaluation, and backpropagation steps sequentially in a large sequence space. A convolutional neural network module is used to output policy distribution and value estimate in the evaluation step of the Monte Carlo tree search module; An independent replay buffer is used to store the state, strategy, and reward information generated during the Monte Carlo tree search module's search process; The network training update module is used to sample training data in batches from an independent replay buffer, update the parameters of the convolutional neural network, and return the updated network to the Monte Carlo tree search module.

8. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the optimal odd complementary sequence pair search method according to any one of claims 1-6.

9. An electronic device comprising a memory and a processor, wherein the memory stores computer instructions executable by the processor, characterized in that, When the processor executes computer instructions, it performs an optimal odd complementary sequence pair search method as described in any one of claims 1-6.

Images