Method and apparatus for training signal modulation recognition model

Abstract

The application provides a training method and device of a signal modulation identification model. The method is implemented by using a blockchain-federated learning framework including a blockchain layer and a calculation layer. A plurality of distributed calculation nodes in the calculation layer are respectively connected with a plurality of block representatives in the blockchain layer. The method locally trains the signal modulation identification model in the calculation layer to obtain updated local parameters of the model, and obtains global parameters through effective evaluation and weighted aggregation of the plurality of local parameters. The local parameters are packaged into transactions in the blockchain layer, and blocks are generated according to the global parameters. The transactions and blocks are broadcasted and verified, and record information is generated.

Description

Technical Field

[0001] This application relates to distributed computing technology, and more particularly to a training method and apparatus for a signal modulation recognition model. Background Technology

[0002] Automatic modulation identification (EMI) is an intermediate process between signal detection and demodulation, playing a crucial role in various civilian and military applications. With the rapid development of wireless communication technology, especially digital communication technology, signal modulation techniques have become increasingly complex, with a growing number of modulation methods. Furthermore, signals are generally encrypted and processed with source and channel coding before modulation, which has significantly limited traditional manual modulation identification methods.

[0003] With the rapid development and widespread application of deep learning technology, methods for automatically training modulation and recognition models using deep learning have emerged. However, training models using deep learning requires a large amount of data to ensure the effectiveness of the training, thus requiring user data to be transmitted to a central server. Using these methods makes it difficult to guarantee data security and privacy, and it cannot effectively prevent attacks on the server from malicious nodes. Summary of the Invention

[0004] In view of this, the purpose of this application is to propose a training method and apparatus for a signal modulation recognition model.

[0005] To achieve the above objectives, this application provides one or more embodiments of a training method for a signal modulation recognition model. The method utilizes a blockchain-federated learning framework including a blockchain layer and a computation layer. Multiple distributed computing nodes in the computation layer are communicatively connected to multiple block representatives in the blockchain layer. The method includes iteratively executing the following operations until the signal modulation recognition model satisfies a first preset condition:

[0006] Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node.

[0007] The block representative packages the received local parameters into a first transaction and stores it in the local first transaction pool, and broadcasts the first transaction in the blockchain layer;

[0008] Upon receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives, in response to determining that the first transaction has passed validity verification, stores the first transaction in its local second transaction pool.

[0009] The block represents the response to determining that the second preset condition is met, and sends all transactions stored in the local first transaction pool to the computing node;

[0010] The computing node uses a preset algorithm to evaluate the validity of all received transactions and perform weighted aggregation to obtain a first global parameter, and then sends the first global parameter to the block representative.

[0011] This block represents the generation of a new block based on the received first global parameters, and the broadcast of the new block in the blockchain layer;

[0012] In response to determining that the blockchain layer has been updated, the computing node downloads the second global parameter from the new block from the block representative as a local parameter of the signal modulation recognition model stored locally.

[0013] Optionally, the preset algorithm includes an objective weighting method and Newton's law of cooling for determining the timeliness of the local parameters.

[0014] Optionally, the validity verification of the first transaction is performed using a hash algorithm and an asymmetric encryption algorithm.

[0015] Optionally, the first preset condition includes: the signal modulation recognition model characterized by the first global parameter converges or reaches a predetermined recognition accuracy.

[0016] Optionally, the second preset condition includes: the first transaction pool is full, or the predetermined block generation time has arrived.

[0017] Optionally, the operation further includes: each of the other block representatives among the plurality of block representatives performing block verification on the new block after receiving the broadcast of the new block.

[0018] Optionally, block verification of the new block includes: performing block verification of the new block using a hash algorithm and an asymmetric encryption algorithm.

[0019] Optionally, the operation further includes: in response to determining that the new block has passed the block verification, the other block represents an update to the blockchain layer.

[0020] Based on the same inventive concept, one or more embodiments of this application also provide a training apparatus for a signal modulation recognition model. The apparatus includes a blockchain layer and a computation layer forming a blockchain-federated learning framework. Multiple distributed computing nodes in the computation layer are communicatively connected to multiple block representatives in the blockchain layer. The computing nodes and the block representatives are configured to train the signal modulation recognition model by iteratively performing the following operations until the signal modulation recognition model satisfies a first preset condition:

[0021] Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node.

[0022] The block representative packages the received local parameters into a first transaction and stores it in the local first transaction pool, and broadcasts the first transaction in the blockchain layer;

[0023] Upon receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives, in response to determining that the first transaction has passed validity verification, stores the first transaction in its local second transaction pool.

[0024] The block represents the response to determining that the second preset condition is met, and sends all transactions stored in the local first transaction pool to the computing node;

[0025] The computing node uses a preset algorithm to evaluate the validity of all received transactions and perform weighted aggregation to obtain a first global parameter, and then sends the first global parameter to the block representative.

[0026] This block represents the generation of a new block based on the received first global parameters, and the broadcast of the new block in the blockchain layer;

[0027] In response to determining that the blockchain layer has been updated, the computing node downloads the second global parameter from the new block from the block representative as a local parameter of the signal modulation recognition model stored locally.

[0028] Optionally, the preset algorithm includes an objective weighting method and Newton's law of cooling for determining the timeliness of the local parameters.

[0029] As can be seen from the above, the signal modulation recognition model training method and apparatus provided in one or more embodiments of this application utilize a blockchain-federated learning framework including a blockchain layer and a computation layer. In the computation layer, the signal modulation recognition model is trained locally to obtain updated local parameters. Global parameters are obtained after effective evaluation and weighted aggregation of the multiple local parameters. In the blockchain layer, transactions are packaged using the local parameters, and blocks are generated based on the global parameters. The transactions and blocks are then broadcast and verified, and record information is generated. Through the federated learning method, a high recognition rate for signal recognition is achieved without uploading data to a central server, meeting users' requirements for data security and privacy. By using the blockchain method and the federated learning value evaluation mechanism, attacks from malicious nodes are weakened, while a decentralized framework is proposed, avoiding the possibility of malicious data tampering. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in this application or related technologies, the drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0031] Figure 1 This is a schematic diagram of a blockchain-federated learning framework for one or more embodiments of this application;

[0032] Figure 2 This is a flowchart illustrating a training method for a signal modulation recognition model according to one or more embodiments of this application.

[0033] Figure 3 This is a flowchart illustrating the application of a training method for a signal modulation recognition model according to one or more embodiments of this application to a computing node.

[0034] Figure 4 This is a flowchart illustrating the application of a signal modulation recognition model training method according to one or more embodiments of this application to a block representative. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0036] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this application should have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," and similar terms used in the embodiments of this application do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0037] As described in the background section, related technologies employ deep learning techniques to train automatic modulation recognition models, reducing the difficulty of modulation recognition and improving signal recognition rates. However, the large amount of data transmitted from the local machine to the central processing unit (CPU) not only increases the CPU's computational burden but also raises issues related to data security and privacy. Furthermore, while deep learning improves the recognition rate of modulation signals, the risk of malicious attacks on the central server or equipment malfunctions remains, potentially leading to information tampering or network paralysis.

[0038] In light of the above considerations, one or more embodiments of this application propose a training method and apparatus for a signal modulation recognition model. The signal modulation recognition model is updated locally on the device, and the global model update is based solely on a weighted aggregation of locally uploaded parameters, ensuring data security and privacy. Simultaneously, the decentralized blockchain-federated learning framework effectively prevents data tampering or network paralysis due to device or central server failures.

[0039] The following describes one or more embodiments of this application in detail with reference to the accompanying drawings.

[0040] Figure 1 This is a schematic diagram of the blockchain-federated learning framework according to an embodiment of this application. The blockchain-federated learning framework is divided into a blockchain layer and a computation layer. In the computation layer, there are many independent devices, called distributed computing nodes. Each node stores a copy of the global model, called a local model, and also stores its own private dataset, i.e., a local dataset. The local model is trained using the local dataset. Each computing node stores the parameters of its local model. In the blockchain layer, there are many block representatives. Each computing node in the computation layer is virtually connected to a block representative.

[0041] refer to Figure 2 The training method for a signal modulation recognition model according to one embodiment of this application includes the following steps:

[0042] Step S101: Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node.

[0043] In this step, each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to obtain the updated local parameters.

[0044] In some embodiments, node D i After receiving the global parameters from the (l-1)th epoch, the model can be updated using the minimization optimization function. Then, the local parameters in the m-th iteration of the l-th epoch... It can be represented as:

[0045]

[0046] Where, f(ω) i ,S i ω is the objective function of the local model, which is to minimize the loss function. i It is node D i Local parameters, These are the local parameters of the (m-1)th iteration in the l-th period. It is the m-th node D i The model updates are based on a locally stored signal dataset, where γ is the learning rate. Each data sample in this dataset can be represented as s. k =(x k ,y k ), where x k It is a high-dimensional vector, y k It is a scalar, representing x k Category tags.

[0047] After training is complete, the computing node sends the updated local parameters to one of the block representatives that is communicatively connected to the computing node.

[0048] In some embodiments, the connection between the computing node and the block representative is fixed. When the computing node or the block representative fails, the connection can be reset and re-established.

[0049] Step S102: The block representative packages the received local parameters into a first transaction and stores it in the local first transaction pool, and broadcasts the first transaction in the blockchain layer.

[0050] In this step, the block representative communicating with the computing node can package the received local parameters, store them in its own transaction pool, and broadcast the transaction in the blockchain layer.

[0051] In some embodiments, a block represents M j Received from node D i After obtaining the local parameters, the local parameters and their basic information can be packaged into a transaction, which can then be broadcast at the blockchain layer. The transaction information includes the local parameters and their basic information. The basic information of the local parameters includes the node number that generated the parameters, the start time, the duration, and the block representation M. j Numbers, etc.

[0052] Step S103: Upon receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives stores the first transaction into its local second transaction pool in response to determining that the first transaction has passed validity verification.

[0053] In this step, each of the other block representatives among the plurality of block representatives verifies the validity of the transaction after receiving the broadcast transaction, and in response to the transaction passing the validity verification, stores the transaction in its local transaction pool.

[0054] There are many methods for validity verification, such as hash algorithms and asymmetric encryption algorithms. As long as different validity verification methods achieve the corresponding purpose, their differences will not affect the scope of protection of this application.

[0055] In some embodiments, the validity verification utilizes a hash algorithm and an asymmetric encryption algorithm. A hash algorithm can map a binary value of arbitrary length to a shorter, fixed-length binary value, i.e., a hash value. Simultaneously, the calculation method of the hash algorithm ensures that when transaction information is maliciously tampered with, the tampered information cannot be recalculated to obtain the original hash value, thus protecting data security through subsequent verification. Using a hash algorithm, a first hash value is calculated based on the transaction information. Then, the hash value is encrypted using a private key to generate signature information. Upon receiving the signature information and transaction information, each of the other block representatives first decrypts the signature information using a public key to verify whether the signature information belongs to that block representative. After the signature information verification passes, a second hash operation is performed based on the transaction information to obtain a second hash value. The validity verification passes if the first hash value is equal to the second hash value; it fails if the first hash value is not equal to the second hash value. If the validity verification passes, the transaction information is stored in the local transaction pool; if it fails, the transaction is discarded.

[0056] Step S104: In response to the determination that the second preset condition is met, the block representative sends all transactions stored in the local first transaction pool to the computing node.

[0057] In this step, as long as the second preset condition is met, the block representative can send all transactions stored in the local first transaction pool to the computing node connected to the communication, without waiting for the parameter information of all computing nodes to be stored before performing this work.

[0058] In implementing this disclosure, the inventors discovered that the training and generation of local model parameters on all computing nodes are asynchronous. When some computing nodes take too long to train or generate local model parameters, the utilization value of the generated parameters gradually decreases over time. To minimize this situation, a second preset condition is set. When the second preset condition is met, transaction information can be sent to the connected computing nodes.

[0059] In some embodiments, the second preset condition is that the transaction pool memory reaches a certain limit or a preset sending time is reached, in which case all transaction information in the transaction pool is sent to the computing node connected to the communication link.

[0060] Step S105: The computing node uses a preset algorithm to evaluate the validity of all received transactions and perform weighted aggregation to obtain a first global parameter, and then sends the first global parameter to the block representative.

[0061] In this step, the received transaction information is evaluated for validity and weighted to obtain a first global parameter, which is then sent to the block representative of the communication connection.

[0062] In realizing this disclosure, the inventors discovered that by establishing a value assessment mechanism, the influence of malicious nodes can be effectively weakened, attacks can be resisted, and the accuracy of parameters can be improved, thereby increasing the model recognition rate.

[0063] There are many methods for validity assessment, such as the objective weighting method and Newton's law of cooling. As long as different validity assessment methods achieve their respective purposes, the different methods will not affect the scope of protection of this application.

[0064] In some embodiments, effectiveness is evaluated using methods such as objective weighting and Newton's law of cooling. Let the set of all transaction information in the transaction be denoted as . Where N T =|T|, N T This represents the total number of transactions. Also, let node D be... i The score in the (l-1)th round of aggregation is Then transaction T j The distributed computing node D corresponding to the parameters in the data. j Number of datasets N j Accuracy A of the local model j Historical score and the local parameter ω j With all the other N transactions mentioned T -1 local parameter ω t (t∈{1,2,…,N T The normalized average correlation of},t≠j) is r j .

[0065] Normalized average correlation r j The calculation process is as follows:

[0066] set up Normalize r: r jt ′=(r jt -min(r)) / (max(r)-min(r));

[0067] Then calculate the average: r j =∑r jt ′ / (N T -1).

[0068] In the following formula, using x ij (i = 1, 2, 3, 4, j = 1, 2, ..., N) T ) represents N jA j , and r j ,Right now

[0069] Then x ij The proportion can be expressed as:

[0070]

[0071] Corresponding average

[0072] Then the standard deviation SD of the i-th indicator i for:

[0073]

[0074] Then, use the correlation coefficient R i Let represent the conflict of the i-th indicator. Based on the Pearson correlation coefficient, we can obtain...

[0075]

[0076] Where, r ik This indicates the degree of correlation between different indicators.

[0077] but

[0078]

[0079] Then based on the standard deviation SD i and correlation coefficient R i Calculate the information content C of the i-th indicator i

[0080]

[0081] Therefore, the objective weight W of the i-th indicator i for

[0082]

[0083] Trading T j The score of the medium parameter can be expressed as

[0084]

[0085] Since the time it takes for each computing node to upload the model's local parameters is uncertain, excessively long local training times can lead to outdated local parameters, reducing the accuracy of the global model. Simultaneously, the score percentage of the older global model also needs to decrease over time. Therefore, Newton's law of cooling is used to adjust the parameter scores of each group during global model aggregation. (Transaction T) jThe included parameter ω j The timeliness at time t can be expressed as:

[0086]

[0087] Assumption For global parameter ω l-1 The total score in round l-1, t l-1 If the upload time is the time when the previous block was uploaded, then

[0088]

[0089] set up For trading T j The parameter ω included j The training start time. Then the parameter ω... j The score should be

[0090]

[0091] Therefore, we can obtain the global parameter update formula.

[0092]

[0093] Get ω l After that, D i ω l Score C l-1 C l C uploaded to this block represents M j M j Pack all transactions, the new global model, and scoring information into a block.

[0094] Step S106: One of the multiple block representatives generates a new block based on the received first global parameters, updates the blockchain, and broadcasts the new block in the blockchain layer.

[0095] In some embodiments, the block representative solves a SHA256-based cryptographic puzzle using a Proof-of-Work (PoW) algorithm. SHA256 is a one-way hash function used for block generation. It has a difficulty metric to determine the range of hash values. The block representative increments a random number (nonce) to calculate a hash value that meets the criteria. When the calculated hash value is within the range, the block representative generates a block. First, block representatives with matching random numbers (nonce) generate new blocks and update the blockchain. Then, the new block is broadcast at the blockchain layer.

[0096] Other block representatives use the new blocks they receive to verify their validity.

[0097] There are many methods for validity verification, such as hash algorithms and asymmetric encryption algorithms. As long as different validity verification methods achieve the corresponding purpose, their differences will not affect the scope of protection of this application.

[0098] In some embodiments, hash algorithms and asymmetric encryption algorithms are used to verify the validity of blocks. The verification process includes verifying the transaction information in the block and verifying the received first global model parameters. In response to determining that the block has passed verification, the new block is updated on the local blockchain and broadcast; in response to determining that the block has failed verification, the block is discarded.

[0099] In some embodiments, to avoid a fork situation arising from multiple block representatives simultaneously calculating and broadcasting new blocks, after a block representative receives the first new block, it determines whether a second new block has been received within a preset time. If it determines that a second new block has been received, the first new block is discarded and a fork signal is returned to the block representative who broadcast the first new block; if it determines that no new block has been received, the first new block is updated in the local blockchain.

[0100] Step S107: In response to determining that the blockchain layer has been updated, the computing node downloads the second global parameter from the new block from the block representative as a local parameter of the signal modulation recognition model stored locally.

[0101] In this step, after the local blockchain represented by all blocks in the blockchain layer is updated, the computing node downloads the second global parameter from the block representative of the new block from the communication connection, as the local parameter of the signal modulation recognition model stored locally.

[0102] refer to Figure 3 As an optional embodiment, the training method of this signal modulation recognition model, applied to the computing node, includes the following steps:

[0103] Step S201: Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node.

[0104] Let the computing node be D. i S i It is node D i The local dataset. Compute node D i Based on the local dataset S i The local model parameters ω of the trained signal modulation recognition model iSend to the multiple block representatives and the computing node D i The block representation of the communication connection is M. j .

[0105] Step S202: Receive all transactions packaged by the block representative according to the local parameters, and use a preset algorithm to evaluate the validity and weighted aggregate the received transactions to obtain the first global parameter, and send the first global parameter to the block representative.

[0106] Receive block representative M j All the transactions packaged together, compute node D i Based on the parameters in the transaction, the parameters contained in each transaction information are calculated using an objective weighting method and Newton's law of cooling to obtain the scores and weights of the parameters. The global parameters are then updated based on these scores and weights to obtain the global parameter ω. l And send the first global parameter to the block representative M j .

[0107] Step S203: In response to determining that the block representative updates the blockchain according to the first global parameter, download the second global parameter from the new block from the block representative as the local parameter of the signal modulation recognition model stored locally.

[0108] In response to determining that the blockchain layer has been updated, from the block representative M j Download the global parameters as local parameters for the signal modulation recognition model stored locally.

[0109] refer to Figure 4 As an optional embodiment, the training method of this signal modulation recognition model, applied to the block representative end, includes the following steps:

[0110] Step S301: Receive local parameters uploaded by one of the multiple computer nodes that is connected to the block representative, package them into a first transaction, store them in the local first transaction pool, and broadcast the first transaction in the blockchain layer.

[0111] Let the block represent M j Receives the computing node D that it communicates with. i The first transaction, packaged with uploaded local parameters, is stored in the local first transaction pool and broadcast on the blockchain layer.

[0112] Step S302: After receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives stores the first transaction into its local second transaction pool in response to determining that the first transaction has passed the validity verification.

[0113] Block representative M j First, the hash value H of the transaction information will be calculated. j Then, it encrypts the hash value using its private key to generate a signature, and subsequently broadcasts the signature and transaction information. When block representative M... i Received block representative M j When broadcasting transactions, first use M j The public key is used to decrypt the signature information and verify whether the signature information belongs to block representative M. j After that, block representative M i Perform a hash operation on the transaction information to obtain the hash value H. i Verify H i and H j Are they equal? ​​This is in response to determining H. i equals H j Then it passes verification; in response to determining H i Not equal to H j If not, the verification will fail.

[0114] Step S303: In response to determining that the second preset condition is met, the block representative sends all transactions stored in the local first transaction pool to the computing node.

[0115] When the block represents M j When the transaction pool reaches its memory limit or a preset time is reached, all transactions stored in the local first transaction pool are sent to the computing node.

[0116] Step S304: One of the multiple block representatives generates a new block based on the received first global parameter, updates the blockchain, and broadcasts the new block in the blockchain layer.

[0117] When generating a new block, all block representatives solve a cryptographic puzzle based on SHA256 using the Proof-of-Work (PoW) algorithm. SHA256 is a one-way hash function used for block generation, with a difficulty index to determine the range of hash values. The block representative increments a random number (nonce) to calculate a hash value that meets the conditions. When the calculated hash value is within the range, the block representative generates a block. Let the block representative be M. u First, find nonces that meet the criteria and generate a new block. M u Broadcast the new block to other block representatives.

[0118] Step S305: Each of the other block representatives among the plurality of block representatives updates the blockchain in response to determining that the new block has passed the validity verification.

[0119] When the block represents M uWhen other block representatives receive the new block, they also verify the new block using validity verification. If the verification passes, the new block is recorded on the local blockchain; if the verification fails, the new block is discarded.

[0120] It should be noted that the method in this embodiment can be executed by a single device, such as a computer or server. The method can also be applied in a distributed scenario, where multiple devices cooperate to complete the task. In such a distributed scenario, one of these devices may execute only one or more steps of the method in this embodiment, and the multiple devices will interact with each other to complete the method described.

[0121] It should be noted that the above description describes some embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in a different order than that shown in the above embodiments and still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0122] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application (including the claims) is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of different aspects of the embodiments of this application as described above, which are not provided in the details for the sake of brevity.

[0123] Additionally, to simplify the description and discussion, and to avoid obscuring the embodiments of this application, the well-known power / ground connections to integrated circuit (IC) chips and other components may or may not be shown in the provided drawings. Furthermore, the apparatus may be shown in block diagram form to avoid obscuring the embodiments of this application, and this also takes into account the fact that the details of the implementation of these block diagram apparatuses are highly dependent on the platform on which the embodiments of this application will be implemented (i.e., these details should be fully understood by those skilled in the art). While specific details (e.g., circuits) have been set forth to describe exemplary embodiments of this application, it will be apparent to those skilled in the art that the embodiments of this application can be implemented without these specific details or with variations thereof. Therefore, these descriptions should be considered illustrative rather than restrictive.

[0124] Although this application has been described in conjunction with specific embodiments thereof, many substitutions, modifications, and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may be used with the embodiments discussed.

[0125] The embodiments of this application are intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this application should be included within the protection scope of this application.

Claims

1. A training method for a signal modulation recognition model, the method being implemented using a blockchain-federated learning framework comprising a blockchain layer and a computation layer, wherein multiple distributed computing nodes in the computation layer are respectively communicatively connected to multiple block representatives in the blockchain layer, the method comprising iteratively executing the following operations until the signal modulation recognition model satisfies a first preset condition: Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node. The block representative packages the received local parameters into a first transaction and stores it in the local first transaction pool, and broadcasts the first transaction in the blockchain layer; Upon receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives, in response to determining that the first transaction has passed validity verification, stores the first transaction in its local second transaction pool. The block represents the response to determining that the second preset condition is met, and sends all transactions stored in the local first transaction pool to the computing node. The computing node uses a preset algorithm to evaluate the validity of all received transactions and perform weighted aggregation to obtain a first global parameter, and then sends the first global parameter to the block representative. This block represents the generation of a new block based on the received first global parameters, and the broadcast of the new block in the blockchain layer; In response to determining that the blockchain layer has been updated, the computing node downloads the second global parameter from the new block from the block representative as a local parameter of the signal modulation recognition model stored locally. The algorithm for the first global parameter is expressed as follows: , This represents the total number of transactions. This represents the information content of the i-th indicator.

2. The method according to claim 1, characterized in that, The preset algorithm includes an objective weighting method and Newton's law of cooling for determining the timeliness of the local parameters.

3. The method according to claim 1 or 2, characterized in that, The validity verification of the first transaction is performed using a hash algorithm and an asymmetric encryption algorithm.

4. The method according to claim 1 or 2, characterized in that, The first preset condition includes: the signal modulation recognition model characterized by the first global parameter converges or reaches a predetermined recognition accuracy.

5. The method according to claim 1 or 2, characterized in that, The second preset condition includes: the first transaction pool is full, or the predetermined block generation time has arrived.

6. The method according to claim 1 or 2, characterized in that, The operation also includes: Each of the other block representatives performs block verification on the new block after receiving the broadcast of the new block.

7. The method according to claim 6, characterized in that, Block verification of the new block includes: The new block is verified using a hash algorithm and an asymmetric encryption algorithm.

8. The method according to claim 6, characterized in that, The operation also includes: In response to the determination that the new block has passed the block verification, the other block represents an update to the blockchain layer.

9. A training apparatus for a signal modulation recognition model, the apparatus comprising a blockchain layer and a computation layer forming a blockchain-federated learning framework, wherein, Multiple distributed computing nodes in the computing layer are respectively connected to multiple block representatives in the blockchain layer. The computing nodes and the block representatives are configured to train the signal modulation recognition model by iteratively performing the following operations until the signal modulation recognition model meets a first preset condition: Each computing node uses the locally stored signal dataset to train the locally stored signal modulation recognition model to update the local parameters of the signal modulation recognition model, and sends the updated local parameters to one of the multiple block representatives that is communicatively connected to the computing node. The block representative packages the received local parameters into a first transaction and stores it in the local first transaction pool, and broadcasts the first transaction in the blockchain layer; Upon receiving the first transaction broadcast, each of the other block representatives among the plurality of block representatives, in response to determining that the first transaction has passed validity verification, stores the first transaction in its local second transaction pool. The block represents the response to determining that the second preset condition is met, and sends all transactions stored in the local first transaction pool to the computing node. The computing node uses a preset algorithm to evaluate the validity of all received transactions and perform weighted aggregation to obtain a first global parameter, and then sends the first global parameter to the block representative. This block represents the generation of a new block based on the received first global parameters, and the broadcast of the new block in the blockchain layer; In response to determining that the blockchain layer has been updated, the computing node downloads the second global parameter from the new block from the block representative as a local parameter of the signal modulation recognition model stored locally. The algorithm for the first global parameter is expressed as follows: , This represents the total number of transactions. This represents the information content of the i-th indicator.

10. The apparatus according to claim 9, characterized in that, The preset algorithm includes an objective weighting method and Newton's law of cooling for determining the timeliness of the local parameters.

Images