Transmission method for uplink multi-cell MIMO semantic communication system with distributed architecture

Abstract

The present invention belongs to the field of wireless communications, and more specifically, is a method for transmitting a distributed uplink multi-cell MIMO semantic communication system, comprising the following steps: (1) allocating a semantic knowledge base to users and base stations in the cell as needed according to the cell; (2) constructing and designing an uplink multi-cell MIMO semantic communication system framework to realize the transmission of semantic-level information; (3) constructing and designing an interference processing module to solve semantic-level interference in a multi-cell scenario; (4) connecting the output of the interference processing module to the uplink multi-cell MIMO semantic communication system framework to realize the separation and extraction of semantic information in the channel; (5) dynamically adjusting the design of the receiver to retrain the network according to the difference in semantic information transmitted by different users; and (6) designing a distributed training framework to share the storage load of the base station training model and improve the reliability of the system.

Description

Technical Field

[0001] The invention belongs to the field of wireless communication, and in particular, is a transmission method for an uplink multi-cell MIMO semantic communication system with a distributed architecture. Background Art

[0002] In recent years, driven by deep learning, natural language processing has achieved great success in analyzing and understanding large amounts of language text. Based on this, people have broken through the communication model of only transmitting text but not conveying the meaning, and developed an end-to-end semantic communication system based on deep learning, which can process data at the semantic level, extract the meaning of the data, filter out useless, irrelevant, and unimportant information, and further compress the data while retaining the meaning, greatly improving the performance of the system. The system is more robust under low signal-to-noise ratio conditions and has advantages in processing large batches of data.

[0003] At present, there have been some studies on semantic communication, including resource allocation and semantic information transmission for semantic communication. However, these works are not perfect in practical application. Existing work can only extract the semantic information behind the sentence, compress and transmit the extracted semantic information, and realize end-to-end semantic communication. Because the actual communication environment is often more complex, involving multiple cells and multiple users, end-to-end communication has great limitations. However, there is no research on multi-cell MIMO semantic communication system. In the multi-cell MIMO communication environment, since the cells are not independent, interference between adjacent cells is inevitable, and the end-to-end semantic communication system cannot solve this problem. At the same time, considering that the meanings of the same words in different disciplines are very different, how to accurately realize the transmission of messages with different semantics in the same cell is also a problem that the end-to-end semantic communication system cannot solve.

[0004] The essence of semantic communication system is composed of neural networks, and neural networks need a large amount of semantic knowledge base to support the model in the training phase to make the model more accurate, which makes the training phase take a lot of time and requires the base station to have sufficient computing power. However, there is currently no solution to the insufficient computing power of base stations in semantic communication scenarios. Therefore, in the multi-cell MIMO communication scenario, how to design a multi-cell MIMO semantic system model with excellent performance and small base station load is a challenging problem. Summary of the invention

[0005] Aiming at the current traditional communication mode of only transmitting text but not conveying the meaning, the present invention dynamically adjusts the design of the decoder and retrains the network according to the difference in semantic information transmitted by different users, and provides a distributed architecture uplink multi-cell MIMO semantic communication system transmission method, which can realize the transmission of semantic level information in a multi-cell MIMO scenario, solve the transmission errors in different semantic environments, share the storage load of the base station training model, and improve the reliability of the system.

[0006] The technical solution adopted by the present invention to solve the above technical problems is as follows:

[0007] A distributed architecture uplink multi-cell MIMO semantic communication system transmission method, the main steps of the method are as follows:

[0008] Step 1: distribute the semantic knowledge base to users and base stations in the cell as needed according to different cells;

[0009] Step 2: Build and design an uplink multi-cell MIMO semantic communication system framework to achieve the transmission of semantic level information;

[0010] Step 3: Build and design an interference processing module to solve semantic interference in multi-cell scenarios;

[0011] Step 4: Connect the output of the interference processing module to the uplink multi-cell MIMO semantic communication system framework to separate and extract semantic information in the channel;

[0012] Step 5: Dynamically adjust the design of the decoder and retrain the network according to the difference in semantic information transmitted by different users;

[0013] Step 6: Design a distributed training architecture to share the storage load of the base station training model and improve the reliability of the system.

[0014] In the above technical solution, step 1, the semantic knowledge base is allocated to users and base stations in the cell as needed according to the different cells. The uplink multi-cell MIMO system includes K cells, each of which has a multi-antenna base station and N single-antenna users. The cell set and the user set in cell k are defined as α = {1, 2, 3, ..., K} and β = {1, 2, 3, ..., N}, respectively, where the base station and the user have the same semantic knowledge base defined as It is defined that the text information transmitted by users in a cell is composed of words, and the information transmitted is different according to the cell, where the information transmitted by the i-th user in the k-th cell is s i =[w 1 , w 2 , w 3 , ..., w L ], where w lrepresents the lth word transmitted by the i-th user.

[0015] Step 2: Build and design an uplink multi-cell MIMO semantic communication system framework to achieve semantic information transmission, using the multi-cell MIMO user side as the transmitter and the base station in the current cell as the receiver to transmit messages. Each user, as an independent individual, has its own encoder consisting of a semantic encoder and a three-layer DNN network. The semantic encoder is used to extract the semantic information behind the message, and the DNN network is used to compress the information; the base station is equipped with a decoder consisting of a three-layer DNN network and a semantic decoder to decode and receive the semantic information transmitted in this cell; the transmission information of the user in the cell is s i As the input of the encoder, the output of the encoder x i It is expressed as:

[0016] x i =C(S(s i ; α i );β i )

[0017] Among them, S(;α i ) is the learning rate of the i-th user α i Semantic encoder of C(β i ) is the learning rate of the i-th user β i DNN network.

[0018] The input signal of the decoder is denoted as The output of the decoder is expressed as:

[0019]

[0020] Among them, S -1 (; γ i ) is the learning rate of the i-th user γ i DNN network; is the learning rate of the i-th user Semantic decoder.

[0021] In order to maximize the system transmission performance, the mean square error is used as the loss function to train the model, which is expressed as:

[0022]

[0023] where s i represents the sentence sent by the i-th user in the k-th cell, It represents the sentence sent by the i-th user received by the base station in the k-th cell.

[0024] Step 3: Build and design an interference processing module to solve the interference at the semantic level in multi-cell scenarios. In multi-cell MIMO scenarios, each cell does not exist independently, and the interference signals between adjacent cells are not useful signals to be received. Therefore, the known transmission signals at the base station receiving end are processed as interference signals of other cells; the interference signals are located and encoded, and the semantic relevance of the signals is extracted using the multi-head attention mechanism. The signals are passed into a three-layer DNN network to achieve semantic compression of the signals, and the interference signals are recorded as s in , then the processed signal x in It is expressed as:

[0025] x in =I(s in ;μ in ) (2)

[0026] Among them, I(;μ in ) is the learning rate μ in Interference processing module.

[0027] Step 4: Connect the output of the interference processing module to the uplink multi-cell MIMO semantic communication system framework to separate and extract semantic information in the channel. When the transmitted signal passes through a MIMO physical channel, the received signal Y is expressed as:

[0028] Y=HX all +N=H(X+X in )+N (3)

[0029] Among them, X is the information transmitted by users in the cell, X in is the received interference signal.

[0030] The received signal Y is processed by MMSE to recover the transmission signal It is expressed as:

[0031]

[0032] Here, H follows a complex Gaussian distribution with a mean of 0 and a variance of 1 / 2.

[0033] Step 5. Dynamically adjust the receiver design to retrain the network based on the difference in semantic information transmitted by different users. Based on the fact that users in the same cell have different semantic knowledge bases, another decoder is designed. This module consists of a three-layer DNN network and multiple semantic decoders. Different semantic decoders are provided for users with different semantic knowledge bases to extract semantic information. The network parameters of the uplink multi-cell MIMO semantic communication system model are updated by minimizing the loss function (1).

[0034] Step 6: Design a distributed training framework to share the storage load of the base station training model. The data transmitted by the user is processed in batches. Only one batch of data is used in each round of communication. A large server is added to the distributed training framework. The server will blend the network parameters of all base stations trained in the multi-cell MIMO scenario:

[0035]

[0036] Among them, ω t+1 represents the network parameters processed by the server, t represents the tth round of communication, k represents the kth base station, and ω after processing by the server t+1 It will be sent to each base station as the network parameter for the next communication.

[0037] The beneficial effects of the present invention are as follows: the present invention is applicable to multi-cell MIMO communication scenarios, and completes the transmission of user data by using a multi-cell semantic communication system model based on a distributed architecture, with the goal of maximizing system transmission, which can effectively improve system performance, accelerate the convergence speed of the neural network, and improve the reliability of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a schematic diagram of an uplink multi-cell MIMO semantic communication system in an embodiment of the present invention.

[0039] Figure 2 The present invention is a flowchart of a transmission method of an uplink multi-cell MIMO semantic communication system based on a distributed architecture in an embodiment of the present invention.

[0040] Figure 3 This is a model diagram of an uplink multi-cell MIMO semantic communication system based on a distributed architecture in an embodiment of the present invention.

[0041] Figure 4 Schematic diagram of BLUE score of a user in a Rayleigh channel in an embodiment of the present invention.

[0042] Figure 5 Schematic diagram of semantic similarity scores of users in a Rayleigh channel in an embodiment of the present invention.

[0043] Specific implementation

[0044] In order to deepen the understanding of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. The embodiments are only used to explain the present invention and do not limit the protection scope of the present invention.

[0045] A distributed architecture uplink multi-cell MIMO semantic communication system transmission method, the method comprising the following steps:

[0046] Step 1: The specific method of allocating the semantic knowledge base to users and base stations in the cell as needed according to the different cells is as follows: In the present invention, the uplink multi-cell MIMO system includes K cells, each of which has a multi-antenna base station and N single-antenna users. We define the cell set and the user set in cell k as α = {1, 2, 3, ..., K) and β = {1, 2, 3, ..., N}, respectively, where the base station and the user have the same semantic knowledge base defined as Each cell has a different semantic knowledge base. We define that the text information transmitted by users in a cell is composed of words, and the information transmitted by the i-th user in the k-th cell is s i =[w 1 , w 2 , w 3 , ..., w L ], where w l represents the lth word transmitted by the i-th user.

[0047] Step two, a transmission method for an uplink multi-cell MIMO semantic communication system based on a distributed framework is designed. The specific method for realizing semantic-level information transmission through the uplink multi-cell MIMO semantic communication system framework is as follows: In the present invention, we use the multi-cell MIMO user side as the transmitter, and the base station in the current cell as the receiver and the message transmission. Each user has its own encoder as an independent individual, and the base station is equipped with a decoder. The encoder here is composed of a semantic encoder and a three-layer DNN network. The semantic encoder is used to extract the semantic information behind the message, and the DNN network is used to compress the information; the decoder is composed of a three-layer DNN network and a semantic decoder, which is used to realize the decoding and reception of the semantic information transmitted in this cell. We convert the transmission information of the users in the cell s i As the input of the encoder, the output of the encoder x i It is expressed as:

[0048] x i =C(S(s i ; α i );β i )

[0049] Among them, S(;α i ) is the learning rate of the i-th user α i Semantic encoder of C(β i ) is the learning rate of the i-th user β i The input signal of the decoder is recorded as The output of the decoder is expressed as:

[0050]

[0051] Among them, S -1 (; γ i ) is the learning rate of the i-th user γ i DNN network; is the learning rate of the i-th user Since the goal of the present invention is to maximize the system transmission performance, in the present invention, we use the mean square error as the loss function to train the model, which is expressed as:

[0052]

[0053] where s i represents the sentence sent by the i-th user in the k-th cell, It represents the sentence sent by the i-th user received by the base station in the k-th cell.

[0054] Step three, build and design an interference processing module to solve the semantic interference in the multi-cell scenario. The specific method is as follows: In the multi-cell MIMO scenario, since each cell does not exist independently, interference between adjacent cells is inevitable, so we designed an interference processing module to solve this problem. However, the interference signal is not the useful signal we want to receive, so we treat the known transmission signal at the base station receiving end as the interference signal of other cells. This module first locates and encodes the interference signal, then uses the multi-head attention mechanism to extract the semantic relevance of the signal, and finally passes it into a three-layer DNN network to achieve semantic compression of the signal. We denote the interference signal as s in , then the processed signal x in It is expressed as:

[0055] x in =I(s in ;μ in )

[0056] Among them, I(;μ in ) is the learning rate μ in Interference processing module.

[0057] Step 4: Connect the output of the interference processing module to the uplink multi-cell MIMO semantic communication system framework. The specific method for separating and extracting semantic information in the channel is as follows: After the transmitted signal passes through a MIMO physical channel, the received signal Y is expressed as:

[0058] Y=HX all +N=H(X+X in )+N

[0059] Among them, X is the information transmitted by users in the cell, X in is the received interference signal. Then, the received signal Y is processed by MMSE to recover the transmission signal It is expressed as:

[0060]

[0061] Here, H follows a complex Gaussian distribution with a mean of 0 and a variance of 1 / 2.

[0062] Step 5, dynamically adjust the receiver design according to the difference in semantic information transmitted by different users. The specific method of retraining the network is as follows: Step 1 assumes that in a multi-cell MIMO scenario, users in the same cell share a common semantic knowledge base, but in reality, users in the same cell may have different semantic knowledge bases. In view of this situation, the present invention designs another decoder, which consists of a three-layer DNN network and multiple semantic decoders, and provides different semantic decoders for users with different semantic knowledge bases to extract semantic information. The network parameters of the uplink multi-cell MIMO semantic communication system model are updated by minimizing the loss function (1).

[0063] Step 6: Design a distributed training framework to share the storage load of the base station training model. The specific method is as follows: First, we process the data transmitted by the user in batches, and only use one batch of data in each round of communication. Secondly, a large server is added to the distributed training framework. The server will blend the network parameters of all base stations in the multi-cell MIMO scenario after each training:

[0064]

[0065] Among them, ω t+1 represents the network parameters processed by the server, t represents the tth round of communication, and k represents the kth base station. t+1 It will be sent to each base station as the network parameters for the next communication. The addition of a distributed training framework reduces the complexity of each base station training. By processing the network parameters uploaded by the base station, the training time can be reduced and the reliability of the system can be improved.

[0066] In the above-mentioned embodiment steps, simulation under Rayleigh channel is performed to illustrate the effect of the present invention. The simulation results are as follows: Figure 4-Figure 5 Here, the method proposed by the present invention is represented by .

[0067] As the simulation results Figure 4 , which compares the changes in BLUE scores under the Rayleigh channel under the base station single training and distributed training framework. Figure 4It can be seen that the system performance after adding the distributed training framework is almost the same as the performance of base station training alone. The addition of the distributed training framework not only ensures the performance of the system, but also reduces the training load of the base station and improves the reliability of the system.

[0068] As the simulation results Figure 5 , which compares the changes in semantic similarity scores under the Rayleigh channel between base station single training and distributed training framework. Figure 5 It can be seen that the system performance after adding the distributed training framework and the base station individual training can retain the semantic information in the transmitted data to a great extent. The addition of the distributed training framework not only guarantees the performance of the system, but also reduces the training load of the base station and improves the reliability of the system.

[0069] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods. These changes and improvements should fall within the scope of the present invention to be protected.

Claims

1. A distributed architecture uplink multi-cell MIMO semantic communication system transmission method, characterized in that: The following steps are involved: Step 1: distribute the semantic knowledge base to users and base stations in the cell as needed according to the differences of uplink multi-cells in the distributed architecture; Step 2: Build and design an uplink multi-cell MIMO semantic communication system framework to realize the transmission of semantic-level information. Use the multi-cell MIMO user side as the transmitter and the base station in the current cell as the receiver to transmit messages. Each user, as an independent individual, has its own encoder consisting of a semantic encoder and a three-layer DNN network. The semantic encoder is used to extract the semantic information behind the message, and the DNN network is used to compress the information. The base station is equipped with a decoder consisting of a three-layer DNN network and a semantic decoder to decode and receive the semantic information transmitted in this cell, and convert the transmission information of users in the cell into s i As the input of the encoder, the output of the encoder x i It is expressed as: x i =C(S(s i ;a i );b i ) Among them, S(;α i ) is the learning rate of the i-th user α i Semantic encoder of C(β i ) is the learning rate of the i-th user β i DNN network, The input signal of the decoder is denoted as The output of the decoder is expressed as: Among them, S -1 (; γ i ) is the learning rate of the i-th user γ i DNN network; is the learning rate of the i-th user The semantic decoder of In order to maximize the system transmission performance, the mean square error is used as the loss function to train the model, which is expressed as: Among them, s i represents the sentence sent by the i-th user in the k-th cell, represents the sentence sent by the i-th user received by the base station in the k-th cell; Step 3: Build and design an interference processing module to solve semantic interference in multi-cell scenarios; Step 4: Connect the output of the interference processing module to the uplink multi-cell MIMO semantic communication system framework to separate and extract semantic information in the channel; Step 5: Dynamically adjust the design of the decoder and retrain the network according to the difference in semantic information transmitted by different users; Step 6: Design a distributed training architecture to share the storage load of the base station training model.

2. The distributed architecture uplink multi-cell MIMO semantic communication system transmission method according to claim 1, characterized in that: In step 1, the uplink multi-cell of the distributed architecture includes K cells, each of which has a multi-antenna base station and N single-antenna users. The cell set and the user set in cell k are defined as α = {1, 2, 3, ..., K} and β = {1, 2, 3, ..., N}, respectively, where the base station and the user have the same semantic knowledge base defined as And each community has a different semantic knowledge base.

3. The transmission method of a distributed uplink multi-cell MIMO semantic communication system according to claim 2, characterized in that: In the step 1, it is defined that the text information transmitted by users in the uplink multi-cell of the distributed architecture is composed of words, and the information transmitted is different according to the cells, where the information transmitted by the i-th user in the k-th cell is s i =[w1,w2,w3,…,w L ], where w l represents the lth word transmitted by the i-th user.

4. The distributed architecture uplink multi-cell MIMO semantic communication system transmission method according to claim 1, characterized in that: In the step 2, the specific method for realizing semantic-level information transmission by building and designing an uplink multi-cell MIMO semantic communication system framework is as follows: multi-cell MIMO users are used as independent individuals with their own encoders consisting of a three-layer DNN network and a semantic decoder, and the base station is equipped with a decoder consisting of a semantic encoder and a three-layer DNN network to realize decoding and reception of semantic information transmitted in this cell.

5. The transmission method of the uplink multi-cell MIMO semantic communication system with a distributed architecture according to claim 4, characterized in that: In step 2, the mean square error is used as the loss function to train the model, which is expressed as: Among them, s i represents the sentence sent by the i-th user in the k-th cell, It represents the sentence sent by the i-th user received by the base station in the k-th cell.

6. The transmission method of a distributed architecture uplink multi-cell MIMO semantic communication system according to claim 1, characterized in that: In step 3, an interference processing module is built and designed to solve the interference at the semantic level in the multi-cell scenario, the interference signal is located and encoded, the semantic relevance of the signal is extracted using the multi-head attention mechanism, and a three-layer DNN network is passed to realize the compression of the signal at the semantic level. The interference signal is recorded as s in , then the processed signal x in It is expressed as: x in =I(s in ;m in ) (2) Among them, I(;μ in ) is the learning rate μ in Interference processing module.

7. The distributed architecture uplink multi-cell MIMO semantic communication system transmission method according to claim 1, characterized in that: In step 4, the output of the interference processing module is connected to the uplink multi-cell MIMO semantic communication system framework to achieve separation and extraction of semantic information in the channel. When the transmitted signal passes through a MIMO physical channel, the received signal Y is expressed as: Y=HX all +N=H(X+X in )+N (3) Among them, X is the information transmitted by users in the cell, X in The received interference signal Y is processed by MMSE to recover the transmission signal It is expressed as: Here, H follows a complex Gaussian distribution with a mean of 0 and a variance of 1 / 2.

8. The transmission method of a distributed architecture uplink multi-cell MIMO semantic communication system according to claim 5, characterized in that: In the step five, the receiver design is dynamically adjusted to retrain the network according to the differences in semantic information transmitted by different users. Another decoder consisting of a three-layer DNN network and multiple semantic decoders is designed. Different semantic decoders are provided for users with different semantic knowledge bases to extract semantic information. The network parameters of the uplink multi-cell MIMO semantic communication system model are updated by minimizing the loss function.

9. The transmission method of the distributed architecture uplink multi-cell MIMO semantic communication system according to claim 8, characterized in that: In step 6, a distributed training framework is designed to share the storage load of the base station training model. The specific method is as follows: first, the data transmitted by the user is processed in batches, and only one batch of data is used in each round of communication. Secondly, a large server is added to the distributed training framework, which will blend the network parameters of all base stations trained in the multi-cell MIMO scenario: Among them, ω t+1 represents the network parameters processed by the server, t represents the tth round of communication, k represents the kth base station, and ω after processing by the server t+1 It will be sent to each base station as the network parameters for the next communication. The addition of a distributed training architecture reduces the complexity of each base station training.

Images