A two-way relay intelligent communication method and device based on deep neural network

Abstract

The present invention discloses a two-way relay intelligent communication method and device based on a deep neural network, comprising: obtaining image data to be transmitted, and inputting the image data to be transmitted into an encoder of a preset deep neural network, thereby obtaining an image transmission signal; transmitting the image transmission signal to a two-way relay wireless channel, and communicating and receiving the image transmission signal output by the two-way relay wireless channel, and then inputting the output image transmission signal into a decoder of a preset deep neural network, thereby obtaining the transmitting end image data; wherein the preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set. The present invention solves the technical problems in the prior art of large one-way communication limitations, low source channel coding efficiency, and limited image processing quality.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a two-way relay intelligent communication method and device based on a deep neural network. Background Art

[0002] Modern communication systems use a two-stage coding process to transmit image / video data: first, the image / video data is compressed using a source coding algorithm to eliminate inherent redundancy and reduce the amount of information transmitted. The compressed bit stream is then channel-coded to combat noise in the wireless channel and then modulated. Shannon's separation theorem proves that these two steps are theoretically optimal in the asymptotic limit of infinitely long source and channel blocks using source and channel coding. However, many emerging applications, from the Internet of Things to autonomous driving, require image / video transmission under extreme latency, bandwidth, or power constraints, which precludes the use of computationally demanding source and channel coding techniques.

[0003] Currently, intelligent communication systems based on deep learning can treat the entire transmission system as a black box model and jointly optimize all components of the transmitter and receiver through neural networks, thereby achieving a globally optimal end-to-end intelligent communication system, including encoding, modulation, demodulation, and decoding processes. Compared with traditional communication systems and communication systems optimized by modules, this end-to-end intelligent communication system can more effectively adapt to unknown communication environments and nonlinearities introduced by communication equipment or channels. At the same time, due to the use of deep learning technology, this communication system can optimize communication algorithms and parameters quickly and cost-effectively.

[0004] However, existing deep learning-based methods have the limitation of one-way communication. Existing technologies generally focus on one-way communication, which limits the efficiency and flexibility of information exchange. In a one-way communication system, information is only transmitted from one source to the receiving end, and two-way or multi-directional information interaction cannot be achieved. This limitation becomes particularly evident in scenarios that require rapid feedback and dynamic adjustment of communication parameters. Secondly, the source channel coding efficiency is low. Traditional digital communication systems often do not have an efficient combination of source channel coding, resulting in low coding efficiency. This is particularly prominent in applications that require large amounts of data transmission, such as image transmission. Thirdly, the image processing quality is limited. Due to the lack of efficient coding and modulation and demodulation processing, existing technologies often cannot effectively improve image quality in image processing, especially when transmitting images, which are susceptible to signal interference and data loss. Summary of the Invention

[0005] The present invention provides a two-way relay intelligent communication method and device based on deep neural network to solve the technical problems in the prior art such as large limitations of one-way communication, low efficiency of source channel coding, and limited image processing quality.

[0006] In order to solve the above technical problems, an embodiment of the present invention provides a two-way relay intelligent communication method based on a deep neural network, comprising:

[0007] Acquire image data to be sent, and input the image data to be sent into an encoder of a preset deep neural network, thereby obtaining an image sending signal;

[0008] Transmitting the image transmission signal to a bidirectional relay wireless channel, communicating and receiving the image transmission signal output by the bidirectional relay wireless channel, and then inputting the output image transmission signal into a decoder of a preset deep neural network, thereby obtaining transmitting end image data;

[0009] The preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set.

[0010] As a preferred solution, the preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set, specifically including:

[0011] Build a two-way relay communication model and obtain a preset data set as the training image data to be sent;

[0012] Initializing an initial deep neural network and inputting the training image data into the preset deep neural network, so that after the training image data is processed by a terminal encoder and a relay encoder, the processed training image data is output to a bidirectional relay wireless channel set according to the bidirectional relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder;

[0013] After the bidirectional relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the relay decoder and the terminal decoder process and output the training image data at the transmitting end;

[0014] According to the difference between the obtained sending-end training image data and the training image data to be sent, the initial deep neural network is iteratively trained to obtain a preset deep neural network.

[0015] As a preferred solution, the two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the construction of the two-way relay communication model specifically includes:

[0016] Set the signal S received by the relay node R for:

[0017]

[0018] Set the first terminal node to receive the signal y from the relay node A for:

[0019]

[0020] Set the first terminal node to remove the self-interference of the received signal to obtain the signal

[0021]

[0022] Set the second terminal node to receive the signal y from the relay node B for:

[0023]

[0024] Set the second terminal node to remove the self-interference of the received signal to obtain the signal

[0025]

[0026] Among them, the signal received by the relay node is S R , the signal received by the encoder of the first terminal node is S A , the encoder of the second terminal node receives the signal S B , the output of the encoder of the first terminal node is x A , the output of the encoder of the second terminal node is x B , are the power of the signals sent by the first terminal node and the second terminal node respectively, Denote the channel coefficients between the first terminal node and the second terminal node and the relay node R, n AR 、n BR They follow complex Gaussian distribution respectively The channel noise, η is the amplification factor;

[0027] Thus, a two-way relay communication model is constructed.

[0028] As a preferred solution, the iterative training of the initial deep neural network based on the difference between the obtained sending end training image data and the training image data to be sent specifically includes:

[0029] Calculate the corresponding mean square error based on the difference between the obtained sending end training image data and the training image data to be sent;

[0030] According to the mean square error, the loss function is obtained: Among them, xi is the i-th training image data to be sent, is the training image data of the i-th transmitter, and N is the number of total image data;

[0031] The loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

[0032] Accordingly, the present invention also provides a two-way relay intelligent communication device based on a deep neural network, comprising: an acquisition module and an input module;

[0033] The acquisition module is used to acquire the image data to be sent and input the image data to be sent into an encoder of a preset deep neural network to obtain an image transmission signal;

[0034] The input module is used to transmit the image transmission signal to a two-way relay wireless channel, and communicate to receive the image transmission signal output by the two-way relay wireless channel, and then input the output image transmission signal into a decoder of a preset deep neural network, thereby obtaining the sending end image data; wherein, the preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set.

[0035] As a preferred solution, the preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set, specifically including:

[0036] Build a two-way relay communication model and obtain a preset data set as the training image data to be sent;

[0037] Initializing an initial deep neural network and inputting the training image data into the preset deep neural network, so that after the training image data is processed by a terminal encoder and a relay encoder, the processed training image data is output to a bidirectional relay wireless channel set according to the bidirectional relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder;

[0038] After the bidirectional relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the relay decoder and the terminal decoder process and output the training image data at the transmitting end;

[0039] According to the difference between the obtained sending-end training image data and the training image data to be sent, the initial deep neural network is iteratively trained to obtain a preset deep neural network.

[0040] As a preferred solution, the two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the construction of the two-way relay communication model specifically includes:

[0041] Set the signal S received by the relay node R for:

[0042]

[0043] Set the first terminal node to receive the signal y from the relay node A for:

[0044]

[0045] Set the first terminal node to remove the self-interference of the received signal to obtain the signal

[0046]

[0047] Set the second terminal node to receive the signal y from the relay node B for:

[0048]

[0049] Set the second terminal node to remove the self-interference of the received signal to obtain the signal

[0050]

[0051] Among them, the signal received by the relay node is S R , the signal received by the encoder of the first terminal node is S A , the encoder of the second terminal node receives the signal S B , the output of the encoder of the first terminal node is x A , the output of the encoder of the second terminal node is x B , are the power of the signals sent by the first terminal node and the second terminal node respectively, Denote the channel coefficients between the first terminal node and the second terminal node and the relay node R, n AR 、n BR They follow complex Gaussian distribution respectively The channel noise, η is the amplification factor;

[0052] Thus, a two-way relay communication model is constructed.

[0053] As a preferred solution, the iterative training of the initial deep neural network based on the difference between the obtained sending end training image data and the training image data to be sent specifically includes:

[0054] Calculate the corresponding mean square error based on the difference between the obtained sending end training image data and the training image data to be sent;

[0055] According to the mean square error, the loss function is obtained: Among them, x i is the i-th training image data to be sent, is the training image data of the i-th transmitter, and N is the number of total image data;

[0056] The loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

[0057] Correspondingly, the present invention also provides a terminal device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the two-way relay intelligent communication method based on deep neural network as described in any one of the above.

[0058] Correspondingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the two-way relay intelligent communication method based on deep neural network as described in any one of the above.

[0059] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0060] The technical solution of the present invention obtains image data to be sent, and inputs the image data into an encoder of a preset deep neural network trained by a preset data set to obtain an image sending signal, which is then transmitted to a two-way relay wireless channel to communicate and receive the image sending signal output by the two-way relay wireless channel, thereby inputting the output image sending signal into a decoder of a preset deep neural network trained by a preset data set to obtain the sending end image data, and provides an efficient joint source channel coding and modulation and demodulation processing method to significantly improve the efficiency and quality of data transmission, and at the same time improve the image restoration capability through the preset deep neural network technology, thereby effectively improving the quality and efficiency of image transmission, avoiding limiting the efficiency and flexibility of information exchange, and realizing two-way information interaction through a two-way relay wireless channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 : A flowchart of the steps of a two-way relay intelligent communication method based on a deep neural network provided by an embodiment of the present invention;

[0062] Figure 2 : A schematic diagram of the structure of a terminal encoder provided in an embodiment of the present invention;

[0063] Figure 3 : A schematic diagram of a two-way relay communication process provided by an embodiment of the present invention;

[0064] Figure 4 : A flow chart of a two-way relay intelligent communication method provided by an embodiment of the present invention;

[0065] Figure 5 : A structural diagram of a two-way relay intelligent communication device based on a deep neural network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0066] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0067] Example 1

[0068] Please refer to Figure 1 , a two-way relay intelligent communication method based on a deep neural network provided by an embodiment of the present invention, comprising the following steps S101-S102:

[0069] Step S101: Acquire image data to be sent, and input the image data to be sent into an encoder of a preset deep neural network to obtain an image sending signal.

[0070] In this embodiment, after the training of a preset deep neural network is complete, a terminal encoder is placed at the transmitting end of a two-way relay communication. The image data to be transmitted is used as input to the encoder, thereby generating an outgoing image transmission signal, which is transmitted over the two-way relay wireless channel. After training is complete, a terminal decoder in the preset deep neural network is placed at the receiving end of the two-way relay communication. After receiving the signal, it is used as input to the terminal decoder, which processes it to obtain the estimated transmitting end image data.

[0071] Step S102: Transmitting the image transmission signal to a bidirectional relay wireless channel, communicating and receiving the image transmission signal output by the bidirectional relay wireless channel, and then inputting the output image transmission signal into a decoder of a preset deep neural network to obtain transmitting end image data. The preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set.

[0072] As a preferred solution of this embodiment, the preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set, specifically including:

[0073] A two-way relay communication model is constructed, and a preset data set is obtained as training image data to be sent; an initial deep neural network is initialized, and the training image data is input into the preset deep neural network, so that after the training image data is processed by the terminal encoder and the relay encoder, the processed training image data is output to a two-way relay wireless channel set according to the two-way relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder and a terminal decoder; after the two-way relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the sending end training image data is obtained through the output after processing by the relay decoder and the terminal decoder; according to the difference between the obtained sending end training image data and the training image data to be sent, the initial deep neural network is iteratively trained to obtain the preset deep neural network.

[0074] In this embodiment, the preset data set can be the CIFAR10 data set, which can be used to supervise the learning and training of deep neural networks. At the same time, each sample in the CIFAR10 data set is equipped with a label value. Therefore, in the training process of the preset deep neural network, the CIFAR10 data set can be divided into a training set, a test set and a validation set. The training set is used to train the network model, the test set is used to detect the training process to save intermediate results, and the validation set is used to test the model effect.

[0075] Furthermore, after obtaining the preset dataset, data preprocessing is performed. Specifically, when using the CIFAR10 dataset, necessary data preprocessing, such as image normalization and data augmentation, is performed to enhance the network's generalization capabilities and noise resistance. Choosing the right architecture for a deep neural network is crucial when initializing it. Therefore, a convolutional neural network (CNN) can be considered as the underlying architecture for both the encoder and decoder to accommodate feature extraction and reconstruction of image data.

[0076] In this embodiment, a preset dataset, namely the CIFAR10 dataset, is used as the image data to be transmitted by the transmitting user. This ensures that the CIFAR10 dataset is selected as the image data to be transmitted by the transmitting user. An initial deep neural network is then initialized, comprising a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder. The CIFAR10 dataset is processed by the terminal encoder and then output to a bidirectional relay wireless channel.

[0077] As a preferred solution of this embodiment, the iterative training of the initial deep neural network based on the difference between the obtained sending-end training image data and the training image data to be sent specifically includes:

[0078] According to the difference between the obtained sending end training image data and the training image data to be sent, the corresponding mean square error is calculated; according to the mean square error, the loss function is obtained: Among them, x i is the i-th training image data to be sent, is the i-th sending end training image data, and N is the total number of image data; the loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

[0079] In this embodiment, after initializing the initial deep neural network, the initial deep neural network can be trained by using the CIFAR10 data set. Preferably, the initial deep neural network is a CNN neural network, which is used to simulate the encoder and decoder. It can be understood that the deep neural network architecture selection, when initializing the deep neural network, choosing a suitable architecture is the key. Convolutional neural network (CNN) and residual network (ResNet) are used as the basic architecture of the head convolution module, upsampling / downsampling module and channel coding module. The head convolution module is used to extract image features to start the joint source channel coding process. The upsampling / downsampling module is used to further capture image features and compress data dimensions. The channel coding module is used to alleviate channel distortion and generate complex channel inputs under bandwidth and power constraints. The activation function uses PreLU, and the terminal encoder structure is as follows: Figure 2 As shown. Then, initialize the network parameters and define the loss function. The loss function uses the mean square error. Calculate the mean square error between the image restored by the terminal and the original image at the sender, and make the network model descend with the loss function as the criterion, as shown in the following formula:

[0080]

[0081] in x i is the i-th original image, is the i-th image restored by the terminal, and N is the total number of samples. Therefore, during network training, the dataset is divided into a training set, a test set, and a validation set. The training set is used to train the network model, the test set is used to detect the training process and save intermediate results, and the validation set is used to test the model's effectiveness. The mean square error is used as the loss function to monitor the effectiveness of network learning. When the loss value of the loss function corresponding to the test set drops to a minimum point, and the preset number of subsequent training times (preferably, the preset number of times is 30) is greater than the loss value of the loss function corresponding to the current validation set, training is stopped and the current model is saved for subsequent testing of the network's effectiveness.

[0082] Furthermore, the two-way relay wireless channel performs processing, and the two-way relay wireless channel scrambles and noises the signals processed by the terminal encoder and the relay encoder according to the simulated communication environment set in the constructed two-way relay communication model, and then sends them to the relay decoder and the terminal decoder. It can be understood that the preset deep neural network is applied to the terminal encoder in the constructed two-way relay communication model, and the two-way relay communication model is used to scramble and noise the signals processed by the terminal encoder, and then sends them to the terminal decoder.

[0083] In this embodiment, the terminal decoder performs the opposite processing from the terminal encoder, thereby obtaining estimated transmit-end image data after initial deep neural network processing. After deep neural network training, the difference between the transmit-end image data and the image data to be sent by the end user is used to calculate the loss function using mean squared error, thereby training the deep neural network and iteratively optimizing the network parameters through gradient descent. It can be understood that, based on the settings of the above steps, supervised learning neural network training is performed, thereby saving the set parameters of the trained model.

[0084] As a preferred solution of this embodiment, the two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the construction of the two-way relay communication model specifically includes:

[0085] Set the signal S received by the relay node R for:

[0086]

[0087] Set the first terminal node to receive the signal y from the relay node A for:

[0088]

[0089] Set the first terminal node to remove the self-interference of the received signal to obtain the signal

[0090]

[0091] Set the second terminal node to receive the signal y from the relay node B for:

[0092]

[0093] Set the second terminal node to remove the self-interference of the received signal to obtain the signal

[0094]

[0095] Among them, the signal received by the relay node is S R , the signal received by the encoder of the first terminal node is S A , the encoder of the second terminal node receives the signal S B , the output of the encoder of the first terminal node is x A , the output of the encoder of the second terminal node is x B , are the power of the signals sent by the first terminal node and the second terminal node respectively, Denote the channel coefficients between the first terminal node and the second terminal node and the relay node R, n AR 、n BR They follow complex Gaussian distribution respectively The channel noise is , and η is the amplification factor; thus a two-way relay communication model is constructed.

[0096] Furthermore, is a complex normal random variable, is a complex normal random variable, σ is the standard deviation.

[0097] It should be noted that the amplification factor η can be defined as:

[0098]

[0099] in, is the transmission signal power of the relay node R.

[0100] In this embodiment, a two-way relay communication model is constructed so that when training a preset deep neural network, the signal processed by the terminal encoder can be scrambled and noised with certainty and precision. It can be understood that the two-way relay communication model determines the communication between the first terminal node A and the second terminal node B to the relay node respectively, and there is also a relay encoder and a relay decoder between the relay nodes. The relay encoder and the relay decoder communicate with each other through a two-way relay wireless channel, thereby realizing two-way communication between the first terminal node A and the second terminal node B through the relay node. Therefore, the two-way relay communication model only stipulates the rules for communication between the first terminal node A, the second terminal node B and the relay node, but the image data is processed by the encoder and decoder in each node, which is the result of the preset deep neural network processing. Among them, the two-way relay communication process is as follows: Figure 3 shown.

[0101] Understandably, see Figure 4 , which is a schematic flow chart of the method of this embodiment. Specifically, it includes the following steps S1-S:

[0102] S1: First, build a two-way relay communication model.

[0103] S2: Perform data preprocessing. When using the CIFAR10 dataset, perform necessary data preprocessing, such as image normalization and data augmentation, to enhance the network's generalization capabilities and noise resistance. Also, select a deep neural network architecture. Consider using a convolutional neural network (CNN) as the underlying architecture for the encoder and decoder to accommodate feature extraction and reconstruction of image data.

[0104] S3: Uses convolutional neural networks (CNN) and residual networks (ResNet) as the infrastructure for the head convolution module, upsampling / downsampling module, and channel coding module. The head convolution module extracts image features to initiate the joint source-channel coding process. The upsampling / downsampling module further captures image features and compresses data dimensions. The channel coding module alleviates channel distortion and generates complex channel inputs under bandwidth and power constraints. The activation function uses PReLU.

[0105] S4: Apply the neural network described in S3 to the terminal encoder in the two-way relay communication model set in S1, scramble and noise the signal processed by the terminal encoder through the two-way relay communication model, and send it to the terminal decoder.

[0106] S5: Perform supervised neural network training according to the settings of S2, S3, and S4, and save the trained model. Specifically, use the CIFAR10 dataset described in S2 to train the CNN neural network, reading two different batches of images from the dataset each time as input to the encoders of terminal nodes A and B. Initialize the network parameters and define the loss function, which uses the mean squared error (MSE). Calculate the mean squared error between the image restored by the terminal and the original image at the sender, and use this loss function as the criterion for gradient descent of the network model, as follows:

[0107]

[0108] in x i is the i-th original image, is the i-th image restored by the terminal, and N is the total number of samples.

[0109] During the network training process, the data set is divided into a training set, a test set, and a validation set. The training set is used to train the network model, the test set is used to detect the training process and save intermediate results, and the validation set is used to test the model effect. The mean square error is used as the loss function to supervise the effect of network learning. When the loss value of the test set drops to a minimum point and the subsequent 30 training times are greater than the current validation set loss value, the training is stopped and the current model is saved for subsequent testing of the network effect.

[0110] S6: Training process monitoring. During the deep neural network training process, the loss values ​​of the training set and validation set are monitored in real time, and learning curves are drawn to adjust the training strategy in time to prevent overfitting or underfitting problems.

[0111] S7: Parameter tuning: Based on the monitoring results of the training process, parameter tuning is performed, such as adjusting the learning rate and optimizing the number of network layers, to improve the network's convergence speed and performance.

[0112] S8: System Performance Evaluation: During the communication system configuration phase, a comprehensive performance evaluation is conducted, including indicators such as signal transmission quality and decoding accuracy. Through testing in real-world scenarios, the system's performance in actual communications is verified.

[0113] Understandably, the above expansion emphasizes several key aspects of the implementation process, including system parameter setting, network architecture selection, data preprocessing, simulation environment setup, training process monitoring, parameter tuning, and system performance evaluation. These considerations will contribute to a more comprehensive and in-depth understanding and implementation of deep learning-based two-way relay intelligent communication methods and systems.

[0114] It can be understood that the embodiment of the present invention is different from the one-way communication with inflexible and inefficient information exchange in that the dynamics and interactivity of information exchange are improved through a two-way relay intelligent communication method, thereby improving communication efficiency and flexibility. At the same time, the present invention provides an efficient joint source channel coding and modulation and demodulation processing method by applying convolutional neural networks (CNN) to significantly improve the efficiency and quality of data transmission. Again, the present invention focuses on improving image restoration capabilities through advanced deep neural network technology, thereby effectively improving the quality and efficiency of image transmission. In summary, the present invention aims to overcome the limitations of the existing technology through these innovative solutions and provide a more efficient and reliable intelligent communication solution, which is particularly suitable for fields such as wireless communications and satellite communications that require efficient and reliable image transmission.

[0115] The implementation of the above embodiment has the following effects:

[0116] The technical solution of the present invention obtains image data to be sent, and inputs the image data into an encoder of a preset deep neural network trained by a preset data set to obtain an image sending signal, which is then transmitted to a two-way relay wireless channel to communicate and receive the image sending signal output by the two-way relay wireless channel, thereby inputting the output image sending signal into a decoder of a preset deep neural network trained by a preset data set to obtain the sending end image data, and provides an efficient joint source channel coding and modulation and demodulation processing method to significantly improve the efficiency and quality of data transmission, and at the same time improve the image restoration capability through the preset deep neural network technology, thereby effectively improving the quality and efficiency of image transmission, avoiding limiting the efficiency and flexibility of information exchange, and realizing two-way information interaction through a two-way relay wireless channel.

[0117] Example 2

[0118] See also Figure 5 , which is a two-way relay intelligent communication device based on deep neural network provided by the present invention, including: an acquisition module 201 and an input module 202;

[0119] The acquisition module 201 is used to acquire image data to be sent and input the image data to be sent into an encoder of a preset deep neural network to obtain an image transmission signal;

[0120] The input module 202 is used to transmit the image transmission signal to a two-way relay wireless channel, and communicate to receive the image transmission signal output by the two-way relay wireless channel, and then input the output image transmission signal into a decoder of a preset deep neural network to obtain the sending end image data; wherein, the preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set.

[0121] As a preferred solution, the preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set, specifically including:

[0122] Build a two-way relay communication model and obtain a preset data set as the training image data to be sent;

[0123] Initializing an initial deep neural network and inputting the training image data into the preset deep neural network, so that after the training image data is processed by a terminal encoder and a relay encoder, the processed training image data is output to a bidirectional relay wireless channel set according to the bidirectional relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder;

[0124] After the bidirectional relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the relay decoder and the terminal decoder process and output the training image data at the transmitting end;

[0125] According to the difference between the obtained sending-end training image data and the training image data to be sent, the initial deep neural network is iteratively trained to obtain a preset deep neural network.

[0126] As a preferred solution, the two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the construction of the two-way relay communication model specifically includes:

[0127] Set the signal S received by the relay node R for:

[0128]

[0129] Set the first terminal node to receive the signal y from the relay node Afor:

[0130]

[0131] Set the first terminal node to remove the self-interference of the received signal to obtain the signal

[0132]

[0133] Set the second terminal node to receive the signal y from the relay node B for:

[0134]

[0135] Set the second terminal node to remove the self-interference of the received signal to obtain the signal

[0136]

[0137] Among them, the signal received by the relay node is S R , the signal received by the encoder of the first terminal node is S A , the encoder of the second terminal node receives the signal S B , the output of the encoder of the first terminal node is x A , the output of the encoder of the second terminal node is x B , are the power of the signals sent by the first terminal node and the second terminal node respectively, Denote the channel coefficients between the first terminal node and the second terminal node and the relay node R, n AR 、n BR They follow complex Gaussian distribution respectively The channel noise, η is the amplification factor;

[0138] Thus, a two-way relay communication model is constructed.

[0139] As a preferred solution, the iterative training of the initial deep neural network based on the difference between the obtained sending end training image data and the training image data to be sent specifically includes:

[0140] Calculate the corresponding mean square error based on the difference between the obtained sending end training image data and the training image data to be sent;

[0141] According to the mean square error, the loss function is obtained: Among them, x i is the i-th training image data to be sent, is the training image data of the i-th transmitter, and N is the number of total image data;

[0142] The loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

[0143] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working process of the device described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0144] The implementation of the above embodiment has the following effects:

[0145] The technical solution of the present invention obtains image data to be sent, and inputs the image data into an encoder of a preset deep neural network trained by a preset data set to obtain an image sending signal, which is then transmitted to a two-way relay wireless channel to communicate and receive the image sending signal output by the two-way relay wireless channel, thereby inputting the output image sending signal into a decoder of a preset deep neural network trained by a preset data set to obtain the sending end image data, and provides an efficient joint source channel coding and modulation and demodulation processing method to significantly improve the efficiency and quality of data transmission, and at the same time improve the image restoration capability through the preset deep neural network technology, thereby effectively improving the quality and efficiency of image transmission, avoiding limiting the efficiency and flexibility of information exchange, and realizing two-way information interaction through a two-way relay wireless channel.

[0146] Example 3

[0147] Correspondingly, the present invention also provides a terminal device, comprising: a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements the two-way relay intelligent communication method based on deep neural network as described in any one of the above embodiments.

[0148] The terminal device of this embodiment includes: a processor, a memory, and a computer program and computer instructions stored in the memory and capable of running on the processor. When the processor executes the computer program, each step in the above embodiment 1 is implemented, such as Figure 1 Alternatively, when the processor executes the computer program, the functions of each module / unit in the above-mentioned device embodiment, such as the input module 202, are implemented.

[0149] Exemplarily, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments that can perform specific functions, and the instruction segments are used to describe the execution process of the computer program in the terminal device. For example, the input module 202 is used to transmit the image transmission signal to a two-way relay wireless channel, and communicate to receive the image transmission signal output by the two-way relay wireless channel, and then input the output image transmission signal into the decoder of a preset deep neural network to obtain the transmitting end image data; wherein the preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set.

[0150] The terminal device may be a computing device such as a desktop computer, laptop, PDA, or cloud server. The terminal device may include, but is not limited to, a processor and memory. Those skilled in the art will appreciate that the schematic diagram is merely an example of a terminal device and does not limit the terminal device. The terminal device may include more or fewer components than shown, or a combination of certain components, or different components. For example, the terminal device may also include input / output devices, network access devices, buses, and the like.

[0151] The processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, etc. The processor is the control center of the terminal device, connecting various parts of the entire terminal device using various interfaces and lines.

[0152] The memory can be used to store the computer programs and / or modules. The processor realizes various functions of the terminal device by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, at least one application required for a function, etc.; the data storage area can store data created according to the use of the mobile terminal, etc. In addition, the memory can include a high-speed random access memory and can also include a non-volatile memory, such as a hard disk, a memory, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), at least one disk storage device, a flash memory device, or other volatile solid-state storage device.

[0153] Wherein, if the module / unit integrated in the terminal device is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present invention implements all or part of the process in the above-mentioned embodiment method, and can also be completed by a computer program to instruct the relevant hardware. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by a processor, it can implement the steps of each of the above-mentioned method embodiments. Wherein, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electric carrier signals and telecommunication signals.

[0154] Example 4

[0155] Accordingly, the present invention also provides a computer-readable storage medium, which includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the two-way relay intelligent communication method based on deep neural network as described in any one of the above embodiments.

[0156] The specific embodiments described above further illustrate the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A two-way relay intelligent communication method based on deep neural network, characterized in that: include: Acquire image data to be sent, and input the image data to be sent into an encoder of a preset deep neural network, thereby obtaining an image sending signal; Transmitting the image transmission signal to a bidirectional relay wireless channel, communicating and receiving the image transmission signal output by the bidirectional relay wireless channel, and then inputting the output image transmission signal into a decoder of a preset deep neural network, thereby obtaining transmitting end image data; The preset deep neural network is obtained by training the initial deep neural network using image data in a preset data set; The preset deep neural network is obtained by training the initial deep neural network using the image data in the preset data set, specifically including: Build a two-way relay communication model and obtain a preset data set as the training image data to be sent; Initializing an initial deep neural network and inputting the training image data into the preset deep neural network, so that after the training image data is processed by a terminal encoder and a relay encoder, the processed training image data is output to a bidirectional relay wireless channel set according to the bidirectional relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder; After the bidirectional relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the relay decoder and the terminal decoder process and output the training image data at the transmitting end; Iteratively training the initial deep neural network based on the difference between the obtained sending-end training image data and the training image data to be sent, thereby obtaining a preset deep neural network; When initializing the deep neural network, the convolutional neural network and the residual network are used as the basic architecture of the head convolution module, the upsampling / downsampling module, and the channel coding module. The head convolution module is used to extract image features to start the joint source-channel coding process. The upsampling / downsampling module is used to further capture image features and compress data dimensions. The channel coding module is used to alleviate channel distortion and generate complex channel input under bandwidth and power constraints.

2. A two-way relay intelligent communication method based on deep neural network according to claim 1, characterized in that: The two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the constructing of the two-way relay communication model specifically includes: Set the signal received by the relay node for: Set the first terminal node to receive the signal from the relay node for: Set the first terminal node to remove the self-interference of the received signal to obtain the signal : Set the second terminal node to receive the signal from the relay node for: Set the second terminal node to remove the self-interference of the received signal to obtain the signal : Among them, the signal received by the relay node is , the signal received by the encoder of the first terminal node is , the encoder of the second terminal node receives the signal , the output of the encoder of the first terminal node is , the output of the encoder of the second terminal node is , 、 are the power of the signals sent by the first terminal node and the second terminal node respectively, 、 Respectively represent the first terminal node and the second terminal node to the relay node The channel coefficient between They follow complex Gaussian distribution The channel noise, is the magnification factor; Thus, a two-way relay communication model is constructed.

3. The two-way relay intelligent communication method based on deep neural network according to claim 1, characterized in that: The iterative training of the initial deep neural network based on the difference between the obtained sending end training image data and the training image data to be sent specifically includes: Calculate the corresponding mean square error based on the difference between the obtained sending end training image data and the training image data to be sent; According to the mean square error, the loss function is obtained: ;in, It is training image data to be sent, It is The sending end training image data, is the amount of total image data; The loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

4. A two-way relay intelligent communication device based on deep neural network, characterized in that: include: Get module and input module; The acquisition module is used to acquire the image data to be sent and input the image data to be sent into an encoder of a preset deep neural network to obtain an image transmission signal; The input module is configured to transmit the image transmission signal to a bidirectional relay wireless channel, receive the image transmission signal output by the bidirectional relay wireless channel, and input the output image transmission signal into a decoder of a preset deep neural network, thereby obtaining transmitting-end image data; wherein the preset deep neural network is obtained by training an initial deep neural network using image data in a preset data set; The preset deep neural network is obtained by training the initial deep neural network using the image data in the preset data set, specifically including: Build a two-way relay communication model and obtain a preset data set as the training image data to be sent; Initializing an initial deep neural network and inputting the training image data into the preset deep neural network, so that after the training image data is processed by a terminal encoder and a relay encoder, the processed training image data is output to a bidirectional relay wireless channel set according to the bidirectional relay communication model; wherein the preset deep neural network includes a terminal encoder, a relay encoder, a relay decoder, and a terminal decoder; After the bidirectional relay wireless channel performs scrambling and noise operations on the input training image data, the scrambled and noised training image data is sent to the relay decoder and the terminal decoder, so that the relay decoder and the terminal decoder process and output the training image data at the transmitting end; Iteratively training the initial deep neural network based on the difference between the obtained sending-end training image data and the training image data to be sent, thereby obtaining a preset deep neural network; When initializing the deep neural network, the convolutional neural network and the residual network are used as the basic architecture of the head convolution module, the upsampling / downsampling module, and the channel coding module. The head convolution module is used to extract image features to start the joint source-channel coding process. The upsampling / downsampling module is used to further capture image features and compress data dimensions. The channel coding module is used to alleviate channel distortion and generate complex channel input under bandwidth and power constraints.

5. A two-way relay intelligent communication device based on a deep neural network according to claim 4, characterized in that: The two-way relay communication model includes a first terminal node, a second terminal node and a relay node; and the constructing of the two-way relay communication model specifically includes: Set the signal received by the relay node for: Set the first terminal node to receive the signal from the relay node for: Set the first terminal node to remove the self-interference of the received signal to obtain the signal : Set the second terminal node to receive the signal from the relay node for: Set the second terminal node to remove the self-interference of the received signal to obtain the signal : Among them, the signal received by the relay node is , the signal received by the encoder of the first terminal node is , the encoder of the second terminal node receives the signal , the output of the encoder of the first terminal node is , the output of the encoder of the second terminal node is , 、 are the power of the signals sent by the first terminal node and the second terminal node respectively, 、 Respectively represent the first terminal node and the second terminal node to the relay node The channel coefficient between They follow complex Gaussian distribution The channel noise, is the magnification factor; Thus, a two-way relay communication model is constructed.

6. A two-way relay intelligent communication device based on a deep neural network according to claim 5, characterized in that: The iterative training of the initial deep neural network based on the difference between the obtained sending end training image data and the training image data to be sent specifically includes: Calculate the corresponding mean square error based on the difference between the obtained sending end training image data and the training image data to be sent; According to the mean square error, the loss function is obtained: ;in, It is training image data to be sent, It is The sending end training image data, is the amount of total image data; The loss function is used to supervise the training of the initial deep neural network, so that after the loss function drops to the lowest point and is greater than the value of the current loss function within a preset number of subsequent trainings, the model data of the current deep neural network is saved, thereby completing the iterative training of the initial deep neural network.

7. A terminal device, characterized in that: It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, it implements the two-way relay intelligent communication method based on deep neural network according to any one of claims 1 to 3.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium includes a stored computer program, wherein when the computer program is running, the device where the computer-readable storage medium is located is controlled to execute the two-way relay intelligent communication method based on deep neural network according to any one of claims 1 to 3.

Images