Adaptive channel feedback method and apparatus based on image prediction performance

Abstract

The application discloses an adaptive channel feedback method and device based on image prediction performance, wherein the method comprises the following steps: acquiring an original image; calculating image prediction performance of the original image through an evaluator; determining a compression rate of a CSI matrix according to the image prediction performance, and sending the compression rate to a receiving end; receiving a CSI compression matrix acquired by the receiving end according to the compression rate, and recovering the CSI matrix according to the CSI compression matrix; and sending the original image according to the CSI matrix. The application can effectively reduce channel feedback overhead while ensuring better image reconstruction quality.

Description

Technical Field

[0001] The present application relates to the field of wireless communication technology, and in particular to an adaptive channel feedback method and device based on image prediction performance. Background Art

[0002] According to Shannon's theory, modern data transmission can be decomposed into a two-step separate coding scheme - source coding and channel coding. There is literature proving that when the block length of the data tends to infinity, the separate coding scheme can achieve the theoretical optimality. However, in the scenario of finite block length, the joint source channel coding scheme is superior to the separate source channel coding. In recent years, inspired by the success of deep learning, the JSCC (Joint-Source Channel Coding) system is implemented using a neural network parameterized autoencoder architecture, which performs better than the separate scheme. Specifically, the input source data is directly mapped to the channel symbols by using a DNN (Deep Neural Network), and then decoded by another DNN at the receiving end.

[0003] MIMO (Multiple-Input Multiple-Output) is now widely deployed in practical communication systems for transmitting a variety of sources, such as images, text, and video. It is considered a key technology for current and future wireless systems because it provides high spectral efficiency and reduces interference by fully utilizing spatial resources.

[0004] However, these advantages are highly dependent on CSI (Channel State Information), which requires the user equipment to feed back CSI to the base station via a feedback link. The large number of antennas used in massive MIMO results in a large dimensional CSI matrix, significantly increasing feedback overhead.

[0005] In downlink MIMO systems, CSI needs to be sent to the base station via a feedback link. To overcome the high feedback overhead in MIMO, assuming perfect CSI estimation, researchers have proposed a number of methods to compress the CSI matrix. These methods primarily exploit the spatial and temporal correlation of CSI. Specifically, existing methods fall into two categories:

[0006] (1) Compressed sensing methods: These methods are based on a sparsity prior, i.e., the CSI matrix is ​​approximately sparse. The CSI matrix is ​​first converted to a sparse domain and then compressed into a low-dimensional codeword. The base station then uses the received codeword to reconstruct the CSI matrix using a compressed sensing algorithm.

[0007] (2) DL (Deep Learning)-based CSI feedback methods: The basic idea of ​​this type of method is to borrow the architecture of the autoencoder. In these works, the original CSI matrix is ​​compressed by the encoder to generate vectors as its representation. Usually, the dimension of these vectors is much lower than that of the original CSI matrix. Then, the receiver uses the decoder to recover the CSI matrix based on the received representation vector. Similar to JSCC, the encoder and decoder are also jointly optimized using an end-to-end training strategy.

[0008] Existing JSCC methods have achieved significant performance improvements; however, most of them focus only on additive white Gaussian noise channels. Therefore, to design more practical systems for existing communication scenarios, it is necessary to extend JSCC to MIMO scenarios. In addition, the performance achieved by JSCC models for image transmission in these samples is usually different, mainly due to the different complexity between different images. Complexity measures redundancy; specifically, images with high complexity usually have low redundancy, high texture complexity, and low image prediction performance. Complexity is useful in many applications, such as determining the compression level of the signal source. This complexity can also guide the design of channel feedback, because images with lower complexity usually have higher image prediction performance and thus higher tolerance to interference, that is, these images are more robust to noise, and the use of this robustness can save communication resources. Summary of the Invention

[0009] The present application aims to solve one of the technical problems in the related art at least to a certain extent.

[0010] To this end, the purpose of this application is to reduce the overhead of channel feedback while ensuring the quality of image reconstruction to save communication resources, and proposes an adaptive channel feedback method based on image prediction performance.

[0011] Another object of the present application is to propose an adaptive channel feedback device based on image prediction performance.

[0012] To achieve the above objectives, the present application proposes an adaptive channel feedback method based on image prediction performance, including:

[0013] Get the original image;

[0014] Calculating the image prediction performance of the original image by an evaluator;

[0015] Determining a compression ratio of a CSI matrix according to the image prediction performance, and sending the compression ratio to a receiving end;

[0016] receiving a CSI compression matrix obtained by the receiving end according to the compression rate, and restoring the CSI matrix according to the CSI compression matrix;

[0017] The original image is sent according to the CSI matrix.

[0018] In a possible implementation, calculating the image prediction performance of the original image by an evaluator includes:

[0019] Training a lightweight model based on the knowledge distillation principle to obtain the evaluator, wherein the training uses the performance of the JSCC system on different images as labels;

[0020] The image prediction performance of the original image is calculated by the evaluator.

[0021] In a possible implementation, the sending the original image according to the CSI matrix includes:

[0022] Encoding the original image by an encoder to obtain encoding symbols;

[0023] Pre-encoding the coded symbols based on the CSI matrix to obtain channel symbols;

[0024] The channel symbols are transmitted.

[0025] In a possible implementation manner, after sending the original image according to the CSI matrix, the method further includes:

[0026] Receiving the channel symbols through the receiving end, and performing linear precoding on the channel symbols to obtain a received signal;

[0027] The received signal is decoded by a decoder to obtain a restored image.

[0028] In a possible implementation, the method further includes:

[0029] Based on the MSE loss function, training the initial encoder and the initial decoder by a joint training method to obtain the encoder and the decoder;

[0030] Among them, the formula of the MSE loss function is as follows:

[0031]

[0032] Wherein, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

[0033] To achieve the above objectives, the present application proposes an adaptive channel feedback device based on image prediction performance, comprising:

[0034] An acquisition module is used to acquire the original image;

[0035] a calculation module, configured to calculate the image prediction performance of the original image through an evaluator;

[0036] a determination module, configured to determine a compression ratio of a CSI matrix according to the image prediction performance, and send the compression ratio to a receiving end;

[0037] a recovery module, configured to receive the CSI compression matrix obtained by the receiving end according to the compression rate, and restore the CSI matrix according to the CSI compression matrix;

[0038] A sending module is used to send the original image according to the CSI matrix.

[0039] In a possible implementation manner, the calculation module includes:

[0040] a first acquisition unit, configured to train a lightweight model based on the knowledge distillation principle to obtain the evaluator, wherein the training uses the performance of the JSCC system on different images as labels;

[0041] A calculation unit is used to calculate the image prediction performance of the original image through the evaluator.

[0042] In a possible implementation manner, the sending module includes:

[0043] A second acquiring unit, configured to encode the original image through an encoder to acquire encoding symbols;

[0044] A third acquiring unit, configured to precode the coded symbols based on the CSI matrix to acquire channel symbols;

[0045] A sending unit is configured to send the channel symbol.

[0046] In a possible implementation manner, the device further includes:

[0047] a precoding module, configured to receive the channel symbols through the receiving end, and perform linear precoding on the channel symbols to obtain a received signal;

[0048] The decoding module is used to decode the received signal through a decoder to obtain a restored image.

[0049] In a possible implementation manner, the device further includes:

[0050] A training module, configured to train an initial encoder and an initial decoder by a joint training method based on an MSE loss function to obtain the encoder and the decoder;

[0051] Among them, the formula of the MSE loss function is as follows:

[0052]

[0053] Wherein, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

[0054] Beneficial effects of this application:

[0055] In an embodiment of the present application, an original image is obtained, and then an evaluator calculates the image prediction performance of the original image. The compression ratio of the CSI matrix is ​​determined based on the image prediction performance and the compression ratio is sent to the receiving end. The receiving end then receives the CSI compression matrix obtained based on the compression ratio, restores the CSI matrix based on the CSI compression matrix, and finally sends the original image based on the CSI matrix. This application can effectively reduce channel feedback overhead while ensuring good image reconstruction quality.

[0056] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0058] Figure 1 Flowchart of an adaptive channel feedback method based on image prediction performance according to an embodiment of the present application;

[0059] Figure 2 Schematic diagram of a framework of an adaptive channel feedback mechanism based on image prediction performance according to an embodiment of the present application;

[0060] Figure 3 is a schematic diagram of an evaluator according to an embodiment of the present application;

[0061] FIG4( a ) is a schematic diagram of an interruption threshold and interference tolerance according to an embodiment of the present application;

[0062] FIG4( b ) is a schematic diagram of a water injection method in a JSCC according to an embodiment of the present application;

[0063] Figure 5 1 is a schematic structural diagram of a JSCC system under MIMO according to an embodiment of the present application;

[0064] Figure 6 Schematic diagram of a process of adaptive channel feedback based on image prediction performance according to an embodiment of the present application;

[0065] Figure 7 Schematic diagram comparing the performance of JSCC under different numbers of antennas at different signal-to-noise ratios;

[0066] Figure 8 Schematic diagram comparing the transmission success rate of JSCC under different transmission bit numbers;

[0067] Figure 9 Schematic diagram comparing the transmission success rate of JSCC under different interruption thresholds;

[0068] Figure 10 Schematic diagram of the number of feedback bits required by JSCC to achieve the same performance;

[0069] Figure 11 Schematic diagram of the structure of an adaptive channel feedback device based on image prediction performance according to an embodiment of the present application. DETAILED DESCRIPTION

[0070] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0071] In order to enable those skilled in the art to better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of this application.

[0072] The following describes the adaptive channel feedback method and device based on image prediction performance proposed in accordance with the embodiments of the present application with reference to the accompanying drawings. First, the adaptive channel feedback method based on image prediction performance proposed in accordance with the embodiments of the present application will be described with reference to the accompanying drawings.

[0073] Figure 1 Flowchart of an adaptive channel feedback method based on image prediction performance according to an embodiment of the present application.

[0074] like Figure 1 As shown, the adaptive channel feedback method based on image prediction performance includes:

[0075] Step S110: obtaining an original image.

[0076] In the embodiment of the present application, an original image may be obtained. The original image may be an image to be transmitted. It is understood that the original image may have different complexities and thus may have different image prediction performances. Figure 2 FIG. 1 is a schematic diagram of a framework of an adaptive channel feedback mechanism based on image prediction performance according to an embodiment of the present application. Figure 2 As shown, images can be divided into high-complexity images and low-complexity images.

[0077] Step S120 : calculating the image prediction performance of the original image through an evaluator.

[0078] In the embodiment of the present application, after the original image is acquired, the image prediction performance of the original image can be calculated by an evaluator. Figure 3 is a schematic diagram of an evaluator according to an embodiment of the present application, as shown in FIG. Figure 3 As shown, the evaluator can be deployed at the transmitting end. Before the original image is sent, the image prediction performance of the original image can be calculated by the evaluator, and the image prediction performance can be used for channel resource allocation considerations.

[0079] Step S130: Determine the compression rate of the CSI matrix according to the image prediction performance, and send the compression rate to the receiving end.

[0080] In the embodiment of the present application, the compression rate of the CSI matrix can be determined according to the image prediction performance, and the compression rate can be sent to the receiving end. Figure 2 As shown, for high-complexity images and low-complexity images, compression decisions can be made based on the corresponding image prediction performance, the compression ratio of the CSI matrix can be determined, and the compression ratio can be sent to the CSI encoder at the receiving end.

[0081] Step S140: receiving the CSI compression matrix obtained by the receiving end according to the compression ratio, and restoring the CSI matrix according to the CSI compression matrix.

[0082] In the embodiment of the present application, the CSI compression matrix obtained by the receiving end according to the compression rate can be received, and the CSI matrix can be restored according to the CSI compression matrix. Figure 2As shown in Figure 1, the CSI encoder at the receiving end can obtain the CSI channel matrix estimated by the CSI estimator and compress the CSI channel matrix according to the compression ratio to obtain a CSI compressed matrix. After obtaining the CSI compressed matrix, the receiving end can send the CSI compressed matrix to the CSI decoder at the transmitting end via a feedback link. After obtaining the CSI compressed matrix, the CSI decoder at the transmitting end can restore the CSI matrix based on the CSI compressed matrix and send the restored CSI matrix to the precoder at the transmitting end.

[0083] It should be noted that the CSI matrix is ​​restored according to the CSI compression matrix, and its restoration error can be calculated using the normalized mean square error (NMSE). The formula of NMSE is as follows:

[0084]

[0085] Where H represents the estimated CSI matrix, Represents the recovered CSI matrix.

[0086] It is understandable that a higher CSI compression rate can lead to a larger reconstruction error, that is, the higher the CSI compression rate, the larger the NMSE value. The impact of a larger NMSE in JSCC is specifically reflected in the reduction of the reconstruction quality of the image. Although reducing the CSI compression rate can effectively reduce the NMSE, it will increase the feedback overhead of the channel. Therefore, the purpose of the embodiment of the present application is to adaptively adjust the number of bits according to the image prediction performance of the original image, thereby reducing the feedback overhead of the channel. That is to say, for the original image with higher complexity, that is, the original image with lower image prediction performance, the number of feedback bits allocated will be higher, and for the original image with lower complexity, that is, the original image with higher image prediction performance, the number of feedback bits allocated will be lower.

[0087] Step S150: Send the original image according to the CSI matrix.

[0088] In an embodiment of the present application, after the CSI matrix is ​​restored according to the CSI compression matrix, the original image may be transmitted according to the CSI matrix. Exemplarily, after the CSI matrix is ​​restored according to the CSI compression matrix, the original image may be encoded by an encoder at the transmitting end, and the encoding of the original image may be transmitted according to the CSI matrix.

[0089] In an embodiment of the present application, an original image is obtained, and then an evaluator calculates the image prediction performance of the original image. The compression ratio of the CSI matrix is ​​determined based on the image prediction performance and the compression ratio is sent to the receiving end. The receiving end then receives the CSI compression matrix obtained based on the compression ratio, restores the CSI matrix based on the CSI compression matrix, and finally sends the original image based on the CSI matrix. This application can effectively reduce channel feedback overhead while ensuring good image reconstruction quality.

[0090] In one possible implementation, calculating the image prediction performance of the original image by the evaluator includes:

[0091] Based on the knowledge distillation principle, a lightweight model is trained to obtain an evaluator, where the performance of the JSCC system on different images is used as the label;

[0092] The image prediction performance of the original image is calculated by the evaluator.

[0093] In an embodiment of the present application, a lightweight model can be trained based on the principle of knowledge distillation to obtain an evaluator, wherein the training uses the performance of the JSCC system on different images as a label. Exemplarily, the performance of the JSCC system on different images can be expressed as Peak Signal to Noise Ratio (PSNR), that is, based on the principle of knowledge distillation, a lightweight model can be trained with the Peak Signal to Noise Ratio as a label to obtain an evaluator. After obtaining the evaluator, the image prediction performance of the original image can be calculated by the evaluator. In this way, an evaluator can be obtained, and then the image prediction performance of the original image can be calculated by the evaluator, and adaptive channel feedback can be implemented based on the image prediction performance, thereby saving channel feedback resources and effectively reducing channel feedback overhead.

[0094] It should be noted that Figure 4(a) is a schematic diagram of the interruption threshold and interference tolerance according to an embodiment of the present application. As shown in Figure 4(a), a lower limit can be set for the reconstruction quality of the image. When the reconstruction quality of the image exceeds this lower limit, the transmission of the image can be considered successful. This lower limit can be regarded as the interruption threshold in JSCC. Images with higher reconstruction performance (lower complexity) usually have higher interference tolerance, that is, images with higher reconstruction performance are more robust to interference. Therefore, the allocation of transmission resources in the JSCC system can be re-determined based on the image prediction performance. Still as shown in Figure 4(a), through the gap between the reconstruction quality of the image and the interruption threshold, t i , to estimate the image redundancy. Then, a similar idea to the water filling method can be used to allocate transmission resources. As shown in Figure 4(b), less transmission resources can be allocated to images with higher tolerance and more transmission resources can be allocated to images with lower tolerance.

[0095] It should be noted that the performance of the JSCC system depends not only on the transmitted image itself, but also on the system capabilities. The reason for the above phenomenon is that there will be training bias during training, which leads to better performance of the system on some images. Therefore, it can be Figure 3 As shown in the figure, based on the knowledge distillation principle, we can use the peak signal-to-noise ratio as a label to train a lightweight model and obtain an evaluator.

[0096] In a possible implementation, sending the original image according to the CSI matrix includes:

[0097] Encode the original image through the encoder to obtain the encoding symbol;

[0098] Pre-encode the coded symbols based on the CSI matrix to obtain channel symbols;

[0099] Send channel symbols.

[0100] In the embodiment of the present application, the original image can be encoded by the encoder to obtain the encoding symbols, and then the encoding symbols can be pre-encoded based on the CSI matrix to obtain the channel symbols, and finally the channel symbols can be sent. Figure 2 As shown in the figure, the original image can be encoded using the JSCC encoder. Then, the precoder precodes the coded symbols based on the CSI matrix to obtain channel symbols, which can then be transmitted. In this way, the original image can be transmitted based on the recovered CSI matrix, ensuring the quality of the original image reconstruction.

[0101] In a possible implementation manner, after sending the original image according to the CSI matrix, the method further includes:

[0102] A receiving end receives channel symbols and linearly precodes the channel symbols to obtain a received signal;

[0103] The received signal is decoded by a decoder to obtain a restored image.

[0104] In the embodiment of the present application, after the original image is sent according to the CSI matrix, the channel symbols can be received by the receiving end, and the channel symbols can be linearly pre-coded to obtain the received signal, and then the received signal can be decoded by the decoder to obtain the restored image. Figure 2 As shown, a combiner at the receiving end receives channel symbols and linearly precodes the channel to obtain a received signal. This signal is then sent to a decoder, which decodes the received signal to obtain a restored image. In this way, the decoder can obtain a restored image corresponding to the original image, ensuring image reconstruction quality.

[0105] It should be noted that Figure 5 FIG. 1 is a structural diagram of a JSCC system under MIMO according to an embodiment of the present application. Figure 5 As shown in the figure, the JSCC system includes an encoder, a MIMO channel, and a decoder. The encoder and decoder can be represented by DNN. The specific network structure is as follows: Figure 5 When an original image needs to be transmitted, the original image can be represented by s∈R N×1 Indicates that, where N represents the vector length, the encoding function can be expressed as Where θ represents a trainable parameter. The encoder can map s to complex-valued channel symbols. The process can be expressed as:

[0106]

[0107] Where K represents the number of transmission channel symbols. Then the power constraint is applied to the symbol z.

[0108] Then, the coded channel symbols will be transmitted via MIMO. Considering the frequency division duplex system, the image is transmitted by a t The base station with N transmitting antennas sends r The user of the receiving antenna. represents the precoding vector at the base station, where d represents the number of data streams. By adopting the singular value decomposition precoding scheme, the channel matrix The original image is encoded as a vector of length K. For transmission, it is decomposed into a set of signals, each of which has a length of d. Select one of the signals For example, the received signal can be expressed as:

[0109] y=HVx+n,

[0110] in represents Gaussian white noise.

[0111] At the receiving end, considering linear precoding, part of the received signal can be expressed as:

[0112]

[0113] Where U represents the precoding matrix and H represents the conjugate transpose. Correspondingly, after multiple transmissions, the received signal is obtained.

[0114] Finally, the decoder decodes it to get the restored image:

[0115]

[0116] where φ represents the trainable parameters of the decoder.

[0117] It should be noted that the channel model can adopt a narrowband millimeter wave channel model, which includes N cl clusters, each containing N ray propagation paths. Each path contains the channel's sending and receiving directions (sending angle, arrival angle), as well as the path complex gain. The channel matrix is ​​expressed as

[0118]

[0119] in is the complex gain of the lth path in the i-th cluster, and represent the arrival angle and transmission angle of the receiving end and the sending end respectively. and denote the receive and transmit steering vectors, respectively. For a linear array of N antennas and an angle φ, the steering vector can be written as

[0120]

[0121] Where d and λ represent the distance between adjacent antennas and the wavelength of the carrier, respectively. It is understandable that the embodiments of the present application can support various numbers of transmit antennas, and the larger the number of antennas, the more significant the reduction in channel feedback overhead.

[0122] In one possible implementation, the adaptive channel feedback method based on image prediction performance further includes:

[0123] Based on the MSE loss function, the initial encoder and initial decoder are trained by a joint training method to obtain the encoder and decoder;

[0124] Among them, the formula of the MSE loss function is as follows:

[0125]

[0126] Among them, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

[0127] In the embodiment of the present application, considering that JSCC is an end-to-end system, the initial encoder and initial decoder can be trained by a joint training method based on the MSE loss function to obtain the encoder and decoder. The formula of the MSE loss function is as follows:

[0128]

[0129] Among them, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

[0130] In this way, the encoder and decoder of the JSCC system can be obtained, which is conducive to ensuring the reconstruction quality of the image.

[0131] It should be noted that because JSCC is an end-to-end system, source coding and channel coding can be integrated. Specifically, the encoding, decoding, and transmission processes can be parameterized using a DNN and trained end-to-end using a data-driven approach. Since the transmission process involves non-differentiable operations, end-to-end training is difficult. However, setting the gradient of some channels to 1 allows for end-to-end training.

[0132] In order to make the description of the embodiments of the present application more detailed and clear, Figure 6 The adaptive feedback process of the embodiment of the present application is further explained. Figure 6 FIG. 1 is a schematic diagram of a process of adaptive channel feedback based on image prediction performance according to an embodiment of the present application, as shown in FIG. Figure 6 As shown in the figure, the initial phase is the offline network training phase, followed by the online network use phase. First, after image prediction performance evaluation and compression level determination, the base station sends the CSI compression level to the user. The user compresses the CSI matrix according to the compression level and sends the CSI compression codeword to the base station. The base station recovers the CSI matrix based on the CSI compression codeword and pre-encodes the image processed by the encoder based on the recovered CSI, thus transmitting the image to the user. After receiving the received signal, the user reconstructs the image based on the received signal.

[0133] For the specific experimental results of the embodiments of the present application, Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 Be informed. Figure 7 Schematic diagram comparing the performance of JSCC under different numbers of antennas at different signal-to-noise ratios; Figure 8 Schematic diagram comparing the transmission success rate of JSCC under different transmission bit numbers; Figure 9 Schematic diagram comparing the transmission success rate of JSCC under different interruption thresholds; Figure 10 Schematic diagram of the number of feedback bits required by JSCC to achieve the same performance. Figure 7 、 Figure 8 、 Figure 9 、 Figure 10 It can be seen that compared with the traditional solution without adaptation, the adaptive channel feedback method based on image prediction performance in the embodiment of the present application can significantly reduce the channel feedback overhead while ensuring better image reconstruction performance.

[0134] In order to implement the above embodiment, Figure 11 As shown, this embodiment further provides an adaptive channel feedback device 11 based on image prediction performance. The device 11 includes: an acquisition module 110 , a calculation module 111 , a determination module 112 , a recovery module 113 , and a sending module 114 .

[0135] An acquisition module 110 is used to acquire an original image;

[0136] A calculation module 111, configured to calculate the image prediction performance of the original image through an evaluator;

[0137] a determination module 112, configured to determine a compression ratio of the CSI matrix according to the image prediction performance, and send the compression ratio to a receiving end;

[0138] The recovery module 113 is configured to receive the CSI compression matrix obtained by the receiving end according to the compression ratio, and restore the CSI matrix according to the CSI compression matrix;

[0139] The sending module 114 is configured to send the original image according to the CSI matrix.

[0140] According to the adaptive channel feedback device based on image prediction performance in an embodiment of the present application, the acquisition module obtains the original image, and then the calculation module calculates the image prediction performance of the original image through an evaluator. The determination module then determines the compression ratio of the CSI matrix based on the image prediction performance and sends the compression ratio to the receiving end. Then, the recovery module receives the CSI compression matrix obtained by the receiving end based on the compression ratio and restores the CSI matrix based on the CSI compression matrix. Finally, the sending module sends the original image based on the CSI matrix. This application can effectively reduce channel feedback overhead while ensuring good image reconstruction quality.

[0141] In a possible implementation, the calculation module 111 includes:

[0142] The first acquisition unit is used to train a lightweight model based on the knowledge distillation principle to obtain an evaluator, wherein the training uses the performance of the JSCC system on different images as labels;

[0143] A computing unit for calculating the image prediction performance of the original image through the evaluator.

[0144] In a possible implementation, the sending module 114 includes:

[0145] A second acquiring unit is configured to encode the original image through an encoder to acquire encoding symbols;

[0146] A third acquisition unit is configured to precode the coded symbols based on the CSI matrix to obtain channel symbols;

[0147] The sending unit is configured to send channel symbols.

[0148] In a possible implementation manner, the adaptive channel feedback device 11 based on image prediction performance further includes:

[0149] A precoding module is used to receive channel symbols through a receiving end and perform linear precoding on the channel symbols to obtain a received signal;

[0150] The decoding module is used to decode the received signal through a decoder to obtain a restored image.

[0151] In a possible implementation manner, the adaptive channel feedback device 11 based on image prediction performance further includes:

[0152] A training module is used to train the initial encoder and the initial decoder through a joint training method based on the MSE loss function to obtain the encoder and the decoder;

[0153] Among them, the formula of the MSE loss function is as follows:

[0154]

[0155] Among them, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

[0156] It should be noted that the aforementioned explanation of the embodiment of the adaptive channel feedback method based on image prediction performance is also applicable to the adaptive channel feedback device based on image prediction performance in this embodiment, and will not be repeated here.

[0157] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of such features. Throughout the description of this application, "plurality" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0158] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0159] Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limitations on the present application. Ordinary technicians in this field can change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. An adaptive channel feedback method based on image prediction performance, characterized in that: include: Get the original image; Calculating the image prediction performance of the original image by an evaluator; Determining a compression ratio of a CSI matrix according to the image prediction performance, and sending the compression ratio to a receiving end; receiving a CSI compression matrix obtained by the receiving end according to the compression rate, and restoring the CSI matrix according to the CSI compression matrix; The original image is sent according to the CSI matrix.

2. The adaptive channel feedback method based on image prediction performance according to claim 1, characterized in that: The calculating the image prediction performance of the original image by the evaluator includes: Training a lightweight model based on the knowledge distillation principle to obtain the evaluator, wherein the training uses the performance of the JSCC system on different images as labels; The image prediction performance of the original image is calculated by the evaluator.

3. The adaptive channel feedback method based on image prediction performance according to claim 1, characterized in that: The sending the original image according to the CSI matrix includes: Encoding the original image by an encoder to obtain encoding symbols; Pre-encoding the coded symbols based on the CSI matrix to obtain channel symbols; The channel symbols are transmitted.

4. The adaptive channel feedback method based on image prediction performance according to claim 3, characterized in that: After sending the original image according to the CSI matrix, the method further includes: Receiving the channel symbols through the receiving end, and performing linear precoding on the channel symbols to obtain a received signal; The received signal is decoded by a decoder to obtain a restored image.

5. The adaptive channel feedback method based on image prediction performance according to claim 4, characterized in that: The method further comprises: Based on the MSE loss function, training the initial encoder and the initial decoder by a joint training method to obtain the encoder and the decoder; Among them, the formula of the MSE loss function is as follows: Wherein, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

6. An adaptive channel feedback device based on image prediction performance, characterized in that: include: An acquisition module is used to acquire the original image; a calculation module, configured to calculate the image prediction performance of the original image through an evaluator; a determination module, configured to determine a compression ratio of a CSI matrix according to the image prediction performance, and send the compression ratio to a receiving end; a recovery module, configured to receive the CSI compression matrix obtained by the receiving end according to the compression rate, and restore the CSI matrix according to the CSI compression matrix; A sending module is used to send the original image according to the CSI matrix.

7. The adaptive channel feedback device based on image prediction performance according to claim 6, characterized in that: The computing module includes: a first acquisition unit, configured to train a lightweight model based on the knowledge distillation principle to obtain the evaluator, wherein the training uses the performance of the JSCC system on different images as labels; A calculation unit is used to calculate the image prediction performance of the original image through the evaluator.

8. The adaptive channel feedback device based on image prediction performance according to claim 6, characterized in that: The sending module includes: A second acquiring unit, configured to encode the original image through an encoder to acquire encoding symbols; A third acquiring unit, configured to precode the coded symbols based on the CSI matrix to acquire channel symbols; A sending unit is configured to send the channel symbol.

9. The adaptive channel feedback device based on image prediction performance according to claim 8, characterized in that: The device further comprises: a precoding module, configured to receive the channel symbols through the receiving end, and perform linear precoding on the channel symbols to obtain a received signal; The decoding module is used to decode the received signal through a decoder to obtain a restored image.

10. The adaptive channel feedback device based on image prediction performance according to claim 9, characterized in that: The device further comprises: A training module, configured to train an initial encoder and an initial decoder by a joint training method based on an MSE loss function to obtain the encoder and the decoder; Among them, the formula of the MSE loss function is as follows: Wherein, s represents the original image, represents the restored image, s i represents the i-th element of s, express The i-th element of N represents s and The number of elements.

Images