Method for adjusting compression rate of encoder, feedback method and device for data

Abstract

The application relates to the field of data processing and provides an encoder compression rate adjusting method and a data feedback method and device. The method comprises the following steps: inputting target data into a first neural network model constructed in advance to obtain the optimal compression rate of the target data; the first neural network model is obtained by training a neural network based on target data samples marked with optimal compression rate labels; and the compression rate of an encoder is adjusted to the optimal compression rate. By inputting the target data into the first neural network model, the optimal compression rate of the target data can be obtained, and then the compression rate of the encoder is adjusted to the optimal compression rate, so that the adaptive adjustment of the compression rate is realized, the data compression process is more flexible and reliable, and the problems that the existing data compression mode cannot adapt to dynamic changes of the environment and the compression process is not flexible and reliable are solved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for adjusting encoder compression rate, a data feedback method, and an apparatus. Background Technology

[0002] In traditional autoencoders, the encoder part can generally only achieve data compression with a fixed compression rate. Since the compression rate cannot be adjusted, the data compression process cannot adapt to dynamic changes in the environment, and the compression process is not flexible and reliable enough. Summary of the Invention

[0003] This invention provides a method for adjusting the compression ratio of an encoder, a method for feeding back data, and an apparatus to address the shortcomings of existing technologies where the compression ratio is not adjustable, the data compression process cannot adapt to dynamic environmental changes, and the compression process is not flexible or reliable enough.

[0004] In a first aspect, the present invention provides a method for adjusting the compression ratio of an encoder, the method comprising:

[0005] The target data is input into a pre-built first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression ratio.

[0006] Adjust the encoder's compression ratio to the optimal compression ratio.

[0007] According to the encoder compression ratio adjustment method provided by the present invention, the step of inputting the target data into a pre-constructed first neural network model to obtain the optimal compression ratio of the target data includes:

[0008] The sparsity of the target data is analyzed to obtain the sparsity analysis results;

[0009] Based on the sparsity analysis results, the optimal compression ratio of the target data is determined.

[0010] According to the encoder compression ratio adjustment method provided by the present invention, determining the optimal compression ratio of the target data based on the sparsity analysis results includes:

[0011] Based on the sparsity analysis results, the theoretical compression ratio with the lowest feedback bit count and the highest feedback accuracy among the preset theoretical compression ratios is selected as the optimal compression ratio.

[0012] According to the encoder compression rate adjustment method provided by the present invention, the training process of the first neural network model includes:

[0013] Multiple target data samples with different sparsity are obtained, and the optimal compression ratio label corresponding to each target data sample is determined respectively;

[0014] The first neural network model is obtained by training a pre-built neural network with target data samples labeled with the optimal compression ratio.

[0015] According to the encoder compression ratio adjustment method provided by the present invention, the step of training a pre-constructed neural network using target data samples labeled with the optimal compression ratio includes:

[0016] A weighted term is introduced into the loss function of the neural network to obtain a weighted loss function;

[0017] Based on the weighted loss function, the neural network is trained using target data samples labeled with the optimal compression ratio.

[0018] According to the encoder compression ratio adjustment method provided by the present invention, adjusting the encoder compression ratio to the optimal compression ratio includes:

[0019] The number of nodes in the fully connected layer of the second neural network model corresponding to the encoder is adjusted to adjust the compression ratio of the encoder to the optimal compression ratio.

[0020] Secondly, the present invention also provides a data feedback method, the method comprising:

[0021] The encoder compression ratio is adjusted to the optimal compression ratio according to any of the encoder compression ratio adjustment methods described above;

[0022] The target data is compressed according to the optimal compression ratio.

[0023] The compressed target data and the preset identification information corresponding to the optimal compression ratio are fed back to the base station;

[0024] The base station is used to determine the optimal compression ratio based on the preset identification information, and to decompress the compressed target data using the optimal compression ratio.

[0025] According to the data feedback method provided by the present invention, the target data is downlink channel state information.

[0026] Thirdly, the present invention also provides an encoder compression rate adjustment device, the device comprising:

[0027] The first processing module is used to input the target data into a pre-constructed first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression ratio.

[0028] An adjustment module is used to adjust the encoder's compression ratio to the optimal compression ratio.

[0029] Fourthly, the present invention also provides a data feedback device, the device comprising:

[0030] The second processing module is used to adjust the encoder compression rate to the optimal compression rate according to any of the encoder compression rate adjustment methods described above.

[0031] A compression module is used to compress the target data according to the optimal compression ratio;

[0032] The feedback module is used to feed back the compressed target data and the preset identification information corresponding to the optimal compression ratio to the base station;

[0033] The base station is used to determine the optimal compression ratio based on the preset identification information, and to decompress the compressed target data using the optimal compression ratio.

[0034] The encoder compression rate adjustment method, data feedback method, and apparatus provided by this invention can obtain the optimal compression rate of the target data by inputting the target data into a first neural network model, and then adjust the encoder compression rate to the optimal compression rate, thereby achieving adaptive adjustment of the compression rate and making the data compression process more flexible and reliable. Attached Figure Description

[0035] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0036] Figure 1 This is a flowchart illustrating the encoder compression rate adjustment method provided by the present invention;

[0037] Figure 2 It is a schematic diagram illustrating the principles of encoder compression rate adjustment, data compression, feedback, and decompression;

[0038] Figure 3 These are schematic diagrams of the structures of various neural network models;

[0039] Figure 4This is a flowchart illustrating the data feedback method provided by the present invention;

[0040] Figure 5 This is a schematic diagram of the encoder compression rate adjustment device provided by the present invention;

[0041] Figure 6 This is a schematic diagram of the data feedback device provided by the present invention;

[0042] Figure 7 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0044] The following is combined with Figures 1 to 7 This invention describes an encoder compression ratio adjustment method, a data feedback method, an encoder compression ratio adjustment device, a data feedback device, and an electronic device built based on the above encoder compression ratio adjustment method, all provided in embodiments of the present invention.

[0045] Figure 1 This invention illustrates a method for adjusting the encoder compression ratio according to an embodiment of the invention. The method includes:

[0046] Step 110: Input the target data into the pre-built first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on the target data samples labeled with the optimal compression ratio.

[0047] Step 120: Adjust the encoder's compression ratio to the optimal compression ratio.

[0048] It is understood that in this embodiment, the target data refers to the data to be compressed. When the above encoder compression rate adjustment method is applied to a frequency division duplex system, the target data can be downlink channel state information, i.e., the channel matrix.

[0049] In an exemplary embodiment, the process of inputting target data into a pre-built first neural network model to obtain the optimal compression ratio of the target data may specifically include:

[0050] The sparsity of the target data is analyzed to obtain the sparsity analysis results;

[0051] Based on the sparsity analysis results, the optimal compression ratio of the target data is determined.

[0052] Furthermore, the process of determining the optimal compression ratio of the target data based on the sparsity analysis results may specifically include:

[0053] Based on the sparsity analysis results, the theoretical compression ratio with the lowest feedback bit count and the highest feedback accuracy among several preset theoretical compression ratios is selected as the optimal compression ratio.

[0054] In this embodiment, there are multiple preset theoretical compression ratios, that is, the number of selectable compression ratios can be K. The main function of the first neural network model is to select the optimal compression ratio from the above K theoretical compression ratios that can balance feedback overhead and feedback accuracy as the best compression ratio. In other words, in this embodiment, the best compression ratio can be defined as the compression ratio with the fewest possible number of feedback bits while ensuring that the feedback accuracy is higher than the preset accuracy threshold.

[0055] Understandably, once the precision threshold corresponding to the feedback precision is determined, the higher the sparsity of the target data, the fewer bits need to be fed back, and the lower the feedback overhead; conversely, the lower the sparsity of the target data, the more bits need to be fed back, and the higher the feedback overhead.

[0056] Figure 2 The implementation principle of the encoder compression rate adjustment method is shown. Figure 2 In this diagram, H represents the input target data, such as a channel matrix. After passing through the first neural network model, the optimal compression ratio is obtained. This optimal compression ratio is input to the adaptive module, which then adjusts the compression ratio of the encoder. Specifically, the number of nodes in the fully connected layers of the second neural network model can be adjusted to achieve this adjustment. Thus, the second neural network model corresponding to the encoder can compress the target data according to the optimal compression ratio. S represents the codeword obtained after the target data is compressed by the encoder. The codeword S and the preset identifier information corresponding to the optimal compression ratio are input to the decoder. The third neural network model corresponding to the decoder can decompress the codeword S according to the optimal compression ratio to obtain the decompressed data.

[0057] also, Figure 2 The method of inputting a fixed compression ratio to the adaptive module in the dashed box is a traditional compression method, mainly used for comparison with the improved adaptive compression ratio method provided in this embodiment.

[0058] In an exemplary embodiment, the training process of the first neural network model described above may specifically include:

[0059] Obtain multiple target data samples with different sparsity and determine the optimal compression ratio label for each target data sample;

[0060] The first neural network model is obtained by training a pre-built neural network with target data samples labeled with the optimal compression ratio.

[0061] Furthermore, the process of training a pre-built neural network using target data samples labeled with the optimal compression ratio can specifically include:

[0062] By introducing a weighting term into the loss function of the neural network, a weighted loss function is obtained.

[0063] Based on the weighted loss function, the neural network is trained using target data samples labeled with the optimal compression ratio.

[0064] See appendix Figure 3 In this embodiment, the first neural network model can also be called the Sparsity Analysis Model (SAM). This model mainly consists of 4 fully connected layers. The activation function of the last layer is Softmax, which can output K numbers in the interval [0,1]. Each number represents the probability that the corresponding compression ratio is the optimal compression ratio. The compression ratio with the highest probability will be adopted, that is, as the optimal compression ratio.

[0065] For this classification network, the number of neurons in the output layer is M, and the activation function is the Softmax function, which can be expressed as:

[0066]

[0067] Where z represents the output vector of the last layer, and p represents the number of output data, which is the same as the number of optional compression ratios, i.e., len(z) equals K.

[0068] However, the aforementioned network inevitably makes mistakes. To increase the probability of keeping the feedback accuracy above the accuracy threshold, this embodiment weights the loss function during network training, i.e., introduces a weighting term into the loss function. In this embodiment, the loss function is the cross-entropy function. By weighting the cross-entropy function, the erroneous judgment of misclassifying the input target data as having a low compression ratio can be suppressed. The expression of the cross-entropy function after introducing the weighting term is as follows:

[0069]

[0070] Among them, y ik This represents the k-th element of the vector representing the compression ratio label of the i-th target data under one-hot encoding. This represents the neural network's estimate of the one-hot encoded vector element, γ. kThis represents the weight of the penalty for misclassifying the target data as having a lower compression ratio. For a smaller compression ratio, a larger penalty weight can be used.

[0071] It is not difficult to see that when training the first neural network model, by appropriately weighting the loss function, without reducing the network's classification accuracy, we can suppress the network's behavior of incorrectly matching the target data with a compression rate smaller than the optimal value, thereby increasing the probability of keeping the feedback accuracy above the preset accuracy threshold.

[0072] In an exemplary embodiment, the process of adjusting the encoder's compression ratio to the optimal compression ratio may specifically include:

[0073] The number of nodes in the fully connected layer of the second neural network model corresponding to the encoder is adjusted to optimize the compression ratio of the encoder.

[0074] Therefore, the encoder compression rate adjustment method provided in this embodiment of the invention analyzes the sparsity of the target data through a first neural network model, and then outputs the optimal compression rate that can balance the feedback accuracy above the accuracy threshold and the feedback overhead, thereby adjusting the encoder compression rate to the optimal compression rate and realizing adaptive dynamic adjustment of the encoder compression rate, making the data compression process more flexible and reliable.

[0075] Figure 4 This invention illustrates a data feedback method provided by an embodiment of the present invention, which can be applied to a user terminal, including:

[0076] Step 410: Adjust the encoder compression rate to the optimal compression rate according to the above encoder compression rate adjustment method;

[0077] Step 420: Compress the target data according to the optimal compression ratio;

[0078] Step 430: Feed back the compressed target data and the preset identification information corresponding to the optimal compression ratio to the base station;

[0079] The base station is used to determine the optimal compression ratio based on preset identification information, and then decompresses the compressed target data using the optimal compression ratio.

[0080] In an exemplary embodiment, the target data mentioned above may be downlink channel state information.

[0081] Since the information between the uplink and downlink channels is not directly reciprocal in a frequency-division duplex (FDD) system, the base station cannot directly obtain the downlink channel through the uplink channel. The user terminal needs to provide feedback on the channel to the base station. Therefore, the feedback of downlink channel status information (CSI) is the key to a frequency-division duplex massive multiple-input multiple-output (MIMO) system.

[0082] Thanks to the development of artificial intelligence technology, wireless communication technologies for 6G are beginning to adopt a feedback display CSI (Code Image Submission) scheme. This involves using an autoencoder at the terminal to compress the displayed CSI into a codeword. The user terminal feeds this codeword back to the base station, where a decoder reconstructs the displayed CSI. By utilizing the CSI fed back from the user terminal, the base station can implement more efficient beamforming algorithms and mitigate channel effects.

[0083] However, traditional encoders are only suitable for fixed compression rate feedback. Since the communication environment is constantly changing, relying on a fixed compression rate may lead to unnecessary bits and wasted resources.

[0084] Considering that in changing environments, a higher compression ratio should be used when the channel matrix is ​​sparse, and vice versa, this embodiment adjusts the encoder's compression ratio according to the sparsity of the downlink channel state information. The downlink channel state information is compressed using the optimal compression ratio, and the preset identifier information corresponding to the optimal compression ratio, along with the compressed codeword, is fed back to the base station. The base station decompresses the codeword using the optimal compression ratio to restore the information, thus realizing an adaptive multi-compression ratio downlink state information feedback process.

[0085] In practical applications, taking downlink channel state information as the target data as an example, the user terminal can provide feedback of the preset identification information corresponding to the optimal compression ratio along with the compressed target data. For example, the preset identification information can be the category code corresponding to the compression ratio. The corresponding category codes can be pre-assigned to multiple selectable compression ratios, and the mapping relationship between compression ratio and category code can be pre-configured on the user terminal and the base station side. In this way, the base station can directly determine the corresponding compression ratio based on the received category code.

[0086] When the number of selectable compression ratios is K, the selection of the compression ratio (i.e. the preset identification information corresponding to the optimal compression ratio) requires log2 K bits. Since K is small in this embodiment, usually below 10, the number of bits required for the selection of the compression ratio can often be ignored. Therefore, in the above data feedback process, the preset identification information corresponding to the optimal compression ratio will not consume too many bits.

[0087] See appendix Figure 3 The convolutional layer in the second neural network model corresponding to the encoder on the user equipment side (i.e., the user end) Figure 3 The encoder in a CNN (convolutional neural network) consists of an inception block, i.e. Figure 3 The leftmost convolutional block, ConvBn (Convolutional Batch Norm), has a 3×3 kernel size. This inception block has three branches, each processed by a convolutional kernel of a different size. For larger convolutional kernels, an asymmetric decomposition method is used to decompose them into two smaller convolutional kernels to reduce the number of parameters.

[0088] Figure 3 Of the three branches shown, the first branch consists of three convolutions with kernel sizes of 1×5, 5×1, and 1×1, the second branch also consists of three convolutions with kernel sizes of 1×7, 7×1, and 1×1, and the third branch consists of two convolutions with kernel sizes of 3×3 and 1×1.

[0089] Finally, the outputs of the multiple branches of the Inception block are merged along the channel dimension. In this embodiment, taking downlink channel state information as the target data as an example, since the channel matrix has 2 channels, a 1×1 convolution is needed to compress the number of channels of the merged data to 2 before feeding it into the fully connected layer. Figure 3 The encoder FC (fully connected encoder layer) in the code allows for the configuration of multiple nodes, each corresponding to a different selectable compression ratio. Figure 3 The diagram shows compression ratios CR=4 and CR=32. The compression ratio adjustment process in the encoder is described in... Figure 3 The principle can be viewed as a switch. The selection of the encoder compression rate is determined by the arrow between the optimal compression rate and the node. After compression by the encoder, the codeword S is obtained.

[0090] It is easy to see that in the encoder part of this embodiment, each compression ratio shares a convolutional neural network to save the load of storing network parameters. By adjusting the number of nodes in the fully connected layer in the second neural network model, the encoder compression ratio can be adjusted. Therefore, this encoder can be called an adaptive multi-compression ratio encoder.

[0091] On the base station side, data decompression can be achieved through a decoder. In the decoder part, considering that the base station has a large storage space, different fully connected layers and convolutional neural networks can be used for different compression ratios.

[0092] Specifically, in this embodiment, the decoder part at the base station (BS) end can be configured with multiple third neural network models to achieve decompression processes at different compression ratios. See Appendix. Figure 3 Taking one of the third neural network models as an example, the basic building block of this network model is RDBlock, which outputs the result after GFF (Global Feature Fusion) operation and GRL (Global Residual Learning).

[0093] Each RDBlock consists of three densely connected layers and an attention mechanism module. Figure 3 The three densely connected layers are three 5×5 convolutions. The attention mechanism module is specifically an SE (Squeeze-and-Excitation) module, which mainly includes an average pooling layer and a fully connected layer (FC). The attention mechanism module leverages the significant differences between rows of the processed data. Specifically, in the application of the channel matrix, this characteristic inherits the sparsity of the channel matrix in the time-delay domain. By applying weights to each row, the neural network can focus more attention on the more information-rich rows, which correspond to the non-zero rows of the original channel matrix.

[0094] Simultaneously, dense connections are employed between each RDBlock to improve the performance of the neural network. Finally, the three densely connected layers and the attention mechanism module output data after LRL (Local Residual Learning). In this embodiment, the decompressed data is obtained after being decompressed by the decoder on the base station side.

[0095] When training the above neural network, the second neural network model in the autoencoder needs to be trained separately from the first neural network model. Taking the target data as the channel matrix as an example, the training process can specifically include:

[0096] First, obtain multiple channel matrix samples with different sparsity, and use these samples to train the second neural network model to ensure that the encoders at each compression rate are fully trained.

[0097] Secondly, the channel matrix samples are tested in a way that balances feedback accuracy and feedback overhead to determine the optimal compression ratio of each channel matrix sample. The samples are then labeled according to their optimal compression ratios to obtain the channel matrix samples with the optimal compression ratio labels.

[0098] Finally, the labeled channel matrix samples are fed into the neural network for training, resulting in the first trained neural network model.

[0099] In practical applications, taking downlink channel state information as the target data as an example, the feedback method for the above data can specifically include:

[0100] Step 1: The frequency division duplex system operates using an adaptive compression rate.

[0101] Step 2: The user terminal transforms the downlink channel state information in the spatial frequency domain to the angle delay domain, and compresses the downlink channel state information according to the corresponding compression ratio;

[0102] Step 3: The user terminal sends the compressed codewords and the category code corresponding to the optimal compression ratio back to the base station;

[0103] Step 4: The base station decompresses the downlink channel state information according to the compression rate information corresponding to the received codeword and category code, and restores the original downlink channel state information;

[0104] Step 5: The base station uses the recovered downlink channel state information to perform beamforming, and after a period of time notifies the user terminal to provide the next channel information feedback, then returns to step 1.

[0105] Therefore, the data feedback method provided in this embodiment of the invention automatically adjusts the encoder compression rate according to changes in the environment through the first neural network model, so that the data to be fed back can be compressed at the optimal compression rate and fed back to the base station. This method has good adaptability and utilizes a flexible and reliable compression link to minimize feedback overhead while ensuring a certain level of feedback accuracy.

[0106] The encoder compression ratio adjustment device provided by the present invention will be described below. The encoder compression ratio adjustment device described below and the encoder compression ratio adjustment method described above can be referred to in correspondence.

[0107] Figure 5 An encoder compression rate adjustment device according to an embodiment of the present invention is shown, the device comprising:

[0108] The first processing module 510 is used to input the target data into a pre-constructed first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on the target data samples labeled with the optimal compression ratio.

[0109] The adjustment module 520 is used to adjust the compression ratio of the encoder to the optimal compression ratio.

[0110] In an exemplary embodiment, the first processing module 510 described above can be specifically used for:

[0111] The sparsity of the target data is analyzed to obtain the sparsity analysis results;

[0112] Based on the sparsity analysis results, the optimal compression ratio of the target data is determined.

[0113] Furthermore, the first processing module 510 described above can specifically determine the optimal compression ratio of the target data based on the sparsity analysis results in the following ways:

[0114] Based on the sparsity analysis results, the theoretical compression ratio with the lowest feedback bit count and the highest feedback accuracy among several preset theoretical compression ratios is selected as the optimal compression ratio.

[0115] In an exemplary embodiment, the encoder compression rate adjustment device described above may further include:

[0116] The model training module is used to acquire multiple target data samples with different sparsity and determine the optimal compression ratio label for each target data sample; the pre-built neural network is trained using the target data samples labeled with the optimal compression ratio to obtain the first neural network model.

[0117] Furthermore, the aforementioned model training module can specifically train the pre-built neural network using target data samples labeled with the optimal compression ratio, as follows:

[0118] By introducing a weighting term into the loss function of the neural network, a weighted loss function is obtained.

[0119] Based on the weighted loss function, the neural network is trained using target data samples labeled with the optimal compression ratio.

[0120] In an exemplary embodiment, the adjustment module 520 can be specifically used to: adjust the number of nodes in the fully connected layer of the second neural network model corresponding to the encoder, so as to adjust the compression ratio of the encoder to the optimal compression ratio.

[0121] Figure 6An embodiment of the present invention illustrates a data feedback device, which includes:

[0122] The second processing module 610 is used to adjust the encoder compression rate to the optimal compression rate using the above-mentioned encoder compression rate adjustment method.

[0123] Compression module 620 is used to compress the target data at the optimal compression ratio;

[0124] Feedback module 630 is used to feed back the compressed target data and the preset identification information corresponding to the optimal compression ratio to the base station;

[0125] The base station is used to determine the optimal compression ratio based on preset identification information, and then decompresses the compressed target data using the optimal compression ratio.

[0126] In an exemplary embodiment, the target data mentioned above may be downlink channel state information.

[0127] Therefore, the data feedback device provided in this embodiment of the invention can reduce unnecessary bits and resource waste during data compression and feedback by adjusting the encoder's compression rate to the optimal compression rate, thus better meeting the needs of practical applications.

[0128] Figure 7 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 7 As shown, the electronic device may include a processor 710, a communications interface 720, a memory 830, and a communication bus 740, wherein the processor 710, communications interface 720, and memory 730 communicate with each other via the communication bus 740. The processor 710 can call logical instructions in the memory 730 to execute a method for adjusting the encoder compression rate. This method includes: inputting target data into a pre-constructed first neural network model to obtain the optimal compression rate of the target data; wherein the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression rate; and adjusting the encoder compression rate to the optimal compression rate.

[0129] Furthermore, the logical instructions in the aforementioned memory 730 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0130] On the other hand, the present invention also provides a computer program product, the computer program product including a computer program stored on a non-transitory computer-readable storage medium, the computer program including program instructions, and when the program instructions are executed by a computer, the computer is able to execute the encoder compression rate adjustment method provided in the above embodiments, the method including: inputting target data into a pre-constructed first neural network model to obtain the optimal compression rate of the target data; wherein, the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression rate; and adjusting the encoder compression rate to the optimal compression rate.

[0131] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the encoder compression rate adjustment method provided in the above embodiments. The method includes: inputting target data into a pre-constructed first neural network model to obtain the optimal compression rate of the target data; wherein the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression rate; and adjusting the encoder compression rate to the optimal compression rate.

[0132] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0133] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0134] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for adjusting the compression ratio of an encoder, characterized in that, include: The target data is input into a pre-built first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression ratio. Adjust the encoder's compression ratio to the optimal compression ratio; The step of inputting the target data into a pre-built first neural network model to obtain the optimal compression ratio of the target data includes: The sparsity of the target data is analyzed to obtain the sparsity analysis results; Based on the sparsity analysis results, the theoretical compression ratio with the highest feedback accuracy and the smallest number of feedback bits among the preset theoretical compression ratios is selected as the optimal compression ratio. Adjusting the encoder's compression ratio to the optimal compression ratio includes: The number of nodes in the fully connected layer of the second neural network model corresponding to the encoder is adjusted to adjust the compression ratio of the encoder to the optimal compression ratio; The number of feedback bits is negatively correlated with the sparsity of the target data; different compression ratios share a single convolutional neural network in the encoder.

2. The encoder compression rate adjustment method according to claim 1, characterized in that, The training process of the first neural network model includes: Multiple target data samples with different sparsity are obtained, and the optimal compression ratio label corresponding to each target data sample is determined respectively; The first neural network model is obtained by training a pre-built neural network with target data samples labeled with the optimal compression ratio.

3. The encoder compression rate adjustment method according to claim 2, characterized in that, The training of the pre-built neural network using target data samples labeled with the optimal compression ratio includes: A weighted term is introduced into the loss function of the neural network to obtain a weighted loss function; Based on the weighted loss function, the neural network is trained using target data samples labeled with the optimal compression ratio.

4. A data feedback method, characterized in that, include: The encoder compression ratio is adjusted to the optimal compression ratio according to the encoder compression ratio adjustment method as described in any one of claims 1 to 3; The target data is compressed according to the optimal compression ratio. The compressed target data and the preset identification information corresponding to the optimal compression ratio are fed back to the base station; The base station is used to determine the optimal compression ratio based on the preset identification information, and to decompress the compressed target data using the optimal compression ratio. The number of bits fed back is negatively correlated with the sparsity of the target data.

5. The data feedback method according to claim 4, characterized in that, The target data is downlink channel state information.

6. A device for adjusting the compression ratio of an encoder, characterized in that, include: The first processing module is used to input the target data into a pre-constructed first neural network model to obtain the optimal compression ratio of the target data; wherein, the first neural network model is obtained by training the neural network based on target data samples labeled with the optimal compression ratio. An adjustment module is used to adjust the compression ratio of the encoder to the optimal compression ratio; The adjustment module is specifically used to analyze the sparsity of the target data and obtain sparsity analysis results. The adjustment module is further configured to, based on the sparsity analysis results, select the theoretical compression ratio with the highest feedback accuracy and the smallest number of feedback bits among a plurality of preset theoretical compression ratios as the optimal compression ratio. The adjustment module is further configured to adjust the number of nodes in the fully connected layer of the second neural network model corresponding to the encoder, so as to adjust the compression ratio of the encoder to the optimal compression ratio. The number of feedback bits is negatively correlated with the sparsity of the target data; different compression ratios share a single convolutional neural network in the encoder.

7. A data feedback device, characterized in that, include: The second processing module is used to adjust the encoder compression rate to the optimal compression rate according to the encoder compression rate adjustment method as described in any one of claims 1 to 3. A compression module is used to compress the target data according to the optimal compression ratio; The feedback module is used to feed back the compressed target data and the preset identification information corresponding to the optimal compression ratio to the base station; The base station is used to determine the optimal compression ratio based on the preset identification information, and to decompress the compressed target data using the optimal compression ratio. The number of bits fed back is negatively correlated with the sparsity of the target data.

Images