Joint source-channel coding channel state information feedback method based on dynamic resource selection

Abstract

The application discloses a joint source channel coding channel state information feedback method based on dynamic resource selection. A base station deploys a resource selection network to evaluate each candidate uplink resource by using a history reconstructed downlink channel state information sequence and a history estimated uplink channel state information sequence, and outputs a target uplink resource to a user terminal. The user terminal performs joint source channel coding on the channel state information on the resource, and sends an uplink feedback signal. The base station receives and decodes the downlink channel state information to reconstruct the downlink channel state information, and writes the reconstruction result and the estimated uplink channel state information into a history cache to form an online closed loop deployment. The application jointly trains and optimizes the resource selection network, an encoder and a decoder, jointly optimizes reconstruction, selection and sorting losses, and improves resource selection and channel reconstruction accuracy. Compared with no dynamic resource selection, the application can reduce feedback loss and improve closed loop performance.

Description

Technical Field

[0001] This invention relates to a joint source channel coding channel state information feedback method based on dynamic resource selection, belonging to the field of communication technology. Background Technology

[0002] In mobile communication closed-loop transmission systems, base stations typically need to optimize the downlink based on channel state information fed back from the user end. This optimization may involve precoding, beamforming, modulation and coding configuration, or resource reselection for the next feedback cycle. Joint source channel coding methods can directly map the channel state information to be fed back into feedback codewords suitable for uplink transmission, and perform decoding and reconstruction at the base station, thus showing promising application prospects.

[0003] However, most existing joint source channel coding channel state information feedback schemes focus on optimizing the coding and decoding process under given uplink transmission resources, while neglecting to consider which set of uplink resources should be selected for feedback transmission in the current feedback period. Without considering the uplink transmission resource selection issue, it is difficult to adapt to transmission performance fluctuations caused by changes in uplink state at different times, resulting in excessive feedback loss, decreased reconstruction accuracy, and impact on transmission performance. Therefore, how to achieve dynamic resource selection in the current feedback period using historical information while maintaining feedback performance, and further improve resource selection accuracy and feedback reconstruction accuracy through joint training, is a pressing technical problem that needs to be solved. Summary of the Invention

[0004] Objective: This invention proposes a joint source channel coding (JCD) channel state information feedback method based on dynamic resource selection. A resource selection network is deployed at the base station. Using historical reconstructed downlink channel state information sequences and historical estimated uplink channel state information sequences, candidate uplink resources for the current feedback period are evaluated, and a target uplink resource is output. The user terminal performs JCD feedback transmission on the target uplink resource, and the base station receives and decodes the reconstructed downlink channel state information for the current feedback period. Simultaneously, the reconstructed downlink channel state information and the estimated uplink channel state information for the current feedback period are written into a historical cache, thus forming an online closed-loop deployment process for subsequent feedback periods. Furthermore, this invention employs a joint training method to uniformly optimize the resource selection network, the JCD encoder, and the JCD decoder. This ensures that the resource selection result is simultaneously constrained by the channel state information reconstruction performance and the uplink resource transmission performance, thereby reducing feedback transmission losses and improving closed-loop feedback performance.

[0005] Technical solution: The present invention provides a joint source channel coding channel state information feedback method based on dynamic resource selection, comprising the following steps:

[0006] (1) The base station extracts the historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence from the historical buffer;

[0007] (2) The base station inputs the historical reconstructed downlink channel state information sequence, the historical estimated uplink channel state information sequence and the estimated uplink channel state information of the current feedback period into the resource selection network, outputs the resource evaluation results corresponding to each candidate uplink resource in the current feedback period, and determines the target uplink resource;

[0008] (3) The base station configures or indicates the target uplink resources to the user terminal, and the user terminal performs joint source channel coding on the current channel state information on the target uplink resources to obtain the uplink feedback signal and sends it to the base station.

[0009] (4) The base station receives the uplink feedback signal on the target uplink resource and reconstructs the current channel state information through the joint source channel decoder to obtain the reconstructed downlink channel state information for the current feedback period;

[0010] (5) The base station writes the reconstructed downlink channel state information of the current feedback period and the estimated uplink channel state information of the current feedback period into the historical cache, and repeats steps (1) to (4) in the next feedback period to form an online closed-loop deployment process.

[0011] The resource evaluation results output by the resource selection network represent the estimated transmission loss of each candidate uplink resource in the current feedback period. The target uplink resource is determined according to the following formula:

[0012]

[0013] in, Represents the set of candidate uplink resources. This represents the estimated transmission loss value corresponding to the candidate uplink resource r. Using the above method, a more suitable uplink resource can be adaptively selected for the current channel state information feedback transmission based on historical feedback results and historical uplink channel state information.

[0014] The resource selection network includes a historical reconstructed downlink channel state information coding branch, a historical estimated uplink channel state information coding branch, a feature fusion module, and a resource scoring head. The historical reconstructed downlink channel state information coding branch is used to extract the temporal features of the historical reconstructed downlink channel state information sequence, the historical estimated uplink channel state information coding branch is used to extract uplink related features, the feature fusion module is used to fuse downlink related features and uplink related features, and the resource scoring head is used to output the resource evaluation results of each candidate uplink resource.

[0015] The joint source channel coding feedback transmission at the user terminal on the target uplink resource includes: mapping the current channel state information into encoded complex-valued or real-valued feedback codewords, transmitting them to the base station through the uplink channel corresponding to the selected target uplink resource, and having the base station recover the reconstructed downlink channel state information for the current feedback period upon receiving the information. The currently estimated uplink channel state information can be obtained by the base station via uplink pilots at a time close to the current feedback period and used for resource selection in the next feedback period.

[0016] Furthermore, the method employs a joint training approach to train the resource selection network, joint source channel encoder, and joint source channel decoder, ensuring that the selection result of the target uplink resource is simultaneously constrained by the reconstruction performance of the current channel state information and the transmission performance of the uplink resource. The joint training includes: constructing training samples, which include a historical reconstructed downlink channel state information sequence, a historical estimated uplink channel state information sequence, and current channel state information; inputting the historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence into the resource selection network to obtain resource evaluation results corresponding to each candidate uplink resource; determining the target uplink resource to be used during training based on the resource evaluation results, and performing joint source channel coding, channel transmission, and joint source channel decoding of the current channel state information on the target uplink resource to obtain the reconstructed downlink channel state information for the current feedback period; and jointly optimizing the resource selection network, joint source channel encoder, and joint source channel decoder based on the difference between the reconstructed downlink channel state information for the current feedback period and the current channel state information, as well as the resource evaluation results corresponding to the candidate uplink resources.

[0017] The joint training employs the following total loss function:

[0018]

[0019] in, This represents the reconstruction loss between the reconstructed downlink information and the current channel state information during the current feedback period. Indicates the loss of resource selection. This represents the ranking loss of candidate uplink resources. , , The weighting coefficients are used. The resource selection loss is obtained by transmitting the current channel state information on the candidate uplink resources and calculating the reconstruction loss label corresponding to each candidate uplink resource; the candidate uplink resource ranking loss is constructed according to the size pattern of the reconstruction loss label corresponding to each candidate uplink resource, and is used to constrain the resource evaluation results output by the resource selection network to be consistent with the actual performance ranking of each candidate uplink resource.

[0020] Furthermore, the joint training can adopt a training framework of teacher model and student model. The teacher model is used to periodically update the reconstruction loss label and ranking label corresponding to the candidate uplink resources, and the student model is used to update the network parameters of the resource selection network, the joint source channel encoder and the joint source channel decoder. The teacher model parameters are updated according to the exponential moving average of the student model parameters.

[0021] Compared with existing technologies, the advantages of this invention are as follows: This invention introduces a dynamic resource selection mechanism into the joint source channel coding channel state information feedback link. The base station uses historically reconstructed downlink channel state information sequences and historically estimated uplink channel state information sequences to evaluate candidate uplink resources in the current feedback period and outputs the target uplink resource. This avoids the problem of insufficient adaptability of transmission methods without dynamic resource selection to time-varying uplink states and reduces channel state information feedback transmission losses. Simultaneously, this invention performs joint source channel coding feedback transmission of the current channel state information on the selected target uplink resource and writes back the reconstructed downlink channel state information and the currently estimated uplink channel state information of the current feedback period to the historical buffer, forming a continuous online closed loop between resource selection and channel state information feedback. Furthermore, this invention optimizes the resource selection network and the joint source channel coding feedback link under a unified objective through joint training, simultaneously considering the accuracy of channel state information reconstruction and the performance of dynamic resource selection, thereby improving the overall effect of subsequent closed-loop transmission optimization. Attached Figure Description

[0022] Figure 1 This is a flowchart of the online closed-loop deployment process of the present invention;

[0023] Figure 2 This is a specific implementation of a resource selection network in one embodiment of the present invention;

[0024] Figure 3 This is a flowchart of the joint training process of the present invention. Detailed Implementation

[0025] The embodiments of the present invention will now be described with reference to the accompanying drawings.

[0026] This invention presents a joint source channel coding channel state information feedback method based on dynamic resource selection. In this method, the base station deploys a resource selection network and uses historical reconstructed downlink channel state information sequences and historical estimated uplink channel state information sequences cached in historical feedback cycles to evaluate candidate uplink resources for the current feedback cycle and output the target uplink resource. The user terminal performs joint source channel coding feedback transmission of the current channel state information on the target uplink resource. The base station receives and decodes the reconstructed downlink channel state information for the current feedback cycle. Simultaneously, the base station writes the reconstructed downlink channel state information and the currently estimated uplink channel state information for the current feedback cycle into a historical cache for resource selection in the next feedback cycle, thus forming an online closed-loop deployment process. Compared with transmission methods without dynamic resource selection, this invention can dynamically adjust feedback resources based on historical feedback results and uplink status, reducing channel state information feedback loss and improving closed-loop feedback performance.

[0027] In one embodiment, consider a closed-loop transmission system where the base station and user terminal exchange channel state information via an uplink feedback link. The candidate uplink resource set consists of multiple resource blocks, used to carry the joint source channel-coded feedback signal sent by the user terminal. In a specific implementation, the candidate uplink resources can be set to 20 resource blocks, and the historical window length can be set to H=3, meaning that before feedback, the base station uses historical information cached from the previous H=3 feedback cycles to determine the target uplink resource for the current feedback cycle.

[0028] The method is implemented according to the following steps:

[0029] (1) Online closed-loop deployment process: such as Figure 1 As shown, the online closed-loop deployment process of this invention involves two main entities: the base station and the user. The base station side includes a historical cache, a resource selection network, a demodulation / equalization module, and a joint source channel decoder; the user side includes a joint source channel encoder and a modulation module. Before feedback begins, the base station first extracts the historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence from the historical cache for the previous H feedback cycles. In this embodiment, the base station maintains these two types of information sequences through the historical cache, so that the resource selection in the current feedback cycle no longer depends solely on a single observation at the current moment, but rather on a decision based on historical feedback quality and historical uplink state. This closed-loop structure of historical cache—resource selection—feedback reconstruction—write-back to cache is consistent with the deployment approach discussed earlier, which involves retrieving historical information from the cache at the online moment and then making resource decisions.

[0030] Subsequently, the base station inputs the historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence into the resource selection network, along with the estimated uplink channel state information for the current feedback period. The resource selection network outputs the resource evaluation results for each candidate uplink resource in the current feedback period and determines the target uplink resource based on these results. Determine according to the following formula:

[0031]

[0032] in, Represents the set of candidate uplink resources. This represents the estimated transmission loss value corresponding to the candidate uplink resource r. Through this method, the base station can select the target resource that is more conducive to the transmission and reconstruction of current channel state information feedback from multiple candidate uplink resources.

[0033] After determining the target uplink resource, the base station configures or indicates the target uplink resource to the user terminal. The user terminal performs joint source channel coding on the current channel state information on the target uplink resource, and then generates an uplink feedback signal through the modulation module and sends it to the base station. The joint source channel encoder can directly map the current channel state information into complex or real-valued feedback codewords, and transmit them through the uplink channel corresponding to the target uplink resource.

[0034] After receiving the uplink feedback signal on the target uplink resource, the base station first processes it through a demodulation / equalization module, then sends it to a joint source channel decoder to recover the reconstructed downlink channel state information for the current feedback cycle. Based on the reconstruction result, the base station can perform subsequent closed-loop transmission optimizations, such as precoding, beamforming, modulation and coding configuration, or resource reselection for the next feedback cycle. Simultaneously, the base station can also obtain the currently estimated uplink channel state information through the uplink pilot signal close to the current feedback cycle, and write the currently reconstructed downlink channel state information and the currently estimated uplink channel state information back to the historical buffer for resource selection in the next feedback cycle, thereby forming... Figure 1 The online closed-loop deployment process is shown.

[0035] (2) Resource selection network structure:

[0036] In one embodiment, the resource selection network includes a historical reconstruction downlink channel state information coding branch, a historical estimation uplink channel state information coding branch, a feature fusion module, and a resource scoring head.

[0037] The historical reconstructed downlink channel state information encoding branch is used to extract time-series features from the historical reconstructed downlink channel state information sequence; the historical estimated uplink channel state information encoding branch is used to extract uplink correlation features from the historical estimated uplink channel state information sequence; the feature fusion module is used to fuse downlink time-series features and uplink correlation features; and the resource scoring head is used to output the resource evaluation results corresponding to each candidate uplink resource.

[0038] In a specific implementation, such as Figure 2 As shown, the historical reconstruction downlink channel state information encoding branch can adopt a structure combining temporally distributed convolutional layers and bidirectional long short-term memory networks (BiLSTM). For example, convolutional feature extraction is first performed on the downlink information of each historical feedback cycle, and then the temporal information across feedback cycles is aggregated through the bidirectional long short-term memory network. The historical estimation uplink channel state information encoding branch can use convolutional layers and global pooling layers to extract uplink-related features. The feature fusion module can use splicing, gated fusion, or weighted fusion to map downlink temporal features and uplink-related features to a unified selection feature space. The resource scoring head outputs the transmission loss estimate corresponding to each candidate uplink resource through a fully connected layer.

[0039] (3) Joint training process:

[0040] This invention employs a joint training approach to uniformly optimize the resource selection network, joint source-channel encoder, and joint source-channel decoder.

[0041] like Figure 3 As shown, the joint training process begins with the construction of training samples under offline joint training. During the training phase, training samples are first constructed. Each training sample includes: a sequence of historically reconstructed downlink channel state information for the previous H feedback cycles, a sequence of historically estimated uplink channel state information for the previous H feedback cycles, and the current channel state information. Subsequently, the sequence of historically reconstructed downlink channel state information and the sequence of historically estimated uplink channel state information are input into the resource selection network to obtain the resource evaluation results corresponding to each candidate uplink resource. The candidate uplink resource with the best evaluation is selected as the target uplink resource used in the training process. The user terminal performs joint source channel coding, channel transmission, and joint source channel decoding of the current channel state information on the target uplink resource, and the base station obtains the reconstructed downlink channel state information for the current feedback cycle.

[0042] In this embodiment, the training loss considers not only the difference between the current reconstruction result and the current channel state information, but also the quality of the resource selection result itself, thereby enabling the resource selection network and the joint source channel coding feedback link to be jointly optimized under a unified objective. In one embodiment, the joint training adopts the following total loss function:

[0043]

[0044] in, This represents the reconstruction loss between the reconstructed downlink channel state information in the current feedback period and the current channel state information. Indicates the loss of resource selection. This represents the ranking loss of candidate uplink resources. , , These are the weighting coefficients.

[0045] like Figure 3 As shown, the joint loss consists of the reconstruction loss L rec Resource selection loss L sel Sorting loss L rank The loss is calculated by combining the actual performance label with candidate resource information. Further, the resource selection loss is obtained by transmitting the current channel state information on each candidate uplink resource and calculating the corresponding reconstruction loss label. Let the candidate uplink resource be r, and the reconstructed downlink channel state information obtained after transmitting the current channel state information on the candidate uplink resource r and performing joint source channel decoding is denoted as r. Then the corresponding reconstruction loss label satisfies:

[0046]

[0047] Where D represents the current channel state information, and d(·) represents the loss calculation function. The ranking loss is constructed based on the size pattern of the reconstruction loss labels corresponding to each candidate uplink resource, and is used to ensure that the resource evaluation results output by the resource selection network are consistent with the actual performance ranking of each candidate uplink resource.

[0048] The parameters are updated based on the joint loss, and the final model is obtained after training. In one embodiment, joint training can adopt a training framework of a teacher model and a student model. The teacher model is used to periodically update the reconstruction loss label and ranking label corresponding to the candidate uplink resources, and the student model is used to update the network parameters of the resource selection network, the joint source-channel encoder, and the joint source-channel decoder. The teacher model parameters are updated according to the exponential moving average of the student model parameters.

[0049] The embodiments are merely illustrative of the technical concept of the present invention and should not be construed as limiting the scope of protection of the present invention. Any modifications made to the technical solution based on the technical concept proposed in this invention shall fall within the scope of protection of this invention.

Claims

1. A joint source channel coding channel state information feedback method based on dynamic resource selection, characterized in that, Includes the following steps: (1) The base station extracts the historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence from the historical buffer; (2) The base station inputs the historical reconstructed downlink channel state information sequence, the historical estimated uplink channel state information sequence and the estimated uplink channel state information of the current feedback period into the resource selection network, outputs the resource evaluation results corresponding to each candidate uplink resource in the current feedback period, and determines the target uplink resource; (3) The base station configures or indicates the target uplink resources to the user terminal, and the user terminal performs joint source channel coding on the current channel state information on the target uplink resources to obtain the uplink feedback signal and sends it to the base station. (4) The base station receives the uplink feedback signal on the target uplink resource and reconstructs the current channel state information through the joint source channel decoder to obtain the reconstructed downlink channel state information for the current feedback period; (5) The base station writes the reconstructed downlink channel state information of the current feedback period and the estimated uplink channel state information of the current feedback period into the historical cache, and repeats steps (1) to (4) in the next feedback period to form an online closed-loop deployment process; The resource selection network includes a historical reconstructed downlink channel state information coding branch, a historical estimated uplink channel state information coding branch, a feature fusion module, and a resource scoring head. The historical reconstructed downlink channel state information coding branch is used to extract the temporal features of the historical reconstructed downlink channel state information sequence, the historical estimated uplink channel state information coding branch is used to extract uplink channel state-related features, the feature fusion module is used to fuse downlink-related features and uplink-related features, and the resource scoring head is used to output the resource evaluation results of each candidate uplink resource. The resource selection network, joint source channel encoder, and joint source channel decoder are trained using a joint training method, so that the selection result of the target uplink resource is simultaneously constrained by the current channel state information reconstruction performance and the uplink resource transmission performance. The joint training employs the following total loss function: ； in, This represents the reconstruction loss between the reconstructed downlink channel state information in the current feedback period and the current channel state information. Indicates the loss of resource selection. This represents the ranking loss of candidate uplink resources. , , These are the weighting coefficients.

2. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The resource evaluation results output by the resource selection network represent the estimated transmission loss of each candidate uplink resource in the current feedback period, and the target uplink resource... Determined according to the following formula: ； in, Represents the set of candidate uplink resources. This represents the estimated transmission loss value corresponding to the candidate uplink resource r.

3. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The joint source channel coding feedback transmission of the user terminal on the target uplink resource includes: mapping the current channel state information into encoded complex or real-valued feedback codewords, sending them to the base station through the uplink channel corresponding to the selected target uplink resource, and having the base station restore the reconstructed downlink channel state information for the current feedback period after receiving it.

4. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The joint training includes: Construct training samples, which include a sequence of historical reconstructed downlink channel state information, a sequence of historical estimated uplink channel state information, and current channel state information; The historical reconstructed downlink channel state information sequence and the historical estimated uplink channel state information sequence are input into the resource selection network to obtain the resource evaluation results corresponding to each candidate uplink resource. Based on the resource evaluation results, the target uplink resources to be used in the training process are determined, and joint source channel coding, channel transmission and joint source channel decoding of the current channel state information are performed on the target uplink resources to obtain the reconstructed downlink channel state information of the current feedback period. Based on the difference between the reconstructed downlink channel state information and the current channel state information in the current feedback period, and the resource evaluation results corresponding to the candidate uplink resources, the resource selection network, the joint source channel encoder, and the joint source channel decoder are jointly optimized.

5. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The resource selection loss is obtained by transmitting the current channel state information on the candidate uplink resources and calculating the reconstruction loss label corresponding to each candidate uplink resource.

6. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The candidate uplink resource ranking loss is constructed based on the size relationship of the reconstruction loss labels corresponding to each candidate uplink resource, and is used to constrain the resource evaluation results output by the resource selection network to be consistent with the actual performance ranking of each candidate uplink resource.

7. The joint source channel coding channel state information feedback method based on dynamic resource selection according to claim 1, characterized in that, The joint training adopts a training framework of teacher model and student model. The teacher model is used to periodically update the reconstruction loss label and ranking label corresponding to the candidate uplink resources, and the student model is used to update the network parameters of the resource selection network, the joint source channel encoder and the joint source channel decoder. The teacher model parameters are updated according to the exponential moving average of the student model parameters.

Images