Semantic-aware HARQ communication method and device, electronic equipment and storage medium

Abstract

The invention provides a semantic-perceived HARQ communication method and device, electronic equipment and a storage medium, and the method comprises the steps: carrying out the retransmission detection of a plurality of received transmission semantic features through a receiving end, feeding back the distortion degree evaluation data generated by a semantic feature distortion evaluation network model to a transmitting end when a retransmission detection result is determined that the detection is not passed, and carrying out the detection of the transmission semantic features. According to the method and the device, the distorted transmission semantic features which are identified by the sending end are transmitted again, so that the plurality of transmission semantic features received by the receiving end can meet certain accuracy, and then decoding features corresponding to the plurality of transmission semantic features are predicted by utilizing a prediction algorithm of at least one preset type of semantic task. And at least one more accurate prediction processing result can be obtained, so that the accuracy of the executed semantic task is guaranteed.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a semantically-aware HARQ communication method, device, electronic device and storage medium. Background Art

[0002] Semantic communication is a communication method that is different from grammatical communication. It aims to transmit the meaning of information and improve communication efficiency, and can meet the needs of intelligent communication. However, due to the influence of noisy transmission channels, the accuracy of the transmitted information received by the receiver will be reduced, thereby affecting the accuracy of the semantic tasks performed in the corresponding application scenarios. Summary of the invention

[0003] In view of this, the purpose of the present application is to propose a semantically-aware HARQ communication method, device, electronic device and storage medium to solve the above-mentioned technical problems.

[0004] Based on the above purpose, a first aspect of the present application provides a semantically aware HARQ communication method, which is applied to a semantically aware HARQ communication system, the system including a transmitting end and a receiving end, and the method including:

[0005] The sending end obtains a transmission source data set, generates a plurality of transmission semantic features based on the transmission source data set, and sends the plurality of transmission semantic features to the receiving end;

[0006] The receiving end performs retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result;

[0007] In response to the receiving end determining that the retransmission detection result is failed, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can determine the transmission semantic features to be retransmitted based on the binary vector and resend them to the receiving end. At the same time, when it is determined that the transmission channel meets the preset transmission conditions, multiple new transmission semantic features are incrementally transmitted to the receiving end. The receiving end repeatedly performs the retransmission detection process based on the received multiple transmission semantic features until it is determined that the retransmission detection result is passed, or the preset maximum number of retransmissions is reached; or,

[0008] In response to the receiving end determining that the retransmission detection result is a passed detection, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

[0009] Optionally, sending the multiple transmission semantic features to the receiving end includes:

[0010] The transmitting end obtains feature importance values ​​corresponding to the multiple transmission semantic features;

[0011] The transmitting end determines a plurality of transmission semantic features to be transmitted in each transmission process according to the feature importance value, and sends the determined plurality of transmission semantic features to be transmitted to the receiving end in each transmission process.

[0012] Optionally, the receiving end performs retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result, including:

[0013] The receiving end determines a received signal-to-noise ratio corresponding to the received multiple transmission semantic features, and a cascade semantic feature corresponding to the received multiple transmission semantic features;

[0014] The receiving end determines, based on the cascade semantic feature, uncertainty values ​​corresponding to the received multiple transmission semantic features through an uncertainty algorithm;

[0015] The receiving end converts the uncertainty value into a comparison value corresponding to the received signal-to-noise ratio using a preset conversion coefficient;

[0016] The receiving end compares the received signal-to-noise ratio with the comparison value to obtain a comparison result, and determines the retransmission detection result according to the comparison result.

[0017] Optionally, determining the retransmission detection result according to the comparison result includes:

[0018] In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is less than or equal to the comparison value, determining that the retransmission detection result is a failure to pass the detection; or,

[0019] In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is greater than the comparison value, the retransmission detection result is determined to be a pass detection.

[0020] Optionally, the training process of the semantic feature distortion assessment network model includes:

[0021] The receiving end obtains original semantic features for training and received semantic features for training corresponding to the original semantic features for training, and constructs an initial model, wherein the initial model includes an initial distortion metric module and a gradient estimator module;

[0022] The receiving end inputs the original semantic features for training and the received semantic features for training into the initial distortion measurement module, and the initial distortion measurement module outputs a first contribution degree parameter and a second contribution degree parameter;

[0023] The receiving end constructs a training objective function based on the first contribution degree parameter and the second contribution degree parameter, and performs training adjustment on the initial distortion metric module according to a maximization processing result obtained in the process of maximizing the training objective function, so as to obtain a trained initial distortion metric module;

[0024] The receiving end obtains the received semantic features for training, and performs back-propagation training on the trained initial distortion metric module through the gradient estimator module based on the received semantic features for training, so as to obtain the semantic feature distortion assessment network model.

[0025] Optionally, the transmitting end determines the transmission semantic feature to be retransmitted based on the binarized vector, including:

[0026] The transmitting end searches for the transmission semantic feature whose binarization vector is a first preset value from various transmission semantic features, and uses the transmission semantic feature whose binarization vector is the first preset value as the transmission semantic feature to be retransmitted.

[0027] Optionally, after the receiving end determines that the retransmission detection result fails the detection, the method further includes:

[0028] The receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can search for the transmission semantic feature whose binarized vector is a second preset value from each transmission semantic feature, and use the transmission semantic feature whose binarized vector is the second preset value as the transmission semantic feature that is not retransmitted.

[0029] Based on the same inventive concept, a second aspect of the present application provides a semantically aware HARQ communication device, wherein the device is arranged in a semantically aware HARQ communication system, wherein the system includes a transmitting end and a receiving end, and the device includes:

[0030] The transmitting end is configured to obtain a transmission source data set, generate a plurality of transmission semantic features based on the transmission source data set, and send the plurality of transmission semantic features to the receiving end;

[0031] The receiving end is configured to perform retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result; in response to determining that the retransmission detection result is not passed, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end determines the transmission semantic features to be retransmitted based on the binary vector and resends it to the receiving end, and when it is determined that the transmission channel meets the preset transmission conditions, incrementally transmits multiple new transmission semantic features to the receiving end, and the receiving end repeatedly performs the retransmission detection process based on the received multiple transmission semantic features until it is determined that the retransmission detection result is passed, or a preset maximum number of retransmissions is reached; or, in response to determining that the retransmission detection result is passed, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

[0032] Based on the same inventive concept, the third aspect of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable by the processor, wherein the processor implements the method described in the first aspect above when executing the computer program.

[0033] Based on the same inventive concept, the fourth aspect of the present application provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable a computer to execute the method described in the first aspect above.

[0034] From the above, it can be seen that the semantic-aware HARQ communication method, device, electronic device and storage medium provided by the present application, the receiving end performs retransmission detection on the received multiple transmission semantic features, and when it is determined that the retransmission detection result is not passed, on this basis, the distortion assessment data generated by the semantic feature distortion assessment network model is fed back to the sending end, so that the sending end can retransmit the identified distorted transmission semantic features to be retransmitted, so that the multiple transmission semantic features received at the receiving end can meet a certain accuracy, and then use at least one preset type of semantic task prediction algorithm to predict the decoding features corresponding to the multiple transmission semantic features, and obtain at least one relatively accurate prediction processing result, thereby ensuring the accuracy of the executed semantic task. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the technical solutions in the present application or related technologies, the drawings required for use in the embodiments or related technical descriptions are briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0036] Figure 1A A schematic diagram of a framework of a semantically-aware HARQ communication system according to an embodiment of the present application;

[0037] Figure 1B A flowchart of a semantically-aware HARQ communication method according to an embodiment of the present application;

[0038] Figure 1C A flowchart of a semantically-aware HARQ mechanism according to an embodiment of the present application;

[0039] Figure 1D A schematic diagram of the training process of the semantic feature distortion assessment network model according to an embodiment of the present application;

[0040] Figure 1E A schematic diagram showing a comparison of vehicle re-identification simulation performances according to an embodiment of the present application;

[0041] Figure 1F A schematic diagram showing a comparison of vehicle color classification simulation performances according to an embodiment of the present application;

[0042] Figure 1G A schematic diagram of vehicle type classification simulation performance comparison of an embodiment of the present application;

[0043] Figure 2 A structural block diagram of a semantically-aware HARQ communication device according to an embodiment of the present application;

[0044] Figure 3 A schematic diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to make the objectives, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below in combination with specific embodiments and with reference to the accompanying drawings.

[0046] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present application should be the usual meanings understood by people with ordinary skills in the field to which the present application belongs. The "first", "second" and similar words used in the embodiments of the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing in front of the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0047] It is understandable that before using the technical solutions of each embodiment of the present application, the type, scope of use, usage scenarios, etc. of the personal information involved will be informed to the user in an appropriate manner, and the user's authorization will be obtained.

[0048] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly remind the user that the operation requested to be performed will require obtaining and using the user's personal information. Thus, the user can independently choose whether to provide personal information to the electronic device, application, server, storage medium or other software or hardware that performs the operation of the technical solution of the present application according to the prompt message.

[0049] As an optional but non-limiting implementation, in response to receiving the user's active request, the prompt information may be sent to the user in the form of a pop-up window, in which the prompt information may be presented in text form. In addition, the pop-up window may also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0050] It is understandable that the above notification and the process of obtaining user authorization are merely illustrative and do not constitute a limitation on the implementation method of this application. Other methods that meet relevant laws and regulations may also be applied to the implementation method of this application.

[0051] The related art proposes a semantic communication system based on incremental transmission, a data importance-aware retransmission system, a wireless communication transmission (Hybrid Automatic Repeat Request, HARQ) semantic communication framework considering incremental transmission, and a generalized semantic communication system implementation method.

[0052] Among them, for the semantic communication system based on incremental transmission: the system transmits the encoded semantic features at a preset bit rate. When the preset upper limit of the number of transmissions is not reached or the semantic task is not completed, the receiving end feeds back negative feedback (Negative Acknowledgement, NACK) to the sending end. At this time, the sending end does not consider the specific distortion of the semantic features, but only sends new incremental semantic features at a certain bit rate until the upper limit of the number of transmissions is reached. Finally, the semantic features obtained from multiple transmissions are spliced ​​and input into the decoder to perform intelligent semantic tasks. However, the system only incrementally transmits new semantic features each time it transmits, and does not consider the situation that the received semantic features may not be successfully decoded and need to be retransmitted, resulting in low communication resource utilization and flexibility. In addition, the system corresponds to different codecs for each transmission, resulting in a complex system structure, and each transmission requires retraining, which increases the training complexity.

[0053] In addition, for the data importance-aware retransmission system: after the information is transmitted to the receiving end, it is determined whether it needs to be retransmitted based on the uncertainty of the received information. If the amount of received information does not reach the threshold, the sending end is notified to retransmit the data, and each transmission is repeated. This system repeatedly transmits the received information and cannot identify the information that can be correctly decoded, which may cause a waste of communication resources. At the same time, it does not consider the transmission of incremental information, which limits the amount of semantic information received and reduces the efficiency of semantic task implementation.

[0054] As for the wireless communication transmission (Hybrid Automatic Repeat Request, HARQ) semantic communication framework considering incremental transmission: the framework designs multiple pairs of codecs for multiple transmissions, and each transmission needs to be retrained. At the same time, the comparative document 3 refers to the traditional cyclic redundancy check (CRC) algorithm and proposes the Sim32 retransmission detection algorithm. Specifically, the transmitter adds 32 bits of redundant information when encoding semantic features, and uses an additional Sim32 decoder at the receiving end to decode the semantic features and redundant information together, and outputs the value of positive feedback (Acknowledgement, ACK) (0 or 1) as the basis for whether to transmit the incremental information again. However, the framework corresponds to different codecs for each transmission, resulting in a complex system structure, and each transmission requires retraining, which increases the training complexity. In addition, only new semantic features are incrementally transmitted each time the transmission is performed, without considering the situation that the received semantic features may not be successfully decoded and need to be retransmitted, resulting in low communication resource utilization and flexibility. The Sim32 retransmission detection algorithm needs to add redundant information after the semantic features, which increases the transmission burden. The use of redundant information for retransmission detection requires the design of additional networks, which increases the complexity of the system. In addition, the detection algorithm can only output instructions on whether retransmission is needed, but cannot detect the specific distortion of the semantic features, and has weak error detection capabilities.

[0055] In addition, for the generalized semantic communication system implementation method: after obtaining the semantic task requirements, a constant bit rate coding transmission is adopted, all semantic features are transmitted without distinction, and then directly decoded at the receiving end without further considering the retransmission gain. However, this method cannot control the transmission bit rate according to the communication conditions, the communication efficiency is not high, and it is impossible to select semantic features that are highly related to the semantic task at the sending end for transmission, and it is impossible to achieve elastic coding that is highly coupled with the semantic task. The receiving end only sets up a decoder to perform error correction on the received semantic features, without considering the role of HARQ transmission in improving the system's error correction and detection capabilities, and does not utilize the performance gain brought by retransmission.

[0056] Compared with the semantic communication system based on incremental transmission and the wireless communication transmission (Hybrid Automatic Repeat Request, HARQ) semantic communication framework considering incremental transmission in the related technology, the semantic-aware HARQ communication system constructed by the present application combines feature retransmission with incremental transmission in the order of semantic importance, and retransmits distorted features while supplementing them with incremental transmission, rather than just continuously transmitting incremental information, thereby improving the reception accuracy of important features and thus improving multi-task performance.

[0057] Compared with the data importance-aware retransmission system in the related art, the present application designs a semantic-aware HARQ mechanism. Compared with the data importance-aware retransmission system in the related art that retransmits all features, the present application does not directly retransmit all features, but selects features with severe distortion and that really need to be retransmitted based on the distortion conditions of different features, thereby greatly improving the retransmission efficiency.

[0058] In addition, compared with the generalized semantic communication system implementation method in the related art, the present application performs elastic retransmission of distorted features. Compared with the generalized semantic communication system implementation method in the related art that does not consider the gain of retransmission, the present application greatly improves the system's error correction capability and transmission efficiency, thereby achieving better task performance.

[0059] An embodiment of the present application provides a semantically-aware HARQ communication method, in which a receiving end performs retransmission detection on multiple received transmission semantic features. When it is determined that the retransmission detection result is as failed detection, on this basis, the distortion assessment data generated by the semantic feature distortion assessment network model is fed back to the sending end, so that the sending end can retransmit the identified distorted transmission semantic features to be retransmitted, so that the multiple transmission semantic features received by the receiving end through detection can meet a certain accuracy, and then the decoding features corresponding to the multiple transmission semantic features are predicted and processed using a prediction algorithm of at least one preset type of semantic task, so that at least one relatively accurate prediction processing result can be obtained, thereby ensuring the accuracy of the executed semantic task.

[0060] The method of this embodiment is applied to a semantically aware HARQ communication system, such as Figure 1A A framework diagram of the semantically-aware HARQ communication system provided in an embodiment of the present application is shown, the system includes a transmitter and a receiver, the transmitter is composed of a multi-task joint semantic encoder, a source-channel joint encoder, a HARQ system elastic feature selector and a normalization layer, and the receiver is composed of a retransmission detection module, a source-source joint decoder, semantic decoders corresponding to multiple tasks and a task executor. The method of the present application can be applied to multiple semantic communication scenarios with limited channel resources, such as smart transportation scenarios, smart home scenarios, and Internet of Vehicles application scenarios, etc. It can support the transmission of multiple modal source data such as images, text, audio and video, and solve multiple semantic tasks such as classification, recognition, and prediction.

[0061] like Figure 1B As shown, the method of this embodiment includes:

[0062] Step 101: the transmitting end obtains a transmission source data set, generates a plurality of transmission semantic features based on the transmission source data set, and sends the plurality of transmission semantic features to the receiving end.

[0063] In this step, if Figure 1A As shown in the figure, taking the vehicle picture in the intelligent traffic scene as an example, the transmitter obtains the input transmission source data set, which is the vehicle picture set x at this time. The source semantic feature is extracted by the multi-task joint semantic encoder to obtain the joint transmission semantic feature s, and then the source channel joint encoder is used for channel coding to obtain multiple transmission semantic features f. Then, the HARQ system elastic feature selector is combined with the feature importance value to perform elastic feature transmission under the guidance of channel capacity. When performing retransmission and incremental transmission, important features are selected for priority transmission to obtain multiple selected transmission semantic features. After the power control is performed through the normalization layer, multiple transmission semantic features z that need to be transmitted are obtained j , and send it to the receiving end through the channel, where j is the number of retransmissions.

[0064] Among them, source semantic feature extraction can be regarded as source transformation, that is, transforming the transmission source data set from the original attribute space to the feature attribute space. Source semantic feature extraction is the basis of semantic communication. It uses deep neural networks to combine low-level features to form more abstract high-level semantic features, which improves the source compression efficiency while retaining the semantic meaning of the source.

[0065] Step 102: The receiving end performs retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result.

[0066] In this step, if Figure 1A As shown, the receiving end receives multiple transmission semantic features through the retransmission detection module Perform retransmission detection and obtain the retransmission detection result to decide whether to trigger the next retransmission.

[0067] Step 103: The receiving end determines whether the retransmission detection result passes the detection.

[0068] In this step, if Figure 1A As shown, the receiving end quickly determines whether multiple transmission semantic features received by the receiving end have passed the detection based on the retransmission detection result obtained by the retransmission detection module.

[0069] Step 104, in response to the receiving end determining that the retransmission detection result fails the detection, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector for transmission to the sending end.

[0070] In this step, if Figure 1A As shown, when the receiving end determines that the retransmission detection result is not passed, the received multiple transmission semantic features are Input semantic feature distortion evaluation network model (FDE network), use semantic feature distortion evaluation network model to receive multiple transmission semantic features Distortion measurement is performed to obtain distortion evaluation data, and semantic distortion-aware HARQ is implemented on this basis.

[0071] Step 105: the transmitting end determines the transmission semantic features to be retransmitted based on the binarized vector, and when it is determined that the transmission channel meets the preset transmission condition, incrementally transmits multiple new transmission semantic features to the receiving end.

[0072] In this step, the sender generates the distortion evaluation data p according to the semantic feature distortion evaluation network model. j The identified distorted features (i.e., the transmission semantic features to be retransmitted) are retransmitted, and when the transmission channel meets the preset transmission conditions, multiple new transmission semantic features are incrementally transmitted to the receiving end, so that the multiple transmission semantic features received by the receiving end through detection can meet a certain accuracy.

[0073] The preset transmission condition indicates the transmission performance of the transmission channel, and the preset transmission condition may be that there is a surplus in the channel condition (eg bandwidth) of the transmission channel.

[0074] Step 106, in response to the receiving end determining that the retransmission detection result is a passed detection, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

[0075] In this step, if Figure 1A As shown, when the receiving end determines that the retransmission detection result is passed, based on the multiple transmission semantic features of the received retransmission Concatenate all feature vectors of multiple transmissions cached by the receiver into Decoded by the source-channel joint decoder as After multiple semantic decoders corresponding to multiple tasks, the distribution is restored to Among them, K is the total number of multiple tasks. Finally, after K task executors, each task executor corresponds to the prediction algorithm of the preset type of semantic task, and K probability prediction vectors c1,...,c are generated based on the prediction algorithm of the K task executors corresponding to the preset type of semantic task. K (i.e., predicted processing results) are used to perform different intelligent multi-semantic tasks, such as vehicle re-identification, vehicle color classification, etc., to ensure the accuracy of the executed semantic tasks.

[0076] Through the above scheme, the receiving end performs retransmission detection on the received multiple transmission semantic features. When it is determined that the retransmission detection result is failed, on this basis, the distortion assessment data generated by the semantic feature distortion assessment network model is fed back to the sending end, so that the sending end can retransmit the identified distorted transmission semantic features to be retransmitted, so that the multiple transmission semantic features received by the receiving end through detection can meet a certain accuracy, and then use at least one preset type of semantic task prediction algorithm to predict the decoding features corresponding to the multiple transmission semantic features, so as to obtain at least one relatively accurate prediction processing result, thereby ensuring the accuracy of the executed semantic task.

[0077] In some embodiments, in step 101, sending the plurality of transmission semantic features to the receiving end includes:

[0078] Step A1: The sending end obtains feature importance values ​​corresponding to the multiple transmission semantic features.

[0079] Step A2: the transmitting end determines a plurality of transmission semantic features to be transmitted in each transmission process according to the feature importance value, and sends the determined plurality of transmission semantic features to be transmitted to the receiving end in each transmission process.

[0080] In the above scheme, if Figure 1A As shown, the HARQ system elastic feature selector in the transmitting end of the present application is designed to improve the reliability and efficiency of data transmission through the transmission technology (Hybrid Automatic Repeat Request, HARQ) of the wireless communication system. By combining automatic repeat request (ARQ) and incremental transmission, HARQ performs error detection and retransmission during data transmission to reduce the error rate. The present application adopts a method of combining information retransmission with incremental transmission, which improves the system robustness while increasing the transmission efficiency.

[0081] The HARQ system elastic feature selector combines the feature importance value to perform elastic feature transmission under the guidance of channel capacity. Important features are selected for priority transmission during retransmission and incremental transmission. When selecting feature transmission, multiple most important features are selected according to channel conditions and sent to the receiving end.

[0082] In some embodiments, step 102 includes:

[0083] Step B1, the receiving end determines a received signal-to-noise ratio corresponding to the received multiple transmission semantic features, and a cascade semantic feature corresponding to the received multiple transmission semantic features.

[0084] Step B2: The receiving end determines the uncertainty value corresponding to the received multiple transmission semantic features through an uncertainty algorithm based on the cascade semantic feature.

[0085] Step B3: The receiving end converts the uncertainty value into a comparison value corresponding to the received signal-to-noise ratio using a preset conversion coefficient.

[0086] Step B4: the receiving end compares the received signal-to-noise ratio with the comparison value to obtain a comparison result, and determines the retransmission detection result according to the comparison result.

[0087] In the above scheme, combined with Figure 1A and Figure 1C As shown, first, multiple transmission semantic features z j After being transmitted to the receiving end through the channel, the retransmission detection module is used to detect multiple received transmission semantic features. Perform retransmission detection to determine whether the next retransmission needs to be triggered.

[0088] This application uses multiple transmission semantic features received The corresponding uncertainty is used to make retransmission decisions. Since the received signal-to-noise ratio increases with the number of transmissions, when the received signal-to-noise ratio is small, the system needs to trigger retransmission until it can make up for the multiple transmission semantic features received by the receiver. The uncertainty of the concatenated feature vector after j transmissions is (i.e., cascade semantic features), the receiving end continues to trigger retransmission until the following is shown:

[0089]

[0090] in, For the receive signal-to-noise ratio, Indicates the comparison value, is the value of uncertainty, L(·) is a monotonically increasing function, and this application adopts L(x)=1+x, θ0 is the conversion coefficient between the value of uncertainty and the received SNR (i.e., the preset conversion coefficient), θ SNR is the preset maximum value of the received signal-to-noise ratio SNR. The uncertainty value is measured by the average entropy of the posterior probability distribution corresponding to different tasks, where the uncertainty corresponding to the kth task is given by the following formula:

[0091]

[0092] Among them, c represents the category label of the classification task, and γ is all the parameters involved in the decoder and task executor. After the retransmission detection is completed, if it fails to pass the detection, it will be input into the FDE network (i.e., the semantic feature distortion evaluation network model) to trigger retransmission. The transmitter retransmits the distorted features and uses the remaining channel resources to incrementally transmit new features in order of importance according to the channel state; if it passes the detection, all features are cascaded for decoding and task execution, and the semantic-aware HARQ process ends.

[0093] In some embodiments, in step B4, determining the retransmission detection result according to the comparison result includes:

[0094] Step C1, in response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is less than or equal to the comparison value, determining that the retransmission detection result is a failure to pass the detection.

[0095] Step C2: in response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is greater than the comparison value, determining that the retransmission detection result is a pass detection.

[0096] In the above scheme, combined with Figure 1A and Figure 1C As shown, first, multiple transmission semantic features z j After being transmitted to the receiving end through the channel, the retransmission detection module is used to detect multiple received transmission semantic features. Perform retransmission detection to determine whether the next retransmission needs to be triggered.

[0097] This application uses multiple transmission semantic features received The corresponding uncertainty is used to make retransmission decisions. Since the received signal-to-noise ratio increases with the number of transmissions, when the received signal-to-noise ratio is small, the system needs to trigger retransmission until it can make up for the multiple transmission semantic features received by the receiver. The uncertainty of the concatenated feature vector after j transmissions is (i.e., cascade semantic features), the receiving end continues to trigger retransmission until the following is shown:

[0098]

[0099] in, For the receive signal-to-noise ratio, Indicates the comparison value, is the value of uncertainty, L(·) is a monotonically increasing function, and this application adopts L(x)=1+x, θ0 is the conversion coefficient between the value of uncertainty and the received SNR (i.e., the preset conversion coefficient), θ SNR It is the preset maximum value of the received signal-to-noise ratio SNR.

[0100] That is, at this time, it means that the receiving end determines that the comparison result is that the receiving signal-to-noise ratio is greater than the comparison value, and then determines that the retransmission detection result is passed.

[0101] The value of uncertainty is measured by the average entropy of the posterior probability distribution corresponding to different tasks, where the uncertainty corresponding to the kth task is given by the following formula:

[0102]

[0103] Among them, C represents the category label of the classification task, and γ is all the parameters involved in the decoder and task executor. After the retransmission detection is completed, if it fails to pass the detection, it will be input into the FDE network (i.e., the semantic feature distortion evaluation network model) to trigger retransmission. The transmitter retransmits the distorted features and uses the remaining channel resources to incrementally transmit new features in order of importance according to the channel state; if it passes the detection, all features are cascaded for decoding and task execution, and the semantic-aware HARQ process ends.

[0104] When it looks like this:

[0105]

[0106] That is, at this time, it means that the receiving end determines that the comparison result is that the receiving signal-to-noise ratio is greater than the comparison value, and then determines that the retransmission detection result is passed.

[0107] The retransmission detection result can be quickly determined by receiving the signal-to-noise ratio and the comparison value.

[0108] In some embodiments, in step 104, the training process of the semantic feature distortion assessment network model includes:

[0109] In step D1, the receiving end obtains original semantic features for training and received semantic features for training corresponding to the original semantic features for training, and constructs an initial model, wherein the initial model includes an initial distortion metric module and a gradient estimator module.

[0110] In step D2, the receiving end inputs the original semantic features for training and the received semantic features for training into the initial distortion measurement module, and the initial distortion measurement module outputs a first contribution degree parameter and a second contribution degree parameter.

[0111] Step D3, the receiving end constructs a training objective function based on the first contribution degree parameter and the second contribution degree parameter, and trains and adjusts the initial distortion measurement module according to the maximization processing result obtained in the process of maximizing the training objective function to obtain a trained initial distortion measurement module.

[0112] Step D4, the receiving end obtains the received semantic features for training, and performs back-propagation training on the trained initial distortion metric module through the gradient estimator module based on the received semantic features for training, so as to obtain the semantic feature distortion assessment network model.

[0113] In the above scheme, if Figure 1D As shown, for the initial distortion metric module

[0114] In order to obtain the distortion of the received semantic features (i.e., the distortion evaluation data of the received multiple transmission semantic features), the present application sets an initial distortion measurement module It consists of four dense neural network layers, and the degree of semantic feature distortion is evaluated during the dense layer parameter optimization process.

[0115] Specifically, in order to effectively improve Before the end-to-end overall training, this application freezes the initial parameters of the remaining modules in the system. Since mutual information can characterize the degree of dependence between random variables, in order to learn the difference in semantic features between the sender and receiver, It is possible to learn the correlation between the received semantic features of all categories in the dataset and the original undistorted semantic features. Figure 1A s in (i.e., the original semantic features used for training) and The mutual information between the semantic features received during training is used as The training goal is as follows:

[0116]

[0117] in, for The trainable parameters of The value of the output vector can represent the contribution of the corresponding input vector to the mutual information during the training process, thereby representing its correlation with s. If the input semantic features are severely distorted during the transmission process, they cannot play a role in the mutual information optimization process, and have a low correlation with the undistorted vector of the corresponding category, so the corresponding output value is small. Based on the above analysis, when mutual information is used as the training target, The output of can indirectly reflect the degree of distortion of the input features, that is, the output can form a corresponding distortion evaluation vector. After the independent training is completed, the system further performs overall end-to-end training. The receiving end will receive the feature vector received after the jth retransmission. (i.e., receiving semantic features for training) input The output of dense layer 3 is Vectors of the same dimension And converted into a distortion metric vector d j ,in (i.e. distortion assessment data).

[0118] In order to estimate the mutual information of the semantic features of the transmitting and receiving ends as accurately as possible, this application adopts mutual information neural estimation technology. Mutual information neural estimation: Mutual information is a measure of the degree of association between two random variables. Since mutual information can be regarded as the relative entropy (Kullback-Leibler, KL divergence) between the joint distribution and the independent distribution of two random variables, estimating the mutual information only requires estimating the value of the KL divergence. The specific method is to give a lower bound on the KL divergence and use a neural network to continuously optimize this lower bound value to approximate the true value of the KL divergence, thereby estimating the value of the mutual information. Specifically, first, s and The mutual information calculation formula between is:

[0119]

[0120] in, For s and The joint probability distribution of , p(s) and are s and According to the derivation in the mutual information neural estimation technology, the lower bound of the mutual information is:

[0121]

[0122] Where T can be any integrable function, or it can be approximated by a neural network, and the expectation in the formula can be calculated after sampling. In this case, maximizing the mutual information is the lower bound of maximization. This application adopts The multi-layer neural network in T in , the mutual information estimation (i.e., training objective function) can be obtained as follows:

[0123]

[0124] In summary, when training the FDE network, As the training objective, the parameters of the dense layer are optimized in an unsupervised learning manner, where The input is a joint distribution and marginal distribution s, After the sampling passes through 4 dense layers, the system substitutes the output value of dense layer 4 into This calculates an estimate of the mutual information.

[0125] Among them, the output values ​​of dense layer 4 are and represents the first contribution degree parameter, Represents the second contribution parameter.

[0126] For the Gradient Estimator (Gumbel-softmax) module:

[0127] In order to use the distortion estimation vector d j (i.e., distortion evaluation data) to guide the next retransmission. This application sets a retransmission threshold t, which is d j The values ​​greater than the threshold are set to 1, and those less than the threshold are set to 0, thus j Quantized into a binary vector p j+1 , which is used to guide the j+1th feature retransmission, where 1 means that the semantic feature transmitted at this position needs to be retransmitted, otherwise it does not need to be retransmitted.

[0128] Since the quantization process is discrete and cannot be differentiated, the network is not differentiable here and is difficult to optimize using back propagation. The commonly used direct estimation method treats the binary function as an identity mapping for back propagation, which may cause differences between forward propagation and backward propagation, resulting in high variance of gradient estimation. Therefore, this application considers d in forward propagation. j Direct discretization, and using the gradient estimator in back propagation, replacing the non-differentiable discretization process with a differentiable sampling process, can effectively improve the efficiency and stability of gradient estimation. Specifically, this application proposes a gradient estimator based on a multidimensional Gumbel-softmax distribution, Gumbel-softmax technology: Gumbel-softmax is a technology for approximating discrete sampling, which is often used in probabilistic models and generative models in deep learning. In order to solve the problem that traditional discrete sampling methods are usually not differentiable, resulting in the inability to back propagate, it introduces a combination of Gumbel distribution and Softmax function to estimate the gradient of the system back propagation during discrete sampling, thereby realizing the continuous approximation of the discrete distribution. Gumbel-softmax uses the gradient optimization method for training in discrete problems, so that the operation of discrete variables can be performed in a continuous manner, which makes Gumbel-softmax widely used in training deep generative models, reinforcement learning and other tasks that require discrete variables. Each sample is a differentiable proxy for the corresponding discrete sample. First, suppose that the semantic features that need to be retransmitted are R. This application is based on d j R probability vectors are derived, and R Gumbel-softmax distributions are formed based on these probability vectors, where the rth probability vector It can be expressed as:

[0129]

[0130] in, is to j The largest r-1 values ​​in are set to 0. It represents the distortion of the remaining features when the r-1 features with the largest distortion are not considered. On this basis, Converted to a probability vector through the softmax function but The value with the highest probability corresponds to the feature with the rth largest distortion. After generating R probability vectors according to the above method, the corresponding R Gumbel-softmax sampling vectors can be obtained. Each random variable in the vector obeys the Gumbel-softmax distribution. The rth sampling vector can be expressed as:

[0131]

[0132] Among them, g r is the sampling vector of the independently distributed Gumbel distribution, the softmax function is a continuous and differentiable approximation of the argmax function, and τ is the sampling temperature. The system sets τ to gradually decrease during the training process, at which point the R Gumbel-softmax distributions will tend to be discretized, and at the same time The sampling vector in will also tend to be a one-hot vector (that is, only one value is 1 and the rest are 0). At this time, it is equivalent to approximately sampling out The value with the highest probability is selected and the corresponding position in the sampling vector is marked as 1. Therefore, by merging R sampling vectors, we can get p j+1 The continuous approximate vector of is used for the back propagation of the neural network, namely:

[0133]

[0134] In some embodiments, in step 105, the transmitting end determines the transmission semantic feature to be retransmitted based on the binarized vector, including:

[0135] The transmitting end searches for the transmission semantic feature whose binarization vector is a first preset value from various transmission semantic features, and uses the transmission semantic feature whose binarization vector is the first preset value as the transmission semantic feature to be retransmitted.

[0136] In the above scheme, in order to use the distortion evaluation vector d j (i.e., distortion evaluation data) to guide the next retransmission. This application sets a retransmission threshold t, which is d j The value greater than the threshold is set to 1 (i.e., the first preset value), and vice versa, it is set to 0 (i.e., the second preset value), thereby setting d j Quantized into a binary vector pj+1 , used to guide the j+1th feature retransmission.

[0137] Among them, 1 means that the transmission semantic feature of the position needs to be retransmitted, otherwise it does not need to be retransmitted.

[0138] The transmission semantic features that need to be retransmitted can be quickly determined by quantizing the distortion assessment data and the retransmission threshold into a binary vector.

[0139] In some embodiments, in step 104, after the receiving end determines that the retransmission detection result is a failure to pass the detection, the method further includes:

[0140] The receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can search for the transmission semantic feature whose binarized vector is a second preset value from each transmission semantic feature, and use the transmission semantic feature whose binarized vector is the second preset value as the transmission semantic feature that is not retransmitted.

[0141] In the above scheme, in order to use the distortion evaluation vector d j (i.e., distortion evaluation data) to guide the next retransmission. This application sets a retransmission threshold t, which is d j The value greater than the threshold is set to 1 (i.e., the first preset value), and vice versa, it is set to 0 (i.e., the second preset value), thereby setting d j Quantized into a binary vector p j+1 , used to guide the j+1th feature retransmission.

[0142] Among them, 1 means that the transmission semantic feature of the position needs to be retransmitted, otherwise it does not need to be retransmitted.

[0143] The binary vector quantized by the distortion evaluation data and the retransmission threshold can quickly determine the transmission semantic features that are not retransmitted, avoiding the problem of reduced channel resource utilization caused by erroneous retransmission.

[0144] In some embodiments, in combination Figure 1A , Figure 1C and Figure 1DAs shown, the present application implements a semantically-aware multi-task HARQ semantically-aware HARQ communication system under limited bandwidth and latency, so that the receiving end has the ability to perform distortion evaluation in the dimension of semantic features, thereby improving the robustness of the system. Furthermore, the distortion feature retransmission is combined with feature incremental transmission to improve the system transmission efficiency and task implementation performance. The present invention is simulated and compared with a semantically-aware HARQ system that only considers semantic incremental transmission (IK-S-HARQ), a semantically-aware HARQ system that only considers all feature retransmissions (RT-S-HARQ), and a traditional type II HARQ system (Traditional II-HARQ). Taking the three tasks of vehicle re-identification, vehicle color recognition, and vehicle type recognition in an intelligent transportation system as examples, the performance simulation comparison results are shown in the figure. Figure 1E , Figure 1F and Figure 1G shown.

[0145] It should be noted that the method of the embodiment of the present application can be performed by a single device, such as a computer or server. The method of this embodiment can also be applied to a distributed scenario and completed by multiple devices cooperating with each other. In the case of such a distributed scenario, one of the multiple devices can only perform one or more steps in the method of the embodiment of the present application, and the multiple devices will interact with each other to complete the described method.

[0146] It should be noted that the above describes some embodiments of the present application. Other embodiments are within the scope of the attached claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the above embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-semantic task processing and parallel processing are also possible or may be advantageous.

[0147] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a semantically-aware HARQ communication device.

[0148] refer to Figure 2 The semantically aware HARQ communication device is applied to a semantically aware HARQ communication system, the system comprising a transmitting end and a receiving end, and the device comprises:

[0149] The transmitting end 201 is configured to obtain a transmission source data set, generate a plurality of transmission semantic features based on the transmission source data set, and send the plurality of transmission semantic features to the receiving end;

[0150] The receiving end 202 is configured to perform retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result; in response to determining that the retransmission detection result is not passed, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end determines the transmission semantic features to be retransmitted based on the binary vector and resends it to the receiving end, and when it is determined that the transmission channel meets the preset transmission conditions, incrementally transmits multiple new transmission semantic features to the receiving end, and the receiving end repeatedly performs the retransmission detection process based on the received multiple transmission semantic features until it is determined that the retransmission detection result is passed, or the preset maximum number of retransmissions is reached; or, in response to determining that the retransmission detection result is passed, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

[0151] In some embodiments, the sending end 201 is specifically configured as follows:

[0152] The transmitting end obtains feature importance values ​​corresponding to the multiple transmission semantic features;

[0153] The transmitting end determines a plurality of transmission semantic features to be transmitted in each transmission process according to the feature importance value, and sends the determined plurality of transmission semantic features to be transmitted to the receiving end in each transmission process.

[0154] In some embodiments, the receiving end 202 includes:

[0155] A first determining unit is configured to determine a received signal-to-noise ratio corresponding to the received multiple transmission semantic features and a cascade semantic feature corresponding to the received multiple transmission semantic features;

[0156] A second determining unit is configured to determine uncertainty values ​​corresponding to the received multiple transmission semantic features through an uncertainty algorithm based on the cascade semantic feature;

[0157] A conversion unit, configured to convert the uncertainty value into a comparison value corresponding to the received signal-to-noise ratio using a preset conversion coefficient;

[0158] The comparison unit is configured to compare the received signal-to-noise ratio with the comparison value to obtain a comparison result, and determine the retransmission detection result according to the comparison result.

[0159] In some embodiments, the comparison unit is specifically configured as follows:

[0160] In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is less than or equal to the comparison value, determining that the retransmission detection result is a failure to pass the detection; or,

[0161] In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is greater than the comparison value, the retransmission detection result is determined to be a pass detection.

[0162] In some embodiments, the semantically-aware HARQ communication device further includes a training module, wherein the training module is specifically configured to:

[0163] The receiving end obtains original semantic features for training and received semantic features for training corresponding to the original semantic features for training, and constructs an initial model, wherein the initial model includes an initial distortion metric module and a gradient estimator module;

[0164] The receiving end inputs the original semantic features for training and the received semantic features for training into the initial distortion measurement module, and the initial distortion measurement module outputs a first contribution degree parameter and a second contribution degree parameter;

[0165] The receiving end constructs a training objective function based on the first contribution degree parameter and the second contribution degree parameter, and performs training adjustment on the initial distortion metric module according to a maximization processing result obtained in the process of maximizing the training objective function, so as to obtain a trained initial distortion metric module;

[0166] The receiving end obtains the received semantic features for training, and performs back-propagation training on the trained initial distortion metric module through the gradient estimator module based on the received semantic features for training, so as to obtain the semantic feature distortion assessment network model.

[0167] In some embodiments, the sending end 201 is specifically configured as follows:

[0168] The transmitting end searches for the transmission semantic feature whose binarization vector is a first preset value from various transmission semantic features, and uses the transmission semantic feature whose binarization vector is the first preset value as the transmission semantic feature to be retransmitted.

[0169] In some embodiments, after the receiving end determines that the retransmission detection result is not passed, the receiving end 202 is specifically configured to:

[0170] The receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can search for the transmission semantic feature whose binarized vector is a second preset value from each transmission semantic feature, and use the transmission semantic feature whose binarized vector is the second preset value as the transmission semantic feature that is not retransmitted.

[0171] For the convenience of description, the above device is described in terms of functions divided into various modules. Of course, when implementing the present application, the functions of each module can be implemented in the same or multiple software and / or hardware.

[0172] The apparatus of the above embodiment is used to implement the corresponding semantically-aware HARQ communication method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0173] Based on the same inventive concept, corresponding to any of the above-mentioned embodiments and methods, the present application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the semantically-aware HARQ communication method described in any of the above embodiments is implemented.

[0174] Figure 3 A more specific schematic diagram of the hardware structure of an electronic device provided in this embodiment is shown, and the device may include: a processor 301, a memory 302, an input / output interface 303, a communication interface 304, and a bus 305. The processor 301, the memory 302, the input / output interface 303, and the communication interface 304 are connected to each other in communication within the device through the bus 305.

[0175] The processor 301 can be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of this specification.

[0176] The memory 302 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 302 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented by software or firmware, the relevant program codes are stored in the memory 302 and called and executed by the processor 301.

[0177] The input / output interface 303 is used to connect the input / output module to realize information input and output. The input / output module can be configured in the device as a component (not shown in the figure), or it can be externally connected to the device to provide corresponding functions. The input device can include a keyboard, a mouse, a touch screen, a microphone, various sensors, etc., and the output device can include a display, a speaker, a vibrator, an indicator light, etc.

[0178] The communication interface 304 is used to connect a communication module (not shown) to realize communication interaction between the device and other devices. The communication module can realize communication through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0179] The bus 305 comprises a pathway for transmitting information between the various components of the device (eg, the processor 301 , the memory 302 , the input / output interface 303 , and the communication interface 304 ).

[0180] It should be noted that, although the above device only shows the processor 301, the memory 302, the input / output interface 303, the communication interface 304 and the bus 305, in the specific implementation process, the device may also include other components necessary for normal operation. In addition, it can be understood by those skilled in the art that the above device may also only include the components necessary for implementing the embodiments of the present specification, and does not necessarily include all the components shown in the figure.

[0181] The electronic device of the above embodiment is used to implement the corresponding semantically-aware HARQ communication method in any of the above embodiments, and has the beneficial effects of the corresponding method embodiment, which will not be repeated here.

[0182] Based on the same inventive concept, corresponding to any of the above-mentioned embodiment methods, the present application also provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the semantically-aware HARQ communication method described in any of the above embodiments.

[0183] The computer-readable medium of this embodiment includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, read-only compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.

[0184] The computer instructions stored in the storage medium of the above embodiment are used to enable the computer to execute the semantically-aware HARQ communication method as described in any of the above embodiments, and have the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0185] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present application (including the claims) is limited to these examples. In line with the concept of the present application, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the embodiments of the present application as described above, which are not provided in detail for the sake of simplicity.

[0186] In addition, to simplify the description and discussion, and in order not to make the embodiments of the present application difficult to understand, the known power supply / ground connection with the integrated circuit (IC) chip and other components may or may not be shown in the provided drawings. In addition, the device can be shown in the form of a block diagram to avoid making the embodiments of the present application difficult to understand, and this also takes into account the fact that the details of the implementation of these block diagram devices are highly dependent on the platform to be implemented in the embodiments of the present application (that is, these details should be fully within the scope of understanding of those skilled in the art). In the case of elaborating specific details (e.g., circuits) to describe exemplary embodiments of the present application, it is obvious to those skilled in the art that the embodiments of the present application can be implemented without these specific details or when these specific details are changed. Therefore, these descriptions should be considered to be illustrative rather than restrictive.

[0187] Although the present application has been described in conjunction with specific embodiments of the present application, many replacements, modifications and variations of these embodiments will be apparent to those skilled in the art from the foregoing description. For example, other memory architectures (e.g., dynamic RAM (DRAM)) may use the embodiments discussed.

[0188] The embodiments of the present application are intended to cover all such substitutions, modifications and variations that fall within the broad scope of the appended claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the embodiments of the present application should be included in the scope of protection of the present application.

Claims

1. A semantically aware HARQ communication method, characterized in that: A semantically aware HARQ communication system, the system comprising a transmitting end and a receiving end, the method comprising: The sending end obtains a transmission source data set, generates a plurality of transmission semantic features based on the transmission source data set, and sends the plurality of transmission semantic features to the receiving end; The receiving end performs retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result; In response to the receiving end determining that the retransmission detection result is failed, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can determine the transmission semantic features to be retransmitted based on the binary vector and resend them to the receiving end. At the same time, when it is determined that the transmission channel meets the preset transmission conditions, multiple new transmission semantic features are incrementally transmitted to the receiving end. The receiving end repeatedly performs the retransmission detection process based on the received multiple transmission semantic features until it is determined that the retransmission detection result is passed, or the preset maximum number of retransmissions is reached; or, In response to the receiving end determining that the retransmission detection result is a passed detection, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

2. The method according to claim 1, characterized in that Sending the plurality of transmission semantic features to the receiving end includes: The transmitting end obtains feature importance values ​​corresponding to the multiple transmission semantic features; The transmitting end determines a plurality of transmission semantic features to be transmitted in each transmission process according to the feature importance value, and sends the determined plurality of transmission semantic features to be transmitted to the receiving end in each transmission process.

3. The method according to claim 1, characterized in that The receiving end performs retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result, including: The receiving end determines a received signal-to-noise ratio corresponding to the received multiple transmission semantic features, and a cascade semantic feature corresponding to the received multiple transmission semantic features; The receiving end determines, based on the cascade semantic feature, uncertainty values ​​corresponding to the received multiple transmission semantic features through an uncertainty algorithm; The receiving end converts the uncertainty value into a comparison value corresponding to the received signal-to-noise ratio using a preset conversion coefficient; The receiving end compares the received signal-to-noise ratio with the comparison value to obtain a comparison result, and determines the retransmission detection result according to the comparison result.

4. The method according to claim 3, characterized in that The determining the retransmission detection result according to the comparison result includes: In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is less than or equal to the comparison value, determining that the retransmission detection result is a failure to pass the detection; or, In response to the receiving end determining that the comparison result is that the received signal-to-noise ratio is greater than the comparison value, the retransmission detection result is determined to be a pass detection.

5. The method according to claim 1, characterized in that The training process of the semantic feature distortion assessment network model includes: The receiving end obtains original semantic features for training and received semantic features for training corresponding to the original semantic features for training, and constructs an initial model, wherein the initial model includes an initial distortion metric module and a gradient estimator module; The receiving end inputs the original semantic features for training and the received semantic features for training into the initial distortion measurement module, and the initial distortion measurement module outputs a first contribution degree parameter and a second contribution degree parameter; The receiving end constructs a training objective function based on the first contribution degree parameter and the second contribution degree parameter, and performs training adjustment on the initial distortion metric module according to a maximization processing result obtained in the process of maximizing the training objective function, so as to obtain a trained initial distortion metric module; The receiving end obtains the received semantic features for training, and performs back-propagation training on the trained initial distortion metric module through the gradient estimator module based on the received semantic features for training, so as to obtain the semantic feature distortion assessment network model.

6. The method according to claim 1, characterized in that The transmitting end determines the transmission semantic feature to be retransmitted based on the binarized vector, including: The transmitting end searches for the transmission semantic feature whose binarization vector is a first preset value from various transmission semantic features, and uses the transmission semantic feature whose binarization vector is the first preset value as the transmission semantic feature to be retransmitted.

7. The method according to claim 1, characterized in that After the receiving end determines that the retransmission detection result is a failure to pass the detection, the method further includes: The receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end can search for the transmission semantic feature whose binarized vector is a second preset value from each transmission semantic feature, and use the transmission semantic feature whose binarized vector is the second preset value as the transmission semantic feature that is not retransmitted.

8. A semantically aware HARQ communication device, characterized in that: The device is arranged in a semantically aware HARQ communication system, the system comprising a transmitting end and a receiving end, and the device comprises: The transmitting end is configured to obtain a transmission source data set, generate a plurality of transmission semantic features based on the transmission source data set, and send the plurality of transmission semantic features to the receiving end; The receiving end is configured to perform retransmission detection based on the received multiple transmission semantic features to obtain a retransmission detection result; in response to determining that the retransmission detection result is not passed, the receiving end inputs the received multiple transmission semantic features into a pre-trained semantic feature distortion assessment network model to obtain distortion assessment data, and converts the distortion assessment data into a binary vector and transmits it to the sending end, so that the sending end determines the transmission semantic features to be retransmitted based on the binary vector and resends it to the receiving end, and when it is determined that the transmission channel meets the preset transmission conditions, incrementally transmits multiple new transmission semantic features to the receiving end, and the receiving end repeatedly performs the retransmission detection process based on the received multiple transmission semantic features until it is determined that the retransmission detection result is passed, or a preset maximum number of retransmissions is reached; or, in response to determining that the retransmission detection result is passed, the receiving end determines a decoding feature corresponding to at least one preset type of semantic task based on the received multiple transmission semantic features, and uses a prediction algorithm of at least one preset type of semantic task to predict the decoding feature to obtain at least one prediction processing result.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that: The computer instructions are used to enable a computer to execute the method according to any one of claims 1 to 7.

Images