Model training method, transmitting-end device, receiving-end device, and system

Abstract

The present disclosure provides a model training method, a transmitting-end device, a receiving-end device, and a system. The method comprises: inputting a sample bit sequence into a modulation model, and acquiring a first modulation symbol outputted by the modulation model; transmitting a second modulation symbol to a receiving-end device via a channel, the second modulation symbol being obtained by performing a first operation on the first modulation symbol, and the second modulation symbol being used by the receiving-end device to determine a second gradient value; receiving the second gradient value transmitted by the receiving-end device; training the modulation model on the basis of the second gradient value; and when a first stop condition is met, stopping the training of the modulation model, and obtaining a trained modulation model. In the present disclosure, a modulation model and a demodulation model can be jointly trained online, thereby improving the accuracy of data transmission, improving the performance of a communication system, and achieving high feasibility.

Description

Model training methods, transmitting devices, receiving devices and systems Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and in particular to model training methods, transmitting devices, receiving devices, and systems. Background Technology

[0002] To meet the service demands of high-speed data transmission, massive broadband traffic, and reliable communication in mobile scenarios, and ultimately realize future wireless communication visions such as smart cities and smart transportation, current wireless communication networks still require improvement and enhancement in many aspects. Among these, accurate signal demodulation is one of the fundamental modules for achieving low bit error rate and high-speed transmission in wireless communication systems, and it affects the overall performance of the communication system. Summary of the Invention

[0003] To improve the performance of communication systems, embodiments of this disclosure provide a model training method, a transmitting device, a receiving device, and a system.

[0004] According to a first aspect of the present disclosure, a model training method is provided, the method being executed by a transmitting device, the method comprising:

[0005] Input the sample bit sequence into the modulation model and obtain the first modulation symbol output by the modulation model;

[0006] The receiving device transmits a second modulation symbol through the channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving device to determine a second gradient value, the second gradient value being the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by reverse transmitting the first gradient value within the demodulation model; and receives the second gradient value transmitted by the receiving device.

[0007] The modulation model is trained based on the second gradient value;

[0008] If the first stopping condition is met, training of the modulation model is stopped, and a trained modulation model is obtained.

[0009] According to a second aspect of the present disclosure, a model training method is provided, the method being executed by a receiving device, the method comprising:

[0010] The receiving end device obtains the received signal by transmitting the second modulation symbol through the channel;

[0011] The symbol to be demodulated is input into the demodulation model to obtain the first bit sequence output by the demodulation model; wherein, the symbol to be demodulated is obtained by performing a second operation on the received signal;

[0012] The loss function is determined based on the difference between the first bit sequence and the sample bit sequence;

[0013] The demodulation model is trained based on the first gradient value, and a second gradient value is sent to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model;

[0014] If the second stopping condition is met, training of the demodulation model is stopped, and a trained demodulation model is obtained.

[0015] According to a third aspect of the present disclosure, a transmitting device is provided, comprising:

[0016] The processing module is configured to input the sample bit sequence into the modulation model and obtain the first modulation symbol output by the modulation model;

[0017] The transceiver module is configured to transmit a second modulation symbol to a receiving device via a channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving device to determine a second gradient value, the second gradient value being the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by reverse transmitting the first gradient value within the demodulation model; the transceiver module is also configured to receive the second gradient value transmitted by the receiving device;

[0018] The processing module is further configured to train the modulation model based on the second gradient value;

[0019] The processing module is further configured to stop training the modulation model when the first stopping condition is met, thereby obtaining the trained modulation model.

[0020] According to a fourth aspect of the present disclosure, a receiving device is provided, comprising:

[0021] The transceiver module is configured to receive the second modulation symbol transmitted by the transmitting device through the channel and obtain the received signal;

[0022] The processing module is configured to input the symbol to be demodulated into the demodulation model and obtain the first bit sequence output by the demodulation model; wherein the symbol to be demodulated is obtained by performing a second operation on the received signal;

[0023] The processing module is further configured to determine a loss function based on the difference between the first bit sequence and the sample bit sequence;

[0024] The processing module is further configured to train the demodulation model based on the first gradient value, and the transceiver module is further configured to send the second gradient value to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model;

[0025] The processing module is further configured to stop training the demodulation model when the second stopping condition is met, thereby obtaining the trained demodulation model.

[0026] According to a fifth aspect of the present disclosure, a transmitting end device is provided, comprising:

[0027] One or more processors;

[0028] The processor is used to execute the method described in any one of the first aspects.

[0029] According to a sixth aspect of the present disclosure, a receiving device is provided, comprising:

[0030] One or more processors;

[0031] The processor is used to execute the method described in any one of the second aspects.

[0032] According to a seventh aspect of the present disclosure, a model training system is provided, comprising:

[0033] A transmitting device, the transmitting device being configured to perform the method described in any one of the first aspects;

[0034] A receiving device, the receiving device being used to perform the method described in any one of the second aspects.

[0035] According to an eighth aspect of the present disclosure, a storage medium is provided that stores instructions that, when executed on a communication device, cause the communication device to perform a model training method as described in any one of the first or second aspects.

[0036] According to a ninth aspect of the present disclosure, a computer program product is provided, including a computer program that, when executed by a processor, is used to implement the model training method described in any one of the first or second aspects.

[0037] In this embodiment, the modulation model and demodulation model can be jointly trained online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0038] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0039] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0040] Figure 1A is an exemplary schematic diagram of the architecture of a communication system provided according to an embodiment of the present disclosure.

[0041] Figure 1B is an exemplary scenario diagram of an AI-based demodulation model provided according to an embodiment of the present disclosure.

[0042] Figure 1C is an exemplary schematic diagram of an AI demodulation scheme based on known modulation information provided in an embodiment of this disclosure.

[0043] Figure 1D is an exemplary schematic diagram of an AI demodulation scheme that does not require obtaining modulation scheme information according to an embodiment of the present disclosure.

[0044] Figure 1E is an exemplary schematic diagram of the mapping relationship between bit sequences and modulation symbols under the 16QAM modulation scheme provided in the embodiments of this disclosure.

[0045] Figure 1F is an exemplary scenario diagram of an AI-based modulation and demodulation model provided according to an embodiment of the present disclosure.

[0046] Figure 2 is an exemplary interactive schematic diagram of a model training method provided according to an embodiment of the present disclosure.

[0047] Figure 3A is one of the exemplary flowcharts of a model training method provided according to an embodiment of the present disclosure.

[0048] Figure 3B is a second exemplary flowchart of a model training method provided according to an embodiment of the present disclosure.

[0049] Figure 3C is a third exemplary flowchart of a model training method provided according to an embodiment of the present disclosure.

[0050] Figure 3D is a fourth exemplary flowchart of a model training method provided according to an embodiment of the present disclosure.

[0051] Figure 4 is an exemplary scenario diagram of joint training of modulation and demodulation models using an intermediate model according to an embodiment of the present disclosure.

[0052] Figure 5A is an exemplary block diagram of a transmitting device provided according to an embodiment of the present disclosure.

[0053] Figure 5B is an exemplary block diagram of a receiving device provided according to an embodiment of the present disclosure.

[0054] Figure 6A is an exemplary block diagram of a communication device provided according to an embodiment of the present disclosure.

[0055] Figure 6B is an exemplary block diagram of a chip provided according to an embodiment of the present disclosure. Detailed Implementation

[0056] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present invention. Rather, they are merely examples of apparatuses and methods consistent with some aspects of the invention as detailed in the appended claims.

[0057] This disclosure presents a model training method, a transmitting device, a receiving device, and a system.

[0058] In a first aspect, embodiments of this disclosure propose a model training method, which is executed by a transmitting device. The method includes: inputting a sample bit sequence into a modulation model to obtain a first modulation symbol output by the modulation model; transmitting a second modulation symbol to a receiving device through a channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving device to determine a second gradient value, the second gradient value being the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by back-transmitting the first gradient value within the demodulation model; receiving the second gradient value transmitted by the receiving device; training the modulation model based on the second gradient value; and stopping the training of the modulation model when a first stopping condition is met, thereby obtaining a trained modulation model.

[0059] In the above embodiments, the modulation model and demodulation model can be jointly trained online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0060] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes at least one of the following: randomly generating a set of sample bit sequences; setting the set of sample bit sequences according to a first rule; receiving first indication information sent by a network device; wherein the first indication information is used to indicate the set of sample bit sequences.

[0061] In the above embodiments, the sending device can determine the set of sample bit sequences to provide a training dataset for joint training, thereby improving the reliability of joint training.

[0062] In conjunction with some embodiments of the first aspect, in some embodiments, the first operation includes at least one of the following: a layer mapping operation; a precoding operation.

[0063] In the above embodiments, the transmitting device can perform the first operation on the first modulation symbol to obtain the second modulation symbol, thereby improving the reliability of data transmission.

[0064] In some embodiments of the first aspect, training the modulation model based on the second gradient value includes: backpropagating the second gradient value within the modulation model to update the model parameters of the modulation model.

[0065] In the above embodiments, the transmitting device can perform online training of the modulation model based on the second gradient value sent by the receiving device, thereby achieving the purpose of joint training of the modulation model and the demodulation model, which has high availability.

[0066] In conjunction with some embodiments of the first aspect, in some embodiments, the first stopping condition includes at least one of the following: not receiving a new second gradient value sent by the receiving device; receiving second indication information sent by the receiving device, the second indication information being used to instruct the transmitting device to stop training the modulation model.

[0067] In the above embodiments, the transmitting device can stop training the modulation model when the first stopping condition is met, thereby obtaining a trained modulation model, which improves the performance and availability of the communication system.

[0068] In conjunction with some embodiments of the first aspect, in some embodiments, the second gradient value is the gradient value output by the input layer of the demodulation model during the last round of update when the receiving device reverse-transmits the first gradient value within the demodulation model and performs multiple rounds of updates to the model parameters of the demodulation model.

[0069] In the above embodiments, the model parameters of the demodulation model can be updated in multiple rounds, and then the second gradient value output by the input layer in the last round of update can be sent to the transmitting device. This reduces the signaling overhead caused by the frequent interaction of the second gradient value between the receiving device and the transmitting device during joint training, avoids occupying too much link resources, and helps improve system performance.

[0070] In some embodiments, in conjunction with the first aspect, the method further includes: determining first initialization parameters; wherein the first initialization parameters include structural parameters of the modulation model and / or training parameters of the modulation model; and initializing the modulation model based on the first initialization parameters.

[0071] In the above embodiments, the transmitting device can initialize the modulation model before joint training, which improves the reliability of joint training.

[0072] In conjunction with some embodiments of the first aspect, in some embodiments, determining the first initialization parameter includes at least one of the following: determining the first initialization parameter based on third indication information sent by the network device; wherein the third indication information is used to indicate the first initialization parameter; determining the first initialization parameter based on fourth indication information sent by the receiving device; wherein the fourth indication information is used to indicate the first initialization parameter; and determining the first initialization parameter based on a second rule.

[0073] In the above embodiments, the transmitting device can use at least one of the above methods to determine the first initialization parameters, which is simple to implement and highly usable.

[0074] In conjunction with some embodiments of the first aspect, in some embodiments, the method further includes: determining a second initialization parameter based on a third rule; wherein the second initialization parameter includes the structural parameters of the demodulation model and / or the training parameters of the demodulation model; and sending fifth indication information to the receiving device; wherein the fifth indication information is used to indicate the second initialization parameter.

[0075] In the above embodiments, the sending device can determine the second initialization parameters and send them to the receiving device, which improves the reliability of joint training.

[0076] Secondly, embodiments of this disclosure propose a model training method, which is executed by a receiving device. The method includes: receiving a second modulation symbol transmitted by a transmitting device through a channel to obtain a received signal; inputting the symbol to be demodulated into a demodulation model to obtain a first bit sequence output by the demodulation model; wherein the symbol to be demodulated is obtained after performing a second operation on the received signal; determining a loss function based on the difference between the first bit sequence and the sample bit sequence; training the demodulation model based on a first gradient value; and sending a second gradient value to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model after backpropagating the first gradient value within the demodulation model and updating the model parameters of the demodulation model; and stopping the training of the demodulation model when a second stopping condition is met to obtain a trained demodulation model.

[0077] In the above embodiments, the modulation model and demodulation model can be jointly trained online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0078] In conjunction with some embodiments of the second aspect, in some embodiments, the second operation includes at least one of the following: channel estimation operation; channel equalization operation.

[0079] In the above embodiments, the receiving device can perform the second operation on the received signal to obtain the symbol to be demodulated, thereby improving the reliability of data transmission.

[0080] In some embodiments of the second aspect, training the demodulation model based on the first gradient value includes: backpropagating the first gradient value within the demodulation model to update the model parameters of the demodulation model.

[0081] In the above embodiments, the receiving device can perform online training on the demodulation model based on the loss function, thereby achieving the purpose of joint training of the modulation model and the demodulation model, which has high availability.

[0082] In conjunction with some embodiments of the second aspect, in some embodiments, the second stopping condition includes at least one of the following: reaching the number of training epochs; the loss function decreasing to within the fault tolerance range; and the accuracy of the first bit sequence reaching a first value.

[0083] In the above embodiments, the receiving device can stop training the demodulation model when the second stopping condition is met, thereby obtaining a well-trained demodulation model, which improves the performance and availability of the communication system.

[0084] In conjunction with some embodiments of the second aspect, in some embodiments, the method further includes at least one of the following: satisfying the second stopping condition to stop sending a new second gradient value to the transmitting device; satisfying the second stopping condition to send second indication information to the transmitting device; wherein the second indication information is used to instruct the transmitting device to stop training the modulation model.

[0085] In the above embodiments, after the receiving device stops training, it can stop sending new second gradient values ​​to the sending device and / or send second indication information to the sending device, thereby improving the reliability of joint training.

[0086] In conjunction with some embodiments of the second aspect, in some embodiments, the second gradient value is the gradient value output by the input layer of the demodulation model during the last round of update when the receiving device reverse-transmits the first gradient value within the demodulation model and performs multiple rounds of updates to the model parameters of the demodulation model.

[0087] In conjunction with some embodiments of the second aspect, in some embodiments, the method further includes: determining a second initialization parameter; wherein the second initialization parameter includes the structural parameters of the demodulation model and / or the training parameters of the demodulation model; and initializing the demodulation model based on the second initialization parameter.

[0088] In the above embodiments, the receiving device can initialize the demodulation model before joint training, which improves the reliability of joint training.

[0089] In conjunction with some embodiments of the second aspect, in some embodiments, determining the second initialization parameter includes at least one of the following: determining the second initialization parameter based on a sixth indication information sent by the network device; wherein the sixth indication information is used to indicate the second initialization parameter; determining the second initialization parameter based on a fifth indication information sent by the sending device; wherein the fifth indication information is used to indicate the second initialization parameter; and determining the second initialization parameter based on a fourth rule.

[0090] In the above embodiments, the receiving device can determine the second initialization parameter using at least one of the above methods, which is simple to implement and highly usable.

[0091] In conjunction with some embodiments of the second aspect, in some embodiments, the method further includes: determining a first initialization parameter based on a fifth rule; wherein the first initialization parameter includes the structural parameters of the modulation model and / or the training parameters of the modulation model; and sending fourth indication information to the transmitting device; wherein the fourth indication information is used to indicate the first initialization parameter.

[0092] In the above embodiments, the receiving device can determine the first initialization parameters and send them to the sending device, which improves the reliability of joint training.

[0093] Thirdly, embodiments of this disclosure propose a transmitting end device, comprising: a processing module configured to input a sample bit sequence into a modulation model to obtain a first modulation symbol output by the modulation model; and a transceiver module configured to transmit a second modulation symbol to a receiving end device via a channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving end device to determine a second gradient value, the second gradient value being a gradient value output by the input layer of the demodulation model when the receiving end device updates the model parameters of the demodulation model by reverse transmission of the first gradient value within the demodulation model; the transceiver module is further configured to receive the second gradient value transmitted by the receiving end device; the processing module is further configured to train the modulation model based on the second gradient value; and the processing module is further configured to stop training the modulation model when a first stopping condition is met, thereby obtaining a trained modulation model.

[0094] Fourthly, embodiments of this disclosure propose a receiving device, comprising: a transceiver module configured to receive a second modulation symbol transmitted by a transmitting device through a channel to obtain a received signal; a processing module configured to input a symbol to be demodulated into a demodulation model to obtain a first bit sequence output by the demodulation model; wherein the symbol to be demodulated is obtained after performing a second operation on the received signal; the processing module is further configured to determine a loss function based on the difference between the first bit sequence and the sample bit sequence; the processing module is further configured to train the demodulation model based on a first gradient value; and the transceiver module is further configured to send a second gradient value to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model; the processing module is further configured to stop training the demodulation model when a second stopping condition is met, thereby obtaining a trained demodulation model.

[0095] Fifthly, embodiments of this disclosure provide a transmitting device, comprising: one or more processors; wherein the processors are configured to perform the method described in any one of the first aspects.

[0096] In a sixth aspect, embodiments of this disclosure provide a receiving device comprising: one or more processors; wherein the processors are configured to perform the method described in any one of the second aspects.

[0097] In a seventh aspect, embodiments of this disclosure provide a model training system, comprising: a transmitting device for performing the method described in any one of the first aspects; and a receiving device for performing the method described in any one of the second aspects.

[0098] Eighthly, embodiments of this disclosure provide a storage medium storing instructions that, when executed on a communication device, cause the communication device to perform a model training method as described in any one of the first or second aspects.

[0099] In a ninth aspect, embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, is used to implement the model training method described in any one of the first or second aspects.

[0100] It is understood that the aforementioned transmitting device, receiving device, model training system, storage medium, program product, etc., are all used to execute the method proposed in the embodiments of this disclosure. Therefore, the beneficial effects that can be achieved can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0101] This disclosure provides a model training method, a transmitting device, a receiving device, and a system. In some embodiments, the terms "model training method" and "neural network training method," "communication method," etc., can be used interchangeably.

[0102] This disclosure is not exhaustive, but merely illustrative of some embodiments, and is not intended to limit the scope of protection of this disclosure. Unless otherwise specified, each step in a particular embodiment can be implemented as an independent embodiment, and the steps can be arbitrarily combined. For example, a solution after removing some steps in a particular embodiment can also be implemented as an independent embodiment, and the order of the steps in a particular embodiment can be arbitrarily interchanged. Furthermore, the optional implementation methods in a particular embodiment can be arbitrarily combined; moreover, the embodiments can be arbitrarily combined, for example, some or all steps of different embodiments can be arbitrarily combined, and a particular embodiment can be arbitrarily combined with the optional implementation methods of other embodiments. In all embodiments of this disclosure, unless otherwise specified or logically conflicting, the terminology and / or descriptions between the embodiments are consistent and can be mutually referenced. Technical features in different embodiments can be combined to form new embodiments based on their inherent logical relationships.

[0103] The terminology used in the embodiments of this disclosure is for the purpose of describing particular embodiments only and is not intended to limit the scope of this disclosure.

[0104] In this embodiment of the disclosure, unless otherwise stated, elements expressed in the singular form, such as "a," "an," "the," "the," "the," "the," "the," "the," "this," etc., can mean "one and only one," or "one or more," "at least one," etc. For example, when using articles such as "a," "an," "the," etc. in translation, the noun following the article can be understood as either a singular expression or a plural expression.

[0105] In the embodiments disclosed herein, "multiple" refers to two or more.

[0106] In some embodiments, the terms “at least one of A or B, at least one of A and B”, “one or more”, “a plurality of”, “multiple”, etc., may be used interchangeably.

[0107] In some embodiments, the notation "at least one of A and B", "A and / or B", "A in one case, B in another", "in response to one case A, in response to another case B", etc., may include the following technical solutions depending on the situation: in some embodiments, A (execute A regardless of whether there is a branch B); in some embodiments, B (execute B regardless of whether there is a branch A); in some embodiments, execution is selected from A and B (A and B are selectively executed); in some embodiments, both A and B are executed. The same applies when there are more branches such as A, B, C, etc.

[0108] In some embodiments, the notation "A or B" may include the following technical solutions, depending on the situation: in some embodiments, A (execute A regardless of whether a branch B exists); in some embodiments, B (execute B regardless of whether a branch A exists); in some embodiments, execution is selected from A and B (A and B are selectively executed). The same applies when there are more branches such as A, B, and C.

[0109] The prefixes "first," "second," etc., used in the embodiments of this disclosure are merely for distinguishing different descriptive objects and do not impose restrictions on the position, order, priority, quantity, or content of the descriptive objects. The description of the descriptive objects is found in the claims or the context of the embodiments, and the use of prefixes should not constitute unnecessary restrictions. For example, if the descriptive object is a "field," the ordinal numbers preceding "field" in "first field" and "second field" do not restrict the position or order of the "fields." "First" and "second" do not restrict whether the "fields" they modify are in the same message, nor do they restrict the order of "first field" and "second field." Similarly, if the descriptive object is a "level," the ordinal numbers preceding "level" in "first level" and "second level" do not restrict the priority between "levels." Furthermore, the number of descriptive objects is not limited by ordinal numbers and can be one or more. For example, in "first device," the number of "devices" can be one or more. Furthermore, the objects modified by different prefixes can be the same or different. For example, if the object being described is "device", then "first device" and "second device" can be the same device or different devices, and their types can be the same or different. Similarly, if the object being described is "information", then "first information" and "second information" can be the same information or different information, and their content can be the same or different.

[0110] In some embodiments, “including A,” “containing A,” “for indicating A,” and “carrying A” can be interpreted as directly carrying A or indirectly indicating A.

[0111] In some embodiments, terms such as “in response to…”, “in response to determining…”, “in the case of…”, “when…”, “when…”, “if…”, etc. can be used interchangeably. These descriptions all refer to the device making a corresponding action under certain objective circumstances. They do not necessarily limit the time, nor do they require the device to make a judgment action when implementing it, nor do they mean that there must be other limitations.

[0112] In some embodiments, the terms “greater than,” “greater than or equal to,” “not less than,” “more than,” “more than or equal to,” “not less than,” “higher than,” “higher than or equal to,” “not lower than,” and “above” can be used interchangeably, as can the terms “less than,” “less than or equal to,” “not greater than,” “less than,” “less than or equal to,” “not more than,” “lower than,” “lower than or equal to,” “not higher than,” and “below”.

[0113] In some embodiments, devices, etc., may be interpreted as physical or virtual, and their names are not limited to those described in the embodiments. Terms such as “device,” “equipment,” “circuit,” “network element,” “network function,” “network device,” “function,” “node,” “unit,” “section,” “system,” “network,” “chip,” “chip system,” “entity,” and “subject” are interchangeable.

[0114] In some embodiments, "network" can be interpreted as devices included in a network (e.g., access network devices, core network devices, etc.).

[0115] In some embodiments, the terms "access network device (AN device)," "radio access network device (RAN device)," "base station (BS)," "radio base station," "fixed station," "node," "access point," "transmission point (TP)," "reception point (RP)," "transmission / reception point (TRP)," "panel," "antenna panel," "antenna array," "cell," "macro cell," "small cell," "femto cell," "pico cell," "sector," "cell group," "serving cell," "carrier," "component carrier," and "bandwidth part (BWP)" can be used interchangeably.

[0116] In some embodiments, the terms "terminal", "terminal device", "user equipment (UE)", "user terminal", "mobile station (MS)", "mobile terminal (MT)", "subscriber station", "mobile unit", "subscriber unit", "wireless unit", "remote unit", "mobile device", "wireless device", "wireless communication device", "remote device", "mobile subscriber station", "access terminal", "mobile terminal", "wireless terminal", "remote terminal", "handset", "user agent", "mobile client", and "client" can be used interchangeably.

[0117] In some embodiments, access network devices, core network devices, or network devices can be replaced by terminals. For example, embodiments of this disclosure can also be applied to structures where communication between access network devices, core network devices, or network devices and terminals is replaced by communication between multiple terminals (e.g., device-to-device (D2D), vehicle-to-everything (V2X), etc.). In this case, the structure can also be configured such that the terminal has all or part of the functions of the access network device. Furthermore, terms such as "uplink" and "downlink" can be replaced with terms corresponding to communication between terminals (e.g., "sidelink"). For example, uplink channel, downlink channel, etc., can be replaced with sidelink channel, and uplink link, downlink, etc., can be replaced with sidelink link.

[0118] In some embodiments, the terminal may be replaced by an access network device, a core network device, or a network device. In this case, the access network device, core network device, or network device may also be configured to have all or some of the functions of the terminal.

[0119] In some embodiments, the acquisition of data, information, etc., may comply with the laws and regulations of the country where the location is situated.

[0120] In some embodiments, data, information, etc., may be obtained with the user's consent.

[0121] Furthermore, each element, each row, or each column in the table of this disclosure can be implemented as an independent embodiment, and any combination of any element, any row, or any column can also be implemented as an independent embodiment.

[0122] Figure 1A is a schematic diagram of the architecture of a communication system according to an embodiment of the present disclosure.

[0123] As shown in Figure 1A, the communication system 100 includes at least one of a transmitting end device 101 and a receiving end device 102.

[0124] In some embodiments, the transmitting device 101 can be used to transmit data and / or signals.

[0125] In some embodiments, the transmitting device 101 may include at least one of a terminal and a network device, which is not limited in this disclosure.

[0126] In some embodiments, the receiving device 102 may be used to receive data and / or signals.

[0127] In some embodiments, the receiving device 102 may include at least one of a network device and a terminal, which is not limited in this disclosure.

[0128] In some embodiments, the transmitting device 101 and the receiving device 102 can be used to perform online training of the modulation model and the demodulation model.

[0129] In one example, a modulation model can be used to modulate an input bit sequence to output a modulation symbol.

[0130] In one example, the modulation model can be deployed on the transmitting device 101 to modulate the encoded bit sequence obtained on the transmitting device 102 and output the modulation symbol.

[0131] In one example, a demodulation model can be used to demodulate the input symbol to be demodulated, thereby outputting a demodulated symbol.

[0132] In one example, the demodulation model can be deployed on the receiving device 102 to demodulate the symbols to be demodulated on the receiving device 102 and output the demodulated symbols.

[0133] In some embodiments, the communication system 100 may further include a network device 103.

[0134] In one example, the network device 103 may be a network-side functional unit for AI model management, including but not limited to Access and Mobility Management Function (AMF), Artificial Intelligence Management Function, third-party provided server or network functions that support the management of AI models, and this disclosure does not limit it.

[0135] In some embodiments, the aforementioned terminals include, but are not limited to, at least one of the following: mobile phone, wearable device, Internet of Things device, car with communication function, smart car, tablet computer, computer with wireless transceiver function, virtual reality (VR) terminal device, augmented reality (AR) terminal device, wireless terminal device in industrial control, wireless terminal device in self-driving, wireless terminal device in remote medical surgery, wireless terminal device in smart grid, wireless terminal device in transportation safety, wireless terminal device in smart city, and wireless terminal device in smart home.

[0136] In some embodiments, the network devices described above include at least one of access network devices and core network devices.

[0137] In some embodiments, the access network device is, for example, a node or device that connects a terminal to a wireless network. The access network device may include at least one of the following in a 5G communication system: evolved Node B (eNB), next-generation eNB (ng-eNB), next-generation Node B (gNB), node B (NB), home node B (HNB), home evolved node B (HeNB), radio backhaul device, radio network controller (RNC), base station controller (BSC), base transceiver station (BTS), base band unit (BBU), mobile switching center, base station in a 6G communication system, open RAN, cloud RAN, base station in other communication systems, and access node in a Wi-Fi system, but is not limited thereto.

[0138] In some embodiments, the technical solutions of this disclosure can be applied to the Open RAN architecture. In this case, the interfaces between or within access network devices involved in the embodiments of this disclosure can be transformed into internal interfaces of Open RAN. The processes and information interactions between these internal interfaces can be implemented by software or programs.

[0139] In some embodiments, the access network device may be composed of a central unit (CU) and a distributed unit (DU). The CU may also be called a control unit. The CU-DU structure can separate the protocol layer of the access network device. Some of the protocol layer functions are centrally controlled by the CU, while the remaining part or all of the protocol layer functions are distributed in the DU and centrally controlled by the CU. However, this is not the only possibility.

[0140] In some embodiments, the core network equipment may be a single device comprising multiple network elements, or it may be multiple devices or a group of devices, each comprising some or all of the multiple network elements. Network elements may be virtual or physical. The core network may include, for example, at least one of the Evolved Packet Core (EPC), 5G Core Network (5GCN), and Next Generation Core (NGC).

[0141] It is understood that the communication system described in this disclosure is for the purpose of more clearly illustrating the technical solutions of this disclosure, and does not constitute a limitation on the technical solutions proposed in this disclosure. As those skilled in the art will know, with the evolution of system architecture and the emergence of new business scenarios, the technical solutions proposed in this disclosure are also applicable to similar technical problems.

[0142] The following embodiments of this disclosure can be applied to the communication system 100 shown in FIG1A, or to some of the main bodies, but are not limited thereto. The main bodies shown in FIG1A are illustrative. The communication system may include all or some of the main bodies in FIG1A, or it may include other main bodies outside of FIG1A. The number and form of each main body are arbitrary. Each main body may be physical or virtual. The connection relationship between the main bodies is illustrative. The main bodies may not be connected or may be connected. The connection can be in any way, it can be a direct connection or an indirect connection, it can be a wired connection or a wireless connection.

[0143] The embodiments disclosed herein can be applied to Long Term Evolution (LTE), LTE-Advanced (LTE-A), LTE-Beyond (LTE-B), SUPER 3G, IMT-Advanced, 4th generation mobile communication system (4G), 5th generation mobile communication system (5G), 5G new radio (NR), Future Radio Access (FRA), New-Radio Access Technology (RAT), New Radio (NR), New radio access (NX), Future generation radio access (FX), Global System for Mobile communications (GSM), CDMA2000, Ultra Mobile Broadband (UMB), IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), and IEEE 802.20, Ultra-Wideband (UWB), Bluetooth (a registered trademark), Public Land Mobile Network (PLMN) networks, Device-to-Device (D2D) systems, Machine-to-Machine (M2M) systems, Internet of Things (IoT) systems, Vehicle-to-Everything (V2X) systems, systems utilizing other communication methods, and next-generation systems built upon them, etc. Furthermore, multiple systems can be combined (e.g., a combination of LTE or LTE-A with 5G).

[0144] In some embodiments, the optimal demodulator in a traditional wireless communication system is generally implemented by classical methods such as maximum likelihood estimation and Bayes' theorem. These demodulators typically require accurate Channel State Information (CSI) and channel noise distribution, and their performance depends on the parameter settings of each module, including filters, phase-locked loops, product modulators, analog-to-digital converters, etc.

[0145] The main limitations of traditional demodulators include:

[0146] 1. Large latency and complex implementation: The use of traditional demodulators usually causes significant latency, and the implementation process is relatively complex.

[0147] 2. Unstable module performance: Due to factors such as vibration, acceleration, temperature fluctuation, aging, and instability of discrete components, the module performance may change, resulting in a decrease in the overall performance of the receiver.

[0148] 3. Poor environmental adaptability: Real-world wireless communication channels may suffer from multipath fading, impulse noise, clutter, or discrete interference, which can significantly degrade demodulation performance. For example, some coherent demodulators require carrier synchronization, and phase errors and frequency offsets during synchronization can lead to demodulation errors.

[0149] 4. High dependence on prior knowledge such as CSI: Traditional demodulation methods usually have a high dependence on prior knowledge. However, in actual communication, especially in fast fading scenarios, it is difficult to accurately estimate CSI. Therefore, the channel model may be unknown at the receiver.

[0150] 5. Traditional demodulation methods do not make full use of prior waveform information and time series information.

[0151] Traditional wireless communication systems are generally designed based on rigorous mathematical theories and precise system models. However, with the increasing demands of wireless services such as smartphones, virtual reality, and the Internet of Things (IoT), these systems need to handle more complex and diverse communication requirements, and traditional mathematical models may not be able to adequately address these challenges. In this context, deep learning has been introduced into wireless communication systems as a powerful solution.

[0152] Since the information of a modulated signal is represented by its amplitude and phase, feature extraction is crucial for signal demodulation. Deep learning-based demodulators (or learning-based demodulators) can automatically learn and extract important features from signals using techniques such as neural networks, thereby achieving more accurate and robust demodulation. Their main advantages are:

[0153] 1. By learning from massive datasets, it possesses greater flexibility and adaptability. Compared to traditional methods, it requires less or even completely eliminates the need for prior knowledge;

[0154] 2. Because the model learns features under various undesirable conditions during training, the deep learning-based demodulator has stronger noise resistance.

[0155] 3. The requirements for prior knowledge (such as CSI and channel noise) can be relaxed or even eliminated.

[0156] Therefore, deep learning-based demodulators have great potential in dealing with complex wireless communication environments and improving communication performance.

[0157] In some embodiments, the demodulation model enhancement scheme based on Artificial Intelligence (AI) can use a neural network model to implement data demodulation, as shown in Figure 1B. This demodulation model takes the data to be demodulated (or the symbols to be demodulated) at the receiving end as input and outputs demodulated symbols (i.e., bit sequence b). During model training, the demodulated bit sequence output by this demodulation model is used... The error between the coded bit sequence b at the transmitting end and the output of the demodulation model is used as a loss function to update the model parameters of each layer, so that the output of the demodulation model is close to the target output.

[0158] In some embodiments, various modulation schemes are currently available, such as Quadrature Phase Shift Keying (QPSK), 16 Quadrature Amplitude Modulation (16QAM), and 64 Quadrature Amplitude Modulation (64QAM). The mapping between constellation symbols and data differs, and one data symbol to be demodulated corresponds to different lengths of data bits. For example, in QPSK, 16QAM, and 64QAM, one data symbol to be demodulated corresponds to 2, 4, and 6 demodulated data bits, respectively. Therefore, for received symbols using different modulation schemes, a matching demodulation scheme is required for data demodulation. For demodulation schemes based on AI models, different modulation schemes also need to be considered.

[0159] Training an AI model requires input and output data of a defined dimension. The length of the demodulated data corresponding to a symbol to be demodulated may differ depending on the modulation scheme. For example, in 16QAM modulation, a symbol to be demodulated corresponds to 4 bits of demodulated data, while in 64QAM modulation, a symbol to be demodulated corresponds to 6 bits of demodulated data. Therefore, the output length of the AI ​​demodulation model used for 16QAM and 64QAM should be 4 bits and 6 bits respectively, and it needs to learn the mapping relationship between constellation points and data under the two modulation schemes respectively.

[0160] Therefore, AI-based demodulation schemes can demodulate data under different modulation schemes, but they require known modulation scheme information. This modulation scheme information can be indicated to the receiving device by the transmitting device during data transmission as auxiliary information for selecting the demodulation model, as shown in Figure 1C. The input data of the AI ​​demodulation model consists of n data symbols to be demodulated, and the output data consists of n×M 0 / 1 bits, where M is the modulation order.

[0161] Furthermore, when the transmitting device does not indicate modulation scheme information, it is possible to explore modulation scheme recognition based on AI models. That is, using the data to be demodulated and the corresponding modulation scheme information as training data, the AI ​​modulation scheme recognition model obtained through training can determine the corresponding modulation scheme of the transmitting signal based on the features of the data to be demodulated, and then select the AI ​​demodulation model under the corresponding modulation scheme to demodulate the data, thus realizing an AI demodulation scheme that does not require indication of modulation information, as shown in Figure 1D.

[0162] In digital communication systems, constellation design is an important research direction. The transmitting device can map bit data to appropriate constellation points to complete data modulation, and the receiving device can demodulate the data according to the corresponding decision regions to recover the bit data. The AI-based data demodulation scheme described above only enhances the receiving end's data demodulation function and does not modify the transmitting end's modulation function. That is, the transmitting device still uses 16QAM, 64QAM, etc., to complete the data modulation of coded bits, using fixed rules to map the coded bit sequence to the corresponding constellation points. Taking 16QAM as an example, the mapping relationship between the bit sequence to be modulated and the modulation symbols can be shown in Figure 1E.

[0163] In typical channel environments, standard modulation schemes such as 16QAM and 64QAM can achieve high-precision data transmission, basically meeting the system's operational requirements. However, in real-world applications, the channel environment is often more complex, and the actual channel conditions are unknown, making it impossible to design constellation diagrams mathematically. In such cases, using a fixed data modulation scheme may not achieve optimal data transmission performance. For example, when the signal-to-noise ratio (SNR) is low, i.e., when channel noise is high, using constellation points that are further apart to perform data modulation can reduce the impact of signal distortion caused by noise, allowing the receiving equipment to perform data demodulation with higher accuracy.

[0164] Therefore, using AI models to simultaneously enhance both the modulation and demodulation models is beneficial for further improving data transmission accuracy and enhancing the overall performance of the communication system. An AI-based modulation and demodulation enhancement scheme can be illustrated as shown in Figure 1F. The input to the modulation model at the transmitting end is the coded bit b, and the output is the modulation symbol x; the input to the demodulation model at the receiving end is the symbol y to be demodulated, and the output is the demodulated bit. In the application of the model, a neural network model is used to implement the data modulation function. The modulation model maps the encoded bits b to constellation points x suitable for the current channel environment, and the demodulation model processes the data symbols y to be demodulated, which are affected by factors such as channel conditions and noise, to obtain the demodulated bits.

[0165] Since the transmitting modulation model does not have its own target output, joint training of the models at both ends is required to calculate the demodulated bits output by the demodulation model. The loss function between the modulated and coded bits b is used to update the parameters of the modulation and demodulation models. The impact on signal transmission from the transmitting modulation model to the receiving demodulation model is reflected in the demodulated bits. In the distortion, the demodulated output bits are suppressed by training the modulation and demodulation models. The error between the target bit b and the target bit b.

[0166] The modulation symbol x output by the modulation model needs to be processed by other modules (such as layer mapping, precoding, etc.) after the modulation model at the transmitting end to obtain the transmitted signal s. transmit =f transmit (x). The transmitted signal reaches the receiver after being transmitted through the channel, and the received signal s receive =f channel (s transmit Affected by channel propagation, noise, and interference, the signal is further processed by other modules (such as channel estimation and equalization) before the receiver's demodulation model to obtain the demodulated symbol y = f. receive (s receive ).

[0167] While conventional offline model training methods can acquire the transmitting modulation symbol x and the receiving demodulated symbol y during data acquisition, training requires adding other modules between the modulation and demodulation models, as well as the channel transmission process. These other modules, such as transmitting layer mapping and receiving channel estimation, can be considered as computational operations with fixed rules. However, channel transmission is affected by noise and interference in the actual channel environment, resulting in uncertain channel states that cannot be accurately implemented using fixed computational methods. Consequently, joint training of the modulation and demodulation models cannot be completed.

[0168] To achieve joint training of the modulation and demodulation models and improve data transmission accuracy, this disclosure provides the following model training method, transmitting device, receiving device, and system.

[0169] Figure 2 is an interactive schematic diagram of a model training method according to an embodiment of the present disclosure. As shown in Figure 2, the embodiments of the present disclosure relate to a model training method, which includes:

[0170] In step S2101, the transmitting device 101 determines the first initialization parameters.

[0171] In some embodiments, the transmitting device 101 and the receiving device 102 can be used to jointly train the modulation model and the demodulation model.

[0172] In some embodiments, the modulation model can be used to modulate the input bit sequence to output a modulation symbol.

[0173] In some embodiments, the demodulation model can be used to demodulate the input symbols to be demodulated, thereby outputting demodulated symbols.

[0174] In some embodiments, the sending device 101 may be, for example, a terminal, network device, server, or other device that supports the above-described joint training.

[0175] In some embodiments, the first initialization parameters may include, but are not limited to, at least one of the following: structural parameters of the modulation model; training parameters of the modulation model.

[0176] In one example, the structural parameters of the modulation model depend on the parameter configuration used in the actual application of the modulation model, the complexity of the channel environment, and so on.

[0177] In one example, the training parameters of the modulation model include, but are not limited to, at least one of the following: the loss function type of the modulation model; the learning rate of the modulation model; the optimizer type of the modulation model; the maximum number of training epochs of the modulation model; and the output accuracy of the modulation model.

[0178] The loss function types of the modulation model include, but are not limited to, the L1 function and the L2 function. The L1 function can refer to the minimum absolute value deviation function, and the L2 function can refer to the minimum squared error function.

[0179] The learning rate of the modulation model is the speed at which the model updates its knowledge or parameters during training, and its value can be in the range of (0, 1).

[0180] The optimizer for the modulation model can be used to optimize the weights and / or biases of the modulation model. The optimizer type can include, but is not limited to, at least one of the following: stochastic gradient descent optimizer; adaptive learning rate optimizer; adaptive learning rate and momentum optimizer; gradient normalization optimizer.

[0181] The maximum number of training cycles for the modulation model can refer to the maximum number of training rounds for the modulation model.

[0182] The output accuracy of the modulation model can refer to the degree of accuracy of the output data of the modulation model relative to the labels (training reference data).

[0183] The above is merely an illustrative example, and this disclosure does not limit the content of the first initialization parameter.

[0184] In some embodiments, the transmitting device 101 may determine the first initialization parameters in the following manner:

[0185] Method 1-1: Determine the first initialization parameters based on the instructions of network device 103.

[0186] In one example, network device 103 could be a network-side functional unit for AI model management.

[0187] In one example, network device 103 can configure first initialization parameters of the modulation model and send third indication information to transmitting device 101. The third indication information indicates the first initialization parameters. Transmitting device 101 determines the first initialization parameters based on the third indication information.

[0188] Method 1-2: Determine the first initialization parameters based on the instruction from the receiving device 102.

[0189] In one example, the receiving device 102 can configure the first initialization parameters of the modulation model and send fourth indication information to the transmitting device 101. The fourth indication information indicates the first initialization parameters. The transmitting device 101 determines the first initialization parameters based on the fourth indication information.

[0190] Methods 1-3 determine the first initialization parameter based on the second rule.

[0191] In one example, the sending device 101 itself can determine the first initialization parameters based on the second rule. The second rule can be configured by the network device 103 and / or agreed upon by the protocol, which is not limited herein.

[0192] The above is merely an illustrative example, and this disclosure does not limit the method by which the transmitting device 101 determines the first initialization parameter.

[0193] Step S2102: The transmitting device 101 initializes the modulation model.

[0194] In some embodiments, the transmitting device 101 may initialize the modulation model based on the first initialization parameters.

[0195] In one example, the transmitting device 101 can initialize the network structure of the modulation model based on the structural parameters of the modulation model.

[0196] In one example, the transmitting device 101 can set relevant parameters during the training process of the modulation model based on the training parameters of the modulation model.

[0197] In one example, initializing the modulation model may also include initializing the network parameters of the modulation model, which may include, but is not limited to, at least one of the following methods: zero initialization; random initialization; Xavier initialization; Kaiming initialization.

[0198] Zero initialization can refer to setting the network parameters to 0 when initializing the modulation model.

[0199] Random initialization can refer to assigning random values ​​to the network parameters when initializing the modulation model.

[0200] Xavier initialization can maintain signal stability during forward and backward propagation by setting appropriate initial values ​​for weights and biases, ensuring that the output variance of each neuron is approximately the same as its input variance.

[0201] Kaiming initialization maintains consistent variance of activation values ​​in deep networks by adjusting the initial values ​​of weights, thereby preventing gradient vanishing and making the network easier to train.

[0202] In step S2103, the receiving device 102 determines the second initialization parameters.

[0203] In some embodiments, the receiving device 102 may be, for example, a terminal, network device, server, or other device that supports the above-described joint training.

[0204] In some embodiments, the second initialization parameters may include, but are not limited to, at least one of the following: structural parameters of the demodulation model; training parameters of the demodulation model.

[0205] In one example, the structural parameters of the demodulation model depend on the parameter configuration used in the actual application, the complexity of the channel environment, and so on.

[0206] In one example, the training parameters of the demodulation model include, but are not limited to, at least one of the following: the loss function type of the demodulation model; the learning rate of the demodulation model; the optimizer type of the demodulation model; the maximum number of training epochs of the demodulation model; and the output accuracy of the demodulation model.

[0207] The loss function types of the demodulation model include, but are not limited to, the L1 function and the L2 function. The L1 function can refer to the minimum absolute value deviation function, and the L2 function can refer to the minimum squared error function.

[0208] The learning rate of the demodulation model is the speed at which the model updates its knowledge or parameters during training, and its value can be in the range of (0, 1).

[0209] The optimizer for the demodulation model can be used to optimize the weights and / or biases of the demodulation model. The optimizer type can include, but is not limited to, at least one of the following: stochastic gradient descent optimizer; adaptive learning rate optimizer; adaptive learning rate and momentum optimizer; gradient normalization optimizer.

[0210] The maximum number of training cycles for the demodulation model can refer to the maximum number of training rounds for the demodulation model.

[0211] The output accuracy of the demodulation model refers to the degree of accuracy of the output data of the demodulation model relative to the labels (training reference data).

[0212] The above is merely an illustrative example, and this disclosure does not limit the content of the second initialization parameter.

[0213] In some embodiments, the receiving device 102 may determine the second initialization parameters in the following manner:

[0214] Method 2-1: Determine the second initialization parameters based on the instructions of network device 103.

[0215] In one example, network device 103 could be a network-side functional unit for AI model management.

[0216] In one example, network device 103 can configure the second initialization parameters of the demodulation model and send a sixth indication message to receiving device 102. The sixth indication message indicates the second initialization parameters. Receiving device 102 determines the second initialization parameters based on the sixth indication message.

[0217] Method 2-2: Determine the first initialization parameters based on the instruction from the transmitting device 101.

[0218] In one example, the transmitting device 101 can configure the second initialization parameters of the demodulation model and send fifth indication information to the receiving device 102. The fifth indication information indicates the second initialization parameters. The receiving device 102 determines the second initialization parameters based on the fifth indication information.

[0219] Methods 2-3, based on the fourth rule, determine the second initialization parameters.

[0220] In one example, the receiving device 102 itself can determine the second initialization parameters based on the fourth rule. The fourth rule can be configured by the network device 103 and / or agreed upon by the protocol, which is not limited herein.

[0221] The above is merely an illustrative example, and this disclosure does not limit the manner in which the receiving device 102 determines the second initialization parameters.

[0222] Step S2104: The receiving device 102 initializes the demodulation model.

[0223] In some embodiments, the receiving device 102 may initialize the demodulation model based on the second initialization parameters.

[0224] In one example, the receiving device 102 can initialize the network structure of the demodulation model based on the structural parameters of the demodulation model.

[0225] In one example, the receiving device 102 can set relevant parameters during the training process of the demodulation model based on the training parameters of the demodulation model.

[0226] In one example, initializing the demodulation model may also include initializing the network parameters of the demodulation model, which may include, but is not limited to, at least one of the following methods: zero initialization; random initialization; Xavier initialization; Kaiming initialization. Specific methods will not be elaborated here.

[0227] Step S2105: The sending device 101 obtains the sample bit sequence set.

[0228] In some embodiments, the transmitting device 101 and the receiving device 102 can be used to jointly train the modulation model and the demodulation model.

[0229] In some embodiments, the modulation model can be used to modulate the input bit sequence to output a modulation symbol.

[0230] In some embodiments, the demodulation model can be used to demodulate the input symbols to be demodulated, thereby outputting demodulated symbols.

[0231] In some embodiments, the sending device 101 may be, for example, a terminal, network device, server, or other device that supports the above-described joint training.

[0232] In some embodiments, the sample bit sequence set may include at least one sample bit sequence b. In one example, the sample bit sequence b may be a sequence of "1" and "0" with a length value greater than or equal to 1, for example, the sample bit sequence may be "100010" or "111110000".

[0233] In one example, the sample bit sequence b can be used as input data for the modulation model.

[0234] In one example, the sample bit sequence b is the bit sequence obtained after encoding the sample data. The sample data is the data that the transmitting device 101 needs to send to the receiving device 102.

[0235] In some embodiments, the set of sample bit sequences can be determined by the transmitting device 101 and sent by the transmitting device 101 to the receiving device 102 during the training process.

[0236] In some embodiments, the transmitting device 101 may determine the sample bit sequence set in at least one of the following ways: Method 3-1, randomly generating the sample bit sequence set.

[0237] In one example, the transmitting device 101 may use a random algorithm to generate the sample bit sequence set.

[0238] Method 3-2: Set the sample bit sequence set according to the first rule.

[0239] In one example, the first rule may be agreed upon by a protocol, and / or the first rule may be set by network device 103 and provided to the sending device 101, which is not limited in this disclosure.

[0240] In one example, the first rule may include at least one of the following: the number of sample bit sequence sets; the number of sample bit sequences included in each sample bit sequence set; the length value of each sample bit sequence; and the algorithm for setting the sample bit sequences.

[0241] In one example, the sending device 101 may set the sample bit sequence set according to a first rule.

[0242] Method 3-3, as instructed by network device 103.

[0243] In one example, network device 103 could be a network-side functional unit for AI model management.

[0244] In one example, after determining the set of sample bit sequences, network device 103 sends first indication information to sending device 101, and sending device 101 receives the first indication information to determine the set of sample bit sequences.

[0245] The method by which network device 103 determines the set of sample bit sequences is similar to that described in methods 3-1 or 3-2 above, and will not be repeated here.

[0246] The above is merely an illustrative example. The transmitting device 101 may also determine the sample bit sequence set in other ways, and this disclosure does not limit this.

[0247] Step S2106: The transmitting device 101 acquires the first modulation symbol.

[0248] In some embodiments, the transmitting device 101 can input at least one sample bit sequence b from the sample bit sequence set into the modulation model to obtain the first modulation symbol output by the modulation model, such as the modulation symbol x in FIG1F.

[0249] In step S2107, the transmitting device 101 determines the second modulation symbol.

[0250] In some embodiments, the transmitting device 101 may perform a first operation on the first modulation symbol x to obtain the second modulation symbol.

[0251] In some embodiments, the first operation may include, but is not limited to, at least one of the following: layer mapping operation; precoding operation.

[0252] In one example, the transmitting device 101 can perform layer mapping on the input data through a layer mapping module. The input data of the layer mapping module may include the first modulation symbol x output by the modulation model.

[0253] In one example, layer mapping operations can refer to passing data from the upper layer to the lower layer and ensuring effective communication and coordination between different layers.

[0254] In one example, the transmitting device 101 can precode the input data using a precoding module. The input data to the precoding module may include the output data from the layer mapping module. The precoding module can output a second modulation symbol.

[0255] In one example, precoding can improve signal reliability and stability by transforming and optimizing data at the transmitting end, thereby reducing noise interference and multipath fading during signal transmission.

[0256] The second modulation symbol can also be called the "transmit signal", such as the transmit signal s in Figure 1F. transmit =ftransmit (x).

[0257] In step S2108, the transmitting device 101 sends a second modulation signal to the receiving device 102.

[0258] In some embodiments, the transmitting device 101 transmits a second modulation symbol to the receiving device 102 via a channel.

[0259] In some embodiments, the second modulation symbol can be used by the receiving device 102 to determine a second gradient value. The second gradient value is the gradient value output by the input layer of the demodulation model when the receiving device 102 updates the model parameters of the demodulation model by transmitting the first gradient value in reverse within the demodulation model.

[0260] In step S2109, the receiving device 102 determines the symbol to be demodulated.

[0261] In some embodiments, the receiving device 102 transmits a second modulation symbol transmitted through the channel to obtain a received signal, such as the received signal s in FIG1F. receive =f channel (s transmit ).

[0262] In some embodiments, the receiving device 102 may perform a second operation on the received signal to obtain the demodulated symbol, which may be, for example, the demodulated symbol y in FIG1F.

[0263] In some embodiments, the second operation may include, but is not limited to, at least one of the following: channel estimation operation; channel equalization operation.

[0264] In one example, the receiving device 102 can perform channel estimation on the input data using a channel estimation module, wherein the input data of the channel estimation module can be the received signal s. receive .

[0265] For example, the receiving device 103 can perform channel equalization on the input data through the channel equalization module, wherein the input data of the channel equalization module can be the output data of the channel estimation module, and the output data of the channel equalization module can be the symbol y to be demodulated.

[0266] In some embodiments, the receiving device 102 may perform other operations on the received signal, which are not limited in this disclosure.

[0267] In step S2110, the receiving device 102 acquires the first bit sequence.

[0268] In some embodiments, the receiving device 102 can input the symbol to be demodulated y into the demodulation model to obtain the first bit sequence output by the demodulation model.

[0269] Step S2111: The receiving device 102 determines the loss function.

[0270] In some embodiments, the receiving device 102 can base its signal on the first bit sequence output by the demodulation model. The difference between sample bit sequences b is used to determine the loss function.

[0271] In step S2112, the receiving device 102 trains the demodulation model.

[0272] In some embodiments, the receiving device 102 may calculate a first gradient value based on the loss function described above, wherein the first gradient value may be the partial derivative of the current loss function with respect to the network parameters of the demodulation model, used to indicate the direction of the largest rate of change of the loss function.

[0273] The receiving device 102 can backpropagate the first gradient value within the demodulation model to update the model parameters of the demodulation model, including but not limited to updating the weights and biases of each layer of the modulation model, thereby achieving the purpose of training the demodulation model. Here, the bias is a learnable parameter used in a neural network to adjust the output of each neuron, providing a learnable offset for neuron activation.

[0274] It is understood that backpropagation corresponds to forward propagation. Forward propagation can be understood as propagating from the input layer of the model to the intermediate layer (e.g., convolutional layer, pooling layer, activation function layer, fully connected layer, etc.) and then to the output layer. In the embodiments of this disclosure, forward propagation can propagate in the following directions: input layer of modulation model, intermediate layer of modulation model, output layer of modulation model (layer mapping module, precoding module, channel transmission, channel estimation module, channel equalization module), input layer of demodulation model, intermediate layer of demodulation model, and output layer of demodulation model.

[0275] Backpropagation is the opposite of forward propagation, and it propagates in the following directions: output layer of the demodulation model, intermediate layer of the demodulation model, input layer of the demodulation model, output layer of the (channel) modulation model, intermediate layer of the modulation model, and input layer of the modulation model.

[0276] In step S2113, the receiving device 102 sends the second gradient value to the sending device 101.

[0277] In some embodiments, the transmitting device 101 receives a second gradient value.

[0278] In some embodiments, the receiving device 102 transmits a second gradient value to the transmitting device 101 via a channel.

[0279] In some embodiments, the transmitting device 101 receives a second gradient value transmitted through the channel.

[0280] In some embodiments, the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model.

[0281] In some embodiments, model training typically requires multiple rounds to achieve model convergence. In this embodiment, since joint training is performed between the transmitting device 101 and the receiving device 102, sending the second gradient value to the transmitting device 101 after each round of training of the demodulation model would result in significant data and gradient transmission overhead, consuming substantial communication resources and impacting system operation. Therefore, after multiple rounds of updating the model parameters of the demodulation model, the second gradient value output by the input layer of the demodulation model during the last round (model parameter) update can be sent to the transmitting device 101.

[0282] In one example, the receiving device 101 can backpropagate the first gradient value within the demodulation model, complete the gradient calculation for each layer, update the weights and biases of the tillage layer using an optimization algorithm, such as stochastic gradient descent, and, if the third stopping condition is met, send the second gradient value output by the input layer of the demodulation model during the last round of model parameter update to the sending device 101.

[0283] The third stopping condition can be a condition for pausing the training of the demodulation model, including but not limited to at least one of the following: reaching a specific number of training cycles; the loss function being reduced to a specified fault tolerance range; and the accuracy of the first bit sequence reaching a second value.

[0284] In step S2114, the transmitting device 101 trains the modulation model.

[0285] In some embodiments, the transmitting device 101 may train the modulation model based on the received second gradient value using an optimization algorithm, such as stochastic gradient descent.

[0286] In some embodiments, the transmitting device 101 may backpropagate the second gradient value within the modulation model to update the model parameters of the modulation model, including but not limited to updating the weights and biases of each layer of the modulation model, thereby achieving the purpose of training the modulation model.

[0287] In step S2115, the receiving device 102 obtains the trained demodulation model.

[0288] In some embodiments, the receiving device 102 stops training the demodulation model when the second stopping condition is met, and obtains the trained demodulation model.

[0289] In one example, the second stopping condition is used to stop training the demodulation model.

[0290] In one example, the second stopping condition may include, but is not limited to, at least one of the following: reaching the number of training epochs; the loss function decreasing to within the tolerance range; and the accuracy of the first bit sequence reaching a first value.

[0291] For example, the receiving device 102 can update and adjust the model parameters of the demodulation model in one round, which can increment the number of training cycles by 1. The number of training cycles can be accumulated from 0 until it is equal to a predetermined number of training cycles, and the second stopping condition is met. At this point, the receiving device 102 can stop training the demodulation model.

[0292] The predetermined number of training cycles can be a positive integer, such as 1, 2, 3, etc. This predetermined number of training cycles can be agreed upon by the protocol and / or provided to the receiving device 102 by the network device, and this disclosure does not limit it.

[0293] The receiving device 102 can determine that the second stopping condition is met when the loss function is reduced to a certain range, such as within the first fault tolerance range, and at this time the training of the demodulation model can be stopped.

[0294] The receiving device 102 can determine that the first stopping condition is met when the accuracy of the first bit sequence output by the demodulation model reaches a first value, and at this time the training of the demodulation model can be stopped.

[0295] The receiving device 102 may also determine that the second stopping condition is met under other circumstances, such as when the learning rate of the demodulation model reaches the second value, which is not limited in this disclosure.

[0296] In step S2116, the receiving device 102 sends the second indication information to the sending device 101.

[0297] In some embodiments, the transmitting device 101 receives second indication information.

[0298] In some embodiments, the second indication information is used to instruct the transmitting device 101 to stop training the modulation model.

[0299] In some embodiments, the receiving device 102 may send a second indication message to the sending device 101 when the second stop condition is met.

[0300] In some embodiments, the receiving device 102 may stop sending a new second gradient value to the sending device 101 if the second stopping condition is met.

[0301] In some embodiments, step S2112 is an optional execution step. For example, if the transmitting device 102 stops training the modulation model based on other conditions, step S2112 may not be executed.

[0302] In step S2117, the transmitting device 101 obtains the trained modulation model.

[0303] In some embodiments, the transmitting device 101 stops training the modulation model when the first stopping condition is met, and obtains the trained modulation model.

[0304] In one example, the first stopping condition is used to stop training the modulation model.

[0305] In one example, the first stopping condition may include, but is not limited to, at least one of the following: not receiving a new second gradient value sent by the receiving device 102; receiving second indication information sent by the receiving device 102.

[0306] The second indication information is used to instruct the transmitting device 101 to stop training the modulation model.

[0307] Considering that the transmitting device 101 cannot determine the loss function of the demodulation model during joint training, the transmitting device 101 can determine that the first stopping condition is met and stop training the modulation model if it does not receive a new second gradient value and / or receives the aforementioned second indication information, thereby obtaining the trained modulation model.

[0308] In some embodiments, the names of information, etc., are not limited to the names described in the embodiments. Terms such as "information", "message", "signal", "signaling", "report", "configuration", "indication", "instruction", "command", "channel", "parameter", "domain", "field", "symbol", "symbol", "codebook", "codeword", "codepoint", "bit", "data", "program", and "chip" can be used interchangeably.

[0309] In some embodiments, "acquire," "get," "obtain," "receive," "transmit," "bidirectional transmission," and "send and / or receive" can be used interchangeably and can be interpreted as receiving from other entities, acquiring from protocols, acquiring from higher layers, obtaining through self-processing, or autonomous implementation. Protocols include, for example, at least one of the 3GPP protocol, Wi-Fi protocol, and audio and / or video protocols.

[0310] In some embodiments, terms such as “send,” “transmit,” “report,” “distribute,” “transfer,” “bidirectional transmission,” “send and / or receive” can be used interchangeably.

[0311] In some embodiments, terms such as "certain," "preset," "default," "set," "indicated," "a certain," "any," and "first" can be used interchangeably. "Certain A," "preset A," "default A," "set A," "indicated A," "a certain A," "any A," and "first A" can be interpreted as A pre-defined in a protocol or the like, or as A obtained through setting, configuration, or instruction, or as specific A, a certain A, any A, or first A, but are not limited thereto.

[0312] In some embodiments, the information transmission method involved in the present disclosure may include at least one of steps S2101 to S2117. For example, step S2101 can be implemented as an independent embodiment, step S2102 can be implemented as an independent embodiment, step S2101+S2102 can be implemented as an independent embodiment, step S2103 can be implemented as an independent embodiment, step S2104 can be implemented as an independent embodiment, step S2103+S2104 can be implemented as an independent embodiment, step S2105 can be implemented as an independent embodiment, step S2106 can be implemented as an independent embodiment, step S2105+S2106 can be implemented as an independent embodiment, step S2107 can be implemented as an independent embodiment, step S2105+S2106+S2107 can be implemented as an independent embodiment, step S2108 can be implemented as an independent embodiment, steps S2101 to S2108 can be implemented as independent embodiments, and step S2109 can be implemented as an independent embodiment. In this embodiment, steps S2101 to S2109 can be implemented as independent embodiments, as can step S2110, step S2111, step S2110 + step S2111, step S2112, step S2113, step S2112 + step S2113, step S2114, step S2115, step S2116, step S2117, step S2115 + S2116 + S2117, and steps S2101 to S2117 can be implemented as independent embodiments, but are not limited thereto.

[0313] In some embodiments, steps S2101 to S2117 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0314] In some embodiments, the execution order of steps S2101 to S217 is not limited.

[0315] In the above embodiments, the modulation model and demodulation model can be jointly trained online. Compared with the offline joint training method, the deployment latency of the modulation model and demodulation model is reduced. Moreover, this joint training method effectively improves the accuracy of data transmission, improves the performance of the communication system, and has high availability.

[0316] In some embodiments, the modulation model and / or demodulation model described above may use any of the following models as the backbone network: residual network, visual geometry group (VGG) network, or GoogleNet. The model is updated based on the online adjusted or online updated model parameters. This disclosure does not limit the type of backbone network.

[0317] Figure 3A is a flowchart illustrating a model training method according to an embodiment of the present disclosure. As shown in Figure 3A, this embodiment of the disclosure relates to a model training method, which can be executed by a transmitting device 101 and includes the following steps:

[0318] Step S3101: Obtain the first modulation symbol.

[0319] In some embodiments, step S3101 may refer to steps in other embodiments described before or after this embodiment, such as step S2106 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0320] Step S3102: Send the second modulation symbol.

[0321] In some embodiments, the transmitting device 101 transmits a second modulated signal to the receiving device 102.

[0322] In some embodiments, the receiving device 102 receives a second modulation symbol to obtain a received signal.

[0323] In some embodiments, step S3102 may refer to steps in other embodiments described before or after this embodiment, such as step S2108 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0324] Step S3103: Obtain the second gradient value.

[0325] In some embodiments, the transmitting device 101 receives a second gradient value sent by the receiving device 102, but is not limited thereto. The transmitting device 101 may also receive a second gradient value sent by other entities, such as relay devices or other terminals, in which case step S3103 may be omitted.

[0326] In some embodiments, the transmitting device 101 obtains the second gradient value specified by the protocol, in which case step S3103 is omitted.

[0327] In some embodiments, the transmitting device 101 obtains the second gradient value from the upper layer(s), in which case step S3103 is omitted.

[0328] In some embodiments, the transmitting device 101 processes the data to obtain a second gradient value, in which case step S3103 is omitted.

[0329] In some embodiments, the transmitting device 101 autonomously implements the function indicated by the second gradient value, or the above function is a default or default value, in which case step S3103 is omitted.

[0330] In some embodiments, step S3103 may refer to steps in other embodiments described before or after this embodiment, such as step S2113 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0331] Step S3104: Train the modulation model.

[0332] In some embodiments, step S3104 may refer to steps in other embodiments described before or after this embodiment, such as step S2114 in FIG2 and its optional implementation, as well as other related parts in the specification, which will not be repeated here.

[0333] Step S3105: The trained modulation model is obtained.

[0334] In some embodiments, step S3105 may refer to steps in other embodiments described before or after this embodiment, such as step S2117 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0335] In some embodiments, steps S3101 to S3105 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0336] In some embodiments, the execution order of steps S3101 to S3105 is not limited.

[0337] In the above embodiments, the transmitting device can receive the second gradient value sent by the receiving device, thereby completing the training of the modulation model online, improving the accuracy of data transmission, improving the performance of the communication system, and increasing its availability.

[0338] Figure 3B is a flowchart illustrating a model training method according to an embodiment of the present disclosure. As shown in Figure 3B, this embodiment of the disclosure relates to a model training method, which can be executed by a transmitting device 101, and includes the following steps:

[0339] Step S3201: Determine the first initialization parameters.

[0340] In some embodiments, step S3201 may refer to steps in other embodiments described before or after this embodiment, such as step S2101 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0341] Step S3202: Initialize the modulation model.

[0342] In some embodiments, step S3202 may refer to steps in other embodiments described before or after this embodiment, such as step S2102 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0343] Step S3203: Obtain the set of sample bit sequences.

[0344] In some embodiments, step S3203 may refer to steps in other embodiments described before or after this embodiment, such as step S2105 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0345] Step S3204: Obtain the first modulation symbol.

[0346] In some embodiments, step S3204 may refer to steps in other embodiments described before or after this embodiment, such as step S2106 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0347] Step S3205: Determine the second modulation symbol.

[0348] In some embodiments, step S3205 may refer to steps in other embodiments described before or after this embodiment, such as step S2107 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0349] Step S3206: Send the second modulation signal.

[0350] In some embodiments, the transmitting device 101 transmits a second modulated signal to the receiving device 102.

[0351] In some embodiments, the receiving device 102 receives a second modulation symbol to obtain a received signal.

[0352] In some embodiments, step S3206 may refer to steps in other embodiments described before or after this embodiment, such as step S2108 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0353] Step S3207: Obtain the second gradient value.

[0354] In some embodiments, the transmitting device 101 receives a second gradient value sent by the receiving device 102, but is not limited thereto. The transmitting device 101 may also receive a second gradient value sent by other entities, such as relay devices or other terminals, in which case step S3207 may be omitted.

[0355] In some embodiments, the transmitting device 101 obtains the second gradient value specified by the protocol, in which case step S3207 is omitted.

[0356] In some embodiments, the transmitting device 101 obtains the second gradient value from the upper layer(s), in which case step S3207 is omitted.

[0357] In some embodiments, the transmitting device 101 processes the data to obtain a second gradient value, in which case step S3207 is omitted.

[0358] In some embodiments, the transmitting device 101 autonomously implements the function indicated by the second gradient value, or the above function is default or default, in which case step S3207 is omitted.

[0359] In some embodiments, step S3207 may refer to steps in other embodiments described before or after this embodiment, such as step S2113 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0360] Step S3208: Train the modulation model.

[0361] In some embodiments, step S3208 may refer to steps in other embodiments described before or after this embodiment, such as step S2114 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0362] Step S3209: Obtain the second instruction information.

[0363] In some embodiments, the transmitting device 101 receives second indication information sent by the receiving device 102, but is not limited thereto. The transmitting device 101 may also receive second indication information sent by other executing entities, such as relay devices or other terminals, in which case step S3209 may be omitted.

[0364] In some embodiments, the transmitting device 101 obtains the second indication information specified by the protocol, in which case step S3209 is omitted.

[0365] In some embodiments, the transmitting device 101 obtains the second indication information from the upper layer(s), in which case step S3209 is omitted.

[0366] In some embodiments, the transmitting device 101 processes the information to obtain the second indication information, in which case step S3209 is omitted.

[0367] In some embodiments, the transmitting device 101 autonomously implements the function indicated by the second indication information, or the above function is default or default, in which case step S3209 is omitted.

[0368] In some embodiments, step S3209 may refer to steps in other embodiments described before or after this embodiment, such as step S2116 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0369] Step S3210: The trained modulation model is obtained.

[0370] In some embodiments, step S3210 may refer to steps in other embodiments described before or after this embodiment, such as step S2117 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0371] In some embodiments, steps S3201 to S3210 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0372] In some embodiments, the execution order of steps S3201 to S3210 is not limited.

[0373] In the above embodiments, the transmitting device can complete the training of the modulation model online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0374] Figure 3C is a flowchart illustrating a model training method according to an embodiment of the present disclosure. As shown in Figure 3C, this embodiment relates to a model training method, which can be executed by a receiving device 102, and includes the following steps:

[0375] Step S3301: Obtain the second modulation signal.

[0376] In some embodiments, the receiving device 102 can obtain a received signal by receiving a second modulation signal.

[0377] In some embodiments, the receiving device 102 may receive the second modulation symbol sent by the transmitting device 101, but is not limited thereto. The receiving device 102 may also receive the second modulation symbol sent by other entities, such as relay devices or other terminals, in which case step S3301 may be omitted.

[0378] In some embodiments, the receiving device 102 acquires the second modulation symbol specified by the protocol, in which case step S3301 is omitted.

[0379] In some embodiments, the receiving device 102 obtains the second modulation symbol from the upper layer(s), in which case step S3301 is omitted.

[0380] In some embodiments, the receiving device 102 processes the data to obtain a second modulation symbol, in which step S3301 is omitted.

[0381] In some embodiments, the receiving device 102 autonomously implements the function indicated by the second modulation symbol, or the above function is default or default, in which case step S3301 is omitted.

[0382] In some embodiments, step S3301 can refer to the steps in other embodiments described before or after this embodiment, such as step S2108 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0383] Step S3302: Obtain the first bit sequence.

[0384] In some embodiments, step S3302 may refer to steps in other embodiments described before or after this embodiment, such as step S2110 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0385] Step S3303: Determine the loss function.

[0386] In some embodiments, step S3303 may refer to steps in other embodiments described before or after this embodiment, such as step S2111 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0387] Step S3304: Train the demodulation model.

[0388] In some embodiments, step S3304 may refer to steps in other embodiments described before or after this embodiment, such as step S2112 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0389] Step S3305: Send the second gradient value.

[0390] In some embodiments, the receiving device 102 sends a second gradient value to the sending device 101.

[0391] In some embodiments, the transmitting device 101 receives a second gradient value.

[0392] In some embodiments, step S3305 may refer to steps in other embodiments described before or after this embodiment, such as step S2113 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0393] Step S3306 yields the trained demodulation model.

[0394] In some embodiments, step S3306 may refer to steps in other embodiments described before or after this embodiment, such as step S2114 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0395] In some embodiments, steps S3301 to S3306 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0396] In some embodiments, the execution order of steps S3301 to S3306 is not limited.

[0397] In the above embodiments, the receiving device can complete the training of the demodulation model online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0398] Figure 3D is a flowchart illustrating a model training method according to an embodiment of the present disclosure. As shown in Figure 3D, this embodiment relates to a model training method, which can be executed by a receiving device 102, and includes the following steps:

[0399] Step S3401: Determine the second initialization parameters.

[0400] In some embodiments, step S3401 may refer to steps in other embodiments described before or after this embodiment, such as step S2103 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0401] Step S3402: Initialize the demodulation model.

[0402] In some embodiments, step S3402 may refer to steps in other embodiments described before or after this embodiment, such as step S2104 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0403] Step S3403: Obtain the second modulation signal.

[0404] In some embodiments, the receiving device 102 can obtain a received signal by receiving a second modulation signal.

[0405] In some embodiments, the receiving device 102 may receive the second modulation symbol sent by the transmitting device 101, but is not limited thereto. The receiving device 102 may also receive the second modulation symbol sent by other entities, such as relay devices or other terminals, in which case step S3403 may be omitted.

[0406] In some embodiments, the receiving device 102 acquires the second modulation symbol specified by the protocol, in which case step S3403 is omitted.

[0407] In some embodiments, the receiving device 102 obtains the second modulation symbol from the upper layer(s), in which case step S3403 is omitted.

[0408] In some embodiments, the receiving device 102 processes the data to obtain a second modulation symbol, in which step S3403 is omitted.

[0409] In some embodiments, the receiving device 102 autonomously implements the function indicated by the second modulation symbol, or the above function is default or default, in which case step S3403 is omitted.

[0410] In some embodiments, step S3403 may refer to steps in other embodiments described before or after this embodiment, such as step S2108 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0411] Step S3404: Determine the symbol to be demodulated.

[0412] In some embodiments, step S3404 may refer to steps in other embodiments described before or after this embodiment, such as step S2109 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0413] Step S3405: ​​Obtain the first bit sequence.

[0414] In some embodiments, step S3405 may refer to steps in other embodiments described before or after this embodiment, such as step S2110 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0415] Step S3406: Determine the loss function.

[0416] In some embodiments, step S3406 may refer to steps in other embodiments described before or after this embodiment, such as step S2111 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0417] Step S3407: Train the demodulation model.

[0418] In some embodiments, step S3407 may refer to steps in other embodiments described before or after this embodiment, such as step S2112 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0419] Step S3408: Send the second gradient value.

[0420] In some embodiments, the receiving device 102 sends a second gradient value to the sending device 101.

[0421] In some embodiments, the transmitting device 101 receives a second gradient value.

[0422] In some embodiments, step S3408 may refer to steps in other embodiments described before or after this embodiment, such as step S2113 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0423] Step S3409 yields the trained demodulation model.

[0424] In some embodiments, step S3409 may refer to steps in other embodiments described before or after this embodiment, such as step S2114 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0425] Step S3410: Send the second instruction information.

[0426] In some embodiments, the receiving device 102 sends a second indication information to the sending device 101.

[0427] In some embodiments, the transmitting device 101 receives second indication information.

[0428] In some embodiments, step S3410 may refer to steps in other embodiments described before or after this embodiment, such as step S2116 in FIG2 and its optional implementation, and other related parts in the specification, which will not be repeated here.

[0429] In some embodiments, steps S3401 to S3410 are optional, and one or more of these steps may be omitted or substituted in different embodiments.

[0430] In some embodiments, the execution order of steps S3401 to S3410 is not limited.

[0431] In the above embodiments, the receiving device can complete the training of the demodulation model online, which improves the accuracy of data transmission, enhances the performance of the communication system, and increases its availability.

[0432] The above process is further illustrated with examples below.

[0433] Applying AI technology to both data transmitters and receivers has broad application and research value. Data modulation and demodulation, as fundamental and crucial modules in both ends, directly impact data transmission accuracy and communication system throughput. While neural network models can enhance the data demodulation function at the receiver and improve demodulation accuracy, using AI models to simultaneously enhance both modulation and demodulation models further improves data transmission accuracy and overall communication system performance, thus possessing significant research and application value.

[0434] In this embodiment, the modulation model is deployed at the transmitting end (device), and the demodulation model is deployed at the receiving end (device). The two models require processing by other modules at the transmitting end, channel transmission, and further processing by other modules at the receiving end. To complete the joint training of the modulation and demodulation models, the output of the modulation model needs to be processed by other modules, transmitted through the channel, and then input into the demodulation model. However, during model training, it is impossible to use fixed data processing or calculation methods to reflect the impact of channel transmission on the data; therefore, conventional offline model training methods cannot be used to complete the joint training of the modulation and demodulation models.

[0435] This disclosure addresses the joint training of models in AI-based data modulation and demodulation enhancement technologies. It proposes an online training method for modulation and demodulation models, completing online data acquisition and model training within a real-world model deployment environment. During training, it directly utilizes data from the transmitter, receiver, other modules, and channel transmission modulation and demodulation models within the actual system. Furthermore, the trained modulation and demodulation models can be effectively applied to real-world channel environments. As shown in Figure 4, the method includes the following steps:

[0436] (1) Data preparation.

[0437] A set of data Data is randomly generated or manually set. Each sample data b in this set of data is a bit data sequence, which is used as the encoded bit data of the sending end to complete the subsequent model training.

[0438] The dataset used for model training is transmitted from the sender to the receiver during the training process. Therefore, it is usually determined by the sender or by the network-side functional unit responsible for AI model management and instructed to the sender.

[0439] (2) Model initialization.

[0440] The determination of modulation and demodulation model structures, training parameter settings, and model parameter initialization can be determined by the functional unit responsible for AI model management on the network side and instructed to the transmitting and receiving ends respectively, or it can be determined by the transmitting or receiving end and instructed to the other end before model training.

[0441] The model structure depends on the parameter configuration and channel environment complexity used in the actual application. Training parameter configuration includes, but is not limited to, loss function type, learning rate, optimizer type, maximum training period, and required model output accuracy. Methods for model parameter initialization include, but are not limited to, zero initialization, random initialization, Xavier initialization, and Kaiming initialization.

[0442] (3) Online training of modulation and demodulation models.

[0443] Forward propagation: Data sample b in dataset Data is processed by the modulation model at the transmitting end to obtain modulation symbol x. Modulation symbol x is then processed by other modules at the transmitting end, transmitted through the actual channel, processed by other modules at the receiving end, and input into the demodulation model to obtain demodulated bits. Computational model output The error between the target output b and the target output b, i.e., the loss function;

[0444] Backpropagation: The gradient is calculated based on the loss function and the gradient value is backpropagated. The gradient value is backpropagated from the output layer to each hidden layer. During this process, the receiver needs to transmit the gradient value to the sender.

[0445] Parameter update: After completing the gradient transfer and calculation of each layer of the modulation and demodulation models, the weights and biases of each layer in the modulation and demodulation models are updated using optimization algorithms (such as stochastic gradient descent).

[0446] The model is trained in multiple rounds. In each round, forward propagation, back propagation, and data and gradient transfer are completed until the model converges, such as when the loss function is reduced to a certain threshold or the demodulated data accuracy reaches a certain requirement, and then the model parameters are updated.

[0447] Since model convergence typically requires multiple training rounds, sending data from the sender to the receiver during forward propagation and sending gradients from the receiver to the sender during backpropagation in each training round would incur significant data and gradient transmission overhead, consuming substantial communication link resources and impacting system performance. Therefore, the following method can be considered to achieve a similar effect to online model updates.

[0448] (1) Data preparation (2) Model initialization remains unchanged

[0449] (3) Online training of modulation and demodulation models

[0450] Forward propagation: The transmitter takes data sample b from the dataset Data, passes it through the transmitter's modulation model to obtain modulation symbol x, processes modulation symbol x through other modules at the transmitter, transmits it through the actual channel, processes it through other modules at the receiver, and inputs it into the demodulation model to obtain demodulated bits. Computational model output The error between the target output b and the target output b, i.e., the loss function;

[0451] Backpropagation of the demodulation model: The receiver calculates the gradient based on the loss function and backpropagates the gradient value. The gradient value is backpropagated from the output layer of the demodulation model to each hidden layer and input layer of the demodulation model.

[0452] Demodulation model parameter update: After completing the gradient calculation of each layer of the demodulation model, use an optimization algorithm (such as stochastic gradient descent) to update the weights and biases of each layer in the demodulation model; when the loss function reaches a certain range, the error between the output of the demodulation model and the target output reaches a certain value, or a certain number of rounds of demodulation model training are performed, the demodulation model training is stopped.

[0453] The receiving end transmits the gradient of the input layer of the demodulation model during the last round of parameter update to the sending end;

[0454] Backpropagation of the modulation model: The backpropagation of the modulation model is completed based on the gradient values ​​from the receiver, from the input layer of the modulation model to each hidden layer and the input layer of the modulation model;

[0455] Modulation model parameter update: After completing the gradient calculation of each layer of the modulation model, the weights and biases of each layer in the modulation model are updated using an optimization algorithm (such as stochastic gradient descent).

[0456] The model training process involves multiple rounds, including forward propagation, modulation, demodulation, backpropagation, and parameter updates in each round until the model converges. This process is repeated when the loss function is reduced to a certain threshold or the demodulated data accuracy meets certain requirements, at which point the model parameters are updated. Each round of model training requires only one gradient transfer from the receiver to the transmitter, reducing the data transmission overhead for online model training.

[0457] (4) Application of modulation and demodulation models

[0458] In practical model applications, modulation and demodulation models are deployed. The encoded bit b is processed by the modulation model, then processed by other transmitting and receiving modules in the actual system, and transmitted through the channel. Finally, the demodulated bit is obtained through the demodulation model.

[0459] In the above embodiments, for technical solutions based on AI models to achieve modulation at the transmitting end and demodulation at the receiving end, and addressing the situation where the channel transmission cannot be reflected during the joint training of the modulation and demodulation models, an online training method for modulation and demodulation models is proposed. This method completes the training data acquisition, forward propagation, and backward propagation processes in the actual model deployment environment to update the model parameters, thereby completing the online training of the modulation and demodulation models. This helps to promote the application of AI-based modulation and demodulation schemes in actual communication systems and improve the accuracy of data transmission and the overall performance of the communication system.

[0460] This disclosure also proposes an apparatus (also referred to as a communication device, etc.) for implementing any of the above methods. For example, an apparatus is proposed, which includes units or modules for implementing the steps performed by the transmitting end device or the receiving end device in any of the above methods.

[0461] It should be understood that the division of units or modules in the above device is only a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, the units or modules in the device can be implemented by a processor calling software: for example, the device includes a processor connected to a memory containing instructions. The processor calls the instructions stored in the memory to implement any of the above methods or to implement the functions of the units or modules in the above device. The processor can be, for example, a general-purpose processor, such as a Central Processing Unit (CPU) or a microprocessor, and the memory can be internal or external to the device. Alternatively, the units or modules in the device can be implemented in the form of hardware circuits. The functionality of some or all of the units or modules can be achieved through the design of these hardware circuits, which can be understood as one or more processors. For example, in one implementation, the hardware circuit is an application-specific integrated circuit (ASIC). The functionality of some or all of the units or modules is achieved through the design of the logical relationships between the components within the circuit. In another implementation, the hardware circuit can be implemented using a programmable logic device (PLD). Taking a field-programmable gate array (FPGA) as an example, it can include a large number of logic gates. The connection relationships between the logic gates are configured through configuration files, thereby achieving the functionality of some or all of the units or modules. All units or modules of the above device can be implemented entirely through processor-called software, entirely through hardware circuits, or partially through processor-called software with the remaining parts implemented through hardware circuits.

[0462] In this embodiment, the processor is a circuit with signal processing capabilities. In one implementation, the processor can be a circuit with instruction read and execute capabilities, such as a Central Processing Unit (CPU), a microprocessor, a graphics processing unit (GPU) (which can be understood as a microprocessor), or a digital signal processor (DSP). In another implementation, the processor can implement certain functions through the logical relationships of hardware circuits. The logical relationships of the aforementioned hardware circuits are fixed or reconfigurable. For example, the processor is a hardware circuit implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD), such as an FPGA. In a reconfigurable hardware circuit, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the process of the processor loading instructions to implement the functions of some or all of the above units or modules. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as an ASIC, such as a Neural Network Processing Unit (NPU), a Tensor Processing Unit (TPU), or a Deep Learning Processing Unit (DPU).

[0463] Figure 5A is a schematic diagram of the structure of the transmitting device according to an embodiment of this disclosure. The transmitting device 5100 is used to perform any of the above methods. In some embodiments, as shown in Figure 5B, the transmitting device 5100 may include at least one of a processing module 5101 and a transceiver module 5102.

[0464] In some embodiments, the processing module 5101 is used to input the sample bit sequence into the modulation model, obtain the first modulation symbol output by the modulation model; train the modulation model based on the second gradient value; stop training the modulation model when the first stopping condition is met, and obtain the trained modulation model.

[0465] In some embodiments, the transceiver module 5102 is used to receive a second gradient value sent by the receiving device; wherein the second modulation symbol is used by the receiving device to determine the second gradient value, and the second gradient value is the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by transmitting the first gradient value in reverse within the demodulation model.

[0466] Optionally, the processing module 5101 is used to execute at least one of the other steps (such as steps S2101, S2102, S2103, S2110, and S2113, but not limited thereto) executed by the transmitting device 5100 in any of the above methods, which will not be described in detail here.

[0467] Optionally, the transceiver module 5102 is used to perform at least one of the communication steps such as receiving and / or sending performed by the transmitting device 5100 in any of the above methods (e.g., step S2104, step S2109, step S2112, but not limited thereto), which will not be described in detail here.

[0468] Figure 5B is a schematic diagram of the structure of the receiving device according to an embodiment of this disclosure. The receiving device 5200 is used to perform any of the above methods. In some embodiments, as shown in Figure 5B, the receiving device 5200 may include at least one of a transceiver module 5201 and a processing module 5202.

[0469] In some embodiments, the transceiver module 5201 is used to receive a second modulation symbol transmitted by the transmitting device through the channel to obtain a received signal; and to send a second gradient value to the transmitting device; wherein the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model.

[0470] In some embodiments, the processing module 5202 is configured to input the symbol to be demodulated into the demodulation model and obtain a first bit sequence output by the demodulation model; wherein the symbol to be demodulated is obtained after performing a second operation on the received signal; determine a loss function based on the difference between the first bit sequence and the sample bit sequence; and train the demodulation model based on a first gradient value; wherein the first gradient value is determined based on the loss function.

[0471] Optionally, the transceiver module 5201 is used to perform at least one of the communication steps (such as step S2104, step S2109, step S2112, but not limited thereto) performed by the receiving device 5200 in any of the above methods, which will not be described in detail here.

[0472] Optionally, the processing module 5202 is used to execute at least one of the other steps (such as steps S2105, S2106, S2107, S2108, and S2111, but not limited thereto) executed by the receiving device 5200 in any of the above methods, which will not be elaborated here.

[0473] In some embodiments, the transceiver module may include a transmitting module and / or a receiving module, which may be separate or integrated. Optionally, the transceiver module may be interchangeable with a transceiver.

[0474] In some embodiments, the processing module may be a single module or may include multiple sub-modules. Optionally, the multiple sub-modules may each perform all or part of the steps required by the processing module.

[0475] In some embodiments, the processing module can be replaced by the processor, and the transceiver module can be replaced by the transceiver.

[0476] Figure 6A is a schematic diagram of the structure of the communication device 6100 proposed in an embodiment of this disclosure. The communication device 6100 can be a first device, a transmitting device, or a receiving device, or it can be a chip, chip system, or processor that supports the first device, transmitting device, or receiving device in implementing any of the above methods. The communication device 6100 can be used to implement the methods described in the above method embodiments, and for details, please refer to the description in the above method embodiments.

[0477] As shown in Figure 6A, the communication device 6100 is used to execute any of the above methods. In some embodiments, the communication device 6100 includes one or more processors 6101. The processor 6101 may be a general-purpose processor or a special-purpose processor, such as a baseband processor or a central processing unit. The baseband processor may be used to process communication protocols and communication data, and the central processing unit may be used to control communication devices (e.g., base stations, baseband chips, terminal devices, terminal device chips, DUs or CUs, etc.), execute programs, and process program data. Optionally, the communication device 6100 is used to execute any of the above methods. Optionally, one or more processors 6101 are used to invoke instructions to cause the communication device 6100 to execute any of the above methods.

[0478] In some embodiments, the communication device 6100 further includes one or more transceivers 6102. When the communication device 6100 includes one or more transceivers 6102, the transceivers 6102 perform at least one of the communication steps such as sending and / or receiving in the above method (e.g., steps S2104, S2109, and S2112, but not limited thereto), and the processor 6101 performs at least one of other steps (e.g., steps S2101, S2102, S2103, S2105, S2106, S2107, S2108, S2110, S2113, and not limited thereto). In optional embodiments, the transceivers may include a receiver and / or a transmitter, which may be separate or integrated together. Optionally, terms such as transceiver, transceiver unit, transceiver, transceiver circuit, interface circuit, and interface can be used interchangeably; terms such as transmitter, transmitter unit, transmitter, and transmitter circuit can be used interchangeably; and terms such as receiver, receiver unit, receiver, and receiver circuit can be used interchangeably.

[0479] In some embodiments, the communication device 6100 further includes one or more memories 6103 for storing data and / or instructions. Optionally, one or more processors 6101 are used to invoke instructions stored in the memory 6103 to cause the communication device 6100 to perform any of the above methods. Optionally, all or part of the memory 6103 may also be located outside the communication device 6100. In an optional embodiment, the communication device 6100 may include one or more interface circuits 6104. Optionally, the interface circuit 6104 is connected to the memory 6103 and can be used to receive data and / or instructions from the memory 6103 or other devices, and can be used to send data and / or instructions to the memory 6103 or other devices. For example, the interface circuit 6104 can read data and / or instructions stored in the memory 6103 and send the data and / or instructions to the processor 6101.

[0480] The communication device 6100 described in the above embodiments may be a network device or a terminal, but the scope of the communication device 6100 described in this disclosure is not limited thereto, and the structure of the communication device 6100 may not be limited by FIG. 6A. The communication device may be a standalone device or a part of a larger device. For example, the communication device may be: (1) a standalone integrated circuit IC, or chip, or chip system or subsystem; (2) a collection of one or more ICs, optionally, the IC collection may also include storage components for storing data, programs and / or instructions; (3) an ASIC, such as a modem; (4) a module that can be embedded in other devices; (5) a receiver, terminal device, smart terminal device, cellular phone, wireless device, handheld device, mobile unit, vehicle device, network device, cloud device, artificial intelligence device, etc.; (6) others, etc.

[0481] Figure 6B is a schematic diagram of the structure of chip 6200 according to an embodiment of this disclosure. For cases where the communication device 6100 can be a chip or a chip system, please refer to the schematic diagram of chip 6200 shown in Figure 6B, but it is not limited thereto.

[0482] Chip 6200 includes one or more processors 6201. Chip 6200 is used to perform any of the methods described above.

[0483] In some embodiments, chip 6200 further includes one or more interface circuits 6202. Optionally, terms such as interface circuit, interface, and transceiver pin can be used interchangeably. In some embodiments, chip 6200 further includes one or more memories 6203 for storing data and / or instructions. Optionally, all or part of the memories 6203 may be located outside of chip 6200. Optionally, interface circuit 6202 is connected to memory 6203, and interface circuit 6202 can be used to receive data and / or instructions from memory 6203 or other devices, and interface circuit 6202 can be used to send data and / or instructions to memory 6203 or other devices. For example, interface circuit 6202 can read data and / or instructions stored in memory 6203 and send the data and / or instructions to processor 6201.

[0484] In some embodiments, the interface circuit 6202 performs at least one of the communication steps such as sending and / or receiving in the above method (e.g., steps S2104, S2109, and S2112, but not limited thereto). The interface circuit 6202 performing the communication steps such as sending and / or receiving in the above method refers, for example, to the interface circuit 6202 performing data and / or instruction interaction between the processor 6201, the chip 6200, the memory 6203, or the transceiver device. In some embodiments, the processor 6201 performs at least one of other steps (e.g., steps S2101, S2102, S2103, S2105, S2106, S2107, S2108, S2110, S2113, but not limited thereto).

[0485] The modules and / or devices described in the various embodiments, such as virtual devices, physical devices, and chips, can be combined or separated arbitrarily as needed. Optionally, some or all steps can also be performed collaboratively by multiple modules and / or devices, which is not limited here.

[0486] This disclosure also proposes a storage medium storing instructions that, when executed on a communication device, cause the communication device to perform any of the above methods. Optionally, the storage medium is an electronic storage medium. Optionally, the storage medium is a computer-readable storage medium, but not limited thereto; it may also be a storage medium readable by other devices. Optionally, the storage medium may be a non-transitory storage medium, but not limited thereto; it may also be a temporary storage medium.

[0487] This disclosure also proposes a program product, including a program and / or instructions, which, when executed by a communication device, cause the communication device to perform any of the above methods. Optionally, the program product is a computer program product. Optionally, the program product is stored on the storage medium.

[0488] This disclosure also proposes a computer program that, when run on a computer, causes the computer to perform any of the above methods.

[0489] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0490] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A model training method, characterized in that, The method is executed by the sending device, and the method includes: Input the sample bit sequence into the modulation model and obtain the first modulation symbol output by the modulation model; The second modulation symbol is transmitted to the receiving device through the channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving device to determine a second gradient value, the second gradient value being the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by reverse transmitting the first gradient value within the demodulation model; Receive the second gradient value sent by the receiving device; The modulation model is trained based on the second gradient value; If the first stopping condition is met, training of the modulation model is stopped, and a trained modulation model is obtained.

2. The method according to claim 1, characterized in that, The method further includes at least one of the following: randomly generating a set of sample bit sequences; According to the first rule, set up the sample bit sequence set; Receive first indication information sent by a network device; wherein the first indication information is used to indicate a set of sample bit sequences.

3. The method according to claim 1 or 2, characterized in that, The first operation includes at least one of the following: Layer mapping operations; Precoding operation.

4. The method according to any one of claims 1-3, characterized in that, The step of training the modulation model based on the second gradient value includes: The second gradient value is backpropagated within the modulation model to update the model parameters of the modulation model.

5. The method according to any one of claims 1-4, characterized in that, The first stopping condition includes at least one of the following: No new second gradient value was received from the receiving device; The receiving end device receives a second indication message, which is used to instruct the transmitting end device to stop training the modulation model.

6. The method according to any one of claims 1-5, characterized in that, The second gradient value is the gradient value output by the input layer of the demodulation model during the last round of updating the model parameters of the demodulation model after the receiving device transmits the first gradient value in reverse within the demodulation model.

7. The method according to any one of claims 1-6, characterized in that, The method further includes: Determine the first initialization parameters; wherein the first initialization parameters include the structural parameters of the modulation model and / or the training parameters of the modulation model; The modulation model is initialized based on the first initialization parameters.

8. The method according to claim 7, characterized in that, The determination of the first initialization parameter includes at least one of the following: The first initialization parameter is determined based on the third indication information sent by the network device; wherein the third indication information is used to indicate the first initialization parameter. Based on the fourth indication information sent by the receiving device, the first initialization parameter is determined; wherein, the fourth indication information is used to indicate the first initialization parameter; Based on the second rule, the first initialization parameter is determined.

9. The method according to any one of claims 1-8, characterized in that, The method further includes: Based on the third rule, a second initialization parameter is determined; wherein the second initialization parameter includes the structural parameters of the demodulation model and / or the training parameters of the demodulation model; Send a fifth indication message to the receiving device; wherein the fifth indication message is used to indicate the second initialization parameter.

10. A model training method, characterized in that, The method is executed by the receiving device, and the method includes: The receiving end device obtains the received signal by transmitting the second modulation symbol through the channel; The symbol to be demodulated is input into the demodulation model to obtain the first bit sequence output by the demodulation model; wherein, the symbol to be demodulated is obtained by performing a second operation on the received signal; The loss function is determined based on the difference between the first bit sequence and the sample bit sequence; The demodulation model is trained based on the first gradient value, and a second gradient value is sent to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model; If the second stopping condition is met, training of the demodulation model is stopped, and a trained demodulation model is obtained.

11. The method according to claim 10, characterized in that, The second operation includes at least one of the following: Channel estimation operation; Channel equalization operation.

12. The method according to claim 10 or 11, characterized in that, The step of training the demodulation model based on the first gradient value includes: The first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model.

13. The method according to any one of claims 10-12, characterized in that, The second stopping condition includes at least one of the following: Reach the required number of training cycles; The loss function is reduced to a fault-tolerant range; The accuracy of the first bit sequence reaches the first value.

14. The method according to any one of claims 10-13, characterized in that, The method further includes at least one of the following: If the second stopping condition is met, the transmission of a new second gradient value to the transmitting device is stopped. If the second stopping condition is met, a second indication message is sent to the transmitting device; wherein the second indication message is used to instruct the transmitting device to stop training the modulation model.

15. The method according to any one of claims 10-14, characterized in that, The second gradient value is the gradient value output by the input layer of the demodulation model during the last round of updating the model parameters of the demodulation model after the receiving device transmits the first gradient value in reverse within the demodulation model.

16. The method according to any one of claims 10-15, characterized in that, The method further includes: Determine the second initialization parameters; wherein the second initialization parameters include the structural parameters of the demodulation model and / or the training parameters of the demodulation model; The demodulation model is initialized based on the second initialization parameters.

17. The method according to claim 16, characterized in that, The determination of the second initialization parameter includes at least one of the following: The second initialization parameter is determined based on the sixth indication information sent by the network device; wherein the sixth indication information is used to indicate the second initialization parameter. The second initialization parameter is determined based on the fifth indication information sent by the transmitting device; wherein the fifth indication information is used to indicate the second initialization parameter. Based on the fourth rule, the second initialization parameter is determined.

18. The method according to any one of claims 10-17, characterized in that, The method further includes: Based on the fifth rule, a first initialization parameter is determined; wherein the first initialization parameter includes the structural parameters of the modulation model and / or the training parameters of the modulation model; Send a fourth indication message to the sending device; wherein the fourth indication message is used to indicate the first initialization parameter.

19. A transmitting device, characterized in that, include: The processing module is configured to input the sample bit sequence into the modulation model and obtain the first modulation symbol output by the modulation model; The transceiver module is configured to transmit a second modulation symbol to a receiving device via a channel; wherein the second modulation symbol is obtained by performing a first operation on the first modulation symbol; wherein the second modulation symbol is used by the receiving device to determine a second gradient value, the second gradient value being the gradient value output by the input layer of the demodulation model when the receiving device updates the model parameters of the demodulation model by reverse transmitting the first gradient value within the demodulation model; the transceiver module is also configured to receive the second gradient value transmitted by the receiving device; The processing module is further configured to train the modulation model based on the second gradient value; The processing module is further configured to stop training the modulation model when the first stopping condition is met, thereby obtaining the trained modulation model.

20. A receiving device, characterized in that, include: The transceiver module is configured to receive the second modulation symbol transmitted by the transmitting device through the channel and obtain the received signal; The processing module is configured to input the symbol to be demodulated into the demodulation model and obtain the first bit sequence output by the demodulation model; wherein the symbol to be demodulated is obtained by performing a second operation on the received signal; The processing module is further configured to determine a loss function based on the difference between the first bit sequence and the sample bit sequence; The processing module is further configured to train the demodulation model based on the first gradient value, and the transceiver module is further configured to send the second gradient value to the transmitting device; wherein the first gradient value is determined based on the loss function, and the second gradient value is the gradient value output by the input layer of the demodulation model when the first gradient value is backpropagated within the demodulation model to update the model parameters of the demodulation model; The processing module is further configured to stop training the demodulation model when the second stopping condition is met, thereby obtaining the trained demodulation model.

21. A transmitting device, characterized in that, include: One or more processors; The processor is used to execute the method according to any one of claims 1-9.

22. A receiving device, characterized in that, include: One or more processors; The processor is used to execute the method according to any one of claims 10-18.

23. A model training system, characterized in that, include: A transmitting device, the transmitting device being configured to perform the method according to any one of claims 1-9; A receiving device, the receiving device being used to perform the method according to any one of claims 10-18.

24. A storage medium storing instructions, characterized in that, When the instruction is executed on the communication device, the communication device performs the model training method as described in any one of claims 1-9 or 10-18.

25. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program is used to implement the model training method according to any one of claims 1-9 or 10-18.

Images