Communication method and system, storage medium, electronic device and computer program product

Abstract

Embodiments of the present application provide a communication method and system, a storage medium, an electronic device, and a computer program product. The method comprises: performing coding processing on original bit data to obtain N channels of coded data, wherein N is an integer greater than or equal to 2; on the basis of N modulation and coding schemes (MCS), converting the N channels of coded data into a time domain signal by means of a pre-trained modulation model and a preset physical resource mapping table, wherein the modulation model is used for implementing multi-level hybrid modulation processing of the N MCSs; and sending the time domain signal by means of a wireless channel.

Description

Communication methods, systems, storage media, electronic devices and computer program products

[0001] Cross-reference of related applications

[0002] This disclosure is based on and claims priority to Chinese Patent Application No. 2025102126854, filed on February 25, 2025, entitled “Communication Method, System, Storage Medium, Electronic Device and Computer Program Product”, and incorporates the entire contents of that patent application by reference. Technical Field

[0003] This application relates to the field of communications, and more specifically, to a communication method, system, storage medium, electronic device, and computer program product. Background Technology

[0004] In wireless mobile communication systems, the characteristics of the wireless channel significantly impact system performance. Transmitted signals typically experience substantial distortion after passing through the channel. Therefore, to ensure accurate signal recovery at the receiver, channel equalization techniques are generally employed to compensate for channel fading. Channel equalization requires prior knowledge of Channel State Information (CSI). Thus, wireless mobile communication systems typically obtain accurate CSI by transmitting pilot symbols for channel estimation, and then use the acquired CSI for channel equalization to eliminate signal waveform distortion caused by the channel. Performing channel estimation before equalization is a crucial step in building a robust wireless communication system. Currently, Least Squares (LS) and Minimum Mean Square Error (MMSE) algorithms are commonly used for channel estimation. Zero Forcing (ZF) and MMSE equalization algorithms are commonly used for channel equalization. However, all of these non-blind channel estimation algorithms require pilot signal transmission, resulting in pilot overhead and reduced communication efficiency. Therefore, researching pilot-free communication systems can avoid the impact of pilot overhead and further improve data transmission efficiency.

[0005] There is no good solution to the above problems in the relevant technologies. Summary of the Invention

[0006] This application provides a communication method, system, storage medium, electronic device, and computer program product.

[0007] According to one embodiment of this application, a communication method is provided, comprising: encoding raw bit data to obtain N coded data, wherein N is an integer greater than or equal to 2; converting the N coded data into a time-domain signal according to N modulation and coding strategies (MCS) through a pre-trained modulation model and a preset physical resource mapping table, wherein the modulation model is used to implement multi-level hybrid modulation processing of N MCS; and transmitting the time-domain signal through a wireless channel.

[0008] According to another embodiment of this application, a communication method is provided, comprising: receiving a time-domain signal through a wireless channel and performing waveform decoding on the time-domain signal; performing demapping on the waveform-decoded data according to a preset physical resource mapping table corresponding to N modulation and coding schemes (MCS) to obtain N channels of frequency-domain received data, wherein N is an integer greater than or equal to 2; converting the N channels of frequency-domain received data into N channels of analytical data using a pre-trained analytical model according to the N MCS, wherein the analytical model is used to implement multi-level hybrid demodulation processing of the N MCS; and performing decoding and merging processing on the N channels of analytical data to obtain analytical bit data.

[0009] According to another embodiment of this application, a communication system is provided, including a transmitter and a receiver, wherein the transmitter is used to communicate according to the steps in any of the above method embodiments; and the receiver is used to communicate according to the steps in any of the above method embodiments.

[0010] According to yet another embodiment of this application, a computer-readable storage medium is also provided, which stores a computer program, wherein the computer program, when executed by a processor, implements the steps in any of the above method embodiments.

[0011] According to yet another embodiment of this application, an electronic device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in any of the above method embodiments.

[0012] According to yet another embodiment of this application, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments. Attached Figure Description

[0013] Figure 1 is a hardware structure block diagram of the base station operated in the embodiment of the method of this application;

[0014] Figure 2 is a flowchart of a communication method of the sending end according to an embodiment of this application;

[0015] Figure 3 is a flowchart of a communication method of a receiving end according to an embodiment of this application;

[0016] Figure 4 is a structural block diagram of a communication system according to an embodiment of this application;

[0017] Figure 5 is a schematic diagram of the architecture of a communication system in one embodiment of this application (I);

[0018] Figure 6 is a schematic diagram of the architecture of a communication system in one embodiment of this application (II);

[0019] Figure 7 is a schematic diagram of a physical resource mapping neural network model in one embodiment of this application;

[0020] Figure 8 is a schematic diagram of a modulation model in one embodiment of this application;

[0021] Figure 9 is a schematic diagram of the analytical model in one embodiment of this application;

[0022] Figure 10 is a schematic diagram of a hierarchical analysis model in one embodiment of this application;

[0023] Figure 11 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS single-transmission-layer hybrid modulation transmissions;

[0024] Figure 12 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS single-transmission-layer hybrid modulation transmissions;

[0025] Figure 13 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS dual-transmission-layer hybrid modulation transmissions;

[0026] Figure 14 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS dual-transmission-layer hybrid modulation transmissions;

[0027] Figure 15 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS four-transmission-layer hybrid modulation transmissions.

[0028] Figure 16 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS four-transmission-layer hybrid modulation transmissions. Detailed Implementation

[0029] The embodiments of this application will be described in detail below with reference to the accompanying drawings and examples.

[0030] It should be noted that the terms "first," "second," etc., in the specification, claims, and drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence.

[0031] The method embodiments provided in this application can be applied to communication nodes in a wireless communication network. Communication nodes include, but are not limited to, base stations (BS), relay nodes, and user equipment (UE). Taking operation on a base station as an example, Figure 1 is a hardware structure block diagram of the base station operated by the method embodiments of this application. As shown in Figure 1, the base station 100 may include one or more (only one is shown in Figure 1) processors 102 (processors 102 may include, but are not limited to, microprocessors (MCUs) or programmable logic devices (FPGAs) and a memory 104 for storing data. The base station may also include transmission devices and input / output devices for communication functions. Those skilled in the art will understand that the structure shown in Figure 1 is merely illustrative and does not limit the structure of the base station. For example, the base station may include more or fewer components than shown in Figure 1, or have a different configuration than shown in Figure 1.

[0032] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the communication method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, thus implementing the above-described method. The memory 104 may include high-speed random access memory and non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to a base station via a network. Examples of the aforementioned networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof. The transmission device is used to receive or send data via a network. Specific examples of the aforementioned networks may include wireless networks provided by the communication provider of the mobile terminal. In one instance, the transmission device includes a Network Interface Controller (NIC), which can be connected to other network devices via a base station to communicate with the Internet. In one instance, the transmission device may be a radio frequency (RF) module used to communicate with the Internet wirelessly.

[0033] In one embodiment of this application, a communication method is provided, applied to a sending end. Figure 2 is a flowchart of the communication method of the sending end according to an embodiment of this application. As shown in Figure 2, the process includes the following steps:

[0034] Step S202: Encode the original bit data to obtain N-channel encoded data;

[0035] Step S204: Based on N modulation and coding strategies (MCS), convert N coded data into time-domain signals using a pre-trained modulation model and a preset physical resource mapping table.

[0036] Step S206: Transmit time-domain signals via a wireless channel.

[0037] In this embodiment, N represents the number of MCSs used in the multi-level hybrid modulation / demodulation of this application, and N is an integer greater than or equal to 2. The modulation model is used to implement multi-level hybrid modulation processing of N types of MCSs.

[0038] Existing pilotless communication systems employ a single modulation transmission method. Since the system itself lacks pilot signal transmission, improving the performance of pilotless systems in high-order modulation and multi-stream transmission is challenging. Through steps S202 to S206, by employing N different MCSs, multi-order hybrid modulation of the data is achieved. This enables high-order modulation and multi-stream transmission without using pilot signals, thus solving the problem of performance improvement in high-order modulation and multi-stream transmission in pilotless communication systems in related technologies. This results in improved data transmission efficiency and quality.

[0039] The entities that perform the above steps can be base stations, user equipment, etc., but are not limited to these.

[0040] In some embodiments, the physical resource mapping table may include resource mapping patterns for N MCSs, wherein the resource mapping patterns are used to indicate the time-domain and frequency-domain positions of the resource units corresponding to each MCS within the resource block. For example, the physical resource mapping table can be generated according to preset resource allocation generation criteria (including constraints on resource allocation for each MCS), or it can be generated using a trained neural network model.

[0041] In some embodiments, the N MCSs are different types and / or different orders of modulation schemes. The MCSs may include: Quadrature Phase Shift Keying (QPSK), Quadrature Amplitude Modulation (QAM), such as 16-QAM, 64-QAM, 256-QAM, etc., but this application is not limited to these.

[0042] In some embodiments, the first MCS among the N MCSs is the lowest-order MCS among the N MCSs; the last MCS among the N MCSs is the highest-order MCS among the N MCSs.

[0043] In some embodiments, prior to step S202, the method may further include at least one of the following: determining the value of N; determining the configuration of the N types of MCS; determining the number of layers of the transmission mode; and the physical resource allocation parameters of the N types of MCS (e.g., resource percentage, number of resource units occupied, etc.).

[0044] In one exemplary embodiment, the configuration of N MCSs may include the modulation type and modulation order of each MCS. The values ​​of the modulation type and modulation order N for each MCS, as well as the configuration of the N MCSs, can be set according to data transmission requirements, and this application does not impose any limitations on this. In one exemplary embodiment, the configuration of the N MCSs can be indicated by different indication information / index numbers / orders.

[0045] In this embodiment, the number of transmission mode layers is the number of data streams transmitted simultaneously in a Multiple-Input Multiple-Output (MIMO) system. MIMO technology can utilize multiple antennas at the transmitting and receiving ends to achieve spatial diversity and spatial multiplexing, thereby improving data transmission rate and reliability. The number of transmission mode layers can be determined based on the number of antennas at the transmitting and receiving ends; for example, the minimum number of antennas at both ends can be determined as the number of transmission mode layers.

[0046] In some embodiments, step S202 may include: splitting the original bit data into N paths of bit data according to the configuration of N types of MCS and the physical resource allocation parameters of the N types of MCS; and encoding the N paths of bit data separately to obtain N paths of encoded data. In this embodiment, the N paths of data are encoded independently, which can better adapt to the MIMO system and improve the system's anti-interference capability in complex environments.

[0047] In other embodiments, step S202 may include: determining the total bit data corresponding to the N MCSs from the original bit data based on the configuration of the N MCSs and the physical resource allocation parameters of the N MCSs; encoding the total bit data to obtain total encoded data; and splitting the total encoded data into N paths of encoded data based on the configuration of the N MCSs and the physical resource allocation parameters of the N MCSs. In this embodiment, the data is uniformly encoded before being split into multiple paths, which can reduce encoding complexity and improve data processing efficiency.

[0048] In some embodiments, step S204 may include the following steps:

[0049] Step S2042: Based on N types of MCS, perform N-channel modulation processing on the N-channel coded data through the modulation model to obtain N-channel modulated data, where each channel of modulated data corresponds to one type of MCS;

[0050] Step S2044: Map the N channels of modulated data to resource blocks using a physical resource mapping table;

[0051] Step S2046: Perform waveform processing on the mapped data to generate a time-domain signal.

[0052] In some embodiments, step S2042 may include: inputting the N types of MCSs and the N coded data into the modulation model, and obtaining the N modulated data output by the modulation model, wherein the modulation model inputs one channel of coded data and the corresponding MCS each time and outputs one channel of modulated data. In this embodiment, each data transmission requires modulation by the modulation model. The modulation model is trained offline, or it can be periodically fine-tuned and updated online, which can better adapt to the characteristics of frequent and rapid changes in the communication environment and improve data transmission performance. This application does not limit this aspect.

[0053] In other embodiments, step S2042 may include: generating N irregular modulation constellation diagrams corresponding to the N types of MCS using the modulation model, and modulating the N coded data according to the N irregular modulation constellation diagrams to obtain the N modulated data, wherein one irregular modulation constellation diagram corresponds to one coded data path and one modulated data path. In this embodiment, each data transmission can be directly modulated using a preset irregular modulation constellation diagram without using a modulation model to process the data to be transmitted, which can improve modulation efficiency and reduce the computational performance requirements of user equipment and base stations.

[0054] In one exemplary embodiment, the irregular modulation constellation diagrams corresponding to various MCSs can be updated periodically, and / or the modulation model can be updated and trained periodically, thereby improving data transmission performance.

[0055] In some embodiments, prior to step S2044, the method may further include the following steps:

[0056] Step S2043-2: Determine the number of resource units that each of the N types of MCS can occupy in the resource block, wherein the resource block includes multiple resource units;

[0057] Step S2043-4: Based on the N types of MCS and the number of resource units of the N types of MCS, generate N resource mapping matrices corresponding to the N types of MCS through a pre-trained physical resource mapping neural network model, wherein the physical resource mapping table includes the N resource mapping matrices.

[0058] The above-mentioned resource mapping matrix generation process (i.e. steps S2043-2 and S2043-4) can occur before the start of the entire data transmission process, or it can occur at any time before step S2044, for example, between step S2042 and step S2044, but this application is not limited to this.

[0059] In this embodiment, the physical resource mapping table can be automatically generated using a physical resource mapping neural network model. In this embodiment, the resource mapping matrix is ​​used to indicate the location (e.g., time domain location, frequency domain location) of the resource unit occupied by each MCS within a resource block.

[0060] In one exemplary embodiment, the physical resource mapping neural network model can process the input data for each MCS individually, or it can automatically process multiple MCSs sequentially (e.g., in ascending order of order). For example, the input data of the physical resource mapping neural network model may include N MCSs, the number of resource units in the N MCSs, and an initial physical resource mappable matrix; the output data may include N resource mapping matrices. Alternatively, the input data may include one MCS, the number of resource units in that MCS, and the current physical resource mappable matrix; the output data may include the resource mapping matrix of that MCS.

[0061] In some embodiments, step S2043-4 may include the following steps:

[0062] 1) In response to N equaling 2, the first type of MCS, the number of resource units of the first type of MCS, and the initial physical resource mappable matrix are input into the physical resource mapping neural network model to obtain the first resource mapping matrix output by the physical resource mapping neural network model, wherein the first resource mapping matrix corresponds to the first type of MCS, and the initial physical resource mappable matrix is ​​an all-zero matrix.

[0063] 2) Invert the first resource mapping matrix to generate a second resource mapping matrix corresponding to the second type of MCS;

[0064] The N resource mapping matrices include the first resource mapping matrix and the second resource mapping matrix.

[0065] In some other embodiments, step S2043-4 may include the following steps: In response to N being greater than 2, for i = 1,...,N, the N resource mapping matrices are generated sequentially according to the following manner:

[0066] 1) In response to i being less than N, the i-th type of MCS, the number of resource units of the i-th type of MCS, and the current physical resource mappable matrix are input into the physical resource mapping neural network model to obtain the i-th resource mapping matrix output by the physical resource mapping neural network model corresponding to the i-th type of MCS. The initial value of the physical resource mappable matrix is ​​an all-zero matrix, and the current physical resource mappable matrix is ​​the complete set of the first to the (i-1)-th resource mapping matrices.

[0067] 2) In response to i being less than or equal to N, update the current physical resource mappable matrix according to the i-th resource mapping matrix;

[0068] 3) In response to i being equal to N, the entire set of the first resource mapping matrix to the (i-1)th resource mapping matrix is ​​inverted to obtain the Nth resource mapping matrix.

[0069] In one exemplary embodiment, the resource block includes 12×14 resource units. The physical resource mappable matrix can be represented as a 12×14 matrix, where each row of the matrix corresponds to a different frequency domain resource (such as a subcarrier), and each column of the matrix corresponds to a different time domain resource (such as a time domain symbol). The element values ​​in the matrix are used to indicate whether the corresponding time-frequency domain resource has been allocated. For example, an element value of 1 indicates that the resource unit has been allocated, and an element value of 0 indicates that the resource unit has not been allocated. When performing resource mapping processing for the lowest-order MCS (i.e., the first type of MCS), the initial value of the physical resource mappable matrix input to the physical resource mapping neural network model is a matrix of all zeros. Correspondingly, the resource unit corresponding to the position of the element value of 1 in the resource mapping matrix of each MCS is the resource allocated to that MCS. However, this application is not limited to this, and the reverse can also be implemented.

[0070] In another exemplary embodiment, an element value of 0 can also indicate that the resource unit cannot be occupied (i.e., it has been allocated), and an element value of 1 can also indicate that the resource unit can be occupied (i.e., it has not been allocated). Correspondingly, the resource unit corresponding to the position of the element value of 0 in the resource mapping matrix of each MCS is the resource allocated to that MCS. When performing resource mapping processing of the lowest order MCS, the initial value of the physical resource mappable matrix of the input physical resource mapping neural network model is a matrix of all 1s. Furthermore, the current update method of the physical resource mappable matrix can be changed to performing a logical AND operation on the two matrices.

[0071] In this embodiment, updating the current physical resource mappable matrix according to the i-th resource mapping matrix may include: determining the updated physical resource mappable matrix by the union of the i-th resource mapping matrix and the current physical resource mappable matrix.

[0072] In this embodiment, the transmission resources allocated to the Nth type of MCS are all the remaining resource units after the resource allocation of the 1st to N-1th types of MCS. This allocation method can maximize the efficiency of resource utilization and ensure that no resource units are idle, thereby avoiding waste of transmission resources.

[0073] In an exemplary embodiment, the number of resource units in the low-order MCS (such as the first to N-1 types of MCS) can be set to a small value, that is, the resource occupancy ratio is small, which is used to realize the corresponding function of the pilot signal and improve the accuracy and efficiency of channel estimation. Correspondingly, the resource occupancy ratio of the highest-order MCS (i.e. the Nth type of MCS) is larger, so that the overall data transmission efficiency of multi-order hybrid MCS modulation can be further improved on the basis of the high data transmission efficiency of the high-order MCS.

[0074] In some embodiments, the method further includes step S201, whereby the joint receiving end performs end-to-end training on multiple machine learning models participating in the communication. This step occurs before the entire data transmission (i.e., communication) process begins. However, this application is not limited to this; model training and updates can be performed periodically to better adapt to changes in the communication environment.

[0075] In some embodiments, step S201 may include the following steps:

[0076] Step S2012: The receiving end performs the first stage of end-to-end training on the modulation model and the analytical model.

[0077] Step S2014: The receiving end performs a second-stage end-to-end training on the physical resource mapping neural network model and the parsing model.

[0078] In this embodiment, the physical resource mapping neural network model is used to generate the physical resource mapping table. The physical resource mapping neural network model is not activated in the first stage of end-to-end training (i.e., step S2012) but is activated in the second stage of end-to-end training (i.e., step S2014). For example, the resource mapping method in the first stage can be fixed as a preset resource mapping table. The first stage training is performed separately for different MCSs; therefore, different MCSs can use the same resource mapping table in the first stage training.

[0079] In some embodiments, step S2012 may include the following steps:

[0080] 1) For each of the M types of MCS, the encoded first data is input into the modulation model to be trained to obtain the modulated second data output by the modulation model to be trained, where M is an integer greater than or equal to N.

[0081] 2) Perform layer mapping processing, resource mapping processing, and waveform processing on the second data to generate a first transmission signal, and transmit the first transmission signal to the receiving end through a wireless channel;

[0082] 3) Obtain the parsed third data generated by the receiving end based on the first transmission signal and the parsing model to be trained, and calculate the binary cross-entropy based on the first data and the third data to obtain the first loss function of the end-to-end training in the first stage.

[0083] 4) The receiving end adjusts the modulation model and the analytical model together until the first loss function converges.

[0084] In this embodiment, for each of the M types of MCS, the modulation model needs to be trained separately (i.e., each MCS needs to repeat each sub-step of step S2012 above). Each MCS can be input into the modulation model to be trained in the form of indication information / index value / modulation order.

[0085] In this embodiment, each acquisition of the output data (i.e., the third data) of the first-stage end-to-end training requires the data to undergo a complete data transmission process (from the receiving end to the sending end). In the first-stage end-to-end training, the sending end and the receiving end need to fix a certain MCS. The sending end performs modulation processing (using the modulation model to be trained), layer mapping processing, resource mapping processing, and waveform processing on the data in sequence for this MCS. Among them, the layer mapping processing, resource mapping processing, and waveform processing can adopt a fixed method. Correspondingly, the receiving end also needs to perform waveform de-processing, resource mapping de-processing, and parsing processing (using the parsing model to be trained) on the data received from the sending end (i.e., the first transmission signal) in sequence for this MCS. During the model training process, in order to improve training efficiency and reduce data processing volume, the encoding processing of the sending end and the decoding processing of the receiving end can be omitted, and the encoded data can be directly used as the input data of the entire training process, but this application is not limited to this. If encoding processing and decoding processing are added in the training process, the encoding and decoding methods also need to be fixed.

[0086] In some embodiments, step S2014 may include the following steps:

[0087] 1) Generate M modulation mapping tables corresponding to M types of MCS based on the modulation model trained in the first stage, and determine N modulation mapping tables corresponding to the N types of MCS according to the selected N types of MCS, where M is an integer greater than or equal to N;

[0088] 2) Modulate and layer map the N coded data according to the N modulation mapping tables to generate N layer mapping data;

[0089] 3) Based on the preset resource proportion of the resource units corresponding to the N types of MCS, generate N-way resource mapping tables through the physical resource mapping neural network model to be trained.

[0090] 4) Perform resource mapping processing and waveform processing on the N-way layer mapping data according to the N-way resource mapping table to generate a second transmission signal, and send the second transmission signal to the receiving end through a wireless channel;

[0091] 5) Obtain the N-way parsing data generated by the receiving end based on the second transmission signal, the N-way resource mapping table and the parsing model, and calculate the binary cross-entropy based on the N-way encoded data and the N-way parsing data to obtain the second loss function for the end-to-end training in the second stage;

[0092] 6) Combine the receiving end to adjust the physical resource mapping neural network model and the analytical model until the second loss function converges.

[0093] In this embodiment, the modulation model needs to be solidified based on the training results of the first stage, that is, solidified into N modulation mapping tables. In the second stage of training, modulation processing and demodulation processing are both performed based on the N modulation mapping tables.

[0094] In this embodiment, each acquisition of the output data (i.e., N-channel parsed data) from the second-stage end-to-end training requires the data to undergo a complete data transmission process (from the receiver to the transmitter). In the second-stage end-to-end training, the transmitter and receiver employ a hybrid modulation scheme using N different MCSs. The transmitter sequentially performs modulation processing (using N fixed modulation mapping tables), layer mapping processing, resource mapping processing (each transmission requires generating N resource mapping tables through a physical resource mapping neural network model), and waveform processing on the N-channel coded data for each MCS. The modulation processing, layer mapping processing, and waveform processing use fixed methods. Correspondingly, the receiver also needs to sequentially perform waveform de-shaping processing, resource de-mapping processing (using the N resource mapping tables generated by the transmitter through the physical resource mapping neural network model in this transmission), and parsing processing (using the parsing model trained in the first stage but not yet fully trained) on the data received from the transmitter (i.e., the second transmission signal) for each of the N MCSs. In the model training process, to improve training efficiency and reduce data processing volume, the encoding process at the sending end and the decoding process at the receiving end can be omitted, and the encoded data can be directly used as the input data for the entire training process. However, this application is not limited to this. If encoding and decoding processes are added to the training process, the encoding and decoding methods also need to be fixed.

[0095] In this embodiment, multi-order hybrid modulation of data is achieved based on N different MCSs without using pilot signals. This leverages both the high robustness of low-order MCSs and the high transmission efficiency of high-order MCSs, thus solving the problem of performance improvement in high-order modulation and multi-stream in pilotless communication systems in related technologies. This achieves improved data transmission efficiency and quality, enhancing not only spectral efficiency and transmission quality but also the system's robustness and adaptability. Furthermore, this application uses machine learning models to learn the modulation processing, resource mapping processing, and parsing processing flows, thereby automatically processing data using trained modulation and parsing models, further improving data transmission quality.

[0096] In one embodiment of this application, a communication method is also provided, applied to a receiving end. Figure 3 is a flowchart of the communication method of the receiving end according to an embodiment of this application. As shown in Figure 3, the process includes the following steps:

[0097] Step S302: Receive time-domain signals through a wireless channel and perform waveform decoding on the time-domain signals;

[0098] Step S304: Based on N modulation and coding strategies (MCS), the data after waveform decoding is converted into N streams of analytical data through a preset physical resource mapping table and a pre-trained analytical model.

[0099] Step S306: Decode the N-channel parsing data to obtain parsing bit data.

[0100] In this embodiment, N is the number of MCSs used in multi-level hybrid parsing, and N is an integer greater than or equal to 2.

[0101] In this embodiment, the analytical model is used to implement multi-level hybrid demodulation processing of N MCSs. For example, the analytical model can perform channel estimation based on the data of the lowest-order MCS, and can determine the channel estimation results of other MCSs based on the channel estimation results of the lowest-order MCS, thereby enabling demodulation processing of the data from the other MCSs.

[0102] In this embodiment, the time-domain signal is generated at the transmitting end through a hybrid modulation of multiple MCSs and then transmitted to the receiving end via a wireless channel. The receiving end needs to perform hybrid parsing of the corresponding multiple MCSs to achieve channel estimation and data recovery. Those skilled in the art will understand that the N types of MCSs and the physical resource mapping table used by the receiving end and the transmitting end during modulation and demodulation are the same.

[0103] In this embodiment, the physical resource mapping table is used to indicate the resources occupied by the transmission data corresponding to each of the N MCSs in a resource block (RB). The total resources occupied by the N MCSs constitute a complete resource block. The resource block is the basic unit of physical resources allocated to the user in the frequency and time domains. A resource block contains multiple resource elements (REs). Taking the LTE system as an example, an RB in the time domain consists of 12 subcarriers within a time slot (containing 7 or 14 OFDM symbols, depending on the subframe configuration), and an RE consists of one subcarrier and one OFDM symbol.

[0104] Through the above steps S302 to S306, by using N different MCSs, multi-order hybrid analysis of data is achieved, enabling high-order modulation and multi-stream transmission without the use of pilot signals. This solves the problem that it is difficult to improve the performance of pilotless communication systems in high-order modulation and multi-stream transmission, and achieves the technical effect of improving data transmission efficiency and transmission quality.

[0105] The entities that perform the above steps can be base stations, user equipment, etc., but are not limited to these.

[0106] In some embodiments, the N MCSs are different types and / or different orders of modulation schemes. The MCSs may include: Binary Phase Shift Keying (BPSK), Quadrature Phase Shift Keying (QPSK), Quadrature Amplitude Modulation (QAM), such as 16-QAM, 64-QAM, 256-QAM, etc., but this application is not limited to these.

[0107] In some embodiments, prior to step S302, the method may further include at least one of the following: determining the value of N; determining the configuration of the N types of MCS; determining the number of layers of the transmission mode; and the physical resource allocation parameters of the N types of MCS (e.g., resource percentage, number of resource units occupied, etc.).

[0108] In one exemplary embodiment, the configuration of N MCSs may include the modulation type and modulation order of each MCS. The values ​​of the modulation type and modulation order N for each MCS, as well as the configuration of the N MCSs, can be set according to data transmission requirements, and this application does not impose any limitations on this. In one exemplary embodiment, the configuration of the N MCSs can be indicated by different indication information / index numbers / orders.

[0109] In this embodiment, the number of transmission mode layers is the number of data streams transmitted simultaneously in a Multiple-Input Multiple-Output (MIMO) system. MIMO technology can utilize multiple antennas at the transmitting and receiving ends to achieve spatial diversity and spatial multiplexing, thereby improving data transmission rate and reliability. The number of transmission mode layers can be determined based on the number of antennas at the transmitting and receiving ends; for example, the minimum number of antennas at both ends can be determined as the number of transmission mode layers.

[0110] In some embodiments, step S304 may include the following steps:

[0111] Step S3042: Determine N resource mapping matrices corresponding to the N types of MCS based on the physical resource mapping table;

[0112] Step S3044: Demapping the waveform-decoded data based on the N resource mapping matrices to obtain N channels of frequency domain received data, wherein each channel of frequency domain received data corresponds to one MCS.

[0113] Step S3046: Based on the N types of MCS and the N resource mapping matrices, the N channels of frequency domain received data are analyzed through the analytical model to obtain the N channels of analytical data output by the analytical model, wherein the analytical data is log-likelihood ratio data, and each channel of analytical data corresponds to one type of MCS.

[0114] In this embodiment, the physical resource mapping table records the resource locations of N types of MCSs, from which resource units corresponding to each type of MCS, and even each transport layer, can be extracted. The de-waveformed data corresponds to a complete resource block. By performing demapping processing on the de-waveformed data, one channel of frequency domain received data corresponding to each MCS can be extracted based on the relationship between the resource block and the resource units occupied by each MCS.

[0115] In some embodiments, the input data of the analytical model includes the N frequency domain received data and the indications of the N types of MCS, wherein the N frequency domain received data are also used to implicitly indicate the N resource mapping matrices.

[0116] In an exemplary embodiment, if the N resource mapping matrices are implicit indicators, the analytical model can also determine the N resource mapping matrices based on the N frequency domain received data. For example, the N resource mapping matrices can be obtained by analyzing the time domain position and frequency domain position of the N frequency domain received data in the resource block.

[0117] In other embodiments, the input data for the analytical model includes the N frequency domain received data, the indications of the N MCSs, and the N resource mapping matrices.

[0118] In some embodiments, the analytical model is a hierarchical analytical model. For example, the analytical model may include the following structure: a first type of analytical model corresponding to the first type of MCS, a data reconstruction unit, a channel estimation model, and N-1 second type of analytical models corresponding to other MCSs besides the first type of MCS, wherein the first type of MCS is the lowest order modulation scheme among the N types of MCSs.

[0119] In this embodiment, after the N frequency domain received data are input into the analytical model, one frequency domain received data corresponding to each MCS is input into the corresponding first-type analytical model and second-type analytical model. First, the first-type analytical model performs calculations on the frequency domain received data of the lowest-order MCS to obtain the analytical data of the lowest-order MCS. This data then passes through the data reconstruction unit and the channel estimation model in sequence to obtain the channel estimation results for each MCS in the entire resource block. Then, each second-type analytical model performs analytical calculations on its own frequency domain received data based on the channel estimation results for each MCS to obtain the analytical data for the other MCSs.

[0120] In some embodiments, based on the hierarchical parsing model described above, step S3046 may include the following steps:

[0121] 1) Input the first frequency domain received data corresponding to the first type of MCS and the indication of the first type of MCS into the first type of analytical model to obtain the first log-likelihood ratio data output by the first type of analytical model;

[0122] 2) The data reconstruction unit reconstructs the transmitting end data corresponding to the first frequency domain received data;

[0123] 3) Input the first frequency domain received data, the reconstructed transmitter data, and the N resource mapping matrices into the channel estimation model to obtain N channel estimation results corresponding to the N types of MCS;

[0124] 4) In response to N equaling 2, the second frequency domain received data corresponding to the second type of MCS, the indication of the second type of MCS, and the second channel estimation result corresponding to the second type of MCS are input into the second type of analytical model to obtain the second log-likelihood ratio data, wherein the N analytical data includes the first log-likelihood ratio data and the second log-likelihood ratio data.

[0125] 5) In response to N being greater than 2, the j-th frequency domain received data corresponding to the j-th MCS, the indication of the j-th MCS, and the j-th channel estimation result corresponding to the j-th MCS are input into the (j-1)-th second-type analytical model to obtain the j-th log-likelihood ratio data, where j = 2, ..., N, and the N analytical data include the first log-likelihood ratio data to the Nth log-likelihood ratio data.

[0126] In an exemplary embodiment, the channel estimation model in the analytical model can obtain N channel estimation results corresponding to the N types of MCS in the following ways: (1) Determine the channel estimation result of MCS1 (i.e., the first type of MCS) based on the first log-likelihood ratio data and the reconstructed transmitter data; (2) Reconstruct the complete channel estimation result of the entire resource block based on the resource mapping matrix of MCS1 and the channel estimation result of MCS1; (3) Extract the channel estimation result of the j-th type of MCS from the complete channel estimation result based on the resource mapping matrix of the j-th type of MCS, where j = 2,...,N.

[0127] In some embodiments, reconstructing the transmitter data corresponding to the first channel frequency domain received data through the data reconstruction unit may include: performing hard decision processing on the first channel log-likelihood ratio data, and remodulating the hard decision processed data according to the first type of MCS to obtain the reconstructed transmitter data.

[0128] In other embodiments, the reconstructing of the transmitter data corresponding to the first frequency domain received data by the data reconstruction unit may include: obtaining the first decoded data generated by decoding the first log-likelihood ratio data, and re-encoding, modulating and resource mapping the first decoded data according to the first type of MCS to obtain the reconstructed transmitter data.

[0129] In this embodiment, the data reconstruction unit can reconstruct the data at the sending end based on either of the two methods described above. The difference is that if the second method (non-hard decision processing) is used, the first log-likelihood ratio data (i.e., the first parsed data) needs to be decoded first, and then reconstructed based on the first decoded data. If the first method (hard decision processing) is used, the first log-likelihood ratio data does not need to be decoded separately, and the N parsed data can be decoded together after the entire parsing model is processed.

[0130] In some embodiments, step S306 may include: decoding the N-channel parsed data separately to obtain N-channel decoded data; and merging the N-channel decoded data to obtain parsed bit data. In this embodiment, the N-channel parsed data are decoded independently, which can better adapt to MIMO systems and improve the system's anti-interference capability in complex environments.

[0131] In other embodiments, step S306 may include: merging the N-way parsing data and decoding the merged parsing data to obtain parsed bit data. In this embodiment, merging multiple data streams and then decoding them uniformly can reduce decoding complexity and improve data processing efficiency.

[0132] In some embodiments, the method further includes step S301, whereby the joint sending end performs end-to-end training on multiple machine learning models participating in the communication. This step occurs before the entire data transmission (i.e., communication) process begins. However, this application is not limited to this; model training and updates can be performed periodically to better adapt to changes in the communication environment.

[0133] In some embodiments, step S301 may include the following steps:

[0134] Step S3012: The transmitting end performs the first stage of end-to-end training on the modulation model and the analytical model.

[0135] Step S3014: The sending end performs a second-stage end-to-end training on the physical resource mapping neural network model and the parsing model.

[0136] In this embodiment, the physical resource mapping neural network model is used to generate the physical resource mapping table. The physical resource mapping neural network model is not activated in the first stage of end-to-end training (i.e., step S3012) but is activated in the second stage of end-to-end training (i.e., step S3014). For example, the resource mapping method in the first stage can be fixed as a preset resource mapping table. The first stage training is performed separately for different MCSs; therefore, different MCSs can use the same resource mapping table in the first stage training.

[0137] In some embodiments, step S3012 may include the following steps:

[0138] 1) For each of the M types of MCS, receive the first transmission signal from the transmitting end via a wireless channel;

[0139] Wherein, the first transmission signal is obtained by the transmitting end through modulation processing of the coded data by the modulation model to be trained, and layer mapping processing, resource mapping processing and waveform processing of the modulated data, where M is an integer greater than or equal to N;

[0140] 2) Perform waveform de-processing and resource de-mapping processing on the first transmission signal to obtain frequency domain received data, and input the frequency domain received data and the corresponding MCS indication into the analytical model to be trained to obtain the analytical data output by the analytical model to be trained, wherein the analytical data is log-likelihood ratio data;

[0141] 3) Obtain the encoded data from the sending end, and calculate the binary cross-entropy based on the encoded data and the parsed data to obtain the first loss function for the end-to-end training in the first stage;

[0142] 4) The transmitting end adjusts the modulation model and the analytical model together until the first loss function converges.

[0143] In this embodiment, for each of the M types of MCS, the modulation model needs to be trained separately (i.e., each MCS needs to repeat each sub-step of step S2012 above). Each MCS can be input into the modulation model to be trained in the form of indication information / index value / modulation order.

[0144] In this embodiment, each acquisition of the output data (i.e., parsed data) in the first stage of end-to-end training requires the data to undergo a complete data transmission process (from the receiving end to the sending end). In the first stage of end-to-end training, the sending end and the receiving end need to fix a certain MCS. The sending end performs modulation processing (using the modulation model to be trained), layer mapping processing, resource mapping processing, and waveform processing on the data in sequence for this MCS. Among them, the layer mapping processing, resource mapping processing, and waveform processing can adopt a fixed method. Correspondingly, the receiving end also needs to perform waveform de-shaping processing, de-resource mapping processing, and parsing processing (using the parsing model to be trained) on the data (i.e., the first transmission signal) received from the sending end in sequence for this MCS. During model training, in order to improve training efficiency and reduce data processing volume, the encoding processing at the sending end and the decoding processing at the receiving end can be omitted, and the encoded data can be directly used as the input data of the entire training process, but this application is not limited to this. If encoding processing and decoding processing are added in the training process, the encoding and decoding methods also need to be fixed.

[0145] In some embodiments, step S3014 may include the following steps:

[0146] 1) Based on the preset resource proportion of the resource units corresponding to the N types of MCS, generate N-way resource mapping tables through the physical resource mapping neural network model to be trained.

[0147] 2) Receive the second transmission signal from the transmitting end via a wireless channel, and perform waveform decoding on the second transmission signal;

[0148] The second transmission signal is obtained by the transmitting end modulating and layer mapping N coded data based on N modulation mapping tables corresponding to the N types of MCS, and then performing resource mapping and waveform processing on the layer-mapped data according to the N resource mapping tables. The N modulation mapping tables are selected by the transmitting end from M modulation mapping tables corresponding to M types of MCS, and the M modulation mapping tables are generated based on the modulation model trained in the first stage, where M is an integer greater than or equal to N.

[0149] 3) Perform de-resource mapping processing on the de-waveform processed data according to the N-way resource mapping table to obtain N-way frequency domain received data, and input the N-way frequency domain received data and the indications of the N types of MCS into the analytical model to obtain N-way analytical data output by the analytical model, wherein the analytical data is log-likelihood ratio data;

[0150] 4) Obtain the N-way encoded data from the transmitting end, and calculate the binary cross-entropy based on the N-way encoded data and the N-way parsed data to obtain the second loss function for the end-to-end training in the second stage;

[0151] 5) Combine the sending end to adjust the physical resource mapping neural network model and the analytical model until the second loss function converges.

[0152] In this embodiment, the modulation model needs to be solidified based on the training results of the first stage, that is, solidified into N modulation mapping tables. In the second stage of training, modulation processing and demodulation processing are both performed based on the N modulation mapping tables.

[0153] In this embodiment, each acquisition of the output data (i.e., N-channel parsed data) from the second-stage end-to-end training requires the data to undergo a complete data transmission process (from the receiver to the transmitter). In the second-stage end-to-end training, the transmitter and receiver employ a hybrid modulation scheme using N different MCSs. The transmitter sequentially performs modulation processing (using N fixed modulation mapping tables), layer mapping processing, resource mapping processing (each transmission requires generating N resource mapping tables through a physical resource mapping neural network model), and waveform processing on the N-channel coded data for each MCS. The modulation processing, layer mapping processing, and waveform processing use fixed methods. Correspondingly, the receiver also needs to sequentially perform waveform de-shaping processing, resource de-mapping processing (using the N resource mapping tables generated by the transmitter through the physical resource mapping neural network model in this transmission), and parsing processing (using the parsing model trained in the first stage but not yet fully trained) on the data received from the transmitter (i.e., the second transmission signal) for each of the N MCSs. In the model training process, to improve training efficiency and reduce data processing volume, the encoding process at the sending end and the decoding process at the receiving end can be omitted, and the encoded data can be directly used as the input data for the entire training process. However, this application is not limited to this. If encoding and decoding processes are added to the training process, the encoding and decoding methods also need to be fixed.

[0154] According to the various embodiments of this application, efficient and reliable data transmission can be achieved through hybrid modulation and demodulation of multi-order MCS without using pilot signals. This not only improves spectral efficiency and transmission quality but also enhances the robustness and adaptability of the system. Starting from low-order MCS, the data of each MCS is analyzed using the channel estimation results of the low-order MCS. High-order modulation and multi-stream transmission of the pilotless communication system are achieved while ensuring data transmission quality. This solves the problem of difficulty in improving the performance of high-order modulation and multi-stream in pilotless communication systems in related technologies, achieving the technical effect of improving data transmission efficiency and quality. Furthermore, it not only improves spectral efficiency and transmission quality but also enhances the robustness and adaptability of the communication system. This application also uses machine learning models to learn the processing flows such as modulation processing, resource mapping processing, and parsing processing, thereby automatically processing data using the trained modulation and parsing models, further improving data transmission quality.

[0155] One embodiment of this application provides a communication system. Figure 4 is a structural block diagram of the communication system according to an embodiment of this application. As shown in Figure 4, the system includes the following structure:

[0156] Transmitter 42 is used to encode the original bit data to obtain N coded data, where N is an integer greater than or equal to 2; according to N modulation and coding strategies (MCS), the N coded data are converted into time-domain signals through a pre-trained modulation model and a preset physical resource mapping table, wherein the modulation model is used to implement multi-level hybrid modulation processing of the N MCS; and the time-domain signals are transmitted through a wireless channel.

[0157] The receiver 44 is used to receive time-domain signals through a wireless channel and perform waveform demodulation processing on the time-domain signals; according to N modulation and coding schemes (MCS), it converts the waveform-demodulated data into N-channel analytical data through a preset physical resource mapping table and a pre-trained analytical model, where N is an integer greater than or equal to 2, and the analytical model is used to implement multi-level hybrid demodulation processing of the N MCS; and decodes the N-channel analytical data to obtain analytical bit data.

[0158] In this embodiment, the transmitting end 42 and the receiving end 44 can be any communication node in the wireless communication network, including but not limited to base stations, user equipment, relay nodes, service nodes, etc.

[0159] In this embodiment, the sending end 42 is further configured to communicate according to the steps in any of the above-described sending end-side method embodiments. The receiving end 44 is further configured to communicate according to the steps in any of the above-described receiving end-side method embodiments.

[0160] The communication system described in this application enables pilot-free communication, avoiding pilot overhead and dedicating all symbols to data transmission, thereby further improving data transmission efficiency. The data RE employs multi-level hybrid modulation, enabling the prediction of channel information for the complete transmission RE based on the REs of lower-order MCSs. This, combined with the predicted channel estimation results, enhances data parsing of REs for other higher-order MCSs, thus improving the high-order transmission performance of the pilot-free communication system and significantly increasing its spectral efficiency. The communication system in this application is versatile across various channel environments.

[0161] This application proposes a novel wireless communication system whose transmission architecture achieves pilot-free communication. By eliminating the need to transmit pilot signals during communication, this application avoids pilot overhead, allowing all symbols to transmit data signals, thus further improving data transmission efficiency. The specific improvement in transmission efficiency depends on the pilot configuration and channel conditions. This novel wireless communication system is versatile across various channel environments.

[0162] Figure 5 is a schematic diagram of the architecture of a communication system according to one embodiment of this application (I). As shown in Figure 5, the system includes: a transmitting end and a receiving end, wherein,

[0163] The transmitting end performs a series of processes on the original bit data, including N-way encoding, N-way modulation, layer mapping, N-way physical resource mapping, and waveform processing, and then transmits the processed data to the receiving end through the wireless channel.

[0164] The receiving end performs a series of processes on the received data, including waveform de-processing, N-way physical resource mapping de-processing, N-way parsing processing, and N-way decoding processing, to obtain the final parsed bit data.

[0165] In this embodiment, N-way modulation processing, N-way physical resource mapping processing, and N-way parsing processing can all be implemented using artificial intelligence (AI) models. For example, the modulation model, physical resource mapping neural network model, and parsing model in the above method embodiments can be used.

[0166] In an exemplary embodiment, the parsing model can be referred to as an AI receiver, or the N-way physical resource mapping processing can be merged with the parsing processing into an AI model and referred to as an AI receiver, which is used to implement the relevant processing flow of the receiving end.

[0167] In one exemplary embodiment, the N-way physical resource mapping processing at the transmitting end can be performed by first generating a physical resource mapping table using an AI model of physical resource mapping, and then using the physical resource mapping table for resource mapping processing; alternatively, it can be performed directly using the AI ​​model of physical resource mapping. Furthermore, the physical resource mapping table used by the receiving end when performing N-way physical resource mapping demapping is the same as that used by the transmitting end.

[0168] In this embodiment, the specific processes of the sending end and the receiving end can be referred to the descriptions in the above method embodiments, and will not be repeated here.

[0169] In this embodiment, the original bit data is directly split into multiple paths, and each path is encoded separately. Different encoding methods can be adopted according to the actual communication environment and scenario requirements, making it more adaptable to different scenarios.

[0170] Figure 6 is a schematic diagram of the architecture of a communication system according to one embodiment of this application (II). As shown in Figure 6, the system includes: a transmitting end and a receiving end, wherein,

[0171] The transmitting end performs a series of processes on the original bit data, including encoding, splitting into N data streams, modulation, layer mapping, physical resource mapping, and waveform processing, and then transmits the processed data to the receiving end through a wireless channel.

[0172] After a series of processing including waveform demodulation processing, N-channel physical resource demapping processing, N-channel parsing processing, N-channel data combining processing and decoding processing performed sequentially on the received data by the receiving end, the finally parsed bit data is obtained.

[0173] In this embodiment, N-channel modulation processing, N-channel physical resource mapping processing and N-channel parsing processing can all be implemented by AI models; for example, the modulation model, physical resource mapping neural network model, parsing model and the like in the foregoing method embodiments may be used.

[0174] In an exemplary embodiment, the parsing model may be referred to as an AI receiver, or N-channel physical resource demapping processing may be combined with parsing processing to be implemented in one AI model, which is referred to as an AI receiver and is configured to implement related processing procedures at the receiving end.

[0175] In an exemplary embodiment, for the N-channel physical resource mapping processing at the transmitting end, the physical resource mapping table may be first generated by using the AI model for physical resource mapping, and then resource mapping processing is performed by using the physical resource mapping table, or resource mapping processing may be directly performed by using the AI model for physical resource mapping. Further, the physical resource mapping table used by the receiving end when performing N-channel physical resource demapping processing is consistent with that used by the transmitting end.

[0176] In this embodiment, for the specific procedures of the transmitting end and the receiving end, reference may be made to the descriptions in the foregoing method embodiments, and details are not described herein again.

[0177] In this embodiment, the original bit data is first encoded, and then the encoded data is split into multiple channels for multi-level hybrid modulation, which can improve the efficiency of data encoding processing.

[0178] In the embodiment of the present application, the multi-level hybrid modulation transmitted data of the communication system is composed of N types of MCS: MCS1 ~ MCSN. For example, the configurable values of N are [1 / 2 / 3 / 4 / 5], but the present application is not limited thereto. The order relationship of each modulation mode is MCS1<MCS2<...<MCSN, and the modulation modes corresponding to different MCSs are different.

[0179] In an embodiment of the present application, before data transmission is performed, the following information needs to be determined first:

[0180] (1) Determine the value of N, where the value of N can be determined by high-level scheduling, and the present application does not impose any limitation on the specific determination manner.

[0181] (2) Determine the resource proportions a1~a of transmission REs of all MCS1~MCSN in the current transmission N(That is, the ratio of the number of REs in each MCS to the total number of REs contained in a unit RB). The specific value of the resource ratio can be obtained through simulation, or it can be set according to the actual transmission requirements. The receiving end and the transmitting end can obtain the value of the resource ratio in a predefined known way. This application does not restrict the specific method of generating the resource ratio.

[0182] In this embodiment, the resource ratio of MCS1 to MCSN needs to meet the following requirements.

[0183] (3) Determine all MCS1 to MCS N The resource mapping pattern can be generated by a trained physical resource mapping neural network model, which is learned through end-to-end training. During the execution phase, the sending and receiving ends can be implemented according to a predefined storage table.

[0184] Figure 7 is a schematic diagram of a physical resource mapping neural network model in one embodiment of this application. As shown in Figure 7, the model architecture is as follows:

[0185] Input features of the physical resource mapping neural network model:

[0186] (1) Modulation method MCS i Number of mappable resource units (REs) i .

[0187] Among them, RE i =a i ·12·Symbol·RB, where a i The resource percentage is represented by RB, the number of resource blocks is represented by 12, the number of frequency domain resources contained in a single resource block is represented by 12, such as the number of subcarriers, and Symbol is represented by the number of time domain resources contained in a single resource block is represented by 12, such as the number of OFDM symbols.

[0188] (2) Current physical resource mappable matrix [N] RE ]12*RB,Symbol.

[0189] The matrix has dimensions [12*RB, Symbol]. If the current modulation scheme is the lowest-order modulation scheme MCS1, then the current physical resource mapping matrix is ​​an all-zero matrix (initial value); if the current modulation scheme is a non-lowest-order modulation scheme MCS... i If i is greater than or equal to 2, then the current physical resource mappable matrix is ​​MCS1~MCS i-1 The complete set, i.e., MCS1 to MCS in the matrix. i-1 Enter 1 for mapped resource locations and 0 for other remaining resource locations.

[0190] (3) Modulation method MCSi Instruction Q i Configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., corresponding to Q values ​​of [1 / 2 / 4 / 6 / 8].

[0191] Output of the physical resource mapping neural network model: Current modulation scheme (MCS) i The matrix of resource mapping patterns [MCS] RE,i The matrix has dimensions [12*RB, Symbol], and its elements consist of 0s and 1s. An element of 0 indicates no mapping, and an element of 1 indicates a mapping. The total number of elements of 1 in the matrix satisfies the MCS (Mean Cross-Sectional Calculation). i The number of mappable REs, i.e., REs i =a i ·12·Symbol·RB.

[0192] In some embodiments, if it is a multi-layer transmission and the multi-layer data is configured to map the same RE resource, then the multi-layer data is simultaneously mapped to the RE resource position of 1 in the above matrix. If it is a multi-layer transmission and the multi-layer data is configured to map different RE resources, then the multi-layer data is mapped to the RE resource position of 1 in the above matrix in a sequentially spaced mapping manner, and the number of REs mapped by each layer is RE. i / L, where L is the number of transmission layers.

[0193] This application does not limit the specific type of neural network model. For example, any neural network model such as Convolutional Neural Network (CNN) and Multilayer Perceptron (MLP) can be used.

[0194] In this embodiment, the process for generating resource mapping matrices for various MCSs is as follows:

[0195] Example 1, when N is 2 in the current transport block:

[0196] 1) Call the physical resource mapping model to generate the MCS1 physical resource mapping matrix.

[0197] 2) Invert the MCS1 physical resource mapping matrix to generate the MCS2 physical resource mapping matrix.

[0198] Example 2, in the current transport block, when the value of N is greater than 2:

[0199] 1) Call the physical resource mapping model to generate the MCS1 physical resource mapping matrix.

[0200] 2) Using the MCS1 physical resource mapping matrix as input to the current physical resource mappable matrix, the physical resource mapping model is called to generate the MCS2 physical resource mapping matrix; combining the used resource locations of MCS1 and MCS2, the MCS3 input mappable matrix is ​​constructed, and MCS3 through MCS are generated sequentially according to the above process. N-1 Physical resource mapping matrix.

[0201] 3) MCS1 to MCS N-1 All mapped locations are set to 0, and the remaining physical resource locations are set to 1, which constitutes MCS. N The corresponding resource mapping matrix.

[0202] In the embodiments of this application, the resource mapping matrices of the various MCSs mentioned above can be generated before the transmission process begins, and the physical resource mapping table (containing the resource mapping patterns of each MCS) corresponding to the entire resource block can be used directly during the data transmission process. Alternatively, the resource mapping matrix can be generated directly through the physical resource mapping neural network model during the transmission process.

[0203] Figure 8 is a schematic diagram of a modulation model in one embodiment of this application. As shown in Figure 8, the model architecture is as follows:

[0204] Input features of the modulation model:

[0205] (1) The input dimension of the data bit to be modulated can be determined according to the modulation method. For example, if the configurable values ​​of the modulation method are BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., then the input dimensions of the data bit to be modulated are [1 / 2 / 4 / 6 / 8].

[0206] (2) Optional input, modulation mode MCS i Instruction Q i Configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., corresponding to Q values ​​of [1 / 2 / 4 / 6 / 8].

[0207] Output of the modulation model: in the modulation scheme MCS i The mapped complex value has a dimension of [1].

[0208] In this embodiment, the modulation scheme can be implicitly indicated by the input dimension of the data bits to be modulated, or it can be directly input into the modulation model.

[0209] In this embodiment, the N channels of data to be modulated can be input into the modulation model for modulation processing, but this application is not limited to this. It is understood that in order to improve modulation efficiency, the modulation processing of multiple MCS can be processed simultaneously by the modulation model.

[0210] In this embodiment, the modulation model can be implemented by a neural network model, such as any neural network model like CNN or MLP, but this application is not limited to this.

[0211] Figure 9 is a schematic diagram of the analytical model in one embodiment of this application. As shown in Figure 9, the model architecture is as follows:

[0212] Input features of the analytical model:

[0213] (1) N channels of received frequency domain data, in the format [12*RB,Symbol,Ka,2], where RB is the number of resource blocks that the analytical model can process each time, Symbol is the number of symbols, Ka is the number of antennas on the receiving side (i.e., the transmission layer), and 2 corresponds to the splitting of the real and imaginary parts of the complex number.

[0214] (2) N physical resource mapping matrices corresponding to N MCSs, in the format [12*RB,Symbol,N]. The matrix consists of 0 and 1, where 1 indicates the physical resource mapping RE position corresponding to the current modulation scheme MCSi.

[0215] (3) N modulation methods, with the format [1, N].

[0216] The output of the analytical model is N-way L-layer analytical data, where the analytical data is log-likelihood ratio (LLR) data with dimensions [12*RB, Symbol, L, N, Q]. i ], where Q i This is the indication value for the current i-th (i-th level) modulation scheme. Configurable values ​​include BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding data dimension values ​​of [1 / 2 / 4 / 6 / 8].

[0217] In the output data matrix, the data at position 1 of the corresponding modulation method resource mapping matrix is ​​a valid value, and the data at position 0 is an invalid value.

[0218] In this embodiment, the analytical model can be implemented using a neural network model, such as any neural network model like CNN, MLP, or Transformer, but this application is not limited to this.

[0219] Figure 10 is a schematic diagram of a hierarchical parsing model in one embodiment of this application. As shown in Figure 10, the hierarchical parsing model includes three AI function models: a low-order MCS parsing model (equivalent to the first type of parsing model in the above embodiment), a channel estimation model, and a high-order MCS parsing model (equivalent to the second type of parsing model in the above embodiment). The low-order MCS parsing model is only used to parse the frequency domain received data of MCS1, while each high-order MCS parsing model is used to parse the frequency domain received data of other MCSs respectively.

[0220] In this embodiment, the structure of the analytical model of the low-order MCS is as follows:

[0221] Model input:

[0222] 1) The frequency domain received data corresponding to the lowest order MCS is in the format [12*RB,Symbol,Ka,2], where RB is the number of processed RBs, Symbol is the number of symbols, Ka is the number of antennas on the receiving side, and 2 is the value after splitting the real and imaginary parts. Data at the RE position that is not mapped by this modulation method is 0.

[0223] 2) Physical resource mapping matrix corresponding to the lowest order MCS: [12*RB, Symbol]. This matrix consists of 0 and 1. 1 indicates the physical resource mapping RE position corresponding to the lowest order MCS. The data of the RE position mapped by this modulation method is 1, and the data of the RE position mapped by other modulation methods is 0.

[0224] 3) The lowest order MCS indicator Q1. Configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, etc., corresponding to [1 / 2 / 4 / 6].

[0225] Model output:

[0226] 1) The LLR value (i.e., the parsed data) after the L layer is parsed corresponding to the lowest order MCS, with dimensions [12*RB, Symbol, L, Q1].

[0227] In the output data matrix, the data at position 1 in the lowest-order MCS modulation resource mapping matrix is ​​a valid value, and the data at position 0 is an invalid value.

[0228] In this embodiment, the MCS1 data reconstruction module is used to reconstruct the MCS1 transmitter data based on the parsed data of MCS1. The reconstruction process can be implemented in the following ways:

[0229] Method 1: The output data of the MCS1 parsing model (i.e., the parsed data of MCS1) is sent to the decoding module to obtain the original bit information of MCS1 data; then the original bit data is re-decoded and AI-modulated, and resource mapping is performed to obtain the reconstructed MCS1 transmitter data.

[0230] Method 2 involves performing hard decision-making on the data bits based on the output data of the MCS1 parsing model to obtain the parsed encoded data information of the transmitter; then, the data is subjected to AI modulation processing and resource mapping processing to obtain the reconstructed MCS1 transmitter data.

[0231] In one exemplary embodiment, the structure of the channel estimation model is as follows:

[0232] Model input:

[0233] 1) The frequency domain received data of MCS1 is in the format [12*RB, Symbol, Ka, 2]. RB is the number of processed RBs, Symbol is the number of symbols, Ka is the number of receiving antennas, and 2 is the value after splitting the real and imaginary parts. Among them, the data at the RE position that is not mapped by this modulation method is 0.

[0234] 2) The MCS1 transmitter data reconstructed from the MCS1 parsing data output by the low-order parsing model is in the format [12*RB,Symbol,Ka,2], where data at RE positions that are not mapped by this modulation scheme are 0.

[0235] 3)(Optional input) N resource mapping matrices corresponding to N types of MCS, or only the resource mapping matrix of MCS1.

[0236] Model output: Complete channel estimation results for the entire resource dimension [12*RB,Symbol,L,2].

[0237] In this embodiment, the resource mapping matrix of MCS1 can be explicitly indicated or implicitly indicated by receiving data in the frequency domain.

[0238] In this embodiment, the structure of the analytical model of the higher-order MCS corresponding to MCS2 to MCSN is as follows:

[0239] Model input:

[0240] 1) The frequency domain received data for the current modulation scheme is in the format [12*RB, Symbol, Ka, 2]. RB is the number of processed RBs, Symbol is the number of symbols, Ka is the number of receiving antennas, and 2 is the value after splitting the real and imaginary parts. Data at RE positions that are not mapped to this modulation scheme is 0.

[0241] 2) Physical resource mapping matrix corresponding to the current modulation mode: [12*RB, Symbol]. This matrix consists of 0 and 1, where 1 indicates the physical resource mapping RE position corresponding to the current modulation mode.

[0242] 3) Modulation mode indication Q i Configurable values ​​are QPSK, 16QAM, 64QAM, 256QAM, etc., corresponding to numerical dimensions of [2 / 4 / 6 / 8].

[0243] 4) Channel estimation results for the entire resource dimension [12*RB,Symbol,L,2].

[0244] Model output:

[0245] 1) The LLR value after L-layer parsing of the current modulation scheme MCSi, with dimensions [12*RB, Symbol, L, Q] i ], Q i This indicates the modulation scheme. The configurable values ​​for the modulation scheme are QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding numerical dimensions of [2 / 4 / 6 / 8]. Data at the RE position that is not mapped to this modulation scheme is 0.

[0246] In some embodiments, the channel estimation model can first determine the channel estimation result of MCS1 (corresponding to physical resources) based on the frequency domain received data and reconstructed transmitter data of MCS1, and then recover the complete channel estimation result of the entire resource dimension based on the channel estimation result of MCS1 (corresponding to physical resources) and the resource mapping matrix of MCS1. At this time, the analytical model of MCSi needs to recover the channel estimation result of MCSi based on the physical resource mapping matrix of MCSi, and then obtain the analytical data of MCSi based on the channel estimation result of MCSi and the frequency domain received data of MCSi.

[0247] In other embodiments, the input to the channel estimation model may also include resource mapping matrices for N MCSs, and correspondingly, the output of the channel estimation model can be the channel estimation results for each MCS. That is, the channel estimation process for each high-order MCS can be centrally processed within the channel estimation model.

[0248] In the embodiments of this application, the neural network models in the above embodiments can be trained and learned in a phased, end-to-end joint training manner.

[0249] Phase 1: End-to-end training of the modulation model and the analytical model.

[0250] 1) Bypass the physical resource mapping neural network model. The current system is a single modulation mode and a full physical resource mapping mode.

[0251] 2) Encode the data Bit tx After passing through the modulation model, the data is mapped to modulated complex data, and then undergoes layer mapping, resource mapping, and waveform processing before being transmitted through the wireless channel.

[0252] 3) The received signal undergoes waveform de-processing and resource mapping, converting the frequency-domain received data into an analytical model input, with the output being analytical bits (LLR result). llr .

[0253] 4) Calculate the encoded data bits at the sending end. tx Bit of LLR result at the receiving end llr The binary cross entropy (BCE) is used as the loss function value (i.e., the first loss function) of the neural network, and the neural network models of the sending and receiving ends are adjusted according to the loss function value.

[0254] The loss function for the first stage is: Loss = BCE(Bit) tx Bit llr ).

[0255] 5) After the training convergence state, determine the modulation constellation diagram under the current modulation mode and save it as a table (such as a modulation mapping table) for use in the execution phase.

[0256] 6) Repeat steps 1) to 5) to complete the learning of modulation constellation diagrams for all modulation modes.

[0257] Phase 2: End-to-end training of the physical resource mapping model and the analytical model.

[0258] 1) Fixed modulation model, using the modulation mapping table obtained from stage 1 training.

[0259] 2) Transform the encoded N-way data into Bits tx,i Perform the corresponding modulation processing according to the modulation method in the modulation mapping table, map it into N channels of modulated complex data, and then perform layer mapping;

[0260] 3) Based on the predefined transmission ratios a1 to a2 for each path N According to the resource mapping matrix generation process in this application, the physical resource mapping neural network model is called respectively to obtain the resource mapping tables of each path, the data of each path is mapped to the corresponding physical resources and waveform processing is performed, and then transmitted through the wireless channel.

[0261] 4) The received signal undergoes waveform de-processing and resource mapping, using the frequency-domain received data as input to an analytical model, and outputting the analyzed N-bit LLR result. llr,i .

[0262] 5) Calculate the N-way encoded data bits tx,i Bit of N-way LLR result llr,i The BCE is used as the loss function value of the neural network (equivalent to the second loss function in the above embodiment), and the neural network models of the sending and receiving ends are adjusted according to the loss function value.

[0263] The loss function for the second stage is:

[0264] 6) After the training convergence state, determine the neural network models for the sending and receiving ends.

[0265] In one embodiment of this application, based on the communication system in Figure 5, the processing flow of the sending end may include the following steps:

[0266] Step S1: Based on the configuration of the N-way MCS and their respective physical resource allocation parameters (such as resource proportion), the original bit data is split into N-way bit data. MCSi , where i = 1, ..., N.

[0267] Step S2, for N-way Bit MCSi Perform encoding processing separately to obtain N-channel encoded data X. di .

[0268] Step S3, according to MCS1(Q1) to MCSN(Q) respectively N The modulation scheme is used to perform N-channel modulation processing to obtain the modulation data Y. di The modulation method uses the irregular modulation constellation pattern obtained from Phase 1 training, or it can directly perform modulation processing using a trained modulation model.

[0269] Modulation processing: Y di =Modulate(X) d1 Q i ), i = 1, 2, ... N.

[0270] Step S4: The N-channel modulated data are subjected to layer mapping processing to obtain Y. di,L .

[0271] Step S5: Based on the physical resource mapping table obtained from Phase 2 training, respectively, Y... di,L This is mapped to the corresponding physical resource (RE). Alternatively, resource mapping can be performed directly using a pre-trained physical resource mapping model.

[0272] Step S6: After waveform processing, the data from the transmitting end is converted into a time-domain transmission signal (i.e., time-domain data) for transmission.

[0273] In one embodiment of this application, based on the communication system in Figure 5, the processing flow of the receiving end may include the following steps:

[0274] Step S7: De-waveform processing is performed on the received signal to obtain the corresponding frequency domain received data Rx, and the data format [12*RB,Symbol,Ka,2] is constructed.

[0275] Step S8: Based on the physical resource mapping table obtained from Phase 2 training, construct the resource mapping matrix [12*RB, Symbol, N] for each path, as well as the modulation method indication information [1, N] for each path.

[0276] Step S9: Perform data parsing processing on the analytical model trained in stage 2 using the above three input features to obtain N-way L-layer analytical data bits. LLR [12*RB,Symbol,L,N,Qi], where Qi is the modulation scheme, and configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding numerical dimensions of [1 / 2 / 4 / 6 / 8].

[0277] Step S10: Perform N-way decoding on the N-way parsed data to obtain the decoded data Bit. MCSi The decoded data is concatenated and combined with the CRC check result, and then sent to the higher layer.

[0278] In another embodiment of this application, based on the communication system in FIG6, the processing flow of the sending end may include the following steps:

[0279] Step S1': Based on the N-way MCS configuration and their respective physical resource allocation parameters (such as resource proportion), determine the total bit data BitMCS corresponding to the N types of MCS from the original bit data; and encode the BitMCS to obtain the encoded data X. d ;

[0280] Step S2': Based on the N-way MCS configuration and their respective physical resource allocation parameters, adjust X... d Perform splitting processing to obtain N-way encoded data X di .

[0281] Step S3, according to MCS1(Q1) to MCSN(Q) respectively N The modulation scheme is used to perform N-channel modulation processing to obtain the modulation data Y. di The modulation method can adopt the irregular modulation constellation pattern obtained from Phase 1 training.

[0282] Y di =AIModulate(X) d1 Q i), i = 1, 2, ... N.

[0283] Step S4: The N-channel modulated data are subjected to layer mapping processing to obtain Y. di ,L.

[0284] Step S5: According to the predefined physical resource mapping table, respectively, Y... di,L Mapped to the corresponding physical resource (RE).

[0285] Step S6: After waveform processing, the data from the transmitting end is converted into a time-domain transmission signal (i.e., time-domain data) for transmission.

[0286] The difference between this embodiment and the previous embodiment is that this example first performs encoding processing, splitting the encoded data into N-channel encoded data, which can reduce the complexity of the encoding processing and improve the encoding efficiency. Other processing steps, such as modulation processing, layer mapping processing, physical resource mapping processing, waveform processing, etc., are the same as those described in the previous embodiment, and will not be repeated here.

[0287] In another embodiment of this application, based on the communication system in FIG6, the processing flow of the sending end may include the following steps:

[0288] Step S7: De-waveform processing is performed on the received signal to obtain the corresponding frequency domain received data Rx, and the data format [12*RB,Symbol,Ka,2] is constructed.

[0289] Step S8: Based on the physical resource mapping table obtained from Phase 2 training, construct the resource mapping matrix [12*RB, Symbol, N] for each path, and the modulation method indication information [1, N] for each path.

[0290] Step S9: Perform data parsing processing on the analytical model trained in stage 2 using the above three input features to obtain N-way L-layer analytical data bits. LLR [12*RB,Symbol,L,N,Qi], where Qi is the modulation scheme, and configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding numerical dimensions of [1 / 2 / 4 / 6 / 8].

[0291] The analytical model in step S9 can be a hierarchical analytical model, which is composed of the following cascaded processing models.

[0292] 1) Perform low-order analytical model processing for MCS1:

[0293] Obtain the resource mapping matrix corresponding to MCS1, [12*RB, Symbol]. This matrix consists of 0 and 1, where 1 indicates the physical resource mapping RE position corresponding to the current modulation mode.

[0294] The frequency domain received data [12*RB, Symbol, Ka, 2] corresponding to MCS1 is obtained according to the MCS1 resource mapping matrix. Here, only the MCS1 resource mapping position has valid data, and the data at other positions is 0.

[0295] Obtain the modulation scheme Q1 corresponding to MCS1. The configurable values ​​are BPSK, QPSK, 16QAM, 64QAM, etc., with corresponding numerical dimensions of [1 / 2 / 4 / 6].

[0296] Input the above data matrix into the MCS1 parsing model, execute the MCS1 parsing model processing, and obtain the MCS1 parsing data with dimensions [12*RB,Symbol,L,Q1].

[0297] 2) Reconstruct the parsed data of MCS1.

[0298] Here, method 2 is adopted to perform bit hard decision processing on the MCS1 parsed data, and according to the data bits after the decision, the data reconstruction modulation is performed using the MCS1 modulation method to obtain the MCS1 reconstructed transmitter modulation data, [12*RB,Symbol,Ka,2]. Here, there is valid data at the RE position corresponding to the MCS1 mapping, and the RE data at the other positions is 0.

[0299] 3) Perform AI channel estimation processing.

[0300] The frequency domain received data [12*RB,Symbol,Ka,2] corresponding to MCS1 and the modulated data [12*RB,Symbol,Ka,2] reconstructed from MCS1 are used as input data for the model. AI channel estimation processing is performed to obtain the channel estimation value [12*RB,Symbol,L,2] corresponding to the entire physical resource location.

[0301] 4) Perform parallel processing of the MCS2-MCSN high-order modulation analytical model.

[0302] Obtain the resource mapping matrix [12*RB, Symbol] corresponding to each MCS. This matrix consists of 0 and 1, where 1 indicates the physical resource mapping RE position corresponding to the current modulation mode.

[0303] The frequency domain received data [12*RB, Symbol, Ka, 2] corresponding to each MCS resource mapping matrix is ​​obtained. Here, only the MCSi resource mapping position has valid data, and the data at other positions is 0.

[0304] Obtain the modulation scheme Qi corresponding to each MCSi. The configurable values ​​are QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding numerical dimensions of [2 / 4 / 6 / 8].

[0305] Obtain the channel estimation results for the entire resource dimension [12*RB,Symbol,L,2].

[0306] Input the above data matrix into each MCS analytical model, perform analytical model processing, and obtain the analytical bit LLR information [12*RB,Symbol,L,Qi] under each MCS modulation. Qi is the modulation mode, and the configurable values ​​are QPSK, 16QAM, 64QAM, 256QAM, etc., with corresponding numerical dimensions of [2 / 4 / 6 / 8].

[0307] Step S10': Combine and decode the N-way parsed data to obtain the decoded data Bit. MCS The decoded data, along with the CRC check result, is sent to the higher layer.

[0308] To better illustrate the communication method in this application, it will be further described below in conjunction with specific parameter configurations and scenarios.

[0309] Example 1: Assuming a single-input single-output (SISO) OFDM communication system, the data is a single stream, the base station and user equipment are configured with a single antenna, and the number of scheduled RBs is 1, the initial parameter determination process of the transmitter in this application is as follows:

[0310] 1. Determine the value of N in this transmission block; N is 2.

[0311] 2. The MCS level and modulation scheme are determined by the higher-level scheduling; MCS1 corresponds to QPSK modulation, and MCS2 corresponds to 64QAM modulation; the method by which the higher-level scheduling determines the MCS level is not limited. For example, it can be represented by MCS1 (Q1=2) and MCS2 (Q2=6).

[0312] 3. Phase 1 - End-to-end training of AI modulation and parsing models.

[0313] 1) Bypass the physical resource mapping neural network model; the current system is a single modulation mode physical resource full mapping mode.

[0314] 2) Encode the data Bit tx After passing through the AI ​​modulation model, the data is mapped into modulated complex data, and then undergoes layer mapping, resource mapping, and waveform processing before being transmitted through the wireless channel.

[0315] 3) The received signal undergoes waveform de-processing and resource mapping, converting the frequency-domain received data into an AI receiver input, with the output being an LLR result in bits after parsing. llr .

[0316] 4) Calculate the BCE of the transmitted bits and the received LLR, and use it as the loss value of the neural network. Adjust the transceiver neural network model according to the loss value.

[0317] First-stage loss function: Loss = BCE(Bit) tx Bit llr ).

[0318] 5) After the training convergence state, determine the AI ​​modulation constellation diagram under the current modulation mode and save it as a table for use in the execution phase.

[0319] 6) Repeat steps 1)-5) to complete the learning of AI modulation constellation diagrams for all modulation modes.

[0320] 4. Determine the RE resource ratio for MCS1 transmission a1 = 12 / 168; and the RE resource ratio for MCS2 transmission a2 = 156 / 168;

[0321] 5. Training phase 2: Physical resource neural network model.

[0322] The input features are:

[0323] 1) Total number of mappable REs in MCS1 modulation scheme.

[0324] RE i =a i ·12·Symbol·RB=12 / 168*12*14*1=12;

[0325] 2) Current overall physical resource mappable matrix: [N RE A 12*RB, Symbol matrix, a zero-matrix of size [12, 14];

[0326] 3) Modulation mode indication, Q1 = 2.

[0327] The output of this model is the RE matrix mapped under the current modulation scheme, with 0 indicating no mapping and 1 indicating mapping.

[0328] [MCS RE,i ]12·RB,Symbol=[MCS RE,1 ] 12,14 There are a total of 12 REs with a value of 1.

[0329] 6. Obtain the resource mapping matrix of MCS1 based on the physical resource neural network model. Figure 11 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS single-transmission layer hybrid modulation transmissions. As shown in Figure 11, the total number of mapping REs of MCS1 is 12. The element 1 in the matrix indicates the RE position of the MCS1 mapping.

[0330] 7. Invert the matrix corresponding to MCS1 to obtain the resource mapping pattern corresponding to MCS2. Figure 12 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS single-transmission-layer hybrid modulation transmissions. As shown in Figure 12, the matrix of MCS2 is obtained by inverting the matrix of MCS1 in Figure 11.

[0331] In this embodiment, Figures 11 and 12 are merely possible examples and not the fixed output of the physical resource mapping neural network. The resource mapping matrices shown in Figures 11 and 12 can be used to map the data of MCS1 and MCS2 to the corresponding physical resources REs, respectively.

[0332] In this embodiment of the application, the sending end processing flow is as follows:

[0333] 1. The original bits are split into two data streams, BitMCS1 and BitMCS2, according to the configuration of the two MCSs and their respective physical resource allocation values.

[0334] 2. Transfer the two bit paths MCS1 And Bit MCS2 The data is encoded separately to obtain two encoded data streams X. d1 and X d2 ;

[0335] 3. Perform two-channel AI modulation processing according to the modulation schemes of MCS1 (Q1=2) and MCS2 (Q2=6) respectively to obtain the modulation data Y. d1 Y d2 .

[0336] Y d1 =AIModulate(X) d1 Q1) = AIModulate(X d1 ,2);

[0337] Y d2 =AIModulate(X) d2 Q2) = AIModulate(X d2 ,6);

[0338] 4. Obtain the resource mapping matrix corresponding to MCS1 based on the resource mapping neural network, and invert the resource mapping matrix of MCS1 to obtain the resource mapping matrix corresponding to MCS2. Then, respectively, set Y... d1 Y d2 Mapped to the corresponding physical resource (RE).

[0339] 5. After waveform processing, the data at the transmitting end is converted into a time-domain transmission signal for transmission.

[0340] In this embodiment of the application, the receiving end processing flow is as follows:

[0341] 1. The received signal undergoes waveform demodulation to obtain frequency domain received data;

[0342] 2. Obtain the input data features of the analytical model:

[0343] 1) Frequency domain received data, in the format [12*RB,Symbol,Ka,2]=[12,14,1,2].

[0344] 2) Current 2-way physical resource mapping matrix: [12*RB,Symbol,N]=[12,14,2], this matrix is ​​composed of 0 / 1, 1 indicates the physical resource mapping RE position corresponding to the current MCSi modulation mode.

[0345] 3) N-channel modulation scheme [1,N]=[1,2]=[2,6]. MCS1 corresponds to QPSK modulation Q=2; MCS2 corresponds to 64QAM modulation, Q=6.

[0346] 3. Input the above data information into the analytical model network, perform analytical model processing, and output the LLR values ​​after N=2 L layers of analysis, with dimensions [12*RB, Symbol, L, N, Qi]. The corresponding data dimensions for each path are MCS1 dimension: [12, 14, 1, 1, 2] and MCS2 dimension: [12, 14, 1, 1, 6].

[0347] 4. Based on the physical resource mapping tables of MCS1 and MCS2, extract the received signal LLR information respectively to obtain... and

[0348] 5. Perform two-way decoding processing to obtain two-way decoded data, and concatenate the two-way decoded data and send the combined CRC check result to the higher layer.

[0349] Example 2: Assuming a two-layer data transmission system in an OFDM MIMO communication system, with dual antennas configured for both the base station and the user, and a scheduled RB count of 1, the process for determining the initial parameters of the transmitter in this application is as follows.

[0350] 1) Determine the value of N in this transport block; N is 2.

[0351] 2) The MCS level and modulation scheme are determined by the higher-level scheduling; MCS1 corresponds to QPSK modulation, and MCS2 corresponds to 64QAM modulation; there are no restrictions on how the higher-level scheduling determines the MCS level.

[0352] 3) Determine the RE resource ratio for MCS1 transmission a1 = 12 / 168; and the RE resource ratio for MCS2 transmission a2 = 156 / 168;

[0353] 4) Training Phase 2: Physical Resource Neural Network Model

[0354] The model input features are: (1) the total number of REs that can be mapped by the MCS1 modulation scheme, and the mapping positions of each layer are consistent. i =a i ·12·Symbol·RB=12 / 168*12*14*1=12;(2) Current overall physical resource mappable matrix: [N RE ]12*RB, Symbol matrix, a 0 matrix of size [12,14]; (3) Modulation mode Qi = 2.

[0355] The model output is the RE matrix mapped under the current modulation scheme, with 0 indicating no mapping and 1 indicating mapping.

[0356] [MCS RE,i ]12·RB,Symbol=[MCS RE,1 ] 12,14 The total number of REs with a value of 1 is 12.

[0357] 5) Obtain the resource mapping matrix of MCS1 based on the physical resource neural network model.

[0358] One possible mapping is shown in Figure 13. Figure 13 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS dual-transmission-layer hybrid modulation transmissions. Figure 13 is only one possible example and not a fixed output definition of the neural network. MCS1 data is mapped to physical resources (REs) as shown in this diagram.

[0359] 6) By inverting the matrix corresponding to MCS1, the resource mapping matrices corresponding to layers 1 and 2 of MCS2 can be obtained.

[0360] Figure 14 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS dual transmission layer hybrid modulation transmissions. As shown in Figure 14, the resource mapping matrix of MCS2 is the inverse of the resource mapping matrix of MCS1 (as shown in Figure 13).

[0361] In this embodiment, the sending end processing flow is as follows:

[0362] 1. The raw bit data is encoded based on the 2-channel MCS configuration and their respective physical resource allocation values, and the encoding rate is calculated according to the Bit... all After completing the data encoding process, X is obtained. d ;

[0363] Encoding processing: X d =Coder(Bit) all );

[0364] 2. Regarding X dThe data is split to obtain two coded data X channels under different modulation schemes. d1 and X d2 ;

[0365] 3. Perform two-channel AI modulation processing according to the modulation schemes of MCS1 (Q1=2) and MCS2 (Q2=6) respectively to obtain the modulation data Y. d1 Y d2 .

[0366] Y d1 =AIModulate(X) d1 Q1) = AIModulate(X d1 ,2);

[0367] Y d2 =AIModulate(X) d2 Q2) = AIModulate(X d2 ,6);

[0368] 4. Y d1 Y d2 Layered processing is performed to obtain two channels of two-layer transmission data Y. d1，L1 / L2 Y d2，L1 / L2 .

[0369] 5. Obtain the resource mapping matrix corresponding to MCS1 based on the resource mapping neural network, and invert the resource mapping matrix of MCS1 to obtain the resource mapping matrix corresponding to MCS2. Then, respectively... d1 Y d2 Mapped to the corresponding physical resource (RE).

[0370] 6. After waveform processing, the data at the transmitting end is converted into a time-domain transmission signal for data transmission.

[0371] Example 3: Assuming an OFDM MIMO communication system with four layers of data, both the base station and user equipment are configured with four antennas, and the number of scheduled RBs is 1, the initial parameter determination process of the transmitter in this application is as follows.

[0372] 1) Determine the value of N in this transport block; N is 2.

[0373] 2) The MCS level and modulation scheme are determined by the higher-level scheduling; MCS1 corresponds to QPSK modulation, and MCS2 corresponds to 64QAM modulation; there are no restrictions on how the higher-level scheduling determines the MCS level.

[0374] 3) Determine the RE resource ratio for MCS1 transmission a1 = 12 / 168; and the RE resource ratio for MCS2 transmission a2 = 156 / 168;

[0375] 4) Training Phase 2: Physical Resource Neural Network Model

[0376] The input features of the model include: (1) the total number of mapping REs for MCS1 modulation, with staggered mapping positions for each layer, and a single layer mapping RE number of 3. i =a i ·12·Symbol·RB=12 / 168*12*14*1=12;(2) Current overall physical resource mappable matrix: [N RE ]12*RB, Symbol matrix, a zero matrix of size [12,14]; (3) Modulation method Q i =2.

[0377] The model output is the RE matrix mapped under the current modulation scheme, where 0 indicates no mapping and 1 indicates mapping.

[0378] [MCS RE,i ]12·RB,Symbol=[MCS RE,1 ] 12,14 There are a total of 12 REs with a value of 1. Each layer is mapped and loaded according to the sequential index.

[0379] 5) Obtain the resource mapping matrix under MCS1 based on the physical resource neural network model. One possible mapping is shown in Figure 15. Figure 15 is a schematic diagram of the resource mapping matrix of MCS1 in two MCS four-transmission-layer hybrid modulation transmissions. Figure 15 is only a possible example and not a fixed output definition of the neural network. Map the MCS1 data to the physical resources RE layer by layer according to the diagram.

[0380] 6) Invert the matrix corresponding to MCS1 to obtain the resource mapping matrix corresponding to each layer of MCS2. Figure 16 is a schematic diagram of the resource mapping matrix of MCS2 in two MCS four-transmission-layer hybrid modulation transmission. As shown in Figure 16, it is obtained by inverting the resource mapping matrix of MCS1 in Figure 15.

[0381] In this embodiment, the sending end processing flow is basically the same as that in embodiment 2 above. The difference is that the number of transmission layers in embodiment 3 is 4, and the resource RE locations mapped by the four transmission layers of MCS1 are different.

[0382] According to the communication methods and communication systems proposed in the various embodiments of this application, multi-level hybrid modulation of transmitted data can be realized, achieving the technical effect of predicting the channel information of the complete transmission RE part based on the low-order modulation RE, and enhancing the data parsing of other high-order modulation RE parts by combining the predicted channel estimation, thereby improving the high-order transmission performance of the pilotless communication system and greatly improving the spectral efficiency of the communication system.

[0383] This application implements a novel pilotless communication system based on neural networks. By combining the end-to-end training method of the neural network model, it can not only learn the optimal hybrid modulation RE distribution pattern, but also the optimal analytical model, thereby greatly improving the performance of the pilotless communication system and increasing the spectral efficiency of the communication system.

[0384] This application proposes a multi-stage end-to-end joint training method, which can realize the joint training of multiple models of the transmitter and receiver, and obtain the optimal joint training performance of the transmitter and receiver, thereby improving the performance of the end-to-end pilotless communication system.

[0385] This application employs a multi-level hybrid modulation method for transmission REs, and for different transmission modes, it designs the optimal resource mapping pattern under different modulation methods based on the AI ​​physical resource mapping neural network model. This ensures that more accurate channel state prediction information can be obtained through low-order modulation REs when transmitting without pilots, thereby improving the system performance of MIMO multi-stream transmission and high-order modulation.

[0386] This application proposes an analytical model based on multi-level hybrid modulation, which can realize the analytical processing of received data from multi-level hybrid modulation and improve the demodulation performance of the receiver. This analytical model can be an architecture design based on a full neural network or a cascaded neural network, both of which can realize the analytical processing of received data from multi-level hybrid modulation. Furthermore, the cascaded neural network model can be trained and learned in modules, making model training more convenient and effective, and enhancing interpretability.

[0387] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods according to the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the system described in the various embodiments of this application.

[0388] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the steps in any of the above method embodiments.

[0389] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0390] Embodiments of this application also provide an electronic device including a memory and a processor, wherein the memory stores a computer program and the processor is configured to run the computer program to perform the steps in any of the above method embodiments.

[0391] In one exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor and the input / output device is connected to the processor.

[0392] Embodiments of this application also provide a computer program product, including a computer program that, when executed by a processor, implements the steps in any of the above method embodiments.

[0393] Specific examples in this embodiment can be found in the examples described in the above embodiments and exemplary implementations, and will not be repeated here.

[0394] Obviously, those skilled in the art should understand that the modules or steps of this application described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. They can be implemented using computer-executable program code, and thus can be stored in a storage device for execution by a computing device. In some cases, the steps shown or described can be performed in a different order than those presented here, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular hardware and software combination.

[0395] The above description is merely an exemplary embodiment of this application and is not intended to limit this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A communication method, comprising: The original bit data is encoded to obtain N-channel encoded data, where N is an integer greater than or equal to 2; According to N modulation and coding strategies (MCS), the N coded data are converted into time-domain signals through a pre-trained modulation model and a preset physical resource mapping table. The modulation model is used to implement multi-level hybrid modulation processing of the N MCS. The time-domain signal is transmitted via a wireless channel.

2. The method according to claim 1, wherein, The step of converting N coded data into time-domain signals based on N modulation and coding strategies (MCS) using a pre-trained modulation model and a preset physical resource mapping table includes: Based on the N types of MCS, the N coded data are modulated using the modulation model to obtain N modulated data, wherein each modulated data corresponds to one type of MCS; The N-channel modulation data are mapped to resource blocks using the physical resource mapping table; The mapped data is subjected to waveform processing to generate the time-domain signal.

3. The method according to claim 2, wherein, The step of performing N-path modulation processing on the N-path coded data according to the N types of MCS and the modulation model to obtain N-path modulated data includes: The N types of MCS and the N coded data are input into the modulation model, and the N modulated data output by the modulation model are obtained. The modulation model inputs one path of coded data and its corresponding MCS each time and outputs one path of modulated data; or... The modulation model generates N irregular modulation constellation diagrams corresponding to the N types of MCS, and modulates the N coded data according to the N irregular modulation constellation diagrams to obtain the N modulated data. Each irregular modulation constellation diagram corresponds to one coded data path and one modulated data path.

4. The method according to claim 2, wherein, Before mapping the N-channel modulated data to resource blocks via the physical resource mapping table, the method further includes: Determine the number of resource units that each of the N types of MCS can occupy in the resource block, wherein the resource block includes multiple resource units; Based on the N types of MCS and the number of resource units of the N types of MCS, N resource mapping matrices corresponding to the N types of MCS are generated by a pre-trained physical resource mapping neural network model, wherein the physical resource mapping table includes the N resource mapping matrices.

5. The method according to claim 4, wherein, The step of generating N resource mapping matrices corresponding to the N MCSs based on the N types of MCSs and the number of resource units in the N types of MCSs, using a pre-trained physical resource mapping neural network model, includes: In response to N equaling 2, the first type of MCS, the number of resource units of the first type of MCS, and the initial physical resource mappable matrix are input into the physical resource mapping neural network model to obtain the first resource mapping matrix output by the physical resource mapping neural network model, wherein the first resource mapping matrix corresponds to the first type of MCS, and the initial physical resource mappable matrix is ​​an all-zero matrix. Invert the first resource mapping matrix to generate a second resource mapping matrix corresponding to the second type of MCS; The N resource mapping matrices include the first resource mapping matrix and the second resource mapping matrix.

6. The method according to claim 4, wherein, The step of generating N resource mapping matrices corresponding to the N MCSs based on the N types of MCSs and the number of resource units in the N types of MCSs, using a pre-trained physical resource mapping neural network model, includes: In response to N being greater than 2, for i = 1, ..., N, the N resource mapping matrices are generated sequentially according to the following method: In response to i being less than N, the i-th type of MCS, the number of resource units of the i-th type of MCS, and the current physical resource mappable matrix are input into the physical resource mapping neural network model to obtain the i-th resource mapping matrix output by the physical resource mapping neural network model corresponding to the i-th type of MCS. The initial value of the physical resource mappable matrix is ​​a matrix of all zeros, and the current physical resource mappable matrix is ​​the complete set of the first to the (i-1)-th resource mapping matrices. In response to i being less than or equal to N, the current physical resource mappable matrix is ​​updated according to the i-th resource mapping matrix; In response to i being equal to N, the entire set of the first resource mapping matrix to the (i-1)th resource mapping matrix is ​​inverted to obtain the Nth resource mapping matrix.

7. The method according to claim 1, wherein, The method further includes: The joint receiver performs the first stage of end-to-end training on the modulation model and the analytical model; The receiving end is combined with the physical resource mapping neural network model and the analytical model to perform a second stage of end-to-end training; The physical resource mapping neural network model is used to generate the physical resource mapping table. The physical resource mapping neural network model is not activated in the first stage of end-to-end training, but is activated in the second stage of end-to-end training.

8. The method according to claim 7, wherein, The joint receiver performs a first-stage end-to-end training of the modulation model and the analytical model, including: For each of the M types of MCS, the encoded first data is input into the modulation model to be trained to obtain the modulated second data output by the modulation model to be trained, where M is an integer greater than or equal to N. The second data is subjected to layer mapping processing, resource mapping processing, and waveform processing to generate a first transmission signal, and the first transmission signal is transmitted to the receiving end through a wireless channel. The receiver obtains the parsed third data generated by the receiver based on the first transmitted signal and the parsing model to be trained, and calculates the binary cross-entropy based on the first data and the third data to obtain the first loss function of the end-to-end training in the first stage. The receiving end adjusts the modulation model and the analytical model together until the first loss function converges.

9. The method according to claim 7, wherein, The joint receiving end performs a second-stage end-to-end training of the physical resource mapping neural network model and the analytical model, including: Based on the modulation model trained in the first stage, M modulation mapping tables corresponding to M types of MCS are generated, and N modulation mapping tables corresponding to the N types of MCS are determined according to the selected N types of MCS, where M is an integer greater than or equal to N. Based on the N modulation mapping tables, the N coded data are modulated and layer mapped to generate N layer mapped data. Based on the preset resource proportion of the resource units corresponding to the N types of MCS, N-way resource mapping tables are generated respectively through the physical resource mapping neural network model to be trained; According to the N-way resource mapping table, the N-way layer mapping data is processed for resource mapping and waveform processing to generate a second transmission signal, and the second transmission signal is sent to the receiving end through a wireless channel; The receiver obtains N-way parsing data generated by the receiver based on the second transmission signal, the N-way resource mapping table and the parsing model, and calculates the binary cross-entropy based on the N-way encoded data and the N-way parsing data to obtain the second loss function for the end-to-end training in the second stage. The receiving end is combined to adjust the physical resource mapping neural network model and the analytical model until the second loss function converges.

10. A communication method, comprising: The time-domain signal is received through a wireless channel, and the time-domain signal is subjected to waveform decoding. Based on N modulation and coding schemes (MCS), the de-waveform data is converted into N streams of analytical data through a preset physical resource mapping table and a pre-trained analytical model, where N is an integer greater than or equal to 2, and the analytical model is used to implement multi-level hybrid demodulation and modulation processing of the N MCS. The N-channel parsed data is decoded to obtain parsed bit data.

11. The method according to claim 10, wherein, The process of converting the decoded waveform data into N streams of analytical data using N modulation and coding schemes (MCS) through a preset physical resource mapping table and a pre-trained analytical model includes: Based on the physical resource mapping table, determine the N resource mapping matrices corresponding to the N types of MCS; Based on the N resource mapping matrices, the de-waveform data is de-mapped to obtain N channels of frequency domain received data, where each channel of frequency domain received data corresponds to a type of MCS. Based on the N types of MCS and the N resource mapping matrices, the N channels of frequency domain received data are analyzed through the analytical model to obtain the N channels of analytical data output by the analytical model. The analytical data is log-likelihood ratio data, and each channel of analytical data corresponds to one type of MCS.

12. The method according to claim 11, wherein, The input data of the analytical model includes the N channels of frequency domain received data and the indications of the N types of MCS, wherein the N channels of frequency domain received data are also used to implicitly indicate the N resource mapping matrices; or, The input data of the analytical model includes the N frequency domain received data, the N MCS indications, and the N resource mapping matrices.

13. The method according to claim 11, wherein, In response to the fact that the analytical model is a hierarchical analytical model, the analytical model includes: a first type of analytical model corresponding to the first type of MCS, a data reconstruction unit, a channel estimation model, and N-1 second type analytical models corresponding to other MCSs besides the first type of MCS, wherein the first type of MCS is the lowest order modulation scheme among the N types of MCSs.

14. The method according to claim 13, wherein, The step of parsing the N frequency domain received data based on the N types of MCS and the N resource mapping matrices using the analytical model to obtain the N parsed data output by the analytical model includes: The first frequency domain received data corresponding to the first type of MCS and the indication of the first type of MCS are input into the first type of analytical model to obtain the first log-likelihood ratio data output by the first type of analytical model. The data reconstruction unit reconstructs the transmitting end data corresponding to the first frequency domain received data. The first frequency domain received data, the reconstructed transmitter data, and the N resource mapping matrices are input into the channel estimation model to obtain N channel estimation results corresponding to the N types of MCS. In response to N equaling 2, the second frequency domain received data corresponding to the second type of MCS, the indication of the second type of MCS, and the second channel estimation result corresponding to the second type of MCS are input into the second type of analytical model to obtain the second log-likelihood ratio data, wherein the N analytical data includes the first log-likelihood ratio data and the second log-likelihood ratio data; In response to N being greater than 2, the j-th frequency domain received data corresponding to the j-th MCS, the indication of the j-th MCS, and the j-th channel estimation result corresponding to the j-th MCS are input into the (j-1)-th second-type analytical model to obtain the j-th log-likelihood ratio data, where j = 2, ..., N, and the N analytical data include the first log-likelihood ratio data to the Nth log-likelihood ratio data.

15. The method according to claim 14, wherein, The step of reconstructing the transmitting end data corresponding to the first frequency domain received data through the data reconstruction unit includes: Hard-decision processing is performed on the first log-likelihood ratio data, and the hard-decision processed data is re-modulated according to the first MCS to obtain the reconstructed transmitter data; or, The first decoded data is obtained by decoding the first log-likelihood ratio data, and the first decoded data is re-encoded, modulated, and resource-mapped according to the first MCS to obtain the reconstructed transmitting data.

16. The method of claim 10, wherein, The method further includes: The joint transmitter performs the first stage of end-to-end training on the modulation model and the analytical model; The sending end and the physical resource mapping neural network model are trained together in the second stage of end-to-end training; The physical resource mapping neural network model is used to generate the physical resource mapping table. The physical resource mapping neural network model is not activated in the first stage of end-to-end training, but is activated in the second stage of end-to-end training.

17. The method according to claim 16, wherein, The joint transmitter performs a first-stage end-to-end training of the modulation model and the analytical model, including: For each of the M types of MCS, a first transmission signal is received from the transmitting end via a wireless channel. The first transmission signal is obtained by the transmitting end through modulation processing of the coded data using the modulation model to be trained, and layer mapping processing, resource mapping processing, and waveform processing of the modulated data. Here, M is an integer greater than or equal to N. The first transmitted signal is subjected to waveform de-processing and resource de-mapping processing to obtain frequency domain received data. The frequency domain received data and the corresponding MCS indication are input into the analytical model to be trained to obtain the analytical data output by the analytical model to be trained. The analytical data is log-likelihood ratio data. The encoded data from the sending end is obtained, and the binary cross-entropy is calculated based on the encoded data and the parsed data to obtain the first loss function for the end-to-end training in the first stage. The transmitting end adjusts the modulation model and the analytical model together until the first loss function converges.

18. The method according to claim 16, wherein, The second phase of end-to-end training of the physical resource mapping neural network model and the analytical model by the sending end includes: Based on the preset resource proportion of the resource units corresponding to the N types of MCS, N-way resource mapping tables are generated respectively through the physical resource mapping neural network model to be trained; The second transmission signal is received from the transmitting end via a wireless channel, and the second transmission signal is subjected to waveform processing. The second transmission signal is obtained by the transmitting end through modulation and layer mapping processing of N coded data based on N modulation mapping tables corresponding to the N types of MCS, and through resource mapping processing and waveform processing of the layer-mapped data according to the N resource mapping tables. The N modulation mapping tables are selected by the transmitting end from M modulation mapping tables corresponding to M types of MCS. The M modulation mapping tables are generated based on the modulation model trained in the first stage, and M is an integer greater than or equal to N. According to the N-way resource mapping table, the data after waveform processing is subjected to de-resource mapping processing to obtain N-way frequency domain received data. The N-way frequency domain received data and the indications of the N types of MCS are input into the analytical model to obtain N-way analytical data output by the analytical model, wherein the analytical data is log-likelihood ratio data. Obtain the N-way encoded data from the transmitting end, and calculate the binary cross-entropy based on the N-way encoded data and the N-way parsed data to obtain the second loss function for the end-to-end training in the second stage; The sending end is combined to adjust the physical resource mapping neural network model and the analytical model until the second loss function converges.

19. A pilotless communication system, comprising a transmitter and a receiver, wherein, The transmitting end is used to perform pilotless communication according to the method described in any one of claims 1 to 9; The receiving end is used for pilotless communication according to the method described in any one of claims 10 to 18.

20. A computer-readable storage medium storing a computer program, wherein, When the computer program is executed by a processor, it implements the steps of the method described in any one of claims 1 to 18.

21. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, When the processor executes the computer program, it implements the steps of the method described in any one of claims 1 to 18.

22. A computer program product comprising a computer program that, when executed by a processor, implements the steps of the method described in any one of claims 1 to 18.

Images