WO2022100493A1 - 数据传输方法、装置及设备 - Google Patents

数据传输方法、装置及设备 Download PDF

Info

Publication number
WO2022100493A1
WO2022100493A1 PCT/CN2021/128386 CN2021128386W WO2022100493A1 WO 2022100493 A1 WO2022100493 A1 WO 2022100493A1 CN 2021128386 W CN2021128386 W CN 2021128386W WO 2022100493 A1 WO2022100493 A1 WO 2022100493A1
Authority
WO
WIPO (PCT)
Prior art keywords
machine learning
ack
learning model
sequence
nack
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/128386
Other languages
English (en)
French (fr)
Inventor
杨现俊
索士强
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Datang Mobile Communications Equipment Co Ltd
Original Assignee
Datang Mobile Communications Equipment Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Datang Mobile Communications Equipment Co Ltd filed Critical Datang Mobile Communications Equipment Co Ltd
Priority to EP21891020.6A priority Critical patent/EP4246850B1/en
Priority to US18/036,577 priority patent/US12615124B2/en
Publication of WO2022100493A1 publication Critical patent/WO2022100493A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L5/00—Arrangements affording multiple use of the transmission path
    • H04L5/003—Arrangements for allocating sub-channels of the transmission path
    • H04L5/0053—Allocation of signalling, i.e. of overhead other than pilot signals
    • H04L5/0055—Physical resource allocation for ACK/NACK
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G06N3/0455—Auto-encoder networks; Encoder-decoder networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0495—Quantised networks; Sparse networks; Compressed networks
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/09—Supervised learning
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/096—Transfer learning
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/12—Arrangements for detecting or preventing errors in the information received by using return channel
    • H04L1/16—Arrangements for detecting or preventing errors in the information received by using return channel in which the return channel carries supervisory signals, e.g. repetition request signals
    • H04L1/18—Automatic repetition systems, e.g. Van Duuren systems
    • H04L1/1829—Arrangements specially adapted for the receiver end
    • H04L1/1854—Scheduling and prioritising arrangements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L1/00—Arrangements for detecting or preventing errors in the information received
    • H04L1/12—Arrangements for detecting or preventing errors in the information received by using return channel
    • H04L1/16—Arrangements for detecting or preventing errors in the information received by using return channel in which the return channel carries supervisory signals, e.g. repetition request signals
    • H04L1/18—Automatic repetition systems, e.g. Van Duuren systems
    • H04L1/1829—Arrangements specially adapted for the receiver end
    • H04L1/1861—Physical mapping arrangements

Definitions

  • the present disclosure relates to the field of communication technologies, and in particular, to a data transmission method, apparatus, and device.
  • the network can support machine densities up to 10 machines per square meter or 100 machines per cubic meter, which usually only sporadically send some small data Packet, using the scheduling-based communication mode in the related art will lead to a large access signaling overhead, while the contention-based scheduling-free communication can reduce the access signaling overhead.
  • PUSCH physical uplink shared channel
  • bitmap bitmap positive acknowledgment/negative acknowledgment
  • ACK/NACK acknowledgement/Negative Acknowledgement
  • the mechanism or the way of adding ACK/NACK to user equipment (user equipment, UE) identifier (identifier, ID) will result in a large downlink feedback overhead.
  • 5G 5th Generation
  • PDSCH 6G physical downlink shared channel
  • the ACK/NACK feedback method for PUSCH in the related art includes:
  • the base station corresponds each UE ID to an ACK/NACK-Inactive bit, that is, the bitmap method. Specifically, 1 can be used to represent ACK, 0 can be used to represent NACK or the terminal is in an inactive state; then the base station sends this long ACK/NACK-Inactive The sequence is broadcast to the terminal;
  • the base station When feeding back the ACK/NACK bit, the base station simultaneously sends and receives the UE ID of the ACK/NACK bit;
  • the base station sends ACK/NACK bits with the Physical Hybrid ARQ Indicator Channel (Physical Hybrid ARQ Indicator Channel, PHICH), and its corresponding relationship is relevant to the uplink resources allocated by the user;
  • Physical Hybrid ARQ Indicator Channel Physical Hybrid ARQ Indicator Channel, PHICH
  • the base station passes the new data indicator (New Data Indicator, PDCCH) in the physical downlink control channel (PDCCH).
  • PDCCH New Data Indicator
  • NDI New Data Indicator
  • the bitmap-based method When applied to 6G ultra-large-scale machine communication, the bitmap-based method will lead to a large signaling overhead due to its large number of users; the feedback method of ACK/NACK bits + UE ID, when applied to 6G ultra-large-scale machine communication However, due to the large number of users, it still has a large signaling overhead; the ACK/NACK feedback method that relies on the uplink resources allocated by the base station is not suitable for the 6G competition-based ultra-large-scale machine communication; the method based on NDI inversion, only It is suitable for scheduling-based data transmission, but not for 6G contention-based hyperscale machine communication.
  • the ACK/NACK feedback method for PDSCH in 5G NR in the related art is:
  • the HARQ-ACK codebook is divided into a semi-static codebook and a dynamic codebook:
  • A) Semi-static codebook The codebook is determined based on the semi-static configuration, which can avoid the situation where the base station and the UE have inconsistent understandings of the size of the codebook due to missed DCI detection, and ensure the reliability of feedback;
  • the dynamic codebook is fed back based on the number of PDSCHs actually scheduled, which can reduce the number of redundant bits fed back.
  • NR supports spatial combining, that is, combining the HARQ-ACK feedback bits of different transport blocks (TBs) of the same PDSCH through an "AND" operation, thereby obtaining less The number of feedback bits.
  • the number of PDSCHs per unit time becomes larger, and the length of the above-mentioned ACK/NACK codebook sequence becomes larger, that is to say, the overhead becomes larger.
  • dynamic codebook and spatial combining are used in NR. None of the methods can effectively reduce its signaling overhead.
  • Embodiments of the present disclosure provide a data transmission method, apparatus, and device.
  • the overhead of ACK/NACK feedback can be reduced.
  • a data transmission method applied to a terminal, the method comprising:
  • the terminal obtains the ACK/NACK sequence to be fed back
  • the terminal selects the machine learning model applied to the ACK/NACK sequence
  • the terminal compresses the ACK/NACK sequence through the machine learning model to obtain an indication sequence
  • the terminal sends the indication sequence to the network device.
  • the terminal selects a machine learning model applied to the ACK/NACK sequence, including:
  • the terminal selects a machine learning model applied to the ACK/NACK sequence according to the machine learning model indication information of the network device.
  • the terminal selects a machine learning model applied to the ACK/NACK sequence, including:
  • the terminal selects a machine learning model to be applied to the ACK/NACK sequence according to the ACK/NACK sequence.
  • the data transmission method further includes: the terminal sending the selected indication information of the machine learning model applied to the ACK/NACK sequence to the network device.
  • the terminal sends the indication sequence to the network device, including:
  • the terminal sends the indication sequence to the network device on the uplink control channel or the uplink shared channel.
  • the data transfer method also includes:
  • the terminal sends the indication sequence and the indication information of the machine learning model corresponding to the indication sequence to the network device together on the uplink shared channel or the uplink control channel.
  • the data transfer method also includes:
  • the terminal sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the network device respectively on the uplink shared channel and the uplink control channel, or respectively on the uplink control channel and the uplink shared channel.
  • the data transmission method further includes: the terminal receives the network equipment through radio resource control (radio resource control, RRC) signaling or media access control-control element (medium access control control element, MAC CE) signaling.
  • Radio resource control radio resource control
  • media access control-control element medium access control control element, MAC CE
  • the machine learning model indication information is sent by one or all of the network devices.
  • the machine learning model is trained and distributed by the following process:
  • the terminal obtains the ACK/NACK sequence to be fed back
  • the terminal stores the ACK/NACK sequence
  • the terminal groups the stored ACK/NACK sequences, performs machine learning model training, generates multiple sets of machine learning models, and reports them to the network device.
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • Embodiments of the present disclosure also provide a method for training and distributing a machine learning model, which is applied to the network device side, and the method includes:
  • the network device acquires the ACK/NACK sequence to be trained, and stores it;
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the training and distribution method of the machine learning model also includes:
  • the network device stores the machine learning model reported by the terminal
  • the network device distributes the stored machine learning models of all different terminals to the terminals using RRC signaling or MAC CE or physical layer signaling.
  • the grouping of the stored ACK/NACK sequences may be performed according to the sequence length and the ratio of ACKs.
  • Embodiments of the present disclosure also provide a data transmission method, which is applied to a network device, and the method includes:
  • the network device obtains the ACK/NACK sequence to be fed back
  • the network device selects the machine learning model applied to the ACK/NACK sequence, and sends the indication information of the selected machine learning model to the terminal;
  • the network device compresses the ACK/NACK sequence through the machine learning model to obtain an indication sequence
  • the network device sends the indication sequence to the terminal.
  • the data transfer method also includes:
  • the network device sends the machine learning model indication information to the terminal through radio resource control RRC signaling or media access control control unit MAC CE signaling or physical layer signaling.
  • the machine learning model is trained and distributed by the following process:
  • the network device obtains the ACK/NACK sequence to be fed back
  • the network device stores the ACK/NACK sequence
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored training machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include: repeating once every fixed period.
  • the stored ACK/NACK sequences are grouped according to the sequence length and the ratio of ACKs.
  • the data transfer method also includes:
  • the network device sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the terminal together in the downlink shared channel or downlink control channel; or,
  • the network device sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the terminal respectively on the downlink shared channel and the downlink control channel, or respectively on the downlink control channel and the downlink shared channel.
  • Embodiments of the present disclosure further provide a data transmission device, including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: acquiring The ACK/NACK sequence to be fed back; select a machine learning model applied to the ACK/NACK sequence; compress the ACK/NACK sequence through the machine learning model to obtain an indication sequence; send the indication sequence to the the network equipment.
  • a data transmission device including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: acquiring The ACK/NACK sequence to be fed back; select a machine learning model applied to the ACK/NACK sequence; compress the ACK/NACK sequence through the machine learning model to obtain an indication sequence; send the indication sequence to the the network equipment.
  • the processor obtains the ACK/NACK sequence to be fed back; the memory stores the ACK/NACK sequence; the processor groups the stored ACK/NACK sequence to perform a machine learning model training to generate sets of machine learning models, which the transceiver reports to the network device.
  • Embodiments of the present disclosure also provide a data transmission device, comprising:
  • the transceiver module is used to obtain the ACK/NACK sequence to be fed back;
  • a processing module for selecting a machine learning model applied to the ACK/NACK sequence; compressing the ACK/NACK sequence through the machine learning model to obtain an indication sequence;
  • the transceiver module is further configured to send the instruction sequence to the network device.
  • the processing module is further configured to obtain the ACK/NACK sequence to be trained
  • the device also includes:
  • the storage module is used to store the ACK/NACK sequence indicating whether the downlink data is received correctly;
  • the processing module is further configured to group the stored ACK/NACK sequences, perform machine learning model training, and generate multiple groups of machine learning models;
  • the transceiver module is further configured to report the machine learning model obtained by training to the network device.
  • Embodiments of the present disclosure further provide a data transmission device, including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: acquiring The ACK/NACK sequence to be fed back; select the machine learning model applied to the ACK/NACK sequence, and send the indication information of the selected machine learning model to the terminal; The ACK/NACK sequence is compressed to obtain an indication sequence; and the indication sequence is sent to the terminal.
  • a data transmission device including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: acquiring The ACK/NACK sequence to be fed back; select the machine learning model applied to the ACK/NACK sequence, and send the indication information of the selected machine learning model to the terminal; The ACK/NACK sequence is compressed to obtain an indication sequence; and the indication sequence is sent to the terminal.
  • the processor obtains the ACK/NACK sequence to be trained; the memory stores the ACK/NACK sequence; the processor groups the stored ACK/NACK sequence, and performs machine learning model training, Multiple groups of machine learning models are generated; the memory stores the machine learning models; the transceiver distributes the stored training machine learning models to terminals using RRC signaling or MAC CE or physical layer signaling.
  • Embodiments of the present disclosure also provide a data transmission device, comprising:
  • the transceiver module is used to obtain the ACK/NACK sequence to be fed back;
  • a processing module for selecting a machine learning model applied to the ACK/NACK sequence
  • the transceiver module is further configured to send the selected indication information of the machine learning model to the terminal;
  • the processing module is further configured to compress the ACK/NACK sequence through the machine learning model to obtain an indication sequence
  • the transceiver module is further configured to send the instruction sequence to the terminal.
  • the processing module is further configured to obtain the ACK/NACK sequence to be fed back;
  • the device also includes:
  • a storage module configured to store the ACK/NACK sequence
  • the processing module is also used to group the stored ACK/NACK sequences, perform machine learning model training, and generate multiple groups of machine learning models;
  • the storage module is further configured to store the machine learning model
  • the transceiver module is further configured to distribute the stored training machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • An embodiment of the present disclosure further provides a network device, including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: The trained ACK/NACK sequences are stored; the stored ACK/NACK sequences are grouped, machine learning model training is performed, and multiple sets of machine learning models are generated; the machine learning models are stored; The machine learning model is distributed to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • Embodiments of the present disclosure also provide an apparatus for training and distributing a machine learning model, which is applied to a network device side, and the apparatus includes:
  • the transceiver module is used to obtain and store the ACK/NACK sequence to be trained
  • a processing module configured to group the stored ACK/NACK sequences, perform machine learning model training, and generate multiple sets of machine learning models
  • a storage module for storing the machine learning model
  • the transceiver module is configured to distribute the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • Embodiments of the present disclosure also provide a data transmission system including the above-mentioned device.
  • Embodiments of the present disclosure also provide a processor-readable storage medium storing processor-executable instructions for causing the processor to execute the above-mentioned steps of the method.
  • the terminal acquires the ACK/NACK sequence to be fed back. Specifically, the terminal receives downlink data sent by at least one network device in the same and/or different time slots; the terminal receives and processes the downlink data, Determine whether the downlink data is received correctly, and use the 0/1 bit sequence to represent; the terminal applies the bit sequence machine learning model according to the machine learning model indication information of the network device or autonomously selects; the terminal uses the machine learning model , compressing the 0/1 bit sequence to obtain an indication sequence; the terminal sends the indication sequence to the network device.
  • the network device receives uplink data sent by at least one terminal; receives and processes the uplink data, determines whether the uplink data is correctly received, and uses a 0/1 bit sequence to represent it; autonomously selects the trained and the bit The machine learning model corresponding to the sequence, and sending the indication information of the selected machine learning model to the terminal; compressing the 0/1 bit sequence through the machine learning model to obtain an indication sequence; An indication sequence is sent to the terminal.
  • Using machine learning reduces the feedback overhead of sparse 0/1 bit sequences (here, sparsity means that there are only a few 0s or a few 1s in the 0/1 sequence).
  • the compression, transmission, and decompression methods for sparse ACK/NACK sequences in the present disclosure are also applicable to other sparse 0/1 bit sequences generated in the communication process.
  • FIG. 1 is a schematic flowchart of a data transmission method on a terminal side of the present disclosure
  • FIG. 2 is a schematic flowchart of the data transmission method on the network side of the present disclosure
  • FIG. 3 is a schematic diagram of a PUSCH-based ACK/NACK feedback method
  • FIG. 4 is a schematic diagram of an ACK/NACK feedback method based on PDSCH
  • FIG. 5 is a schematic diagram of the architecture of the data transmission device of the present disclosure.
  • FIG. 6 is a schematic block diagram of the data transmission apparatus of the present disclosure.
  • an embodiment of the present disclosure provides a data transmission method, which is applied to a terminal, and the method includes:
  • Step 11 the terminal obtains the ACK/NACK sequence to be fed back; in an optional implementation manner, the terminal receives downlink data sent by at least one network device in the same and/or different time slots; the terminal receives and transmits the downlink data. Processing, judging whether the downlink data is correctly received, and using the 0/1 bit sequence to represent;
  • Step 12 the terminal selects the machine learning model applied to the ACK/NACK sequence; here, the terminal may select the machine learning model applied to the ACK/NACK sequence according to the machine learning model indication information of the network device; or The machine learning model applied to the ACK/NACK sequence can be independently selected;
  • Step 13 the terminal compresses the ACK/NACK sequence through the machine learning model to obtain an indication sequence
  • Step 14 The terminal sends the instruction sequence to the network device.
  • each bit in the ACK/NACK sequence corresponds to a data block sent by the base station, which may be different time slots, different channels, data blocks divided in a predetermined manner on the same channel, etc., and combinations thereof;
  • the terminal receives downlink data sent by at least one network device in the same and/or different time slots; receives and processes the downlink data, determines whether the downlink data is correctly received, and uses a 0/1 bit sequence to represent it;
  • the terminal Apply to the bit sequence machine learning model according to the machine learning model indication information of the network device or autonomously select; compress the 0/1 bit sequence through the machine learning model to obtain an indication sequence; the terminal compresses the bit sequence
  • the instruction sequence is sent to the network device, and the above-mentioned sparse bit sequence is compressed, transmitted and decompressed based on a deep learning technology (such as an autoencoder technology), which reduces the feedback overhead of the sparse bit sequence.
  • a deep learning technology such as an autoencoder technology
  • the terminal selects a machine learning model applied to the ACK/NACK sequence, including:
  • the terminal selects a machine learning model applied to the ACK/NACK sequence according to the machine learning model indication information of the network device.
  • the terminal selects a machine learning model applied to the ACK/NACK sequence, including:
  • the terminal selects a machine learning model to be applied to the ACK/NACK sequence according to the ACK/NACK sequence.
  • the above data transmission method may further include:
  • the terminal sends the selected indication information of the machine learning model applied to the ACK/NACK sequence to the network device.
  • the above data transmission method may further include:
  • the terminal receives the machine learning model indication information sent by the network device through radio resource control RRC signaling or medium access control control unit MAC CE signaling or physical layer signaling.
  • the machine learning model here indicates that the information is sent by one or all of the network devices. All network devices here correspond to multiple receiving points in a coordinated multiple point (Coordinated Multiple Points, CoMP) transmission scenario.
  • the machine learning model is trained and distributed by the following process:
  • the network device obtains the ACK/NACK sequence to be trained, and optionally, the network device receives the indication sequence sent by the terminal; the network device decompresses the indication sequence to generate a corresponding ACK/NACK sequence , and store it;
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include: executing on one or all of the network devices; and repeating it every other fixed period.
  • the network device stores the received 0/1 bit sequence; performs level division on the sparsity of the above 0/1 bit sequence with sparsity; for example, the ACK/NACK sequence [0,0,0,0, 0,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,0,0], where 1 represents ACK, 0 represents NACK; in addition, 1 can also be used to represent NACK, 0 Indicates ACK, which is to reflect the sparsity of the ACK/NACK sequence.
  • the sparsity of the bit sequence can be the ratio of the number of 1s in the bit sequence to the total number of bits in the bit sequence.
  • the Nth machine learning model is trained to obtain the Nth machine learning model corresponding to the bit sequence of the Nth level;
  • the N+1th machine learning model is trained to obtain the N+1th machine learning model corresponding to the bit sequence of the N+1th level.
  • the 0/1 bit sequences with different sparse levels generated recently according to the demodulation results are stored; the sparse degrees of the above 0/1 bit sequences with sparseness are graded; the training unit selects a sparse level The 0/1 bit sequence of , train the autoencoder to be used; based on the above trained autoencoder model of a certain sparse level, perform transfer learning to obtain the parameters of the autoencoder at other sparse levels, that is, in training For other sparse level autoencoder parameters, the above-trained machine learning model of a certain sparse level is used as the starting point.
  • bit sequence of the Nth level in the bit sequences of multiple different sparse levels to train the Nth machine learning model to obtain the Nth machine learning model corresponding to the bit sequence of the Nth level;
  • the N+1 th machine learning model is trained by using the bit sequence of the N+1 th level in the bit sequences of the plurality of different sparse levels, and the N+1 th machine learning model corresponding to the bit sequence of the N+1 th level is obtained.
  • the recently received decompressed 0/1 bit sequence is stored; the sparseness of the above stored 0/1 bit sequence is graded; the training unit uses the above 0/1 bit sequence of different sparse levels , train the machine learning device to be used, and obtain machine learning models with different parameters corresponding to different sparsity levels.
  • the machine learning model can also be trained and distributed by the following process:
  • the terminal acquires the ACK/NACK sequence to be fed back.
  • the terminal receives downlink data sent by at least one network device in the same and/or different time slots; the terminal receives and processes the downlink data, and acquires the downlink data to be fed back.
  • Feedback ACK/NACK sequence further, judge whether downlink data is received correctly, and use 0/1 bit sequence to represent;
  • the terminal stores the ACK/NACK sequence
  • the terminal groups the stored ACK/NACK sequences, performs machine learning model training, generates multiple sets of machine learning models, and reports them to the network device;
  • the network device stores the reported machine learning model
  • the network device distributes the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the stored ACK/NACK sequences may be grouped, and may be grouped according to the sequence length and the ratio of ACKs.
  • the ACK/NACK sequence is compressed by using a trained machine learning model corresponding to the sparse level to obtain an indication sequence, including:
  • the encoder in the autoencoder of the trained machine learning model is batch normalized through the convolutional neural network, and then passes through the activation function and the fully connected layer to limit the coefficients and output of the autoencoder to binary bits. Get the instruction sequence.
  • the above-mentioned ACK/NACK sequence is compressed based on the encoder in the selected autoencoder, and the coefficients and output of the autoencoder are also limited to binary 0, 1 bits to obtain a compressed Mx1 binary vector [0,1,1,0,0,1,0];
  • the specific operation of the encoder in the autoencoder is described as: passing the above sparse 0/1 bit sequence through a convolutional neural network, then batch normalization, and then through a leakyReLU activation function, and then through a full Connect the layers, and then limit the output values to 0 or 1, resulting in the compressed binary sequence described above.
  • the terminal sending the indication sequence to the network device includes:
  • the terminal sends the indication sequence and the indication information of the machine learning model corresponding to the indication sequence to the network device together on the uplink shared channel or the uplink control channel; or,
  • the terminal sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the network device respectively on the uplink shared channel and the uplink control channel, or respectively on the uplink control channel and the uplink shared channel; or,
  • the terminal only sends the indication sequence to the network device on the uplink control channel or the uplink shared channel.
  • the ACK/NACK sequence is compressed, transmitted and decompressed on the 6G PDSCH based on the deep learning technology (such as the self-encoder technology).
  • the method on the network side corresponds to the method on the terminal side, and the above method on the terminal side is also applicable to the embodiment on the network side, and can achieve the same technical effect.
  • Embodiments of the present disclosure also provide a method for training and distributing a machine learning model, which is applied to the network device side, and the method includes:
  • the network device acquires the ACK/NACK sequence to be trained, and stores it;
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the method for training and distributing a machine learning model further includes:
  • the network device stores the machine learning model reported by the terminal
  • the network device distributes the stored machine learning models of all different terminals to the terminals using RRC signaling or MAC CE or physical layer signaling.
  • the terminal side reports the model trained by itself to the network device, and after the network device aggregates the models of all other terminals, the aggregated result is sent to the terminal.
  • the grouping of the stored ACK/NACK sequences may be performed according to the sequence length and the ratio of ACKs.
  • the network device can determine the machine learning model, and the feedback overhead of the sparse bit sequence is reduced.
  • an embodiment of the present disclosure further provides a data transmission method, which is applied to a network device, and the method includes:
  • Step 21 The network device obtains the ACK/NACK sequence to be fed back.
  • the network device receives uplink data sent by at least one terminal; receives and processes the uplink data to determine whether the uplink data is correctly received data, represented by a 0/1 bit sequence;
  • Step 22 the network device selects the machine learning model applied to the corresponding ACK/NACK sequence, and sends the indication information of the selected machine learning model to the terminal;
  • Step 23 the network device compresses the ACK/NACK sequence through the machine learning model to obtain an indication sequence
  • Step 24 The network device sends the instruction sequence to the terminal.
  • each bit in the ACK/NACK sequence corresponds to a different terminal, the terminal transmits on a predetermined channel, and the base station feeds back in a centralized manner; in addition to indicating that the data has not been successfully demodulated, the NACK is used to indicate that the data is not successfully demodulated. Also used to indicate that the terminal is in the Inactive (inactive) state.
  • the network device receives the uplink data sent by at least one terminal; the network device receives and processes the uplink data, determines whether the uplink data is correctly received, and uses a 0/1 bit sequence to represent it, such as ACK/ The 0/1 bit sequence corresponding to NACK-Inactive; the network device selects the trained machine learning model corresponding to the bit sequence according to the load level (or sparse level) corresponding to the bit sequence; the network device learns through the machine learning model model, compress the 0/1 bit sequence to obtain an indication sequence; the network device sends the indication sequence to the terminal, and based on deep learning technology (such as auto-encoder technology), the above sparse bit sequence Compression, transmission and decompression are performed to reduce the feedback overhead of sparse bit sequences.
  • deep learning technology such as auto-encoder technology
  • the data transmission method may further include:
  • the network device sends the machine learning model indication information to the terminal through radio resource control RRC signaling or media access control control unit MAC CE signaling or physical layer signaling.
  • the machine learning model is trained and distributed by the following process:
  • the network device obtains the ACK/NACK sequence to be fed back, and optionally, the network device receives the uplink data sent by the terminal; the network device receives and processes the uplink data, and obtains the ACK/NACK sequence to be fed back ;
  • the network device stores the ACK/NACK sequence
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored training machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include: repeating once every fixed period.
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the stored ACK/NACK sequences are grouped according to the sequence length and the ratio of ACKs.
  • sending the indication sequence to the terminal includes:
  • the network device sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the terminal together in the downlink shared channel or downlink control channel; or, the network device sends the instruction sequence and the instruction information corresponding to the instruction sequence to the terminal.
  • the indication information of the machine learning model is sent to the terminal through the downlink shared channel and the downlink control channel, or the downlink control channel and the downlink shared channel respectively; or,
  • the network device only sends the indication sequence to the terminal in the downlink control channel or downlink shared channel.
  • FIG. 3 it is a schematic diagram of the transmission of information bits of bit-map-based ACK/NACK-Inactive for the PUSCH of the 6G ultra-large-scale machine communication.
  • the base station classifies its network load
  • the base station stores the uncompressed ACK/NACK-Inactive sequences under different load levels recently sent to the UE. Specifically, 1 can be used to represent ACK, 0 to represent NACK or the terminal is in the inactive state;
  • the training unit of the base station trains the machine learning device to be used, and obtains machine learning models with different parameters corresponding to different load levels respectively.
  • the training unit transmits the models and parameter sets of the compressor and decompressor (or encoder, decoder) of the trained machine learning device with different load levels to the storage unit of the base station;
  • the base station distributes the stored model and parameter set of the decompressor or decoder of the machine learning device with different load levels and uses RRC signaling or MAC CE or physical layer signaling to the UE;
  • the active UE sends uplink data based on competition, that is, sends PUSCH;
  • the base station demodulates after receiving the PUSCH from the active user
  • the base station forms an ACK/NACK-Inactive sequence according to the demodulation result, for example, each ACK/NACK-Inactive bit corresponds to the UE ID one-to-one;
  • the base station selects a model and parameter set of the compressor and decompressor in a trained machine learning device
  • the base station compresses the above-mentioned ACK/NACK-Inactive sequence based on the selected trained machine learning device;
  • the base station sends the model of the decompressor it selected, the subscript of the parameter set and the ACK/NACK-Inactive sequence to the UE. Specifically, when sending:
  • the base station After the base station sends the model of the decompressor it selected and the subscript of the parameter set to the UE through physical layer signaling, the base station sends the ACK/NACK-Inactive sequence to the UE in broadcast form;
  • the base station sends the model of the decompressor it selected and the subscript of the parameter set together with the ACK/NACK-Inactive sequence to the UE in broadcast form;
  • the active UE performs decompression based on the signaling received by the base station and/or the received compressed ACK/NACK-Inactive sequence to obtain ACK/NACK-Inactive bits corresponding to itself.
  • the training process of the above machine learning and the process of storing and transferring model parameters are not always bound to the use process.
  • the implementation process includes the following two methods:
  • the above model parameters can be applied repeatedly until the communication environment has changed greatly and the model parameters are no longer applicable, instead of training the model before each use.
  • the base station extracts the above data in stages, conducts training offline independently, then compares the trained model with the model currently in use, and selects a more suitable model for online use (when using The newly trained model also includes the transfer process of the model).
  • FIG. 4 it is a schematic diagram of ACK/NACK codebook feedback for PDSCH.
  • the disclosure of the transmission method of ACK/NACK feedback for PDSCH specifically includes:
  • the base station stores the ACK/NACK sequences recovered by decompression sent by the UE in different channel states in the near term;
  • the training unit transmits the trained models and parameter sets of the decompressor or decoder of the machine learning device in different groups of channel states to the storage unit of the base station;
  • the base station configures the stored model and parameter set of the machine learning device compressor or encoder to the UE using RRC signaling or MAC CE or physical layer signaling.
  • the UE receives PDSCHs from different serving cells and/or different time slots, and demodulates them;
  • the UE forms multiple groups of ACK/NACK sequences corresponding to PDSCHs in different time slots from the above-mentioned different serving cells;
  • the UE autonomously selects the model and parameter set of the compressor and decompressor in the trained machine learning device according to the channel states corresponding to different groups of ACK/NACK sequences, or according to the compressor and decompressor in the machine learning device indicated by the primary serving cell
  • the model and parameter set of the device the UE compresses the above ACK/NACK sequence
  • the UE sends the compressed ACK/NACK sequence to the base station side; specifically:
  • the UE sends the ACK/NACK sequence together with the model of the decoder in the selected trained machine learning device and the subscript of the parameter set to the base station side, such as using the uplink shared channel or the uplink control channel, or
  • the UE uses different channels for the ACK/NACK sequence and the model of the decoder in the selected trained machine learning device and the subscript of the parameter set (for example, the ACK/NACK sequence uses the uplink shared channel, and the subscript uses the uplink shared channel). control channel) are sent to the base station side respectively, or
  • the base station side uses the primary serving cell or the machine learning device notified by the UE to decompress the received compressed ACK/NACK sequence, and then obtain the ACK/NACK sequence corresponding to the sent PDSCH.
  • training, storage and transfer parameters can also be the following steps, and the use method is the same as the previous method, and will not be repeated;
  • the above different channel states are grouped, and the training unit uses the UE to generate ACK/NACK sequences in the same group of channel states, and trains the machine learning device to be used until the channel states of all groups are traversed, and different parameters corresponding to different channel states are obtained.
  • machine learning model ;
  • the training unit reports the parameters and/or structures of the compressors or encoders of the trained machine learning devices in different groups of channel states to the base station using RRC signaling or MAC CE or physical layer signaling. It is passed to the storage unit of the base station;
  • the base station configures the stored model and parameter set of the machine learning device compressor or encoder and sends it to the UE using RRC signaling or MAC CE or physical layer signaling.
  • the training process of the above machine learning and the process of storing and transferring model parameters are not always bound to the use process.
  • the implementation process includes the following two methods:
  • model parameters can be applied repeatedly until the communication environment has changed greatly and the model parameters are no longer applicable, instead of training the model before each use.
  • the base station extracts the above data in stages, conducts training offline independently, then compares the trained model with the model currently in use, and selects a more suitable model for online use (when using The newly trained model also includes the transfer process of the model).
  • Embodiment 1 Based on the self-encoder technology in machine learning, the ACK/NACK-Inactive sequence of PUSCH in 6G ultra-large-scale machine communication is compressed, wherein the self-encoder technology uses an encoder to compress data at the transmitting end first, Then, the decoder is used for decompression at the receiving end, in which the parameters of the encoder and the decoder are obtained based on data training by means of machine learning.
  • Step 1 The base station classifies its network load
  • Step 2 The base station stores the recent ACK/NACK-Inactive sequences delivered to the UE at different load levels;
  • Step 3 The training unit selects an ACK/NACK-Inactive sequence of a network load level to train the autoencoder to be used;
  • Step 4 The base station migrates the autoencoder parameters trained based on the above load levels under other network load levels to obtain parameters of the machine learning device under different network load levels;
  • Step 1 The training unit transmits the models and parameter sets of the encoder and decoder in the trained autoencoders with different load levels to the storage unit of the base station;
  • Step 2 The base station configures the stored trained decoder models and parameter sets of autoencoders with different load levels to the UE using RRC signaling.
  • Step 3 Repeat steps 2-4 of training every fixed period, and steps 1-2 of storing and transferring models and parameters, and update the training results.
  • Step 1 In 6G ultra-large-scale machine communication, each UE wakes up at regular intervals to monitor the environment.
  • Step 3 The base station receives the data sent by all K UEs and demodulates them;
  • Step 4 According to the demodulation result, the base station forms an Nx1 ACK/NACK-Inactive sequence [0,0,0,0,0,0,0,0,0,1,0, 0,0,0,0,0,0,1,0,0,0,0,0], where 1 represents ACK, 0 represents NACK or the UE is in an inactive state; in addition, 1 can also be used to represent NACK or the UE is in an inactive state, 0 Indicates ACK.
  • the original ACK/NACK-Inactive sequence can be flipped 0/1 first;
  • Step 5 The base station selects a trained autoencoder parameter and/or structure based on the load level of the current network
  • Step 6 The base station compresses the above-mentioned ACK/NACK-Inactive sequence based on the encoder in the selected autoencoder, and also limits the coefficient and output of the autoencoder to binary 0,1 bits to obtain a compressed binary vector of Mx1 [0,1,1,0,0,1,0];
  • Step 7 The base station broadcasts the selected autoencoder model and parameter subscripts together with the compressed ACK/NACK-Inactive sequence to all UEs.
  • Step 8 The UE decompresses the received compressed Mx1 ACK/NACK-Inactive sequence based on the decoder in the trained self-encoder notified by the base station, and finally outputs the decompressed sequence [0,0,0,0 ,0,0,0,0,0,1,0,0,0,0,0,0,0,1,0,0,0,0] to obtain the ACK/NACK-Inactive bit corresponding to itself.
  • Embodiment 2 Compress the ACK/NACK sequence in PDSCH based on the autoencoder technology in machine learning
  • Step 1 The base station stores the ACK/NACK sequences sent by the UE in different channel states in the near term;
  • Step 2 Group the above-mentioned different channel states, and the training unit uses the ACK/NACK sequences sent by different UEs in the same group of channel states to train the autoencoder to be used, until all the channel state groups are traversed, and different channel states are obtained. Corresponding machine learning models with different parameters.
  • Step 1 The training unit transmits the trained models and parameter sets of the encoder and decoder in the autoencoder under different channel states to the storage unit of the base station;
  • Step 2 The base station configures the stored trained decoder models and parameter sets of the auto-encoder under different channel states to the UE using RRC signaling.
  • Step 3 Repeat steps 1-2 of training, and steps 1-2 of storing and transferring models and parameters at regular intervals, and update the training results.
  • Step 1 Different serving cells in NR send PDSCH to UE;
  • Step 2 UE receives PDSCH from different serving cells on different time slots and demodulates it;
  • Step 3 According to the demodulation result, the UE forms multiple sets of ACK/NACK sequences corresponding to the PDSCH in different time slots of different serving cells, such as [1, 1, 0, 0, 1, 1]; [1, 1, 1, 1, 0, 0]; [1, 1, 1, 1] (for convenience, the length of the sequence is set to 16), here it is assumed that 1 represents ACK and 0 represents NACK, and it is assumed that there are two
  • the bits correspond to the ACK/NACK of a PDSCH. There are three serving cells in total. The first group of bits corresponds to the first serving cell, the second group of bits corresponds to the second serving cell, and the third group of bits corresponds to the third serving cell.
  • Step 4 When the channel conditions are good, the 01 bit in the generated ACK/NACK codebook can be flipped, that is, 0 becomes 1, 1 becomes 0, and the above three sets of sequences are concatenated to obtain the sequence [0, 0,1,1,0,0,0,0,0,0,1,1,0,0,0,0]; when the channel conditions are poor, the above flipping is not required.
  • Step 5 The Primary serving cell notifies the UE of the autoencoder model and parameter subscripts to be compressed with other serving cells;
  • Step 6 The UE compresses the above-mentioned ACK/NACK sequence based on the encoder of the self-encoder notified by the primary serving cell, and also limits the coefficient and output of the self-encoder to binary 0,1 bits to obtain a compressed binary vector [ 1,0,1,1,0,0,0,1];
  • Step 7 The UE feeds back the ACK/NACK sequence to the base station side;
  • Step 8 The base station uses the autoencoder model and parameters notified by the primary serving cell to decompress the received compressed ACK/NACK sequence, and then obtain the bit-flipped ACK/NACK bit sequence [0,0,1, 1,0,0,0,0,0,0,0,1,1,0,0,0,0], after bit flip it becomes [1,1,0,0,1,1,1,1,1 ,1,0,0,1,1,1,1].
  • the above-mentioned embodiments of the present disclosure use machine learning technology to reduce the feedback overhead of sparse 0/1 bit sequences.
  • Specific implementation examples include but are not limited to: 1) Reduced 6G ultra-large-scale machine communication with low complexity The overhead of ACK/NACK feedback of PUSCH; 2) the overhead of ACK/NACK sequence of PDSCH of 6G ultra-high data rate is reduced with low complexity.
  • the method on the terminal side corresponds to the method on the network side, and the above method on the network side is also applicable to the embodiment on the terminal side, and the same technical effect can be achieved.
  • an embodiment of the present disclosure further provides a terminal 50, including: a transceiver 51, a processor 52, and a memory 53, where the memory 53 stores a program executable by the processor 52; the When the processor 52 executes the program, it realizes: acquiring the ACK/NACK sequence to be fed back; selecting a machine learning model applied to the ACK/NACK sequence; compressing the ACK/NACK sequence through the machine learning model to obtain an indication sequence; sending the indication sequence to the network device.
  • selecting the machine learning model applied to the ACK/NACK sequence by the terminal includes: selecting, by the terminal, the machine learning model applied to the ACK/NACK sequence according to the machine learning model indication information of the network device.
  • selecting the machine learning model applied to the ACK/NACK sequence by the terminal includes: selecting the machine learning model applied to the ACK/NACK sequence by the terminal according to the ACK/NACK sequence.
  • the terminal sends the selected indication information of the machine learning model applied to the ACK/NACK sequence to the network device.
  • the terminal sends the indication sequence to the network device, including:
  • the terminal sends the indication sequence to the network device on the uplink control channel or the uplink shared channel.
  • the data transfer method also includes:
  • the terminal sends the indication sequence and the indication information of the machine learning model corresponding to the indication sequence to the network device together on the uplink shared channel or the uplink control channel.
  • the data transfer method also includes:
  • the terminal sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the network device respectively on the uplink shared channel and the uplink control channel, or respectively on the uplink control channel and the uplink shared channel.
  • the machine learning model indication information is sent by one or all of the network devices.
  • the machine learning model training process and distribution process further include: executing on one or all of the network devices; and repeating the process every other fixed period.
  • the machine learning model is trained and distributed by the following process:
  • the terminal obtains the ACK/NACK sequence to be fed back
  • the terminal stores the ACK/NACK sequence
  • the terminal groups the stored ACK/NACK sequences, performs machine learning model training, generates multiple sets of machine learning models, and reports them to the network device.
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the device in this embodiment is a device corresponding to the method shown in FIG. 1 above, and the implementation manners in the above embodiments are all applicable to the embodiments of the device, and the same technical effect can also be achieved.
  • the transceiver 51 and the memory 53, as well as the transceiver 51 and the processor 52 can be communicated and connected through a bus interface, the function of the processor 52 can also be realized by the transceiver 51, and the function of the transceiver 51 can also be realized by the processor 52 realized.
  • an embodiment of the present disclosure further provides a data transmission apparatus 60, including:
  • the transceiver module 61 is used to obtain the ACK/NACK sequence to be fed back;
  • a processing module 62 configured to select a machine learning model applied to the ACK/NACK sequence; compress the ACK/NACK sequence through the machine learning model to obtain an indication sequence;
  • the transceiver module 61 is further configured to send the instruction sequence to the network device.
  • selecting the machine learning model applied to the ACK/NACK sequence by the terminal includes: selecting, by the terminal, the machine learning model applied to the ACK/NACK sequence according to machine learning model indication information of the network device.
  • selecting the machine learning model applied to the ACK/NACK sequence by the terminal includes: selecting the machine learning model applied to the ACK/NACK sequence by the terminal according to the ACK/NACK sequence.
  • the terminal sends the selected indication information of the machine learning model applied to the ACK/NACK sequence to the network device.
  • the terminal sends the indication sequence to the network device, including:
  • the terminal sends the indication sequence to the network device on the uplink control channel or the uplink shared channel.
  • the data transfer method also includes:
  • the terminal sends the indication sequence and the indication information of the machine learning model corresponding to the indication sequence to the network device together on the uplink shared channel or the uplink control channel.
  • the data transfer method also includes:
  • the terminal sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the network device respectively on the uplink shared channel and the uplink control channel, or respectively on the uplink control channel and the uplink shared channel.
  • the machine learning model indication information is sent by one or all of the network devices.
  • the machine learning model training process and distribution process further include: executing on one or all of the network devices; and repeating the process every other fixed period.
  • the machine learning model is trained and distributed by the following process:
  • the terminal obtains the ACK/NACK sequence to be fed back
  • the terminal stores the ACK/NACK sequence
  • the terminal groups the stored ACK/NACK sequences, performs machine learning model training, generates multiple sets of machine learning models, and reports them to the network device.
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the device in this embodiment is a device corresponding to the method shown in FIG. 1 above, and the implementation manners in the above embodiments are all applicable to the embodiments of the device, and the same technical effect can also be achieved. It should be noted here that the above-mentioned device provided by the embodiment of the present disclosure can realize all the method steps realized by the above-mentioned method embodiment, and can achieve the same technical effect, and the same as the method embodiment in this embodiment is not repeated here. The parts and beneficial effects will be described in detail.
  • An embodiment of the present disclosure further provides a network device, including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: The trained ACK/NACK sequences are stored; the stored ACK/NACK sequences are grouped, machine learning model training is performed, and multiple sets of machine learning models are generated; the machine learning models are stored; The machine learning model is distributed to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the storage module stores the machine learning model reported by the terminal
  • the transceiver module is used to distribute the stored machine learning models of all different terminals to the terminals using RRC signaling or MAC CE or physical layer signaling.
  • the grouping of the stored ACK/NACK sequences may be performed according to the sequence length and the ratio of ACKs.
  • Embodiments of the present disclosure also provide an apparatus for training and distributing a machine learning model, which is applied to a network device side, and the apparatus includes:
  • the transceiver module is used to obtain and store the ACK/NACK sequence to be trained
  • a processing module configured to group the stored ACK/NACK sequences, perform machine learning model training, and generate multiple sets of machine learning models
  • a storage module for storing the machine learning model
  • the transceiver module is configured to distribute the stored machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the storage module stores the machine learning model reported by the terminal
  • the transceiver module is used to distribute the stored machine learning models of all different terminals to the terminals using RRC signaling or MAC CE or physical layer signaling.
  • the grouping of the stored ACK/NACK sequences may be performed according to the sequence length and the ratio of ACKs.
  • Embodiments of the present disclosure further provide a network device, including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: Feedback ACK/NACK sequence; select the machine learning model applied to the ACK/NACK sequence, and send the indication information of the selected machine learning model to the terminal; through the machine learning model, the ACK The /NACK sequence is compressed to obtain an indication sequence; the indication sequence is sent to the terminal.
  • a network device including: a transceiver, a processor, and a memory, where a program executable by the processor is stored in the memory; when the processor executes the program, the processor implements: Feedback ACK/NACK sequence; select the machine learning model applied to the ACK/NACK sequence, and send the indication information of the selected machine learning model to the terminal; through the machine learning model, the ACK The /NACK sequence is compressed to obtain an indication sequence; the indication sequence is sent to the terminal.
  • the network device sends the machine learning model indication information to the terminal through radio resource control RRC signaling or media access control control unit MAC CE signaling or physical layer signaling.
  • one or all network devices send machine learning model indication information to the terminal.
  • the machine learning model is trained and distributed by the following process:
  • the network device acquires the ACK/NACK sequence to be fed back, and stores it;
  • the network device groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple groups of machine learning models;
  • the network device stores the machine learning model
  • the network device distributes the stored training machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the stored ACK/NACK sequences are grouped according to the sequence length and the proportion of ACKs.
  • sending the indication sequence to the terminal includes:
  • the network device sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the terminal together in the downlink shared channel or downlink control channel; or,
  • the network device sends the instruction sequence and the instruction information of the machine learning model corresponding to the instruction sequence to the terminal respectively on the downlink shared channel and the downlink control channel, or respectively on the downlink control channel and the downlink shared channel; or,
  • the network device only sends the indication sequence to the terminal in the downlink control channel or downlink shared channel.
  • the device in this embodiment is a device corresponding to the method shown in FIG. 2 above, and the implementation manners in the above embodiments are all applicable to the embodiments of the device, and the same technical effect can also be achieved.
  • the transceiver and the memory, as well as the transceiver and the processor can be communicated and connected through a bus interface, the function of the processor can also be realized by the transceiver, and the function of the transceiver can also be realized by the processor.
  • the above-mentioned device provided by the embodiment of the present disclosure can realize all the method steps realized by the above-mentioned method embodiment, and can achieve the same technical effect, and the same as the method embodiment in this embodiment is not repeated here. The parts and beneficial effects will be described in detail.
  • Embodiments of the present disclosure also provide a data transmission device, comprising:
  • the transceiver module is used to obtain the ACK/NACK sequence to be fed back;
  • a processing module configured to select a machine learning model applied to the ACK/NACK sequence, and send indication information of the selected machine learning model to the terminal;
  • the sequence is compressed to obtain the indicated sequence;
  • the transceiver module is further configured to send the indication sequence to the terminal.
  • the transceiver module sends machine learning model indication information to the terminal through radio resource control RRC signaling or media access control control unit MAC CE signaling or physical layer signaling.
  • one or all network devices send machine learning model indication information to the terminal.
  • the machine learning model is trained and distributed by the following process:
  • the processing module obtains the ACK/NACK sequence to be fed back, and stores it;
  • the processing module groups the stored ACK/NACK sequences, performs machine learning model training, and generates multiple sets of machine learning models;
  • the processing module stores the machine learning model
  • the processing module distributes the stored training machine learning model to the terminal using RRC signaling or MAC CE or physical layer signaling.
  • the machine learning model training process and distribution process further include:
  • the terminal reports the trained model to the network device through the following process:
  • the terminal reports the trained machine learning model to the network device through RRC signaling or MAC CE or physical layer signaling.
  • the stored ACK/NACK sequences are grouped according to the sequence length and the ratio of ACKs.
  • sending the indication sequence to the terminal includes:
  • the sequence is sent to the terminal in the downlink control channel or the downlink shared channel.
  • the device in this embodiment is a device corresponding to the method shown in FIG. 2 above, and the implementation manners in each of the above embodiments are applicable to the embodiments of the device, and the same technical effect can also be achieved. It should be noted here that the above-mentioned device provided by the embodiment of the present disclosure can realize all the method steps realized by the above-mentioned method embodiment, and can achieve the same technical effect, and the same as the method embodiment in this embodiment is not repeated here. The parts and beneficial effects will be described in detail.
  • Embodiments of the present disclosure further provide a data transmission system, including: the device on the network side and the device on the terminal side as described in the foregoing embodiments.
  • Embodiments of the present disclosure also provide a processor-readable storage medium storing processor-executable instructions for causing the processor to execute the above-mentioned Methods. All implementation manners in the foregoing method embodiment are applicable to this embodiment, and the same technical effect can also be achieved.
  • the disclosed apparatus and method may be implemented in other manners.
  • the apparatus embodiments described above are only illustrative.
  • the division of the units is only a logical function division. In actual implementation, there may be other division methods.
  • multiple units or components may be combined or Can be integrated into another system, or some features can be ignored, or not implemented.
  • the shown or discussed mutual coupling or direct coupling or communication connection may be through some interfaces, indirect coupling or communication connection of devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution in this embodiment.
  • each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the functions, if implemented in the form of software functional units and sold or used as independent products, may be stored in a computer-readable storage medium.
  • the computer software product is stored in a storage medium, including several
  • the instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program codes .
  • each component or each step can be decomposed and/or recombined. These disaggregations and/or recombinations should be considered equivalents of the present disclosure.
  • the steps of executing the above-described series of processes can naturally be executed in chronological order in the order described, but need not necessarily be executed in chronological order, and some steps may be executed in parallel or independently of each other.
  • Those of ordinary skill in the art can understand all or any steps or components of the method and device of the present disclosure. , software, or a combination thereof, which can be implemented by those of ordinary skill in the art using their basic programming skills after reading the description of the present disclosure.
  • modules, units, and sub-units can be implemented in one or more Application Specific Integrated Circuits (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSP Device, DSPD) ), Programmable Logic Device (PLD), Field-Programmable Gate Array (FPGA), general-purpose processor, controller, microcontroller, microprocessor, in other electronic units or combinations thereof.
  • ASIC Application Specific Integrated Circuits
  • DSP Digital Signal Processor
  • DSP Device Digital Signal Processing Device
  • DSPD Digital Signal Processing Device
  • PLD Programmable Logic Device
  • FPGA Field-Programmable Gate Array
  • the technologies described in the embodiments of the present disclosure may be implemented through modules (eg, procedures, functions, etc.) that perform the functions described in the embodiments of the present disclosure.
  • Software codes may be stored in memory and executed by a processor.
  • the memory can be implemented in the processor or external to the processor.
  • the objects of the present disclosure can also be achieved by running a program or set of programs on any computing device.
  • the computing device may be a known general purpose device. Therefore, the objects of the present disclosure can also be achieved merely by providing a program product containing program code for implementing the method or apparatus. That is, such a program product also constitutes the present disclosure, and a storage medium in which such a program product is stored also constitutes the present disclosure.
  • the storage medium can be any known storage medium or any storage medium developed in the future.
  • each component or each step can be decomposed and/or recombined. These disaggregations and/or recombinations should be considered equivalents of the present disclosure.
  • the steps of executing the above-described series of processes can naturally be executed in chronological order in the order described, but need not necessarily be executed in chronological order. Certain steps may be performed in parallel or independently of each other.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Signal Processing (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Detection And Prevention Of Errors In Transmission (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开实施例提供了一种数据传输方法、装置及设备，终端侧的方法包括：终端获取待反馈的ACK/NACK序列；终端选择应用于所述ACK/NACK序列的机器学习模型；终端通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；终端将所述指示序列发送给所述网络设备。

Description

数据传输方法、装置及设备
相关申请的交叉引用
本申请主张在2020年11月12日在中国提交的中国专利申请号No.202011261728.1的优先权，其全部内容通过引用包含于此。
技术领域
本公开涉及通信技术领域，尤其涉及一种数据传输方法、装置及设备。
背景技术
在第六代(6 th Generation，6G)超大规模机器通信中，网络可支持的机器密度达到每平方米10个机器或者每立方米100个机器，上述机器通常仅零星地发送一些较小的数据包，采用相关技术中的基于调度的通信方式，将会导致很大的接入信令开销，而基于竞争的免调度通信可以减小接入信令开销。但是，在如此超大规模的基于竞争的机器通信中，如果对物理上行共享信道(physical uplink shared channel，PUSCH)采用bitmap(位图)的肯定确认/否定确认(Acknowledgement/Negative Acknowledgement，ACK/NACK)机制或者ACK/NACK加用户设备(user equipment，UE)标识(identifier，ID)的方式，将会导致很大的下行反馈开销。此外，如果对6G物理下行共享信道(physical downlink shared channel，PDSCH)继续采用第五代(5 th Generation，5G)的ACK/NACK反馈机制，将导致很大的上行反馈开销。
相关技术中的针对PUSCH的ACK/NACK反馈方法包括：
基站将每个UE ID与一个ACK/NACK-Inactive比特对应，即bitmap的方法，具体地，可以用1代表ACK，0代表NACK或者终端处于inactive状态；然后基站把这个长的ACK/NACK-Inactive序列广播给终端；
基站在反馈ACK/NACK比特时，同时发送接收该ACK/NACK比特的UE ID；
LTE中，基站用物理混合自动重传指示信道(Physical Hybrid ARQ Indicator Channel，PHICH)来发送ACK/NACK比特，其对应关系与用户分 配的上行资源有关；
5G新空口(New Radio，NR)与窄带物联网(Narrow Band Internet of Things，NB-IoT)中，基站通过物理下行控制信道(physical downlink control channel，PDCCH)中的新数据指示(New Data Indicator，NDI)是否反转来通知终端重传或者新传。
相关技术中的针对PUSCH的ACK/NACK反馈方法的缺点，如下所述：
当应用到6G超大规模机器通信时，基于bitmap的方法，由于其用户数很大，将导致很大的信令开销；ACK/NACK比特+UE ID的反馈方法，当应用到6G超大规模机器通信时，由于其用户数很大，仍然具有较大的信令开销；依赖于基站分配的上行资源的ACK/NACK反馈方法，不适用6G基于竞争的超大规模机器通信；基于NDI翻转的方法，仅适用于基于调度的数据传输，不适用于6G基于竞争的超大规模机器通信。
相关技术中的5G NR中针对PDSCH的ACK/NACK反馈方法为：
1)基于混合自动重传请求(hybrid automatic repeat request，HARQ)-ACK码本，将不同serving cell与不同时隙的ACK/NACK放在一个码本中，一起发给基站。具体地，HARQ-ACK码本分为半静态码本和动态码本：
A)半静态码本基于半静态配置确定码本，能够避免DCI漏检导致基站和UE对码本大小理解不一致的情况，保证反馈的可靠性；
B)动态码本基于实际调度的PDSCH的个数进行反馈，能减少反馈的冗余比特数。
此外，为了减小HARQ-ACK码本的大小，NR支持空间合并，即通过“与”操作将同一个PDSCH的不同传输块(transport block，TB)的HARQ-ACK反馈比特合并，从而得到较少的反馈比特数。
相关技术中5G NR中针对不同PDSCH的ACK/NACK码本序列的缺点是：
在6G中，单位时间内PDSCH的数目变大，上述ACK/NACK码本序列的长度就变得很大，也就是说开销变得很大，而相关技术中NR中利用动态码本和空间合并的方法都不能有效降低其信令开销。
发明内容
本公开实施例提供了一种数据传输方法、装置及设备。可以降低ACK/NACK反馈的开销。
为解决上述技术问题，本公开的实施例提供如下技术方案：
一种数据传输方法，应用于终端，所述方法包括：
终端获取待反馈的ACK/NACK序列；
终端选择应用于所述ACK/NACK序列的机器学习模型；
终端通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
终端将所述指示序列发送给所述网络设备。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
所述终端根据所述网络设备的机器学习模型指示信息选择应用于所述ACK/NACK序列的机器学习模型。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
终端根据所述ACK/NACK序列选择应用于所述ACK/NACK序列的机器学习模型。
可选的，数据传输方法，还包括：终端将选择的应用于所述ACK/NACK序列的机器学习模型的指示信息发送给所述网络设备。
可选的，终端将所述指示序列发送给所述网络设备，包括：
终端将指示序列，以上行控制信道或上行共享信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以上行共享信道或上行控制信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以上行共享信道和上行控制信道，或者分别以上行控制信道和上行共享信道，发送给所述网络设备。
可选的，数据传输方法，还包括：所述终端接收所述网络设备通过无线资源控制(radio resource control，RRC)信令或者媒体访问控制-控制单元 (medium access control control element，MAC CE)信令或物理层信令发送的机器学习模型指示信息。
可选的，所述机器学习模型指示信息由所述网络设备中的一个或所有网络设备发送。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述终端获取待反馈的ACK/NACK序列；
所述终端将所述ACK/NACK序列进行存储；
所述终端对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，并将其上报给网络设备。
可选的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
本公开的实施例还提供一种机器学习模型的训练与分发方法，应用于网络设备侧，所述方法包括：
所述网络设备获取待训练的ACK/NACK序列，并进行存储；
所述网络设备对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
可选的，机器学习模型的训练与分发方法，还包括：
所述网络设备将所述终端上报的机器学习模型进行存储；
所述网络设备将存储的所有不同终端的机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。可选的，所述对存储的所述ACK/NACK序列进行分组，可以按照序列长度和ACK的占比进行分组。
本公开的实施例还提供一种数据传输方法，应用于网络设备，所述方法 包括：
网络设备获取待反馈的ACK/NACK序列；
网络设备选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；
网络设备通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
网络设备将所述指示序列发送给所述终端。
可选的，数据传输方法，还包括：
所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令向终端发送机器学习模型指示信息。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述网络设备获取待反馈的ACK/NACK序列；
所述网络设备对所述ACK/NACK序列进行存储；
所述网络设备对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：每隔一个固定周期重复进行一次。
可选的，所述对存储的ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
可选的，数据传输方法，还包括：
网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以下行共享信道或下行控制信道发送给所述终端；或者，
网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以下行共享信道和下行控制信道，或者分别以下行控制信道和下行共享信道，发送给所述终端。
本公开的实施例还提供一种数据传输设备，包括：收发机，处理器，存 储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述网络设备。
可选的，所述处理器获取待反馈的ACK/NACK序列；所述存储器将所述ACK/NACK序列进行存储；所述处理器对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，所述收发机将其上报给所述网络设备。
本公开的实施例还提供一种数据传输装置，包括：
收发模块，用于获取待反馈的ACK/NACK序列；
处理模块，用于选择应用于所述ACK/NACK序列的机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
所述收发模块，还用于将所述指示序列发送给所述网络设备。
可选的，所述处理模块，还用于获取待训练的ACK/NACK序列；
所述装置还包括：
存储模块，用于将表示下行数据是否正确接收的ACK/NACK序列进行存储；
所述处理模块，还用于对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述收发模块，还用于将训练得到的机器学习模型上报给所述网络设备。
本公开的实施例还提供一种数据传输设备，包括：收发机，处理器，存储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述终端。
可选的，所述处理器获取待训练的ACK/NACK序列；所述存储器对所述ACK/NACK序列进行存储；所述处理器对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；所述存储器对所述机 器学习模型进行存储；所述收发机将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的实施例还提供一种数据传输装置，包括：
收发模块，用于获取待反馈的ACK/NACK序列；
处理模块，用于选择应用于所述ACK/NACK序列的机器学习模型；
所述收发模块，还用于将选择的所述机器学习模型的指示信息发送给所述终端；
所述处理模块，还用于通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
所述收发模块，还用于将所述指示序列发送给所述终端。
可选的，所述处理模块还用于获取待反馈的ACK/NACK序列；
所述装置还包括：
存储模块，用于对所述ACK/NACK序列进行存储；
所述处理模块，还用于对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述存储模块还用于对所述机器学习模型进行存储；
所述收发模块还用于将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的实施例还提供一种网络设备，包括：收发机，处理器，存储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待训练的ACK/NACK序列，并进行存储；对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；对所述机器学习模型进行存储；将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的实施例还提供一种机器学习模型的训练与分发装置，应用于网络设备侧，所述装置包括：
收发模块，用于获取待训练的ACK/NACK序列，并进行存储；
处理模块，用于对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
存储模块，用于对所述机器学习模型进行存储；
所述收发模块，用于将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的实施例还提供一种数据传输系统，包括如上所述的设备。
本公开的实施例还提供一种处理器可读存储介质，所述处理器可读存储介质存储有处理器可执行指令，所述处理器可执行指令用于使所述处理器执行如上所述的方法的步骤。
本公开实施例的有益效果是：
本公开的上述实施例，终端获取待反馈的ACK/NACK序列，具体的，终端接收至少一个网络设备在相同和/或不同时隙发送的下行数据；终端对所述下行数据进行接收和处理，判断是否正确接收到下行数据，并使用0/1比特序列进行表示；终端根据所述网络设备的机器学习模型指示信息或自主选择应用于所述比特序列机器学习模型；终端通过所述机器学习模型，对所述0/1比特序列进行压缩，得到指示序列；终端将所述指示序列发送给所述网络设备。所述网络设备接收至少一个终端发送的上行数据；对所述上行数据进行接收和处理，判断是否正确接收到上行数据，并使用0/1比特序列进行表示；自主选择训练好的与所述比特序列对应的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；通过所述机器学习模型，对所述0/1比特序列进行压缩，得到指示序列；将所述指示序列发送给所述终端。利用机器学习降低了具有稀疏性的0/1比特序列(这里的稀疏性是指0/1序列中只有少数的0或少数的1)的反馈开销。此外，本公开中对具有稀疏性的ACK/NACK序列的压缩、传输与解压缩方法，还适用于通信过程中所产生的其他具有稀疏性的0/1比特序列。
附图说明
图1为本公开的终端侧的数据传输方法的流程示意图；
图2为本公开的网络侧的数据传输方法的流程示意图；
图3为基于PUSCH的ACK/NACK反馈方法示意图；
图4为基于PDSCH的ACK/NACK反馈方法示意图；
图5为本公开的数据传输设备的架构示意图；
图6为本公开的数据传输装置的模块示意图。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例，然而应当理解，可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反，提供这些实施例是为了能够更透彻地理解本公开，并且能够将本公开的范围完整的传达给本领域的技术人员。
如图1所示，本公开的实施例提供一种数据传输方法，应用于终端，所述方法包括：
步骤11，终端获取待反馈的ACK/NACK序列；一种可选的实现方式中，终端接收至少一个网络设备在相同和/或不同时隙发送的下行数据；终端对所述下行数据进行接收和处理，判断是否正确接收到下行数据，并使用0/1比特序列进行表示；
步骤12，终端选择应用于所述ACK/NACK序列的机器学习模型；这里，所述终端可以根据所述网络设备的机器学习模型指示信息选择应用于所述ACK/NACK序列的机器学习模型；也可以自主选择应用于所述ACK/NACK序列的机器学习模型；
步骤13，终端通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
步骤14，终端将所述指示序列发送给所述网络设备。
本公开的该实施例中，ACK/NACK序列中每一个比特对应于基站发送的数据块，可以是不同时隙、不同信道、同一信道上按预定方式分割的数据块等及其组合；该实施例通过终端接收至少一个网络设备在相同和/或不同时隙发送的下行数据；对所述下行数据进行接收和处理，判断是否正确接收到下行数据，并使用0/1比特序列进行表示；终端根据所述网络设备的机器学习模型指示信息或自主选择应用于所述比特序列机器学习模型；通过所述机器学习模型，对所述0/1比特序列进行压缩，得到指示序列；终端将所述指示序列发送给所述网络设备，基于深度学习技术(如自编码器技术)，对上述具 有稀疏性的比特序列进行压缩、传输与解压缩，降低了具有稀疏性的比特序列的反馈开销。
本公开的一可选的实施例中，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
所述终端根据所述网络设备的机器学习模型指示信息选择应用于所述ACK/NACK序列的机器学习模型。
本公开的一可选的实施例中，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
终端根据所述ACK/NACK序列选择应用于所述ACK/NACK序列的机器学习模型。
本公开的一可选的实施例中，上述数据传输方法还可以包括：
终端将选择的应用于所述ACK/NACK序列的机器学习模型的指示信息发送给所述网络设备。
本公开的一可选的实施例中，上述数据传输方法还可以包括：
所述终端接收所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令发送的机器学习模型指示信息。这里的机器学习模型指示信息由所述网络设备中的一个或所有网络设备发送。这里的所有网络设备，对应多点协作(Coordinated Multiple Points，CoMP)传输场景下的多个接收点。
具体的，所述机器学习模型由以下过程进行训练和分发：
所述网络设备获取待训练的ACK/NACK序列，可选的，网络设备接收所述终端发送的所述指示序列；所述网络设备对所述指示序列进行解压缩，生成对应的ACK/NACK序列，并进行存储；
所述网络设备对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
所述机器学习模型训练过程和分发过程，进一步包括：所述网络设备中 的一个或所有网络设备执行；每隔一个固定周期重复进行一次。
这里，网络设备将接收的0/1比特序列存储下来；对上述具有稀疏性的0/1比特序列的稀疏度进行等级划分；例如，ACK/NACK序列[0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,1,0,0,0]，其中1代表ACK，0代表NACK；另外，也可以用1表示NACK，0表示ACK，这是为了体现ACK/NACK序列的稀疏性，比特序列的稀疏度可以为比特序列中的1的个数与比特序列的总比特数的比值，上述比特序列的稀疏值为：2/20＝0.1，若0.1对应的稀疏等级为1，则确定该比特序列的稀疏等级为1。
利用多个不同稀疏等级的比特序列中的第N等级的比特序列，对第N机器学习模型进行训练，得到第N等级的比特序列对应的第N机器学习模型；
利用所述第N机器学习模型和第N+1等级的比特序列，对第N+1机器学习模型进行训练，得到第N+1等级的比特序列对应的第N+1机器学习模型。
具体实现时，将近期根据解调结果生成的具有不同稀疏等级下的0/1比特序列存储下来；对上述具有稀疏性的0/1比特序列的稀疏度进行等级划分；训练单元选择一个稀疏等级的0/1比特序列，对所要使用的自编码器进行训练；基于上述训练好的某一个稀疏等级的自编码器模型，进行迁移学习，获得其他稀疏等级下自编码器的参数，即在训练其他稀疏等级的自编码器参数时，是以上述训练好的某个稀疏等级的机器学习模型为起始点。
另外，还可以利用多个不同稀疏等级的比特序列中的第N等级的比特序列，对第N机器学习模型进行训练，得到第N等级的比特序列对应的第N机器学习模型；
利用多个不同稀疏等级的比特序列中的第N+1等级的比特序列，对第N+1机器学习模型进行训练，得到第N+1等级的比特序列对应的第N+1机器学习模型。
具体实现时，将近期从接收的解压缩的0/1比特序列存储下来；对上述存储的0/1比特序列的稀疏度进行等级划分；训练单元，利用上述不同稀疏等级的0/1比特序列，对所要使用的机器学习装置进行训练，分别得到与不同稀疏等级对应的不同参数的机器学习模型。
本公开的一可选的实施例中，所述机器学习模型还可以由以下过程进行 训练和分发：
所述终端获取待反馈的ACK/NACK序列，可选的，终端接收至少一个网络设备在相同和/或不同时隙发送的下行数据；所述终端对所述下行数据进行接收和处理，获取待反馈的ACK/NACK序列，进一步的，判断是否正确接收到下行数据，并使用0/1比特序列进行表示；
所述终端将所述ACK/NACK序列进行存储；
所述终端对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，并将其上报给网络设备；
进一步的，所述网络设备将所述上报的机器学习模型进行存储；
所述网络设备将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
进一步的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
这里，可以对存储的所述ACK/NACK序列进行分组，可以按照序列长度和ACK的占比进行分组。
本公开的一可选的实施例中，通过训练好的与所述稀疏等级对应的机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列，包括：
通过训练好的机器学习模型的自编码器中的编码器通过卷积神经网络，进行批量归一化，再经过激活函数以及全连接层，将自编码器的系数和输出限制为二进制的比特，得到指示序列。
这里，基于选择的自编码器(autoencoder)中的编码器对上述ACK/NACK序列进行压缩，并将自编码器的系数和输出也限制为二进制的0，1比特，得到压缩的Mx1的二进制向量[0,1,1,0,0,1,0]；
其中自编码器中的编码器的具体操作的描述为：将上述稀疏的0/1比特序列通过一个卷积神经网络，再进行批量归一化，再经过一个leakyReLU的激活函数，再通过一个全连接层，然后将输出值限制为0或1，即得到上述压缩的二进制序列。
本公开的一可选的实施例中，终端将所述指示序列发送给所述网络设备， 包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以上行共享信道或上行控制信道发送给所述网络设备；或者，
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以上行共享信道和上行控制信道，或者分别以上行控制信道和上行共享信道，发送给所述网络设备；或者，
终端仅将指示序列，以上行控制信道或上行共享信道发送给所述网络设备。
本公开的上述实施例中，基于深度学习技术(如自编码器技术)，对6G PDSCH进行ACK/NACK序列的压缩、传输与解压缩。
基于本公开的该图1所述的实施例，网络侧的方法与终端侧的方法相对应，上述终端侧的方法也同样适用于网络侧的实施例中，也能达到相同的技术效果。
本公开的实施例还提供一种机器学习模型的训练与分发方法，应用于网络设备侧，所述方法包括：
所述网络设备获取待训练的ACK/NACK序列，并进行存储；
所述网络设备对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的一可选的实施例中，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
本公开的一可选的实施例中，机器学习模型的训练与分发方法，还包括：
所述网络设备将所述终端上报的机器学习模型进行存储；
所述网络设备将存储的所有不同终端的机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
这里，终端侧将自己训练的模型上报网络设备，网络设备汇总其他所有终端的模型后，再将汇总的结果发给终端。
本公开的一可选的实施例中，所述对存储的所述ACK/NACK序列进行分组，可以按照序列长度和ACK的占比进行分组。
本公开的该实施例，通过上述训练过程，可以使网络设备能够确定机器学习模型，降低了具有稀疏性的比特序列的反馈开销。
如图2所示，本公开的实施例还提供一种数据传输方法，应用于网络设备，所述方法包括：
步骤21，网络设备获取待反馈的ACK/NACK序列，一种可选的实现方式中，网络设备接收至少一个终端发送的上行数据；对所述上行数据进行接收和处理，判断是否正确接收到上行数据，并使用0/1比特序列进行表示；
步骤22，网络设备选择应用于所述ACK/NACK序列对应的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；
步骤23，网络设备通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
步骤24，网络设备将所述指示序列发送给所述终端。
本公开的该实施例，ACK/NACK序列中每一个比特对应于不同的终端，所述终端在预定的信道上发送，基站集中反馈；这里面的NACK除了用来指示数据没有成功解调以外，还用来指示终端处于Inactive(非激活)状态。
该实施例通过网络设备接收至少一个终端发送的上行数据；网络设备对所述上行数据进行接收和处理，判断是否正确接收到上行数据，并使用0/1比特序列进行表示，比如可以是ACK/NACK-Inactive对应的0/1比特序列；网络设备根据所述比特序列对应的负载等级(或稀疏等级)，选择训练好的与所述比特序列对应的机器学习模型；网络设备通过所述机器学习模型，对所述0/1比特序列进行压缩，得到指示序列；网络设备将所述指示序列发送给所述终端，基于深度学习技术(如自编码器技术)，对上述具有稀疏性的比特序列进行压缩、传输与解压缩，降低了具有稀疏性的比特序列的反馈开销。
本公开的一可选的实施例中，数据传输方法，还可以包括：
所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元 MAC CE信令或物理层信令向终端发送机器学习模型指示信息。
本公开的一可选的实施例中，所述机器学习模型由以下过程进行训练和分发：
所述网络设备获取待反馈的ACK/NACK序列，可选的，网络设备接收所述终端发送的上行数据；所述网络设备对所述上行数据进行接收和处理，获取待反馈的ACK/NACK序列；
所述网络设备对所述ACK/NACK序列进行存储；
所述网络设备对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
本公开的一可选的实施例中，所述机器学习模型训练过程和分发过程，进一步包括：每隔一个固定周期重复进行一次。
本公开的一可选的实施例中，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
本公开的一可选的实施例中，所述对存储的ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
本公开的一可选的实施例中，将所述指示序列发送给所述终端，包括：
网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以下行共享信道或下行控制信道发送给所述终端；或者，网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以下行共享信道和下行控制信道，或者分别以下行控制信道和下行共享信道，发送给所述终端；或者，
网络设备仅将指示序列，以下行控制信道或下行共享信道发送给所述终端。
上述的实施例具体实现时，如图3所示，为针对6G超大规模机器通信 PUSCH的基于bit-map的ACK/NACK-Inactive的信息比特的传输示意图。针对PUSCH的基于机器学习的ACK/NACK-Inactive反馈的传输方法的公开，具体包含：
一、训练
11、基站对其网络负载进行等级划分；
12、基站将近期发给UE的未经压缩的不同负载等级下的ACK/NACK-Inactive序列存储下来，具体地，可以用1代表ACK，0代表NACK或者终端处于inactive状态；
13、分别利用不同负载等级的ACK/NACK-Inactive序列，基站的训练单元对所要使用机器学习装置进行训练，分别得到与不同负载等级对应的不同参数的机器学习模型。
二、存储与传递模型和参数
21、训练单元将训练好的不同负载等级的机器学习装置的压缩器、解压器(或编码器、译码器)的模型和参数集合，传递给基站的存储单元中；
22、基站将存储的训练好的不同负载等级的机器学习装置的解压器或译码器的模型和参数集合，使用RRC信令或MAC CE或者物理层信令分发给UE；
23、每隔一个固定的周期，重复训练的步骤2-3，与存储传递模型和参数的步骤1-2，更新训练的结果。
三、使用
31、活跃的UE基于竞争发送上行数据，即发送PUSCH；
32、基站接收到来自活跃用户的PUSCH后进行解调；
33、基站根据解调结果形成一个ACK/NACK-Inactive序列，比如每个ACK/NACK-Inactive比特与UE ID一一对应；
34、基站选择一个训练好的机器学习装置中压缩器与解压器的模型和参数集合；
35、基站基于选择的训练好的机器学习装置，对上述ACK/NACK-Inactive序列进行压缩；
36、基站将其选择的解压缩器的模型和参数集合的下标以及 ACK/NACK-Inactive序列发送给UE，具体的，在发送时：
A)基站将其选择的解压缩器的模型和参数集合的下标通过物理层信令发送给UE后，将ACK/NACK-Inactive序列以广播形式发送给UE；
或者
B)基站将其选择的解压缩器的模型和参数集合的下标与ACK/NACK-Inactive序列一同以广播形式发送给UE；
37、活跃的UE基于基站接收的信令和/或接收到的压缩的ACK/NACK-Inactive序列进行解压缩，获得与其自身对应的ACK/NACK-Inactive比特。
上述机器学习的训练过程与存储传递模型参数的过程，并不是一直与使用过程绑定在一起的，实施过程中包含如下两种方式：
第一种，一旦训练完成并将模型参数传递后，即可以重复应用上述模型参数，直至通信环境发送了很大变化，模型参数不再适用，而不是每次使用前都进行模型的训练。
第二种，基站阶段性地将上述数据提取出来，线下独立地进行训练，然后将训练好的模型与当前正在使用的模型进行比较，选择更合适的模型进行线上使用(当使用的是新训练的模型时还包含模型的传递过程)。
如图4所示，为针对PDSCH的ACK/NACK码本反馈示意图。针对PDSCH的ACK/NACK反馈的传输方法的公开，具体包含：
一、训练
11、基站将近期内不同信道状态下UE发送的通过解压缩恢复的ACK/NACK序列存储下来；
12、对上述不同信道状态进行分组，训练单元利用同一组信道状态下UE发送的ACK/NACK序列，对所要使用的机器学习装置进行训练，直至遍历所有信道状态组，得到与不同信道状态对应的不同参数的机器学习模型；
二、存储与传递模型和参数
21、训练单元将训练好的不同组信道状态下机器学习装置的解压器或译码器的模型和参数集合，传递给基站的存储单元中；
22、基站将存储的训练好的机器学习装置压缩器或编码器的模型和参数 集合，使用RRC信令或MAC CE或物理层信令配置给UE。
23、每隔一个固定周期，重复训练的步骤1-2，与存储传递模型和参数的步骤1-2；更新训练的结果。
三、使用
31、UE接收来自不同serving cell和/或不同时隙的PDSCH，并进行解调；
32、UE根据解调结果，形成包含与上述不同serving cell不同时隙的PDSCH对应的多组ACK/NACK序列；
33、UE根据不同组ACK/NACK序列所对应的信道状态自主选择训练好的机器学习装置中压缩器与解压器的模型和参数集合，或者根据primary serving cell指示的机器学习装置中压缩器与解压器的模型和参数集合，UE对上述ACK/NACK序列进行压缩；
34、UE用将压缩后ACK/NACK序列发给基站侧；具体的：
A)UE将ACK/NACK序列以及其所选择的训练好的机器学习装置中译码器的模型和参数集合的下标一同发送给基站侧，比如使用上行共享信道或者上行控制信道，或者
B)UE将ACK/NACK序列以及其所选择的训练好的机器学习装置中译码器的模型和参数集合的下标使用不同的信道(比如ACK/NACK序列使用上行共享信道，下标使用上行控制信道)分别发送给基站侧，或者
C)当UE所采用的机器学习装置中译码器的模型和参数集合是由primary serving cell指定时，仅将ACK/NACK序列发送给基站侧，可以使用上行共享信道或者上行控制信道。
35、基站侧利用primary serving cell或UE通知的机器学习装置，对接收到的压缩的ACK/NACK序列进行解压缩，然后获得与所发送PDSCH对应的ACK/NACK序列。
另外，训练、存储传递参数还可以是下面的步骤，使用方法与前述方法一致，不再赘述；
一、训练
不同UE将近期内不同信道状态下生成的未经压缩的ACK/NACK序列存储下来；
对上述不同信道状态进行分组，训练单元利用同一组信道状态下UE生成ACK/NACK序列，对所要使用的机器学习装置进行训练，直至遍历所有组的信道状态，得到与不同信道状态对应的不同参数的机器学习模型；
二、存储与传递参数模型和参数
训练单元将训练好的不同组信道状态下的机器学习装置的压缩器或编码器的参数和/或结构，使用RRC信令或MAC CE或物理层信令上报给基站，基站接收到之后，将其传递给基站的存储单元中；
基站将存储的训练好的机器学习装置压缩器或编码器的模型和参数集合，使用RRC信令或MAC CE或物理层信令配置给UE发送给UE。
每隔一段时间，重复步骤训练的步骤1-2，与存储传递模型和参数的步骤1-2，更新训练的结果。
上述机器学习的训练过程与存储传递模型参数的过程，并不是一直与使用过程绑定在一起的，实施过程中包含如下两种方式：
第一种，一旦训练完成并将模型参数传递后，即可以重复应用上述模型参数，直至通信环境发送了很大变化，模型参数不再适用，而不是每次使用前都进行模型的训练。
第二种，基站阶段性地将上述数据提取出来，线下独立地进行训练，然后将训练好的模型与当前正在使用的模型进行比较，选择更合适的模型进行线上使用(当使用的是新训练的模型时还包含模型的传递过程)。
下面结合具体的实施例，说明上述方法的具体实现：
实施例1：基于机器学习中的自编码器技术，对6G超大规模机器通信中的PUSCH的ACK/NACK-Inactive序列进行压缩，其中自编码器技术在发送端先用编码器对数据进行压缩，再在接收端用译码器进行解压缩，其中编码器与译码器的参数都是通过机器学习的方式基于数据训练得到的。
一、训练
步骤1：基站对其网络负载进行等级划分；
步骤2：基站将近期不同负载等级下发给UE的ACK/NACK-Inactive序列存储下来；
步骤3：训练单元选择一个网络负载等级的ACK/NACK-Inactive序列， 对所要使用的自编码器进行训练；
步骤4：基站在其他网络负载等级下对基于上述负载等级训练的自编码器参数进行迁移，获得不同网络负载等级下机器学习装置的参数；
二、存储与传递模型和参数
步骤1：训练单元将训练好的不同负载等级的自编码器中编码器与译码器的模型和参数集合，传递给基站的存储单元；
步骤2：基站将存储的训练好的不同负载等级的自编码器的译码器的模型和参数集合，使用RRC信令配置给UE。
步骤3：每隔一固定周期，重复训练的步骤2-4，与存储传递模型和参数的步骤1-2，更新训练的结果。
三、使用
步骤1：在6G超大规模机器通信中，每个UE每隔一段时间醒来，对环境进行监测，不同UE醒来的时间可能相同或不同，并根据监测结果判断是否需要发送上行数据，即PUSCH，为表达方便，这里假设N＝20；
步骤2：K个UE需要发送数据，并将PUSCH发送给基站，这里假设K＝2；
步骤3：基站接收所有K个UE发送的数据，并进行解调；
步骤4：基站根据解调的结果，形成一个与UE ID一一对应的Nx1的ACK/NACK-Inactive序列[0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,1,0,0,0]，其中1代表ACK，0代表NACK或者UE处于非激活状态；另外，也可以用1表示NACK或者UE处于非激活状态，0表示ACK，这时为了体现ACK/NACK-Inactive序列的稀疏性，以便于压缩，可以先对原ACK/NACK-Inactive序列进行0/1翻转；
步骤5：基站基于当前网络的负载等级，选择一个训练好的自编码器参数和/或结构；
步骤6：基站基于选择的自编码器(autoencoder)中的编码器对上述ACK/NACK-Inactive序列进行压缩，并将自编码器的系数和输出也限制为二进制的0,1比特，得到压缩的Mx1的二进制向量[0,1,1,0,0,1,0]；
步骤7：基站将其选择的自编码器模型和参数的下标与压缩的 ACK/NACK-Inactive序列一同以广播的形式发送给所有UE。
步骤8：UE基于基站通知的训练好的自编码器中的译码器对接收到的压缩的Mx1的ACK/NACK-Inactive序列进行解压缩最终输出解压缩的序列[0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,1,0,0,0]，获得与其自身对应的ACK/NACK-Inactive比特。
实施例2：基于机器学习中的自编码器技术，对PDSCH中的ACK/NACK序列进行压缩
一、训练
步骤1：基站将近期内不同信道状态下UE发送的ACK/NACK序列存储下来；
步骤2：对上述不同信道状态进行分组，训练单元利用同一组信道状态下不同UE发送的ACK/NACK序列，对所要使用的自编码器进行训练，直至遍历所有信道状态组，得到与不同信道状态对应的不同参数的机器学习模型。
二、存储与传递模型和参数
步骤1：训练单元将训练好的不同组信道状态下自编码器中编码器与译码器的模型和参数集合，传递给基站的存储单元；
步骤2：基站将存储的训练好的不同组信道状态下自编码器的译码器的模型和参数集合，使用RRC信令配置给UE。
步骤3：每隔一段时间，重复训练的步骤1-2，与存储传递模型和参数的步骤1-2，更新训练的结果。
三、使用
步骤1：NR中不同的serving cell向UE发送PDSCH；
步骤2：UE接收来自不同serving cell在不同时隙上的PDSCH，并进行解调；
步骤3：UE根据解调结果，形成包含与上述不同serving cell不同时隙的PDSCH对应的多组ACK/NACK序列，如[1,1,0,0,1,1]；[1,1,1,1,0,0]；[1,1,1,1](为表示方便，设序列的长度为16)，这里假设1代表ACK，0代表NACK，其中假设每个时隙上有两比特对应一个PDSCH的ACK/NACK，共有3个serving cell，第1组比特对应第一个serving cell，第2组比特对应第二个serving  cell，第3组比特对应第3个serving cell。
步骤4：当信道条件很好时，可以将生成的ACK/NACK码本中的01比特进行翻转，即将0变为1，1变为0，并将上述三组序列串联，得到序列[0,0,1,1,0,0,0,0,0,0,1,1,0,0,0,0]；当信道条件很差时，无需上述翻转。
步骤5：Primary serving cell通知UE与其他serving cell进行压缩的自编码器模型和参数下标；
步骤6：UE基于primary serving cell通知的自编码器的编码器对上述ACK/NACK序列进行压缩，并将自编码器的系数和输出也限制为二进制的0,1比特，得到压缩的二进制向量[1,0,1,1,0,0,0,1]；
步骤7：UE将ACK/NACK序列反馈给基站侧；
步骤8：基站侧利用primary serving cell通知的自编码器模型与参数，对接收到的压缩的ACK/NACK序列进行解压缩，然后获得比特翻转后的ACK/NACK比特序列[0,0,1,1,0,0,0,0,0,0,1,1,0,0,0,0]，比特翻转之后变为[1,1,0,0,1,1,1,1,1,1,0,0,1,1,1,1]。
本公开的上述实施例，利用机器学习技术，降低了具有稀疏性的0/1比特序列的反馈开销，具体的实施例子包含但不限于：1)以低复杂度降低了6G超大规模机器通信中PUSCH的ACK/NACK反馈的开销；2)以低复杂度降低了6G超高数据速率的PDSCH的ACK/NACK序列的开销。
基于本公开的该图2所述的实施例，终端侧的方法与网络侧的方法相对应，上述网络侧的方法也同样适用于终端侧的实施例中，也能达到相同的技术效果。
如图5所示，本公开的实施例还提供一种终端50，包括：收发机51，处理器52，存储器53，所述存储器53上存有所述处理器52可执行的程序；所述处理器52执行所述程序时实现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述网络设备。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：所述终端根据所述网络设备的机器学习模型指示信息选择应用于所述 ACK/NACK序列的机器学习模型。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：终端根据所述ACK/NACK序列选择应用于所述ACK/NACK序列的机器学习模型。
可选的，终端将选择的应用于所述ACK/NACK序列的机器学习模型的指示信息发送给所述网络设备。
可选的，接收所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令发送的机器学习模型指示信息。
可选的，终端将所述指示序列发送给所述网络设备，包括：
终端将指示序列，以上行控制信道或上行共享信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以上行共享信道或上行控制信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以上行共享信道和上行控制信道，或者分别以上行控制信道和上行共享信道，发送给所述网络设备。
可选的，所述机器学习模型指示信息由所述网络设备中的一个或所有网络设备发送。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：所述网络设备中的一个或所有网络设备执行；每隔一个固定周期重复进行一次。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述终端获取待反馈的ACK/NACK序列；
所述终端将所述ACK/NACK序列进行存储；
所述终端对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，并将其上报给网络设备。
可选的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模 型上报给所述网络设备。
需要说明的是，该实施例中的设备是与上述图1所示的方法对应的设备，上述各实施例中的实现方式均适用于该设备的实施例中，也能达到相同的技术效果。该设备中，收发机51与存储器53，以及收发机51与处理器52均可以通过总线接口通讯连接，处理器52的功能也可以由收发机51实现，收发机51的功能也可以由处理器52实现。在此需要说明的是，本公开实施例提供的上述设备，能够实现上述方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
如图6所示，本公开的实施例还提供一种数据传输装置60，包括：
收发模块61，用于获取待反馈的ACK/NACK序列；
处理模块62，用于选择应用于所述ACK/NACK序列的机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
所述收发模块61还用于将所述指示序列发送给所述网络设备。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：所述终端根据所述网络设备的机器学习模型指示信息选择应用于所述ACK/NACK序列的机器学习模型。
可选的，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：终端根据所述ACK/NACK序列选择应用于所述ACK/NACK序列的机器学习模型。
可选的，终端将选择的应用于所述ACK/NACK序列的机器学习模型的指示信息发送给所述网络设备。
可选的，接收所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令发送的机器学习模型指示信息。
可选的，终端将所述指示序列发送给所述网络设备，包括：
终端将指示序列，以上行控制信道或上行共享信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以上行共享信道或上行控制信道发送给所述网络设备。
可选的，数据传输方法，还包括：
终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以上行共享信道和上行控制信道，或者分别以上行控制信道和上行共享信道，发送给所述网络设备。
可选的，所述机器学习模型指示信息由所述网络设备中的一个或所有网络设备发送。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：所述网络设备中的一个或所有网络设备执行；每隔一个固定周期重复进行一次。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述终端获取待反馈的ACK/NACK序列；
所述终端将所述ACK/NACK序列进行存储；
所述终端对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，并将其上报给网络设备。
可选的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
需要说明的是，该实施例中的装置是与上述图1所示的方法对应的装置，上述各实施例中的实现方式均适用于该装置的实施例中，也能达到相同的技术效果。在此需要说明的是，本公开实施例提供的上述装置，能够实现上述方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
本公开的实施例还提供一种网络设备，包括：收发机，处理器，存储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待训练的ACK/NACK序列，并进行存储；对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；对所述机器学习模型进行存储；将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
可选的，所述存储模块将所述终端上报的机器学习模型进行存储；
所述收发模块，用于将存储的所有不同终端的机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述对存储的所述ACK/NACK序列进行分组，可以按照序列长度和ACK的占比进行分组。
上述方法实施例中的所有实现方式均适用于该实施例中，也能达到相同的技术效果。
本公开的实施例还提供一种机器学习模型的训练与分发装置，应用于网络设备侧，所述装置包括：
收发模块，用于获取待训练的ACK/NACK序列，并进行存储；
处理模块，用于对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
存储模块，用于对所述机器学习模型进行存储；
所述收发模块，用于将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
可选的，所述存储模块将所述终端上报的机器学习模型进行存储；
所述收发模块，用于将存储的所有不同终端的机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述对存储的所述ACK/NACK序列进行分组，可以按照序列长度和ACK的占比进行分组。
上述方法实施例中的所有实现方式均适用于该实施例中，也能达到相同的技术效果。
本公开的实施例还提供一种网络设备，包括：收发机，处理器，存储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实 现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述终端。
可选的，所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令向终端发送机器学习模型指示信息。
可选的，一个或所有网络设备向终端发送机器学习模型指示信息。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述网络设备获取待反馈的ACK/NACK序列，并进行存储；
所述网络设备对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述网络设备对所述机器学习模型进行存储；
所述网络设备将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
可选的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
可选的，所述对存储的ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
可选的，将所述指示序列发送给所述终端，包括：
网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以下行共享信道或下行控制信道发送给所述终端；或者，
网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以下行共享信道和下行控制信道，或者分别以下行控制信道和下行共享信道，发送给所述终端；或者，
网络设备仅将指示序列，以下行控制信道或下行共享信道发送给所述终 端。
需要说明的是，该实施例中的设备是与上述图2所示的方法对应的设备，上述各实施例中的实现方式均适用于该设备的实施例中，也能达到相同的技术效果。该设备中，收发机与存储器，以及收发机与处理器均可以通过总线接口通讯连接，处理器的功能也可以由收发机实现，收发机的功能也可以由处理器实现。在此需要说明的是，本公开实施例提供的上述设备，能够实现上述方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
本公开的实施例还提供一种数据传输装置，包括：
收发模块，用于获取待反馈的ACK/NACK序列；
处理模块，用于选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
所述收发模块还用于将所述指示序列发送给所述终端。
可选的，所述收发模块通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令向终端发送机器学习模型指示信息。
可选的，一个或所有网络设备向终端发送机器学习模型指示信息。
可选的，所述机器学习模型由以下过程进行训练和分发：
所述处理模块获取待反馈的ACK/NACK序列，并进行存储；
所述处理模块对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
所述处理模块对所述机器学习模型进行存储；
所述处理模块将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
可选的，所述机器学习模型训练过程和分发过程，进一步包括：
所述网络设备中的一个或所有网络设备执行；
每隔一个固定周期重复进行一次。
可选的，所述终端通过以下过程将训练好的模型上报给所述网络设备：
所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模 型上报给所述网络设备。
可选的，所述对存储的ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
可选的，将所述指示序列发送给所述终端，包括：
将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以下行共享信道或下行控制信道发送给所述终端；或者，
将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以下行共享信道和下行控制信道，或者分别以下行控制信道和下行共享信道，发送给所述终端；或者，仅将指示序列，以下行控制信道或下行共享信道发送给所述终端。
需要说明的是，该实施例中的装置是与上述图2所示的方法对应的装置，上述各实施例中的实现方式均适用于该装置的实施例中，也能达到相同的技术效果。在此需要说明的是，本公开实施例提供的上述装置，能够实现上述方法实施例所实现的所有方法步骤，且能够达到相同的技术效果，在此不再对本实施例中与方法实施例相同的部分及有益效果进行具体赘述。
本公开的实施例还提供一种数据传输系统，包括：如上述实施例所述的网络侧的设备以及终端侧的设备。
需要说明的是，上述各实施例中的所有实现方式均适用于该系统的实施例中，也能达到相同的技术效果。
本公开的实施例还提供一种处理器可读存储介质，所述处理器可读存储介质存储有处理器可执行指令，所述处理器可执行指令用于使所述处理器执行如上所述的方法。上述方法实施例中的所有实现方式均适用于该实施例中，也能达到相同的技术效果。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及算法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本公开的范围。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描 述的系统、装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本公开所提供的实施例中，应该理解到，所揭露的装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对相关技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(read-only memory，ROM)、随机存取存储器(random access memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
此外，需要指出的是，在本公开的装置和方法中，显然，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本公开的等效方案。并且，执行上述系列处理的步骤可以自然地按照说明的顺序按时间顺序执行，但是并不需要一定按照时间顺序执行，某些步骤可以并行或彼此独 立地执行。对本领域的普通技术人员而言，能够理解本公开的方法和装置的全部或者任何步骤或者部件，可以在任何计算装置(包括处理器、存储介质等)或者计算装置的网络中，以硬件、固件、软件或者它们的组合加以实现，这是本领域普通技术人员在阅读了本公开的说明的情况下运用他们的基本编程技能就能实现的。
可以理解的是，本公开实施例描述的这些实施例可以用硬件、软件、固件、中间件、微码或其组合来实现。对于硬件实现，模块、单元、子单元可以实现在一个或多个专用集成电路(Application Specific Integrated Circuits，ASIC)、数字信号处理器(Digital Signal Processor，DSP)、数字信号处理设备(DSP Device，DSPD)、可编程逻辑设备(Programmable Logic Device，PLD)、现场可编程门阵列(Field-Programmable Gate Array，FPGA)、通用处理器、控制器、微控制器、微处理器、用于执行本公开所述功能的其它电子单元或其组合中。
对于软件实现，可通过执行本公开实施例所述功能的模块(例如过程、函数等)来实现本公开实施例所述的技术。软件代码可存储在存储器中并通过处理器执行。存储器可以在处理器中或在处理器外部实现。
因此，本公开的目的还可以通过在任何计算装置上运行一个程序或者一组程序来实现。所述计算装置可以是公知的通用装置。因此，本公开的目的也可以仅仅通过提供包含实现所述方法或者装置的程序代码的程序产品来实现。也就是说，这样的程序产品也构成本公开，并且存储有这样的程序产品的存储介质也构成本公开。显然，所述存储介质可以是任何公知的存储介质或者将来所开发出来的任何存储介质。还需要指出的是，在本公开的装置和方法中，显然，各部件或各步骤是可以分解和/或重新组合的。这些分解和/或重新组合应视为本公开的等效方案。并且，执行上述系列处理的步骤可以自然地按照说明的顺序按时间顺序执行，但是并不需要一定按照时间顺序执行。某些步骤可以并行或彼此独立地执行。
以上所述的是本公开的可选实施方式，应当指出对于本技术领域的普通人员来说，在不脱离本公开所述的原理前提下还可以作出若干改进和润饰，这些改进和润饰也在本公开的保护范围内。

Claims (33)

  1. 一种数据传输方法，应用于终端，包括：
    终端获取待反馈的肯定确认/否定确认ACK/NACK序列；
    终端选择应用于所述ACK/NACK序列的机器学习模型；
    终端通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
    终端将所述指示序列发送给所述网络设备。
  2. 根据权利要求1所述的数据传输方法，其中，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
    所述终端根据所述网络设备的机器学习模型指示信息选择应用于所述ACK/NACK序列的机器学习模型。
  3. 根据权利要求1所述的数据传输方法，其中，终端选择应用于所述ACK/NACK序列的机器学习模型，包括：
    终端根据所述ACK/NACK序列选择应用于所述ACK/NACK序列的机器学习模型。
  4. 根据权利要求3所述的数据传输方法，还包括：
    终端将选择的应用于所述ACK/NACK序列的机器学习模型的指示信息发送给所述网络设备。
  5. 根据权利要求1所述的数据传输方法，其中，终端将所述指示序列发送给所述网络设备，包括：
    终端将指示序列，以上行控制信道或上行共享信道发送给所述网络设备。
  6. 根据权利要求4所述的数据传输方法，还包括：
    终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以上行共享信道或上行控制信道发送给所述网络设备。
  7. 根据权利要求4所述的数据传输方法，还包括：
    终端将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以上行共享信道和上行控制信道，或者分别以上行控制信道和上行共享信道，发送给所述网络设备。
  8. 根据权利要求1所述的数据传输方法，还包括：
    所述终端接收所述网络设备通过无线资源控制RRC信令或者媒体访问控制-控制单元MAC CE信令或物理层信令发送的机器学习模型指示信息。
  9. 根据权利要求5所述的数据传输方法，其中，所述机器学习模型指示信息由所述网络设备中的一个或所有网络设备发送。
  10. 根据权利要求1所述的数据传输方法，其中，所述机器学习模型由以下过程进行训练和分发：
    所述终端获取待反馈的ACK/NACK序列；
    所述终端将所述ACK/NACK序列进行存储；
    所述终端对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，并将其上报给网络设备。
  11. 根据权利要求10所述的数据传输方法，其中，所述终端通过以下过程将训练好的模型上报给所述网络设备：
    所述终端通过RRC信令或MAC CE或物理层信令将训练的机器学习模型上报给所述网络设备。
  12. 一种机器学习模型的训练与分发方法，应用于网络设备侧，包括：
    所述网络设备获取待训练的ACK/NACK序列，并进行存储；
    所述网络设备对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
    所述网络设备对所述机器学习模型进行存储；
    所述网络设备将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  13. 根据权利要求12所述的机器学习模型的训练与分发方法，其中，所述机器学习模型训练过程和分发过程，进一步包括：
    所述网络设备中的一个或所有网络设备执行；
    每隔一个固定周期重复进行一次。
  14. 根据权利要求12所述的机器学习模型的训练与分发方法，还包括：
    所述网络设备将所述终端上报的机器学习模型进行存储；
    所述网络设备将存储的所有不同终端的机器学习模型，使用RRC信令或 MAC CE或物理层信令分发给终端。
  15. 根据权利要求12所述的机器学习模型的训练与分发方法，其中，所述对存储的所述ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
  16. 一种数据传输方法，应用于网络设备，包括：
    网络设备获取待反馈的ACK/NACK序列；
    网络设备选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；
    网络设备通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
    网络设备将所述指示序列发送给所述终端。
  17. 根据权利要求16所述的数据传输方法，还包括：
    所述网络设备通过无线资源控制RRC信令或者媒体访问控制控制单元MAC CE信令或物理层信令向终端发送机器学习模型指示信息。
  18. 根据权利要求16所述的数据传输方法，其中，所述机器学习模型由以下过程进行训练和分发：
    所述网络设备获取待反馈的ACK/NACK序列；
    所述网络设备对所述ACK/NACK序列进行存储；
    所述网络设备对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
    所述网络设备对所述机器学习模型进行存储；
    所述网络设备将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  19. 根据权利要求18所述的数据传输方法，其中，所述机器学习模型训练过程和分发过程，进一步包括：
    每隔一个固定周期重复进行一次。
  20. 根据权利要求18所述的数据传输方法，其中，所述对存储的ACK/NACK序列进行分组，按照序列长度和ACK的占比进行分组。
  21. 根据权利要求16所述的数据传输方法，还包括：
    网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，一同以下行共享信道或下行控制信道发送给所述终端；或者，
    网络设备将指示序列以及该指示序列对应的所述机器学习模型的指示信息，分别以下行共享信道和下行控制信道，或者分别以下行控制信道和下行共享信道，发送给所述终端。
  22. 一种数据传输设备，包括：收发机，处理器，存储器，所述存储器上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述网络设备。
  23. 根据权利要求22所述的数据传输设备，其中，
    所述处理器获取待反馈的ACK/NACK序列；所述存储器将所述ACK/NACK序列进行存储；所述处理器对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型，所述收发机将其上报给所述网络设备。
  24. 一种数据传输装置，包括：
    收发模块，用于获取待反馈的ACK/NACK序列；
    处理模块，用于选择应用于所述ACK/NACK序列的机器学习模型；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
    所述收发模块，还用于将所述指示序列发送给所述网络设备。
  25. 根据权利要求24所述的数据传输装置，其中，
    所述处理模块，还用于获取待训练的ACK/NACK序列；
    所述装置还包括：
    存储模块，用于将表示下行数据是否正确接收的ACK/NACK序列进行存储；
    所述处理模块，还用于对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
    所述收发模块，还用于将训练得到的机器学习模型上报给所述网络设备。
  26. 一种数据传输设备，包括：收发机，处理器，存储器，所述存储器 上存有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待反馈的ACK/NACK序列；选择应用于所述ACK/NACK序列的机器学习模型，并将选择的所述机器学习模型的指示信息发送给所述终端；通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；将所述指示序列发送给所述终端。
  27. 根据权利要求26所述的数据传输设备，其中，
    所述处理器获取待训练的ACK/NACK序列；所述存储器对所述ACK/NACK序列进行存储；所述处理器对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；所述存储器对所述机器学习模型进行存储；所述收发机将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  28. 一种数据传输装置，包括：
    收发模块，用于获取待反馈的ACK/NACK序列；
    处理模块，用于选择应用于所述ACK/NACK序列的机器学习模型；
    所述收发模块，还用于将选择的所述机器学习模型的指示信息发送给所述终端；
    所述处理模块，还用于通过所述机器学习模型，对所述ACK/NACK序列进行压缩，得到指示序列；
    所述收发模块，还用于将所述指示序列发送给所述终端。
  29. 根据权利要求28所述的数据传输装置，其中，
    所述处理模块还用于获取待反馈的ACK/NACK序列；
    所述装置还包括：
    存储模块，用于对所述ACK/NACK序列进行存储；
    所述处理模块，还用于对存储的ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
    所述存储模块还用于对所述机器学习模型进行存储；
    所述收发模块还用于将所述存储的训机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  30. 一种网络设备，包括：收发机，处理器，存储器，所述存储器上存 有所述处理器可执行的程序；所述处理器执行所述程序时实现：获取待训练的ACK/NACK序列，并进行存储；对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；对所述机器学习模型进行存储；将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  31. 一种机器学习模型的训练与分发装置，应用于网络设备侧，包括：
    收发模块，用于获取待训练的ACK/NACK序列，并进行存储；
    处理模块，用于对存储的所述ACK/NACK序列进行分组，进行机器学习模型训练，生成多组机器学习模型；
    存储模块，用于对所述机器学习模型进行存储；
    所述收发模块，用于将存储的所述机器学习模型，使用RRC信令或MAC CE或物理层信令分发给终端。
  32. 一种数据传输系统，包括如权利要求22至23中任一项所述的设备，或者如权利要求26至27中任一项所述的设备。
  33. 一种处理器可读存储介质，所述处理器可读存储介质存储有处理器可执行指令，所述处理器可执行指令用于使所述处理器执行如权利要求1至11中任一项所述的方法或者如权利要求12至15中任一项所述的方法或者如权利要求16至21中任一项所述的方法的步骤。
PCT/CN2021/128386 2020-11-12 2021-11-03 数据传输方法、装置及设备 Ceased WO2022100493A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP21891020.6A EP4246850B1 (en) 2020-11-12 2021-11-03 Method, apparatus, and device for data transmission
US18/036,577 US12615124B2 (en) 2020-11-12 2021-11-03 Data transmission method, data transmission apparatus and data transmission device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011261728.1 2020-11-12
CN202011261728.1A CN114499796B (zh) 2020-11-12 2020-11-12 一种数据传输方法、装置及设备

Publications (1)

Publication Number Publication Date
WO2022100493A1 true WO2022100493A1 (zh) 2022-05-19

Family

ID=81490029

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/128386 Ceased WO2022100493A1 (zh) 2020-11-12 2021-11-03 数据传输方法、装置及设备

Country Status (4)

Country Link
US (1) US12615124B2 (zh)
EP (1) EP4246850B1 (zh)
CN (1) CN114499796B (zh)
WO (1) WO2022100493A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
JP7842968B2 (ja) * 2022-04-14 2026-04-09 ソニーグループ株式会社 送信装置、受信装置、送信方法および受信方法
US12356388B2 (en) * 2022-06-03 2025-07-08 Qualcomm Incorporated Network coding with user equipment cooperation
US12470355B2 (en) * 2022-09-07 2025-11-11 Qualcomm Incorporated ACK coalescing performance through dynamic stream selection
US12513043B2 (en) * 2023-07-26 2025-12-30 Rakuten Symphony, Inc. System, method, and computer program for managing machine learning models in a network
WO2026000390A1 (zh) * 2024-06-28 2026-01-02 Oppo广东移动通信有限公司 基于模型的反馈方法、装置、设备、介质和程序产品

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090225700A1 (en) * 2008-03-04 2009-09-10 Texas Instruments Incorporated Transmission of Multiple ACK/NAK Bits with Data
CN103095432A (zh) * 2011-11-07 2013-05-08 北京三星通信技术研究有限公司 一种发送harq-ack反馈信息的方法
CN105187157A (zh) * 2015-08-06 2015-12-23 北京北方烽火科技有限公司 一种反馈ack/nack信息的方法、装置及终端
US10732902B1 (en) * 2018-09-26 2020-08-04 Amazon Technologies, Inc. Using feedback for adaptive data compression

Family Cites Families (56)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8543867B2 (en) * 2010-04-02 2013-09-24 Sharp Laboratories Of America, Inc. Transmission of acknowledgement and negative acknowledgement in a wireless communication system
CN111818569B (zh) * 2015-04-10 2024-04-16 瑞典爱立信有限公司 用于对harq反馈进行压缩的方法和用户设备
CN111800232B (zh) * 2015-08-10 2024-12-20 华为技术有限公司 上行控制信息传输方法和装置
EP3447953B1 (en) * 2016-05-10 2020-07-22 Huawei Technologies Co., Ltd. Feedback information transmission method, relevant device and communication system
WO2017201654A1 (zh) * 2016-05-23 2017-11-30 华为技术有限公司 传输数据的方法、网络设备和终端设备
KR102150246B1 (ko) * 2016-06-30 2020-09-01 후아웨이 테크놀러지 컴퍼니 리미티드 업링크 정보 전송 방법, 관련 디바이스 및 통신 시스템
US10763893B2 (en) * 2016-07-20 2020-09-01 Georges Harik Method for data compression
CN107733558B (zh) * 2016-08-12 2020-07-21 华为技术有限公司 混合自动重传请求确认harq-ack反馈方法和装置
WO2018027997A1 (zh) * 2016-08-12 2018-02-15 华为技术有限公司 上行信号传输方法、终端设备和网络设备
CN108063647B (zh) * 2016-11-05 2021-04-20 华为技术有限公司 数据传输的方法和装置
US11050516B2 (en) * 2017-01-05 2021-06-29 Guangdong Oppo Mobile Telecommunications Corp., Ltd. Method, network device and terminal device for transmitting feedback information
WO2018201369A1 (zh) * 2017-05-04 2018-11-08 华为技术有限公司 一种控制信息传输的方法、终端设备和网络设备
SG11201910586WA (en) * 2017-08-10 2020-02-27 Guangdong Oppo Mobile Telecommunications Corp Ltd Data transmission method, network device and terminal device
CN109391352B (zh) * 2017-08-11 2021-12-10 华为技术有限公司 一种应答信息的传输方法、终端设备和网络设备
CN110048826B (zh) * 2017-09-30 2020-08-21 华为技术有限公司 数据传输方法、终端设备以及网络设备
CN110035531B (zh) * 2018-01-12 2021-12-03 华为技术有限公司 上行控制信息传输方法及装置
CN111566961B (zh) * 2018-01-30 2022-05-06 华为技术有限公司 一种信息传输方法及设备
JP2021517382A (ja) * 2018-02-13 2021-07-15 オッポ広東移動通信有限公司Ｇｕａｎｇｄｏｎｇ Ｏｐｐｏ Ｍｏｂｉｌｅ Ｔｅｌｅｃｏｍｍｕｎｉｃａｔｉｏｎｓ Ｃｏｒｐ．， Ｌｔｄ． サウンディング参照信号の伝送方法、端末機器およびネットワーク機器
KR20200119813A (ko) * 2018-02-14 2020-10-20 광동 오포 모바일 텔레커뮤니케이션즈 코포레이션 리미티드 물리 상향 공유 채널의 전송 방법 및 단말기 디바이스
US11665017B2 (en) * 2018-02-28 2023-05-30 Cisco Technology, Inc. Telemetry reporting in vehicle super resolution systems
CN110324126B (zh) * 2018-03-30 2020-12-04 电信科学技术研究院有限公司 一种数据传输方法、终端设备及网络设备
CN110324117B (zh) * 2018-03-30 2021-10-26 大唐移动通信设备有限公司 一种数据传输方法、终端设备及网络设备
CN110351020B (zh) * 2018-04-03 2021-06-15 华为技术有限公司 一种传输数据的方法、装置和系统
CN111989904B (zh) * 2018-04-04 2022-04-05 华为技术有限公司 数据传输方法、终端设备及网络设备
CN108683901B (zh) * 2018-05-10 2020-11-03 Oppo广东移动通信有限公司 一种数据处理方法、mec服务器及计算机可读存储介质
CN110505040B (zh) * 2018-05-18 2020-05-26 维沃移动通信有限公司 信息传输方法、终端及网络设备
CN112737738A (zh) * 2018-05-21 2021-04-30 华为技术有限公司 上行控制信息的传输方法及设备
CN110557231B (zh) * 2018-06-04 2021-02-12 华为技术有限公司 传输信息的方法和通信设备
CN110830151B (zh) * 2018-08-07 2021-06-15 华为技术有限公司 反馈信息的传输方法和装置
CN112534910A (zh) * 2018-08-17 2021-03-19 Oppo广东移动通信有限公司 Harq信息的传输方法、网络设备和终端设备
WO2020056559A1 (zh) * 2018-09-17 2020-03-26 Oppo广东移动通信有限公司 一种信息传输方法、设备及存储介质
CN110943805B (zh) * 2018-09-21 2021-06-04 电信科学技术研究院有限公司 一种harq-ack的传输方法、终端设备及网络设备
WO2020068498A1 (en) * 2018-09-27 2020-04-02 Google Llc Data compression using integer neural networks
JP7229362B2 (ja) * 2018-12-29 2023-02-27 北京小米移動軟件有限公司 直接通信のデータ伝送方法、装置及びシステム
CN111417191A (zh) * 2019-01-04 2020-07-14 电信科学技术研究院有限公司 一种信息传输方法、终端及网络设备
CN111565093B (zh) * 2019-02-14 2022-08-26 华为技术有限公司 信息传输方法、终端设备及网络设备
CN111585727B (zh) * 2019-02-15 2021-11-05 大唐移动通信设备有限公司 一种harq-ack信息的传输方法、装置、终端及网络设备
CN111586850B (zh) * 2019-02-15 2023-09-19 大唐移动通信设备有限公司 传输方法、接收方法、终端及网络设备
WO2021007685A1 (zh) * 2019-07-12 2021-01-21 Oppo广东移动通信有限公司 用于传输侧行数据的方法、终端设备和网络设备
CN110311762B (zh) * 2019-07-16 2021-04-16 北京紫光展锐通信技术有限公司 反馈信息传输方法、装置、终端及存储介质
CN116667989A (zh) * 2019-07-31 2023-08-29 中兴通讯股份有限公司 信号传输方法、装置、网络设备及存储介质
WO2021026917A1 (zh) * 2019-08-15 2021-02-18 富士通株式会社 信号发送方法、装置和系统
CN114785463B (zh) * 2019-11-06 2023-12-05 Oppo广东移动通信有限公司 传输反馈信息的方法、终端设备和网络设备
EP4398507A3 (en) * 2019-11-08 2024-10-16 Guangdong Oppo Mobile Telecommunications Corp., Ltd. Uplink data transmission method and apparatus, terminal, and storage medium
KR20220100788A (ko) * 2019-11-18 2022-07-18 삼성전자주식회사 Harq-ack 송신 방법 및 장치
CN111079948B (zh) * 2019-12-27 2023-02-24 电子科技大学 一种基于sdn的分布式机器学习训练加速方法
US12413347B2 (en) * 2020-02-11 2025-09-09 Beijing Xiaomi Mobile Software Co., Ltd. Communication method for determining priority of HARQ-ACK information, and device
CN113541876B (zh) * 2020-04-10 2023-07-21 中国移动通信有限公司研究院 一种上行传输方法、终端及网络设备
WO2022027605A1 (zh) * 2020-08-07 2022-02-10 华为技术有限公司 信息传输方法和通信装置
CN115918220A (zh) * 2020-09-29 2023-04-04 Oppo广东移动通信有限公司 信道传输的方法、终端设备和网络设备
CN114362895B (zh) * 2020-10-12 2024-06-25 维沃移动通信有限公司 反馈信息传输方法、终端及网络设备
CN116686361A (zh) * 2021-01-13 2023-09-01 华为技术有限公司 上行确认信息的传输方法及装置
CN115118403B (zh) * 2021-03-19 2024-12-03 北京紫光展锐通信技术有限公司 一种信息传输方法及通信装置
WO2022205103A1 (zh) * 2021-03-31 2022-10-06 北京小米移动软件有限公司 一种传输harq-ack的方法、装置及可读存储介质
CN115175317A (zh) * 2021-04-01 2022-10-11 华为技术有限公司 一种上行控制信息的传输方法及装置
CN117675126A (zh) * 2022-08-10 2024-03-08 华为技术有限公司 数据传输方法及装置

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090225700A1 (en) * 2008-03-04 2009-09-10 Texas Instruments Incorporated Transmission of Multiple ACK/NAK Bits with Data
CN103095432A (zh) * 2011-11-07 2013-05-08 北京三星通信技术研究有限公司 一种发送harq-ack反馈信息的方法
CN105187157A (zh) * 2015-08-06 2015-12-23 北京北方烽火科技有限公司 一种反馈ack/nack信息的方法、装置及终端
US10732902B1 (en) * 2018-09-26 2020-08-04 Amazon Technologies, Inc. Using feedback for adaptive data compression

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4246850A4 *

Also Published As

Publication number Publication date
CN114499796B (zh) 2024-10-15
US20240014981A1 (en) 2024-01-11
US12615124B2 (en) 2026-04-28
CN114499796A (zh) 2022-05-13
EP4246850C0 (en) 2025-07-23
EP4246850B1 (en) 2025-07-23
EP4246850A4 (en) 2024-05-01
EP4246850A1 (en) 2023-09-20

Similar Documents

Publication Publication Date Title
WO2022100493A1 (zh) 数据传输方法、装置及设备
CN111670555B (zh) 数据重传方法及其接收器
CN107431570B (zh) 用于减轻多层协议无线系统中的突发打孔和干扰的码块层面纠错和介质访问控制(mac)层面混合自动重传请求
CN106850129B (zh) 非预期的下行链路子帧的harq-ack处置方法
CN104301077B (zh) 一种混合重传的方法
US20190182809A1 (en) Data transmission method and apparatus, and data processing method and apparatus
CN110635870B (zh) 生成混合自动重传请求harq信息的方法和装置
CN112567857B (zh) 用于调度传输块的方法和装置
EP3627740B1 (en) Methods, system and receiver for efficient harq operation
US20200170033A1 (en) Transmission method, terminal device and base station
JP2018525928A (ja) フィードバック情報送信方法、ユーザ機器、および基地局
CN102340375A (zh) 数据传输管理系统及其相关数据传输管理方法
CN111132358B (zh) 数据包的传输方法、终端及存储介质
WO2016011950A1 (zh) 一种传输、存储下行数据的方法、基站及终端
CN108347295A (zh) 一种数据传输方法及装置
CN110351019A (zh) 通信方法和装置
WO2019134140A1 (zh) 反馈应答信息发送方法、接收方法、装置及系统
CN110034849A (zh) 一种数据传输方法及设备
CN110073622B (zh) 用于数据块组的反馈信息
CN109150380B (zh) 数据传输的方法、网络设备和终端设备
CN120019600A (zh) 基于子帧的混合自动重传请求反馈
CN111294143A (zh) Pdsch软合并方法及装置、存储介质、用户终端、基站
WO2023226702A1 (zh) 一种通信方法及装置
EP4346139A1 (en) Feedback per code block group
CN112584433B (zh) 一种数据发送、接收方法、设备及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21891020

Country of ref document: EP

Kind code of ref document: A1

WWE Wipo information: entry into national phase

Ref document number: 18036577

Country of ref document: US

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2021891020

Country of ref document: EP

Effective date: 20230612

WWG Wipo information: grant in national office

Ref document number: 2021891020

Country of ref document: EP