User equipment and wireless communication method for neural network operation

Abstract

A user equipment for neural network operation is provided. The user equipment can include a processor and a transmitter. The processor performs a neural network operation to generate a complex neural network operation result, wherein the complex neural network operation result is in a data packet, and the complex neural network operation result is intermediate data of the neural network operation. The transmitter transmits the data packet to a base station. The data packet includes a descriptor, and the descriptor includes parameters and settings corresponding to the complex neural network operation result.

Description

Technical Field

[0001] The embodiments of the present invention mainly relate to a neural network computation technique, and particularly to a neural network computation technique for transmitting intermediate data generated by a user equipment (UE) for neural network computation. Background Technology

[0002] Artificial neural networks (ANNs) and neural networks (NNs) have become important machine learning techniques used to provide intelligent solutions in many applications. Furthermore, deep learning methods, utilizing deep structures within neural networks, have shown great potential, achieving excellent performance in machine learning tasks.

[0003] In current technology, deep learning tasks based on neural networks can be divided into different network nodes for distributed processing. The model-level pipeline can be divided into different machine learning tasks corresponding to different network nodes. For example, a mobile device can execute an object detection module to detect objects, and the detected objects (e.g., a human face) can be transmitted to a multi-access edge computing (MEC) node for face recognition.

[0004] In other words, a layer-level pipeline can be divided into different neural network (or deep learning) sub-models within the overall neural network (or deep learning) model, allowing these sub-models to be configured in wireless devices, edge computing nodes, and cloud computing nodes. In this type of neural network, where components are decomposed, some neural network parameters computed at one computing node are transmitted to another node for the next stage of the process. However, current wireless communication methods do not address how to transmit neural network parameters or partial computation results from a neural network model. Summary of the Invention

[0005] In view of the problems of the prior art described above, embodiments of the present invention provide a user device and a wireless communication method for neural network operations.

[0006] According to an embodiment of the present invention, a user equipment for neural network operations is provided. The user equipment may include a processor and a transmitter. The processor performs a neural network operation to generate a complex neural network operation result, wherein the complex neural network operation result is in a data packet, and the complex neural network operation result is intermediate data of the neural network operation. The transmitter transmits the data packet to a base station. The data packet includes a descriptor, and the descriptor includes parameters and settings corresponding to the complex neural network operation result.

[0007] According to some embodiments of the present invention, the data packet further includes a packet header and a data payload. The packet header is used to indicate the data packet. The data payload includes the result of the complex neural network operation.

[0008] According to some embodiments of the present invention, the parameters and settings corresponding to the above-mentioned complex neural network operation results include a neural network type, the number of layers contained in the neural network, the size of the complex neural network operation result, the bit order corresponding to the complex neural network operation result, a sequence number, and a timestamp.

[0009] According to some embodiments of the present invention, a Protocol Data Unit (PDU) type is set in the data packet to indicate that the data packet is used to carry the results of the complex neural network operation.

[0010] According to some embodiments of the present invention, a Quality of Service (QoS) type is set in the data packet to indicate that the data packet is used to carry the neural network operation result with the corresponding QoS characteristics.

[0011] According to some embodiments of the present invention, the above data packet includes a Quality of Service Flow Identifier (QFI) or a 5G Quality of Service Identifier (5QI).

[0012] According to some embodiments of the present invention, the transmitter transmits a scheduling request to the base station for performing the neural network operation. The scheduling request may include a binary indicator indicating that the scheduling request is for the neural network operation, a request type, a request descriptor, a model indicator, and the magnitude of the result of the complex neural network operation. The scheduling request also includes a semi-persistent scheduling descriptor, the number of data transmissions required, and one cycle of an uplink data packet transmission.

[0013] According to some embodiments of the present invention, the transmitter further transmits an uplink buffer status report to the base station to perform the neural network operation, wherein the buffer status report includes an information descriptor, wherein the information descriptor includes a neural network type and the size of the result of the complex neural network operation.

[0014] According to some embodiments of the present invention, the transmitter further transmits a network slice establishment request information to the base station to perform the neural network operation. The network slice establishment request information may include an information descriptor, wherein the information descriptor may include a neural network type, the size of the result of the complex neural network operation, an average rate for transmitting the result of the complex neural network operation, and a peak rate for transmitting the result of the complex neural network operation.

[0015] According to some embodiments of the present invention, the transmitter further transmits radio resource control (RRC) connection setting information to the base station, wherein the radio resource control connection setting information may include a binary indicator, a protocol data unit session type field, and a descriptor for indicating that the radio resource control connection setting information is used for the neural network communication.

[0016] According to an embodiment of the present invention, a wireless communication method for neural network operations is provided. The wireless communication method can be applied to a user equipment. The steps of the wireless communication method include: executing a neural network operation through a processor of the user equipment to generate a complex neural network operation result, wherein the complex neural network operation result is in a data packet, and the complex neural network operation result is intermediate data of the neural network operation; and transmitting the data packet to a base station through a transmitter of the user equipment, wherein the data packet includes a descriptor, and the descriptor includes parameters and settings corresponding to the complex neural network operation result.

[0017] Regarding other additional features and advantages of the present invention, those skilled in the art can make some modifications and refinements based on the user equipment and wireless communication method for neural network operations disclosed in the embodiments of this disclosure without departing from the concept and scope of the present invention. Attached Figure Description

[0018] Figure 1 This is a block diagram showing a wireless communication system 100 according to an embodiment of the present invention.

[0019] Figure 2 This is a schematic diagram illustrating neural network operations according to some embodiments of the present invention.

[0020] Figure 3 This is a schematic diagram of a beam fault recovery (BFR) procedure according to an embodiment of the present invention.

[0021] Figure 4A This is a schematic diagram illustrating a neural network type according to an embodiment of the present invention.

[0022] Figure 4B This is a schematic diagram illustrating a neural network type according to another embodiment of the present invention.

[0023] Figure 5 This is a flowchart illustrating a communication session that supports neural network communication transmission according to some embodiments of the present invention.

[0024] Figure 6 This is a flowchart illustrating neural network communication for dynamic grant according to some embodiments of the present invention.

[0025] Figure 7 This is a flowchart illustrating neural network communication for semi-persistent scheduling (SPS) according to some embodiments of the present invention.

[0026] Figure 8 This is a flowchart illustrating neural network communication during base station-initiated operations according to some embodiments of the present invention.

[0027] Figure 9 This is a flowchart of a wireless communication method for neural network operations according to some embodiments of the present invention.

[0028] Explanation of reference numerals in the attached figures:

[0029] 100: Wireless Communication System

[0030] 110: User Equipment

[0031] 111: Processor

[0032] 112: Teleporter

[0033] 113: Receiver

[0034] 114: Memory device

[0035] 120: Base station

[0036] 130: Core Network

[0037] 300: Data Packet

[0038] 302: Packet header

[0039] 304: Descriptor

[0040] 306: Data Load

[0041] 400, 450: Neural Network Type

[0042] 402, 404, 452, 454, 456, L1, L2…LM: Layers

[0043] L1_n1~L1_n N L2_n1~L2_n N LM_n1~LM_n N computing nodes

[0044] x1~x n Input data

[0045] y1~y n Output data

[0046] Z1, Z2: Intermediate data

[0047] S502~S510, S600~S614, S700~S708, S800~S806, S910~S920: Steps Detailed Implementation

[0048] This section describes preferred embodiments of the invention and is intended to illustrate the concept of the invention rather than to limit the scope of protection of the invention. The scope of protection of the invention shall be determined by the claims.

[0049] Figure 1 This is a block diagram illustrating a wireless communication system 100 according to an embodiment of the present invention. According to an embodiment of the present invention, the wireless communication system 100 can be applied to a machine learning technique, such as a neural network (NN) technique. Figure 1 As shown, the wireless communication system 100 may include a user equipment (UE) 110, a base station 120, and a core network 130. Note that in... Figure 1 The block diagrams shown are for illustrative purposes only and are not intended to represent embodiments of the present invention. Figure 1 Limited to.

[0050] According to embodiments of the present invention, the user equipment 110 may be a smartphone, a wearable device, a tablet computer, a desktop computer, an Internet of Things (IoT) node (e.g., a webcam or a wireless sensor node with processing capabilities for sensed data, but the present invention is not limited thereto), a gateway, or an edge computing device, but the present invention is not limited thereto.

[0051] like Figure 1 As shown, user equipment 110 may include a processor 111, a transmitter 112, a receiver 113, and a memory device 114. Note that in... Figure 1 The user equipment 110 shown is only for illustrative purposes and is not intended to represent an embodiment of the present invention. Figure 1 Limited to.

[0052] According to an embodiment of the present invention, the processor 111 may be a central processing unit (CPU) or a graphics processing unit (GPU). The processor 111 is used to perform neural network operations. According to an embodiment of the present invention, the processor 111 may be used to control the operation of the transmitter 112, the receiver 113, and the memory device 114.

[0053] According to an embodiment of the present invention, transmitter 112 can transmit data packets or information to base station 120, and receiver 113 can receive data packets or information from base station 120. According to one embodiment of the present invention, transmitter 112 and receiver 113 can be integrated into a single transceiver. According to an embodiment of the present invention, transmitter 112 and receiver 113 can be used in 3G, 4G, 5G, or Wi-Fi communication, but the present invention is not limited thereto.

[0054] According to an embodiment of the present invention, the memory device 114 may store software and firmware code, system data, and user data of the user equipment 110. The memory device 114 may be a volatile memory (e.g., random access memory, RAM), a non-volatile memory (e.g., flash memory, read-only memory, ROM), a hard disk, or a combination of the above. According to an embodiment of the present invention, the memory device 114 may store data corresponding to neural network operations.

[0055] According to embodiments of the present invention, base station 120 may be an evolved Node B (eNB) or a generation Node B (gNB). According to embodiments of the present invention, core network 130 may be a 4G core network or a 5G core network, but the present invention is not limited thereto.

[0056] Figure 2 This is a schematic diagram illustrating neural network operations according to some embodiments of the present invention. For example... Figure 2 As shown, the neural network operation includes complex layers L1, L2…LM. Layer L1 includes operation nodes L1_n1, L1_n2…L1_n. N The L2 layer includes computational nodes L2_n1, L2_n2…L2_n. N The LM layer includes computation nodes LM_n1, LM_n2…LM_n. N In this embodiment, the input data are x1, x2…x n The data is transmitted to the computation nodes L1_n1, L1_n2…L1_nN in the L1 layer for neural network operations to obtain intermediate data Z1. Intermediate data Z1 represents the computation nodes L1_n1, L1_n2…L1_nN in the L1 layer. N The corresponding neural network operation results. Furthermore, the intermediate data Z1 will be transmitted to the L2 layer operation nodes L2_n1, L2_n2…L2_n. N Neural network operations are performed to obtain intermediate data Z2. Intermediate data Z2 represents the operation nodes L2_n1, L2_n2…L2_n in layer L2. N The corresponding neural network operation results. Similarly, the intermediate data ZM-1 will be transmitted to the operation nodes LM_n1, LM_n2…LM_n of the LM layer. N Perform neural network operations to obtain intermediate data Z. M (which can be considered as the output data y1, y2...y of a neural network operation) n Intermediate data Z) M This represents the computation nodes LM_n1, LM_n2, ..., LM_n of the LM-th layer. N The corresponding neural network operation results. According to one embodiment of the present invention, the user equipment 110 can process the operations of several layers of the neural network. For example, the user equipment 110 can process the neural network operations of the L1 layer and the L2 layer to generate intermediate data Z1 and intermediate data Z2 (i.e., the neural network operation results corresponding to the neural network operations of the L1 layer and the L2 layer). Then, the user equipment 110 can transmit the intermediate data Z1 and intermediate data Z2 to the base station 120 in sequence to perform the neural network operations corresponding to the remaining other layers.

[0057] Figure 3 This is a schematic diagram showing a data packet in a data structure according to some embodiments of the present invention. According to one embodiment of the present invention, the neural network operation result generated by the user equipment 110 (e.g., Figure 2 The intermediate data Z1 or Z2 shown (but not limited to this invention) can be carried in the data packet 300. Next, the user equipment 110 transmits the data packet 300 to the base station 120 for the neural network operations corresponding to the remaining layers. Figure 3 As shown, the data packet 300 may include a packet header 302, a descriptor 304, and a data payload 306. In this embodiment, the packet header 302 may include an indicator to indicate that the data packet 300 includes the results of a neural network operation, such as the L1 layer operation nodes L1_n1, L1_n2…L1_n. N The corresponding neural network operation result or the operation nodes L2_n1, L2_n2…L2_n in the L2 layer N The corresponding neural network operation result. In this embodiment, descriptor 304 may include the parameters and settings corresponding to the neural network operation result generated by user device 110. In this embodiment, data payload 306 may include the neural network operation result generated by user device 110.

[0058] In some embodiments of the present invention, the parameters and settings corresponding to the neural network operation results may include the type of neural network (e.g.: Figure 4A and Figure 4B The neural network type shown is Model 1 or Model 2, but this invention is not limited thereto; the number of layers contained in the neural network (e.g., 2 layers, L1 layer and L2 layer, but this invention is not limited thereto); the size of the neural network operation result (e.g., ... Figure 3 As shown, the neural network operation result corresponding to the L1 layer in data payload 306 can include 6 elements Z. 11 Z 12 Z 13 Z 14 Z 15 and Z 16However, this invention is not limited to these), the level corresponding to the neural network operation result (e.g., the L1 layer or the L2 layer, but this invention is not limited to these), a sequence number (i.e., the sequence number of the neural network operation result), and a time stamp (i.e., the time point when the neural network operation result was transmitted), but this invention is not limited to these.

[0059] Figure 4A This is a schematic diagram illustrating a neural network type according to an embodiment of the present invention. For example... Figure 4A As shown, neural network type 400 comprises two layers, layer 402 and layer 404. Layer 402 in neural network type 400 may include complex number operation nodes, and layer 404 in neural network type 400 may include two operation nodes. Figure 4B This is a schematic diagram illustrating a neural network type according to another embodiment of the present invention. For example... Figure 4B As shown, neural network type 450 consists of three layers: layer 452, layer 454, and layer 456. Each layer (layers 452, 454, and 456) in neural network type 450 can include two computation nodes. It should be noted that... Figure 4A and Figure 4B The types of neural networks shown are merely illustrative embodiments of the present invention, but the present invention is not limited thereto.

[0060] In some embodiments of the present invention, data packets (e.g., data packet 300) transmitted from user equipment 110 to base station 120 may include a new Protocol Data Unit (PDU) type. The PDU type can be used to indicate that the data packet is used to carry the results of a neural network operation. In some embodiments of the present invention, data packets (e.g., data packet 300) transmitted from user equipment 110 to base station 120 may include a new Quality of Service (QoS) type. The QoS type can be used to indicate that the data packet is used to carry the results of a neural network operation with corresponding QoS characteristics.

[0061] In some embodiments of the present invention, user equipment 110 can control the quality of service (QoS) for neural network operations between user equipment 110 and base station 120. In this embodiment, data packets (e.g., data packet 300) transmitted from user equipment 110 to base station 120 may include a QoS Flow Identifier (QFI) to indicate that the data packet is used to carry the results of neural network operations. Furthermore, in this embodiment, user equipment 110 can map the QoS flow of neural network communication to a data radio bearer (DRB), which can provide the communication quality corresponding to the neural network operation results.

[0062] In some embodiments of the present invention, data packets (e.g., data packet 300) transmitted from user equipment 110 to base station 120 may include a 5G QoS Identifier (5QI). The 5G QoS Identifier can be used to indicate that the data packet is used to carry the results of neural network operations.

[0063] In some embodiments of the present invention, user equipment 110 may place a data packet containing the results of a neural network operation (e.g., data packet 300) into a specific data buffer (not shown in the figure).

[0064] In some embodiments of the present invention, user equipment 110 may set quality of service scheduling rules (QoS scheduling policies) based on the quality of service requirements of neural network communication between user equipment 110 and base station 120.

[0065] In some embodiments of the present invention, user equipment 110 may configure scheduling rules based on the traffic characteristics and application requirements of neural network communication between user equipment 110 and base station 120. Scheduling rules can be set by considering the neural network processing time for data processing. According to the scheduling rules, transmitter 112 of user equipment 110 may transmit a scheduling request to base station 120 for neural network computation. According to one embodiment of the present invention, the scheduling request may include a binary indication indicating that the scheduling request is for neural network computation, a request type, a request descriptor, a model identifier, and the magnitude of the neural network computation result. The binary indication may indicate that the scheduling request is for neural network communication. The request type may indicate the procedure corresponding to the scheduling request (e.g., Dynamic Grant or Semi-Persistent Scheduling (SPS)). The request descriptor may describe the characteristics of the scheduling request. Model indicators can describe the type of neural network (e.g., neural network type 400 and neural network type 450, but the invention is not limited thereto).

[0066] In some embodiments of the present invention, user equipment 110 may transmit an uplink buffer status report (BSR) to base station 120. The BSR may include an identifier indicating whether a set of neural network operation results is available in a specific data buffer used for uplink transmission. The BSR may also include a descriptor for the neural network operation results. This descriptor may indicate the type of neural network (e.g., neural network type 400 and neural network type 450, but the present invention is not limited thereto) and the size of the neural network operation results, but the present invention is not limited thereto.

[0067] In some embodiments of the present invention, user equipment 110 may transmit a network slice establishment request (PSP) to base station 120. The PSP PSP establishment request can be used to establish a network slice for neural network communication. The PSP PSP establishment request may include an indicator to indicate that the requested network slice is for communication supporting neural network operation results. The PSP PSP establishment request may also include a descriptor for neural network operation results. The descriptor may indicate the neural network type (e.g., neural network type 400 and neural network type 450, but the present invention is not limited thereto), the size of the neural network operation results, the average rate of transmitting the neural network operation results (e.g., how many images can be transmitted per second, but the present invention is not limited thereto), and a peak rate of transmitting the neural network operation results (e.g., the maximum number of images that can be transmitted per second, but the present invention is not limited thereto), but the present invention is not limited thereto.

[0068] Figure 5 This is a flowchart illustrating a communication session that supports neural network communication transmission according to some embodiments of the present invention. Figure 5 The flowchart shown can be applied to wireless communication system 100. For example... Figure 5 As shown, User Equipment 110 can establish a Protocol Data Unit (PDU) session for neural network communication with Base Station 120. In step S502, User Equipment 110 can send a signaling message (e.g., a Radio Resource Control (RRC) connection configuration message) to Base Station 102 to initiate a PDU session for neural network communication. In step S504, after receiving the signaling message, Base Station 120 can send an initiating User Equipment information to Core Network 130. In step S506, after receiving the initiating User Equipment information, Core Network 130 can send a PDU session resource configuration request to Base Station 120. In step S508, after receiving the PDU session resource configuration request, Base Station 120 can send a PDU session resource configuration response to Core Network 130. In step S510, a PDU session supporting neural network communication is established between User Equipment 110 and Base Station 120.

[0069] In this embodiment, the signaling information (e.g., an RRC connection configuration information) may include a binary indicator, a PDU session type field, and a descriptor. The binary indicator in the signaling information can be used to indicate that the session requires neural network communication. The PDU session type field in the signaling information is used to indicate the type of neural network (e.g., neural network type 400 and neural network type 450, but this invention is not limited thereto). The descriptor can be used to describe the characteristics of the neural network communication (e.g., the number of layers in the neural network, the size of the neural network operation result, the bit order corresponding to the neural network operation result, a sequence number, and a timestamp, but this invention is not limited thereto).

[0070] Figure 6 This is a flowchart illustrating neural network communication for dynamic grant according to some embodiments of the present invention. Figure 6 The flowchart shown can be applied to wireless communication system 100. For example... Figure 6 As shown, in step S600, user equipment 110 may transmit a scheduling request (or signaling information) to base station 120 to request the transmission of data packets containing the results of neural network operations. In step S602, after receiving the scheduling request, base station 120 may transmit a grant message to agree to the request from user equipment 110. In step S604, after receiving the grant message, user equipment 110 may transmit data packets to base station 120 to complete a cycle of neural network communication for dynamic agreement.

[0071] In step S610, when user equipment 110 wants to transmit the next data packet to base station 120, user equipment 110 can transmit a scheduling request to base station 120 to request the transmission of the next data packet containing the results of neural network operations. In step S612, after receiving the scheduling request, base station 120 can transmit an agreement message to agree to the request from user equipment 110. In step S614, after receiving the agreement message, user equipment 110 can transmit a data packet to base station 120 to complete the next cycle of neural network communication for dynamic agreement. It should be noted that... Figure 6 Only two cycles are shown, but the invention is not limited thereto.

[0072] According to some embodiments of the present invention, the scheduling requirement in steps S600 and S610 may include a binary indicator, a requirement type, a requirement descriptor, a model indicator, and the size of the neural network operation result. The binary indicator can be used to indicate that the scheduling requirement is for neural network communication. The requirement type can be used to indicate that the scheduling requirement is for Dynamic Grant. The requirement descriptor can be used to describe the characteristics of the scheduling requirement. The model indicator can be used to describe the neural network type (e.g., neural network type 400 and neural network type 450, but the present invention is not limited thereto).

[0073] Figure 7 This is a flowchart illustrating neural network communication for semi-persistent scheduling (SPS) according to some embodiments of the present invention. Figure 7 The flowchart shown can be applied to wireless communication system 100. For example... Figure 7 As shown, in step S700, user equipment 110 may transmit a scheduling request (or signaling information) to base station 120 to request the transmission of multiple uplink data packets containing neural network operation results. In step S702, after receiving the scheduling request, base station 120 may transmit a grant message to agree to the request from user equipment 110. In steps S704-S708, after receiving the grant message, user equipment 110 may periodically transmit data packets to base station 120 in each period t.

[0074] According to some embodiments of the present invention, the scheduling requirement in step S700 can be used to indicate the traffic characteristics of data transmission. Furthermore, the scheduling requirement in step S700 may include a semi-persistent scheduling (SPS) descriptor, the number of data transmissions required, and the period of uplink data packet transmission. Additionally, the scheduling requirement in S700 may also include a binary indicator, a requirement type, a requirement descriptor, a model indicator, and the size of the neural network operation result. The binary indicator can be used to indicate that the scheduling requirement is for neural network communication. The requirement type can be used to indicate that the scheduling requirement is for semi-persistent scheduling (SPS). The requirement descriptor can be used to describe the characteristics of the scheduling requirement. The model indicator can be used to describe the neural network type (e.g., neural network type 400 and neural network type 450, but the invention is not limited thereto).

[0075] Figure 8 This is a flowchart illustrating neural network communication during base station-initiated operations according to some embodiments of the present invention. Figure 8The flowchart shown can be applied to wireless communication system 100. For example... Figure 8 As shown, in step S800, base station 120 may transmit a signaling message (e.g., an RRC configuration) to user equipment 110. In steps S802 to S806, user equipment 110 may transmit multiple data packets (or one data packet) containing the results of neural network operations in response to the signaling message transmitted by base station 120.

[0076] According to some embodiments of the present invention, each data packet transmitted in steps S802 to S806 may include a packet header, a descriptor, and a data payload. The packet header in the data packet may include an indicator to indicate that the data packet includes the result of a neural network operation. The descriptor in the data packet may include parameters and settings corresponding to the result of the neural network operation. The parameters and settings corresponding to the result of the neural network operation may include the type of neural network, the number of layers contained in the neural network, the size of the result of the neural network operation, the layer corresponding to the result of the neural network operation, a sequence number, and a timestamp, but the present invention is not limited thereto. The data payload in the data packet may include the result of the neural network operation.

[0077] Figure 9 This is a flowchart illustrating a wireless communication method for neural network operations according to some embodiments of the present invention. This wireless communication method can be applied to user equipment 110 of wireless communication 100. Figure 9 As shown, in step S910, the processor of user equipment 110 performs a neural network operation to generate a complex neural network operation result, wherein the complex neural network operation result may be included in a data packet, and the complex neural network operation result is intermediate data of the neural network operation. In step S920, the transmitter of user equipment 110 may transmit the data packet to the base station, wherein the data packet may include a descriptor, and the descriptor includes parameters and settings corresponding to the neural network operation result.

[0078] According to one embodiment of the present invention, in a wireless communication method, a data packet further includes a packet header and a data payload. The packet header may include an indicator to indicate that the data packet includes the result of a neural network operation. The data payload may include the result of the neural network operation.

[0079] According to an embodiment of the present invention, in a wireless communication method, the parameters and settings corresponding to the neural network operation result may include the type of neural network, the number of layers contained in the neural network, the size of the neural network operation result, the bit level corresponding to the neural network operation result, a sequence number, and a timestamp.

[0080] According to an embodiment of the present invention, in a wireless communication method, a Protocol Data Unit (PDU) type can be set in a data packet to indicate that the data packet is used to carry the results of neural network operations.

[0081] According to an embodiment of the present invention, in a wireless communication method, a Quality of Service (QoS) type can be set in a data packet to indicate that the data packet is used to carry the neural network operation results with corresponding QoS characteristics.

[0082] According to an embodiment of the present invention, in a wireless communication method, a data packet may include a Quality of Service Flow Identifier (QFI) or a 5G Quality of Service Identifier (5QI).

[0083] According to an embodiment of the present invention, in a wireless communication method, the processor of user equipment 110 may also map the quality of service flow of neural network communication to a data radio bearer (DRB), wherein the data radio bearer can be used to provide the communication quality corresponding to the neural network operation results.

[0084] According to one embodiment of the present invention, in the wireless communication method, the transmitter of user equipment 110 may further transmit a scheduling request to base station 120 for performing neural network operations. In one embodiment, the scheduling request may include a binary indication indicating that the scheduling is for neural network operations, a request type, a request descriptor, a model identifier, and the magnitude of the neural network operation result. In another embodiment, the scheduling request may further include a semi-persistent scheduling (SPS) descriptor, the number of data transmissions required, and the uplink data packet transmission period.

[0085] According to an embodiment of the present invention, in the wireless communication method, the transmitter of the user equipment 110 may also transmit an uplink buffer status report (BSR) to the base station for neural network operation. The buffer status report may include an information descriptor, wherein the information descriptor includes the neural network type and the size of the neural network operation result.

[0086] According to an embodiment of the present invention, in a wireless communication method, the transmitter of the user equipment 110 may also transmit a network slice establishment request information to the base station for performing neural network operations. The network slice establishment request information may include an information descriptor, which may include the neural network type, the size of the neural network operation result, the average rate of transmitting the neural network operation result, and a peak rate of transmitting the neural network operation result.

[0087] According to an embodiment of the present invention, in the wireless communication method, the transmitter of the user equipment 110 may further transmit radio resource control (RRC) connection configuration information to the base station. The RRC connection configuration information may include a binary indicator indicating that the RRC connection configuration information is used for neural network communication, a Protocol Data Unit (PDU) session type field, and a descriptor. After receiving the RRC connection configuration information, the base station may establish a PDU session with the user equipment 110 for neural network communication.

[0088] According to an embodiment of the present invention, in a wireless communication method, the receiver of user equipment 110 may receive an RRC configuration from a base station, and then the transmitter of user equipment 110 may transmit data packets to the base station.

[0089] The wireless communication system 100 and wireless communication method proposed according to the present invention can improve the efficiency of neural network transmission and enhance the transmission performance of neural network transmission.

[0090] The serial numbers in this specification and the claims, such as "first," "second," etc., are for ease of explanation only and do not have any sequential relationship with each other.

[0091] The steps of the methods and algorithms disclosed in this specification can be directly applied to hardware and software modules or a combination of both by executing a processor. A software module (including execution instructions and related data) and other data can be stored in a data storage device, such as random access memory (RAM), flash memory, read-only memory (ROM), erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM), registers, hard disks, portable hard disks, optical disc read-only memory (CD-ROM), DVDs, or any other computer-readable storage media format in the art. A storage media can be coupled to a machine device, for example, such as a computer / processor (referred to as processor in this specification for convenience), which can be used to read information (such as program code) and write information to the storage media. A storage media can integrate a processor. An application-specific integrated circuit (ASIC) includes a processor and a storage media. A user equipment includes an application-specific integrated circuit. In other words, the processor and storage media are included in the user equipment in a manner that is not directly connected to the user equipment. Furthermore, in some embodiments, any product suitable for computer programs includes a readable storage medium, wherein the readable storage medium includes program code associated with one or more of the disclosed embodiments. In some embodiments, the product of computer programs may include packaging material.

[0092] The paragraphs above describe the subject using multiple layers. Clearly, the revelations in this paper can be implemented in various ways, and any particular architecture or functionality disclosed in the examples is merely a representative case. Based on the revelations in this paper, anyone skilled in this art should understand that the layers disclosed herein can be implemented independently or two or more layers can be implemented in combination.

[0093] Although this disclosure has been presented above with reference to embodiments, it is not intended to limit this disclosure. Any person skilled in the art may make some changes and modifications without departing from the concept and scope of this disclosure. Therefore, the scope of protection of this invention shall be determined by the claims.

Claims

1. A user device for neural network operations, comprising: A processor executes a neural network operation to produce a complex neural network operation result, wherein the complex neural network operation result is contained in a data packet, and the complex neural network operation result is intermediate data of the neural network operation; and A transmitter transmits the aforementioned data packets to a base station; The aforementioned data packet includes a descriptor, and the descriptor includes parameters and settings corresponding to the results of the complex neural network operation. The parameters and settings corresponding to the above complex neural network operation results include a neural network type, the number of layers in the neural network, the size of the complex neural network operation result, the bit order corresponding to the complex neural network operation result, a sequence number, and a timestamp.

2. The user equipment as claimed in claim 1, wherein the data packet further comprises: A packet header includes an indicator to indicate the aforementioned data packet; as well as A data load, including the results of the complex neural network operations described above.

3. The user equipment as claimed in claim 1, wherein a protocol data unit type is set in the data packet to indicate that the data packet is used to carry the results of the complex neural network operation.

4. The user equipment as claimed in claim 1, wherein a quality of service type is set in the data packet to indicate that the data packet is used to carry the neural network operation result having the corresponding quality of service characteristics.

5. The user equipment as claimed in claim 1, wherein the data packet includes a quality of service stream identifier or a 5G quality of service identifier.

6. The user equipment as claimed in claim 1, wherein the transmitter transmits a scheduling request to the base station for performing the neural network operation, wherein the scheduling request may include a binary indicator, a request type, a request descriptor, a model indicator, and the magnitude of the result of the complex neural network operation, wherein the binary indicator is used to indicate that the scheduling request is for the neural network operation.

7. The user equipment as claimed in claim 6, wherein the scheduling requirements further include a persistent scheduling descriptor, the number of times data transmission is required, and one cycle of an uplink data packet transmission.

8. The user equipment as claimed in claim 1, wherein the transmitter further transmits an uplink buffer status report to the base station to perform the neural network operation, wherein the buffer status report includes an information descriptor, wherein the information descriptor includes a neural network type and the size of the result of the complex neural network operation.

9. The user equipment as claimed in claim 1, wherein the transmitter further transmits a network slice establishment request information to the base station to perform the neural network operation, wherein the network slice establishment request information may include an information descriptor, wherein the information descriptor may include a neural network type, the size of the result of the complex neural network operation, an average rate for transmitting the result of the complex neural network operation, and a peak rate for transmitting the result of the complex neural network operation.

10. The user equipment of claim 1, wherein the transmitter further transmits radio resource control link setting information to the base station, wherein the radio resource control link setting information may include a binary indicator, a protocol data unit session type field, and a descriptor for indicating that the radio resource control link setting information is used for the neural network communication.

11. A wireless communication method for neural network operations, wherein the wireless communication method is applied to a user equipment, comprising: A neural network operation is executed by a processor of the aforementioned user equipment to generate a complex neural network operation result, wherein the complex neural network operation result is contained in a data packet, and the complex neural network operation result is intermediate data of the neural network operation; and The data packets are transmitted to a base station via a transmitter of the aforementioned user equipment. The aforementioned data packet includes a descriptor, and the descriptor includes parameters and settings corresponding to the results of the complex neural network operation. The parameters and settings corresponding to the above complex neural network operation results include a neural network type, the number of layers in the neural network, the size of the complex neural network operation result, the bit order corresponding to the complex neural network operation result, a sequence number, and a timestamp.

12. The wireless communication method of claim 11, wherein the data packet further comprises: A packet header includes an indicator to indicate the aforementioned data packet; as well as A data load, including the results of the complex neural network operations described above.

13. The wireless communication method of claim 11, wherein a protocol data unit type is set in the data packet to indicate that the data packet is used to carry the result of the complex neural network operation.

14. The wireless communication method of claim 11, wherein a quality of service type is set in the data packet to indicate that the data packet is used to carry the neural network operation result having the corresponding quality of service characteristics.

15. The wireless communication method of claim 11, wherein the data packet includes a quality of service stream identifier or a 5G quality of service identifier.

16. The wireless communication method of claim 11, further comprising: A scheduling request is transmitted to the base station via the aforementioned transmitter in order to perform the aforementioned neural network calculations. The scheduling requirements mentioned above may include a binary indicator, a requirement type, a requirement descriptor, a model indicator, and the size of the result of the complex neural network operation. The binary indicator is used to indicate that the scheduling requirements are for the neural network operation.

17. The wireless communication method of claim 16, wherein the scheduling requirements further include a half-persistent scheduling descriptor, the number of times data transmission is required, and one cycle of an uplink data packet transmission.

18. The wireless communication method of claim 11, further comprising: The aforementioned transmitter transmits an uplink buffer status report to the aforementioned base station in order to perform the aforementioned neural network operations. The aforementioned buffer status report information includes an information descriptor, which includes a neural network type and the size of the complex neural network operation result.

19. The wireless communication method as described in claim 11, further comprising: The aforementioned transmitter transmits a network slice creation requirement to the aforementioned base station in order to perform the aforementioned neural network operations. The network slice establishment requirements may include an information descriptor, which may include a neural network type, the size of the complex neural network operation result, an average rate for transmitting the complex neural network operation result, and a peak rate for transmitting the complex neural network operation result.

20. The wireless communication method of claim 11, further comprising: The aforementioned transmitter transmits wireless resource control connection setting information to the aforementioned base station. The aforementioned radio resource control connection setting information may include a binary indicator, a protocol data unit session type field, and a descriptor, used to indicate that the aforementioned radio resource control connection setting information is used for the aforementioned neural network communication.

Images