WO2021129591A1 - 无线通信系统中的电子设备、通信方法和存储介质 - Google Patents

无线通信系统中的电子设备、通信方法和存储介质 Download PDF

Info

Publication number
WO2021129591A1
WO2021129591A1 PCT/CN2020/138206 CN2020138206W WO2021129591A1 WO 2021129591 A1 WO2021129591 A1 WO 2021129591A1 CN 2020138206 W CN2020138206 W CN 2020138206W WO 2021129591 A1 WO2021129591 A1 WO 2021129591A1
Authority
WO
WIPO (PCT)
Prior art keywords
base station
frequency
frequency base
user equipment
communication
Prior art date
Application number
PCT/CN2020/138206
Other languages
English (en)
French (fr)
Inventor
马可
王昭诚
曹建飞
Original Assignee
索尼集团公司
马可
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 索尼集团公司, 马可 filed Critical 索尼集团公司
Publication of WO2021129591A1 publication Critical patent/WO2021129591A1/zh

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W48/00—Access restriction; Network selection; Access point selection
    • H04W48/20—Selecting an access point
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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; CALCULATING OR 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/04—Wireless resource allocation
    • H04W72/044—Wireless resource allocation based on the type of the allocated resource
    • H04W72/046—Wireless resource allocation based on the type of the allocated resource the resource being in the space domain, e.g. beams
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W72/00—Local resource management
    • H04W72/20—Control channels or signalling for resource management
    • H04W72/27—Control channels or signalling for resource management between access points
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W74/00—Wireless channel access
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W74/00—Wireless channel access
    • H04W74/002—Transmission of channel access control information
    • H04W74/006—Transmission of channel access control information in the downlink, i.e. towards the terminal
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04W—WIRELESS COMMUNICATION NETWORKS
    • H04W76/00—Connection management
    • H04W76/10—Connection setup
    • H04W76/11—Allocation or use of connection identifiers

Definitions

  • the present disclosure relates to electronic devices, communication methods, and storage media in wireless communication systems. More specifically, the present disclosure relates to electronic devices, communication methods, and storage media that use low-frequency communication to assist millimeter wave communication access in a high- and low-frequency hybrid network architecture.
  • millimeter wave can greatly enrich the available spectrum resources, which means wider bandwidth and faster transmission rate.
  • the size of the antenna used in millimeter wave communication is also on the order of millimeters. Hundreds or even thousands of millimeter wave antennas can be placed in a small space, which is conducive to the application of massive MIMO technology in actual systems. Therefore, millimeter wave technology has become one of the key technologies in 5G wireless communication systems.
  • millimeter wave communication has the defects of too small coverage and excessive power consumption.
  • the beamforming technology can be used to form a directional spatial beam to gather energy in a specific spatial direction to combat channel path fading, thereby expanding the coverage.
  • the topology of high and low frequency hybrid deployment can also be used to achieve large-scale stable coverage and provide low-rate transmission services for user equipment (UE) through low-frequency base stations (for example, traditional LTE base stations) as anchor points. Switching to a high-frequency base station (for example, a millimeter wave base station) is only considered when high-rate data transmission is required. This kind of high and low frequency mixed deployment can combine the advantages of both low frequency communication and high frequency communication.
  • the UE when deciding to access the millimeter wave base station, since the quality of the wireless channel between the UE and each millimeter wave base station is unknown, the UE cannot know which millimeter wave base station is most suitable for access. If the search for each millimeter wave base station is performed sequentially, the efficiency of access will be low. Even as the worst case, if the channel quality of any millimeter wave base station cannot meet the communication requirements, turning on the millimeter wave communication module rashly will result in waste of power consumption.
  • various aspects of the present disclosure provide solutions suitable for efficiently accessing high-frequency base stations such as millimeter-wave base stations in a high- and low-frequency hybrid network architecture.
  • an electronic device for a low-frequency base station including a processing circuit configured to obtain a channel state information (CSI) matrix based on a reference signal received from a user equipment via low-frequency communication Based on the CSI matrix, use a neural network to determine candidate high-frequency base stations suitable for high-frequency communication with the user equipment from a plurality of high-frequency base stations; determine access assistance information associated with the candidate high-frequency base station; and Sending the access assistance information to the user equipment.
  • CSI channel state information
  • an electronic device for user equipment including a processing circuit configured to send a reference signal to a low-frequency base station via a low-frequency link for the low-frequency base station to obtain channel state information (CSI ) Matrix; receiving access assistance information associated with a candidate high-frequency base station determined by a low-frequency base station, where the candidate high-frequency base station is determined by the low-frequency base station based on the CSI matrix using a neural network and is suitable for communicating with the user equipment A high-frequency base station for high-frequency communication; and using the access auxiliary information to access the candidate high-frequency base station.
  • CSI channel state information
  • an electronic device for a high-frequency base station including a processing circuit configured to receive an identification code of a user equipment from the low-frequency base station and information about the high-frequency base station available for use.
  • the information of the beam of the user equipment, wherein the beam is determined by the low-frequency base station by inputting a CSI matrix acquired based on a reference signal sent by the user equipment via a low-frequency link into a neural network; using the beam Establish high-frequency communication with the user equipment.
  • a method for training a neural network including: receiving a reference signal sent via low-frequency communication from a user equipment; acquiring a channel state information (CSI) matrix based on the reference signal; The identification information of the high-frequency base station for high-frequency communication; the parameters of the neural network are determined by using the CSI matrix as an input and the identification information of the high-frequency base station as an output for deep learning.
  • CSI channel state information
  • a communication method for a low-frequency base station based on a reference signal received from a user equipment via low-frequency communication, a channel state information (CSI) matrix is acquired; based on the CSI matrix, a neural network is used to Among the multiple high-frequency base stations, determine a candidate high-frequency base station suitable for high-frequency communication with the user equipment; determine access assistance information associated with the candidate high-frequency base station; and send the access assistance information to the User equipment.
  • CSI channel state information
  • a communication method for user equipment is provided: a reference signal is sent to a low-frequency base station via low-frequency communication for the low-frequency base station to obtain a channel state information (CSI) matrix; Access assistance information associated with a high-frequency base station, wherein the candidate high-frequency base station is a high-frequency base station that is determined by the low-frequency base station based on the CSI matrix using a neural network and is suitable for high-frequency communication with the user equipment; and uses the Access auxiliary information, access the candidate high-frequency base station.
  • CSI channel state information
  • a communication method for a high-frequency base station receiving an identification code of a user equipment and information about a beam of the high-frequency base station available for the user equipment from a low-frequency base station, wherein The beam is determined by the low-frequency base station by inputting a CSI matrix obtained based on a reference signal sent by the user equipment via a low-frequency link into a neural network; and the beam is used to establish high-frequency communication with the user equipment.
  • non-transitory computer-readable storage medium storing executable instructions that, when executed, implement any of the methods described above.
  • Figure 1 is a simplified diagram illustrating the architecture of the NR communication system
  • Figures 2A and 2B respectively illustrate the NR radio protocol architecture of the user plane and the control plane
  • Figure 3 is a schematic diagram illustrating a high-frequency and low-frequency hybrid network architecture
  • Figure 4 illustrates the frame structure used in 5G NR
  • FIG. 5 is a schematic diagram illustrating the association between the beam and the synchronization signal
  • Figure 6 illustrates the time-frequency structure of the synchronization signal block (SSB) in 5G NR;
  • FIG. 7 illustrates the relationship between the amplitude and the angle of the angle domain CSI
  • Figure 8 illustrates the relationship between the path loss and the distance between the UE and the base station
  • FIG. 9 illustrates an example of a convolutional neural network (CNN) according to the present disclosure
  • Figure 10 illustrates the convolution operation performed by the convolution kernel
  • Fig. 11 illustrates an example of pooling processing
  • FIG. 12 illustrates another example of a convolutional neural network (CNN) according to the present disclosure
  • FIG. 13 illustrates an example of a communication flow of high-frequency communication access according to the present disclosure
  • Figure 14 illustrates the communication flow chart of initial access
  • 15A-15B illustrate another example of the communication flow of high-frequency communication access according to the present disclosure
  • FIG. 16 illustrates another example of the communication flow of high-frequency communication access according to the present disclosure
  • FIG. 17 illustrates a communication flow for collecting training data according to the present disclosure
  • 21A-21B illustrate an electronic device for a low-frequency base station and a communication method thereof according to the present disclosure
  • 22A-22B illustrate an electronic device for UE and a communication method thereof according to the present disclosure
  • 23A-23B illustrate an electronic device for a high-frequency base station and a communication method thereof according to the present disclosure
  • FIG. 24 illustrates a first example of a schematic configuration of a base station according to the present disclosure
  • FIG. 25 illustrates a second example of the schematic configuration of the base station according to the present disclosure
  • FIG. 26 illustrates a schematic configuration example of a smart phone according to the present disclosure
  • FIG. 27 illustrates a schematic configuration example of a car navigation device according to the present disclosure.
  • FIG. 1 is a simplified diagram showing the architecture of a 5G NR communication system.
  • the radio access network (NG-RAN) nodes of the NR communication system include gNB and ng-eNB.
  • the gNB is a newly defined node in the 5G NR communication standard.
  • the interface is connected to the 5G core network (5GC), and provides NR user plane and control plane protocols that terminate with terminal equipment (also referred to as “user equipment”, hereinafter referred to as “UE”); ng-eNB is used to communicate with 4G LTE communication system compatible and defined node, which can be an upgraded Node B (eNB) of the LTE radio access network, connects the device to the 5G core network via the NG interface, and provides an evolved universal terrestrial radio terminal with the UE Access (E-UTRA) user plane and control plane protocol.
  • NG-RAN nodes for example, gNB, ng-eNB
  • gNB and ng-eNB are collectively referred to as "base stations”.
  • the term "base station” used in the present disclosure is not limited to the above two types of nodes, but is an example of a control device in a wireless communication system, having the full breadth of its usual meaning.
  • the “base station” may also be, for example, an eNB or a remote radio head in an LTE or LTE-A communication system.
  • Terminal (RRH) wireless access point
  • relay node drone control tower
  • control node in automated factory, or communication device or its components that perform similar control functions.
  • the term "UE" used in the present disclosure has the full breadth of its usual meaning, including various terminal devices or in-vehicle devices that communicate with a base station.
  • the UE may be a terminal device such as a mobile phone, a laptop computer, a tablet computer, an in-vehicle communication device, a drone, a sensor and an actuator in an automated factory, or an element thereof.
  • a terminal device such as a mobile phone, a laptop computer, a tablet computer, an in-vehicle communication device, a drone, a sensor and an actuator in an automated factory, or an element thereof.
  • Figure 2A shows the radio protocol stack for the user plane of the UE and the base station
  • Figure 2B shows the radio protocol stack for the control plane of the UE and the base station.
  • Layer 1 (L1) of the radio protocol stack is the lowest layer, also known as the physical layer.
  • L1 realizes various physical layer signal processing to provide transparent signal transmission function. For example, when sending data, L1 performs a series of physical layer processing on user data from the MAC layer, such as cyclic redundancy check (CRC), channel coding, rate matching, scrambling, modulation, precoding, resource mapping, etc. Mapped to the transmission channel, on the contrary, a series of inverse processing can be performed when receiving data.
  • CRC cyclic redundancy check
  • L2 is above the physical layer and is responsible for managing the wireless link between the UE and the base station.
  • L2 includes a medium access control (MAC) sublayer, a radio link control (RLC) sublayer, a packet data convergence protocol (PDCP) sublayer, and a service data adaptation protocol (SDAP) sublayer.
  • MAC medium access control
  • RLC radio link control
  • PDCP packet data convergence protocol
  • SDAP service data adaptation protocol
  • L2 includes MAC sublayer, RLC sublayer, and PDCP sublayer.
  • the physical layer provides transmission channels for the MAC sublayer, such as physical uplink shared channel (PUSCH), physical uplink control channel (PUCCH), physical random access channel (PRACH), physical downlink shared channel (PDSCH) , Physical Downlink Control Channel (PDCCH), Physical Broadcast Channel (PBCH);
  • the MAC sublayer provides logical channels for the RLC sublayer; the RLC sublayer provides RLC channels for the PDCP sublayer, and the PDCP sublayer provides radio bearers for the SDAP sublayer.
  • the radio resource control (RRC) sublayer in layer 3 (L3) is also included in the UE and the base station.
  • the RRC sublayer is responsible for obtaining radio resources (ie, radio bearers) and for configuring the lower layers using RRC signaling.
  • the non-access stratum (NAS) control protocol in the UE performs functions such as authentication, mobility management, and security control.
  • Fig. 4 shows a diagram of a frame structure in a 5G communication system.
  • the frame in NR also has a length of 10ms, including 2 half-frames of 5ms in length, and further includes 10 subframes of equal size, each of which is 1ms.
  • the frame structure in NR has a flexible structure that depends on the subcarrier spacing.
  • Each subframe has configurable Time slots, such as 1, 2, 4, 8, 16.
  • Each time slot also has configurable For the normal cyclic prefix, each slot contains 14 consecutive OFDM symbols, and for the extended cyclic prefix, each slot includes 12 consecutive OFDM symbols.
  • each time slot includes several resource blocks, and each resource block includes, for example, 12 consecutive subcarriers in the frequency domain.
  • a resource grid can be used to represent resource elements (RE) in a time slot, as shown in FIG. 4.
  • the 5G NR communication system considers three application scenarios: enhanced mobile broadband (eMBB), massive machine-type communication (mMTC), ultra-reliable low-latency communication (URLLC), each with wider bandwidth (for example, greater than 1Gbps), More user access (1 million connections per square kilometer), lower delay (less than 1 millisecond) and other characteristics, and these all rely on abundant spectrum resources.
  • eMBB enhanced mobile broadband
  • mMTC massive machine-type communication
  • URLLC ultra-reliable low-latency communication
  • FR1 about 450MHz to 6GHz, also known as the sub-6GHz frequency band
  • FR2 about 24GHz to 52GHz.
  • the electromagnetic wave wavelength in FR2 is basically millimeter-level, that is, it belongs to the so-called millimeter wave.
  • the base station and UE have many antennas that support massive MIMO, such as dozens, hundreds, or even thousands of antennas, and the parameters of the antennas can be adjusted to cause the constructive length of the radio signal at certain angles. Interference and destructive interference of radio signals at other angles to form a narrow beam to provide a strong power coverage in a specific direction. This process is also called beamforming.
  • the base station and the UE can use multiple beams with different directions to achieve cell coverage.
  • the use of massive MIMO and beamforming technology can overcome the defect of excessive millimeter wave channel path fading.
  • low frequency refers to a communication frequency band lower than the millimeter wave frequency band, such as the frequency band used by the LTE or LTE-A communication system, the FR1 frequency band (sub-6GHz frequency band) used by the NR communication system, and so on.
  • the advantages of the low frequency band are low frequency, strong diffraction ability, and good coverage effect, so it can be used as a basic coverage frequency band to realize the rapid deployment of 5G networks.
  • High frequency refers to the millimeter wave frequency band or the communication frequency band in the vicinity, such as the FR2 frequency band used by the NR communication system.
  • the advantages of the high frequency band are large bandwidth, clean spectrum, and less interference, so it can be used as a capacity supplementary frequency band to support high-rate applications.
  • Figure 3 shows a schematic diagram of the high and low frequency hybrid network architecture.
  • the communication network includes a hybrid network of more than two layers, such as a macro cell and a micro cell/pico cell.
  • Macro cells are deployed in low-frequency bands, and low-frequency base stations provide stable coverage over a wide range.
  • the UE under this network architecture can support dual connectivity, that is, in most cases, the control signal and low-rate data transmission services are completed through the low-frequency link with the low-frequency base station, and only when the UE has high-rate data transmission It’s difficult to switch to a high-frequency base station (for example, a millimeter wave base station) when it is required by the user. This is because due to power consumption and stability limitations, it is difficult for the UE to always use millimeter waves for communication, especially when the UE is resource affected. Limited equipment time.
  • a high-frequency base station for example, a millimeter wave base station
  • the UE decides whether to turn on the high-frequency communication module, the channel conditions of each high-frequency base station are unknown to the UE.
  • the UE does not know which high-frequency base station is most suitable to provide services for it, nor does it know whether the channel quality of the high-frequency base station meets the transmission requirements.
  • the UE will sequentially detect the frequency points where the cells of each high-frequency base station may reside in all frequency ranges, that is, the UE "blindly searches" for available cells and tries to access.
  • the initial access of 5G NR uses a beam management mechanism.
  • the gNB broadcasts a synchronization signal burst (SS Burst) at a regular cycle T.
  • the SS burst includes one or more synchronization signal/physical broadcast channel blocks (SSB), and each SSB corresponds to In a beam with different directions.
  • SSB synchronization signal/physical broadcast channel blocks
  • the gNB can periodically scan all predetermined directions with the SS burst beam.
  • one SS burst can be sent within a 5ms time window (half frame), repeated in a period of, for example, 20ms.
  • Fig. 6 shows the time-frequency structure of the SSB in the NR communication system.
  • SSB is composed of primary synchronization signal (PSS), secondary synchronization signal (SSS) and PBCH.
  • PSS primary synchronization signal
  • SSS secondary synchronization signal
  • PBCH primary synchronization signal
  • each SSB occupies 4 consecutive OFDM symbols
  • each SSB contains 240 consecutive subcarriers, of which PSS and SSS occupy 1 OFDM symbol and 127 PBCH spans 3 OFDM symbols and 240 subcarriers, but there is an embedded SSS part in the middle of an OFDM symbol (OFDM symbol 2 in Figure 6).
  • Each SSB in the SS burst has a corresponding index (SSB_index), therefore, SSB_index can also be used to identify the corresponding beam.
  • the SSB sent by each high-frequency base station may have mutually different frequency positions SS REF , indicated by the corresponding global synchronization channel number (GSCN).
  • SS REF and GSCN for all frequency ranges are shown in Table 1 below:
  • the UE blindly detects all possible SSB frequency positions within the frequency range used by its public land mobile network (PLMN), when the signal quality of the SSB (for example, reference signal received power (RSRP))
  • PLMN public land mobile network
  • RSRP reference signal received power
  • the UE attempts to access the cell of the high-frequency base station that sends the SSB.
  • PLMN public land mobile network
  • the present disclosure considers the use of low-frequency communication that maintains a network connection to provide information that is beneficial to high-frequency communication access, so as to improve the efficiency and quality of initial access with the high-frequency base station.
  • the low frequency base station can act as an anchor point to achieve a stable connection between the UE and the wireless communication network.
  • the low-frequency base station may be, for example, an eNB of a 4G communication system or an ng-eNB of a 5G communication system, and so on.
  • the low-frequency reference signal can be used to evaluate the channel condition.
  • the low-frequency base station can send a downlink reference signal such as channel state information reference signal (CSI-RS), and the UE can use the measurement reference signal
  • CSI-RS channel state information reference signal
  • the signal obtains the channel state information of the downlink low-frequency channel and feeds it back to the low-frequency base station, or the UE can send uplink reference signals such as sounding reference signals (SRS) or CSI-RS, and the low-frequency base station obtains the uplink low-frequency channel information by measuring the reference signals.
  • SRS sounding reference signals
  • CSI-RS channel state information reference signals
  • Channel state information In particular, in the case of time division duplexing, the channel state information of the uplink channel and the downlink channel can be obtained from measurements in a single direction.
  • the CSI obtained by measuring the reference signal may be a three-dimensional complex number matrix associated with the number of antennas of the base station, the number of antennas of the UE, and the number of subcarriers, and may also be referred to as spatial domain CSI.
  • the elements in the CSI matrix describe the status of the wireless transmission path from each antenna of the UE to each antenna of the low-frequency base station, such as information such as signal scattering and fading.
  • FIG. 7 schematically shows the relationship between the amplitude and the angle of the CSI of three UEs when the spatial domain CSI is transformed into the angular domain CSI, where the horizontal axis is an angle of 0 to ⁇ . Range, the vertical axis is the amplitude value of CSI. It can be clearly seen from Fig. 7 that the CSI has a significant peak in a certain direction. In an environment where a line-of-sight (LOS) path exists, this direction is often the LOS direction.
  • LOS line-of-sight
  • the antenna response vector is a function of the channel angle of arrival (AoA) and the channel departure angle (AoD), so CSI includes angle information such as AoA and AoD.
  • FIG. 8 shows a scatter diagram with the path loss on the vertical axis and the distance between the UE and the base station on the horizontal axis. It can be seen from Figure 8 that the amplitude attenuation of the CSI is related to the distance between the UE and the base station. The greater the distance, the smaller the amplitude of the CSI, which indicates that the CSI contains distance information.
  • the low-frequency CSI contains the position information of the UE.
  • the UE position information hidden in the CSI matrix is difficult to obtain by traditional methods, because the small number of low-frequency antennas limits the performance of the traditional angle estimation method; at the same time, it can also be seen from Figure 8 that the low-frequency CSI and millimeter wave path attenuation The relationship between is difficult to express directly with explicit expressions.
  • a prediction model based on deep learning is introduced to extract the hidden complex relationship between the CSI and the UE position, and then predict out-of-band information that assists the UE in deciding on high-frequency communication access.
  • Convolutional Neural Network is a widely used deep learning model, including convolution calculation and deep structure, with strong nonlinear fitting ability.
  • the non-linear fitting capability of, for example, a convolutional neural network when the CSI matrix acquired via low-frequency communication is input into the convolutional neural network, information about the location of the UE hidden in the low-frequency CSI matrix can be mined.
  • Convolutional neural networks have self-learning capabilities and can determine the parameters of the neural network through real data training without the need for complicated manual parameter design.
  • the out-of-band information obtained by the prediction model includes information about the high-frequency base station that is most suitable for serving the UE, information about the channel conditions between the UE and the high-frequency base station, and information about the optimal beam used by the high-frequency base station to communicate with the UE, etc. Such information is beneficial to the high frequency communication access of the UE, and is also referred to as "access assistance information" in the present disclosure.
  • the present disclosure provides different examples of prediction models based on deep learning.
  • the following description mainly takes a convolutional neural network as an example.
  • the present disclosure may not be limited to a convolutional neural network, but other types of neural networks or any suitable deep learning model can be used, as long as it can be generated based on the same input The desired output is sufficient.
  • Fig. 9 illustrates a schematic configuration diagram of a convolutional neural network according to the first example.
  • CNN consists of five parts: convolutional layer, activation function, fully connected layer, pooling layer and batch normalization layer.
  • the input data of the convolutional neural network is the low-frequency CSI matrix obtained by the low-frequency base station based on the uplink reference signal and the position data of all candidate high-frequency base stations.
  • the CSI matrix is a three-dimensional complex number matrix.
  • the location data of the high-frequency base station can be expressed as a two-dimensional matrix. Generally speaking, the location of the high-frequency base station is fixed, so the location data of the high-frequency base station can be determined in advance and stored in the memory of the low-frequency base station. When a high-frequency base station is added or removed, the location data stored in the low-frequency base station can be updated accordingly.
  • the location data of the high-frequency base station can be expressed in multiple ways, such as relative position with the low-frequency base station as a reference, absolute geographic coordinates, etc. The location data of different representations essentially only involve linear transformation, and there is no essential difference.
  • the batch normalization layer normalizes the input data matrix to a standard distribution with a mean of 0 and a variance of 1, thereby accelerating the convergence of the neural network.
  • the CNN may include one or more batch normalization layers.
  • the batch normalization layer can be used to batch normalize the input data input to the CNN and then input to the convolutional layer, or it can be used to batch normalize the intermediate data of the neural network.
  • Convolutional layer As shown in Figure 9, in the convolutional layer, a filter (ie, a convolution kernel) whose parameters can be self-learned is convolved with a data matrix to extract hidden features in the input data. Considering that the size of the convolution kernel is often much smaller than the data matrix, the convolution kernel moves on the data matrix to traverse the data matrix, and the distance moved is called the step length. In addition, in order to match the movement of the convolution kernel, the data matrix may undergo edge expansion (that is, edge filling). Convolution kernels with different parameters are used to extract different features from the data matrix, and the corresponding convolutional output is called a feature channel. In order to extract richer features, as the number of network layers deepens, the number of feature channels gradually increases.
  • FIG. 10 illustrates a schematic diagram of convolving the same data matrix with different convolution kernels, where the step size of the convolution is 2.
  • Activation function The output of the convolutional layer often passes through the activation function before being input to the next layer.
  • the activation function is usually a non-linear function, so the activation function can introduce a non-linear fitting ability for CNN.
  • Deep learning can exhibit high performance, precisely because high nonlinearity can be obtained by repeating nonlinear transformation with a multilayer structure. If there is no activation function responsible for nonlinear changes and the network only includes linear transformations, then there is only an equivalent single-layer linear transformation regardless of the number of layers, and the multiple structure is useless. Obviously, as the number of layers increases, deep learning shows stronger nonlinearity and higher performance.
  • Pooling layer The pooling layer down-samples the input matrix to reduce the amount of data and calculations of the neural network. Pooling operations include maximum pooling and average pooling.
  • Figure 11 shows a schematic diagram of two pooling treatments respectively. As shown in Figure 11, maximum pooling retains the maximum value of the data matrix, while average pooling retains the average value of the data matrix.
  • the feature vectors obtained by different pooling layers can be combined into a feature vector to facilitate subsequent network structure prediction output.
  • Fully connected layer In the fully connected layer, the input feature vector is linearly fitted to get the output.
  • the fully connected layer can control the output size, so the fully connected layer is usually used to realize the size transformation from the extracted data features to the output.
  • the convolutional neural network can output high-frequency base stations that are most suitable for providing high-frequency communication services for the UE (hereinafter also referred to as “candidate high-frequency base stations"). This means that at the current location of the UE, the high-frequency base station may provide a high-frequency link connection with the best channel condition.
  • the output form of the candidate high-frequency base station can be its identification information, such as an identifier used to uniquely identify the high-frequency base station-high-frequency base station ID, including but not limited to eNodeB ID, gNBID (which is the same as the physical layer cell ID (PCI ) Together form the NR cell ID (NCI) that uniquely identifies the cell) or a simpler base station number, as long as the low-frequency base station can associate the high-frequency base station ID with the corresponding high-frequency base station one by one.
  • an identifier used to uniquely identify the high-frequency base station-high-frequency base station ID
  • gNBID which is the same as the physical layer cell ID (PCI )
  • PCI physical layer cell ID
  • NCI NR cell ID
  • the classifier of the convolutional neural network in Fig. 9 corresponds to the candidate high-frequency base station.
  • the features extracted from the low-frequency CSI matrix and the position of the high-frequency base station are transformed into an output whose size is the number of candidate high-frequency base stations through a fully connected layer, and then through, for example, the Softmax activation function
  • the Softmax activation function normalizes the output of the fully connected layer to probability.
  • the convolutional neural network selects the class of the classifier with the highest probability as the output result of the prediction.
  • the output expression provides a ranking of the predicted priority of the high-frequency base station, that is, the greater the probability, the more likely the user is to access the high-frequency base station.
  • the convolutional neural network can output only one high-frequency base station with the highest probability, or multiple high-frequency base stations that can be sorted by probability.
  • the prediction criterion of the candidate high-frequency base station may be the path loss of the high-frequency link between the UE and the high-frequency base station.
  • the prediction standard of the convolutional neural network depends on the principle of obtaining training data, and different training data may result in the output of candidate high-frequency base stations in different meanings.
  • the convolutional neural network can also output other access auxiliary information.
  • a branch network (branch 1) of the convolutional neural network can output the path loss value corresponding to the candidate high-frequency base station.
  • the prediction of the path loss value helps to estimate whether the channel condition of the high-frequency communication link between the UE and the candidate high-frequency base station meets the connection requirements. If the path loss value indicates that the wireless channel does not meet the requirements, the UE may temporarily not turn on the high-frequency communication module, but wait for a better opportunity.
  • the predicted path loss value can be modeled as a regression problem, that is, the fully connected layer converts the extracted features into a scalar, and the scalar is directly output as the path loss value.
  • the path loss value predicted by the neural network can be expressed in dB.
  • a branch network (branch 2) of the convolutional neural network can output the optimal beam for the candidate high-frequency base station to provide high-frequency communication services for the UE.
  • the AOD and AOA of the optimal beam may be closest to the channel direction, such as the LOS direction.
  • the low-frequency base station can notify the candidate high-frequency base station of the predicted optimal beam, so that the high-frequency base station can use this beam to establish high-frequency communication with the UE, or perform further fine-tuning of the beam for the UE.
  • the convolutional neural network predicts that the optimal beam is essentially selected from a limited number of beams available in the high-frequency base station. Therefore, this type of problem can also be modeled as a multi-class problem.
  • the number of classifiers is the number of candidate beams, and the classifier type is not limited to the limited number of beams of the high-frequency base station.
  • the beam can be indexed by an associated reference signal (eg, SSB).
  • the loss function is used to measure the difference between the predicted result and the target output.
  • the gradient backpropagation algorithm is used to update the parameters of the convolutional neural network.
  • a cross-entropy loss function can be used for multi-class prediction of candidate high-frequency base stations or optimal beams
  • a Smooth-L1 loss function can be used for regression prediction of path loss values.
  • Loss ⁇ BS loss BS + ⁇ beam loss beam + ⁇ path loss path
  • loss BS , loss beam , and loss path respectively represent the loss functions of the three prediction targets of candidate high-frequency base stations, optimal beams, and path loss values, and ⁇ BS , ⁇ beam , and ⁇ path respectively represent the corresponding linear weight coefficients.
  • the convolutional neural network as the prediction model is used to predict the high-frequency base station with the smallest path loss as a candidate high-frequency base station suitable for providing high-frequency communication services for the UE.
  • the transmit power of each high-frequency base station is often different.
  • the high-frequency base station with the smallest path loss may not have the highest signal received power at the UE, and the UE usually selects the one to access by comparing the received signal power. Base station, this may lead to deviations between the predicted candidate high-frequency base station and the actual candidate high-frequency base station.
  • the convolutional neural network of the second example takes this difference in transmit power into account, and the prediction criterion of the candidate high-frequency base station may be the received power (RSRP) of the synchronization signal at the UE.
  • RSRP received power
  • FIG. 12 illustrates a schematic configuration diagram of a convolutional neural network as an example of a prediction model according to the second embodiment.
  • Convolutional neural network includes five parts: convolutional layer, activation function, fully connected layer, pooling layer and batch normalization layer. The following focuses on the differences from the convolutional neural network shown in FIG. 9.
  • the input of the convolutional neural network also includes the transmit power of each high-frequency base station.
  • the transmission power may be the power of the SSB signal broadcast by the high-frequency base station.
  • the position data of each high-frequency base station can be input into the neural network as high-frequency base station information together with the transmission power.
  • the transmit power of a high-frequency base station does not change frequently, so it can be pre-stored at the low-frequency base station like the position data.
  • the convolutional neural network in Figure 12 outputs candidate high-frequency base station IDs in the sense of RSRP.
  • the convolutional neural network further includes a branch network (branch 1) that outputs the corresponding RSRP value and a branch network (branch 2) of the optimal beam of the candidate high-frequency base station.
  • the prediction of the RSRP value helps to estimate whether the channel condition of the high-frequency communication link between the UE and the candidate high-frequency base station meets the connection requirements. If the RSRP value indicates that the wireless channel does not meet the requirements, the UE may temporarily not turn on the high-frequency communication module, but wait for a better opportunity.
  • RSRP is a continuous value
  • predicting RSRP can also be modeled as a regression problem, that is, the fully connected layer converts the extracted features into a scalar, and the scalar is directly used as RSRP Value output.
  • the cross-entropy loss function can be used for the multi-class prediction of the candidate high-frequency base station or the optimal beam
  • the Smooth-L1 loss function can be used for the regression prediction of the RSRP value.
  • linear weighting can be used to involve the overall loss function Loss, namely:
  • Loss ⁇ BS loss BS + ⁇ beam loss beam + ⁇ RSRP loss RSRP
  • loss BS , loss beam , and loss RSRP respectively represent the loss functions of the three prediction targets of candidate high-frequency base stations, optimal beams, and RSRP values, and ⁇ BS , ⁇ beam , and ⁇ RSRP respectively represent the corresponding linear weight coefficients.
  • FIG. 13 is an example of a communication flow showing the initial access of high-frequency communication.
  • the communication process shown in FIG. 13 may start from the UE sending an uplink reference signal (S1) to the low-frequency base station.
  • the uplink reference signal may be, for example, SRS, CSI-RS, etc., and is sent to the low-frequency base station via the low-frequency link.
  • the behavior of the UE sending the uplink reference signal may occur when the UE has a high-rate data transmission requirement.
  • the UE may send a high-frequency link access request (not shown in FIG. 13) to the low-frequency base station to request the low-frequency base station to provide information that facilitates access to the high-frequency link.
  • the UE may use pre-configured communication resources (for example, time-frequency resource blocks) to send separately or together the uplink reference signal and the high-frequency link access request.
  • the UE may first send a high-frequency link access request to the low-frequency base station.
  • the low-frequency base station may allocate communication resources for the UE (for example, through DCI), so that the UE may use the allocated communication resources to send an uplink reference signal.
  • the low-frequency base station may acquire the CSI matrix (S2).
  • the low-frequency base station can measure the reference signals received through its multiple antennas, and estimate the CSI matrix based on the measured values.
  • the CSI matrix can be obtained using traditional estimation methods, and will not be described in detail here.
  • the estimated CSI matrix may be in the form of a three-dimensional matrix as described above.
  • the low-frequency base station uses the acquired CSI matrix to predict the most preferable high-frequency base station for the UE (S3).
  • the low-frequency base station inputs the CSI matrix as input data into a trained prediction model, such as the convolutional neural network shown in FIG. 9 or 12.
  • a trained prediction model such as the convolutional neural network shown in FIG. 9 or 12.
  • the input data also includes the location data of all available high-frequency base stations, and these location data can be stored in the low-frequency base station in advance.
  • the input data also includes the position data and transmission power of all available high-frequency base stations, and these position data and transmission power can be stored in the low-frequency base station in advance.
  • the convolutional neural network can determine as output candidate high-frequency base stations suitable for providing high-frequency communication services for the UE. For example, depending on the prediction model used by the low-frequency base station, the candidate high-frequency base station may have the smallest high-frequency communication link with the UE. A high-frequency base station with path loss, or a high-frequency base station with the largest received signal power at the UE.
  • the low-frequency base station may determine the access assistance information associated with the candidate high-frequency base station (S4). Although it is a feasible way to directly notify the UE of the ID of the candidate high-frequency base station, preferably, the low-frequency base station can determine information that is more convenient to use when the UE accesses the candidate high-frequency base station. ⁇ Into the process.
  • the access assistance information determined by the low-frequency base station includes information that facilitates the UE to identify the synchronization signal of the candidate high-frequency base station, for example, the frequency position SS REF of the SSB broadcast by the candidate high-frequency base station.
  • the frequency position SS REF of SSB can also be indicated by GSCN.
  • the low-frequency base station may send the determined access assistance information (for example, GSCN) to the UE (S5).
  • the UE can use such access assistance information to access the candidate high-frequency base station (S6).
  • the UE may turn on the high-frequency communication module when triggered by receiving the access assistance information associated with the candidate high-frequency base station.
  • the initial access process between the UE and the base station is described below with reference to FIG. 14.
  • the cell search is a process in which the UE obtains time and frequency synchronization with the cell and detects the physical layer cell ID of the cell.
  • the UE performs a cell search by receiving an SSB.
  • the UE can directly search for the synchronization signal at the frequency location indicated in the access assistance information.
  • the UE can receive the SS burst periodically sent by the candidate high-frequency base station , Including one or more SSBs corresponding to different beams.
  • the SSB that directly locates the candidate high-frequency base station is obviously more efficient.
  • the UE can detect whether the signal quality (for example, RSRP) of each SSB meets the threshold requirement, or detect the SSB with the best signal quality.
  • the UE can decode the SSB that meets the threshold requirement to synchronize to the downlink timing, for example, through the following steps:
  • PCI physical layer cell ID
  • the position of the DMRS of the PBCH can be determined.
  • the position offset of the DMRS is PCI mod 4;
  • the SSB index (i SSB ) and half frame information (n hf ) can be obtained, and the UE can obtain 10 ms frame synchronization.
  • the UE can receive cell system information, such as a master information block (MIB) and various system information blocks (SIB), at an appropriate position in the downlink frame.
  • the system information may be periodically broadcast by the base station through a broadcast channel (for example, broadcast channel PBCH, shared channel PDSCH, etc.), and may include information necessary for the UE to access the base station, such as random access related information.
  • a broadcast channel for example, broadcast channel PBCH, shared channel PDSCH, etc.
  • the UE needs to perform a random access procedure.
  • the UE may notify the base station of its access behavior by sending a random access preamble (for example, included in MSG-1) to the candidate high-frequency base station.
  • the transmission of the random access preamble enables the base station to estimate the uplink timing advance of the terminal equipment (Timing Advance).
  • the base station may notify the UE of the aforementioned timing advance by sending a random access response (for example, included in MSG-2) to the UE.
  • the UE can achieve uplink cell synchronization through the timing advance.
  • the random access response may also include uplink resource information, and the UE may use the uplink resource in operation 104.
  • the UE may send the UE identifier and possibly other information (for example, included in MSG-3) through the above-mentioned scheduled uplink resource.
  • the base station can determine the contention resolution result through the UE identifier.
  • the base station can inform the UE of the contention resolution result (e.g. included in MSG-4).
  • the competition is successful, the UE successfully accesses the base station, and the random access process ends; otherwise, the UE needs to repeat the random access process from operations 102 to 105.
  • the UE establishes high-frequency communication with the candidate high-frequency base station, and can perform subsequent communication.
  • the low-frequency base station can respectively determine the access assistance information associated with each high-frequency base station (S4), and send it to the UE (S5 ).
  • the UE can sequentially search for corresponding SS bursts according to the priority of these candidate high-frequency base stations. If none of the SSBs in the SS burst of the candidate high-frequency base station with the highest priority (that is, the highest probability) meets the threshold requirement, the UE can search for the SS burst of the candidate high-frequency base station with the second highest priority. analogy. If no SSB that meets the requirements is found, this initial access fails.
  • FIGS. 15A and 15B Another example of the communication process of high-frequency communication access according to the present disclosure will be described below with reference to FIGS. 15A and 15B.
  • the differences from the communication flow in FIG. 13 will be mainly described, and the rest of the same parts will not be described repeatedly.
  • the low-frequency base station uses the convolutional neural network to predict the candidate high-frequency base station as well as the path loss value or RSRP value corresponding to the candidate high-frequency base station, which depends on the low-frequency base station. Whether the base station uses the convolutional neural network of the first example or the second example. The predicted path loss value or RSRP value can be used to determine whether the predicted candidate high-frequency base station is worthy of access.
  • the convolutional neural network used in S31 may include a branch network (branch 1 in FIG. 9 or 12) for predicting the path loss value or the RSRP value.
  • the low-frequency base station may compare the path loss value (or RSRP value) output by the convolutional neural network with a predetermined threshold (S40). In the case that the path loss value is lower than the predetermined threshold (or the RSRP value exceeds the predetermined threshold), the low-frequency base station can consider that the predicted preferred high-frequency base station can provide a high-frequency communication link that meets the requirements, and determine the associated high-frequency communication link in S4 Access auxiliary information to facilitate the initial access of the UE. In response to receiving the access assistance information, the UE turns on the high-frequency communication module and starts the initial access process.
  • a predetermined threshold S40
  • the low-frequency base station can consider that the predicted preferred high-frequency base station can provide a high-frequency communication link that meets the requirements, and determine the associated high-frequency communication link in S4 Access auxiliary information to facilitate the initial access of the UE.
  • the UE turns on the high-frequency communication module and starts the initial access process.
  • the low-frequency base station may generate the turn-on instruction and send it to the UE together with the access assistance information (shown in brackets in S5 of FIG. 15A) to instruct the UE to turn on the high-frequency communication mode and start the initial access process.
  • the access assistance information shown in brackets in S5 of FIG. 15A
  • the low-frequency base station can consider that the predicted candidate high-frequency base station may not be able to provide a high-frequency communication link that meets the requirements, and the following will not be performed. step. At this time, the low-frequency base station may instead send an indication to the UE that there is currently no suitable high-frequency base station for access.
  • the low-frequency base station may send the determined access assistance information together with the path loss value (or RSRP value) to the UE (S51).
  • the UE can compare the received path loss value (or RSRP value) with a predetermined threshold. When the path loss value is lower than the predetermined threshold (or the RSRP value exceeds the predetermined threshold), the UE can decide to turn on the high-frequency communication mode and pass The initial access process connects to the candidate high-frequency base station.
  • the UE may decide that there is no suitable high-frequency base station for access at present, and try to access at a subsequent time, for example, repeating Figure 15B ⁇ steps S1 to S6.
  • the above-mentioned predetermined threshold for comparison with the path loss value may be an adjustable parameter.
  • the predetermined threshold used by the low-frequency base station or the UE may depend on various factors, such as: the current power of the UE, the more the remaining power of the UE, the higher the predetermined threshold may be, thereby increasing the chance of high-frequency access; the connection preference of the UE, for example
  • the UE or its users can be set to connect to high-frequency base stations with good wireless channel conditions (ie, low path loss) as much as possible; the transmission success rate of high-frequency communication, for example, based on the result of the last high-frequency access or high-frequency communication
  • the predetermined threshold can be dynamically adjusted according to the data transmission success rate; the business urgency, for example, if the business requires high-frequency communication immediately, the UE can temporarily increase the predetermined threshold; and so on.
  • the predetermined threshold used for comparison with the RSRP value may also be an adjustable parameter.
  • the low-frequency base station or UE may adjust this according to, for example, the current power of the UE, connection preference, transmission success rate of high-frequency communication, and business urgency.
  • the predetermined threshold It should be understood that the predetermined threshold regarding path loss and the predetermined threshold regarding RSRP are different thresholds.
  • the low-frequency base station may use a convolutional neural network to predict the optimal beam of the candidate high-frequency base station for the UE.
  • the optimal beam may correspond to the SSB in the SS burst sent by the candidate high-frequency base station.
  • the access assistance information determined by the low-frequency base station in S4 may include information about the time-frequency resources of the SSB, such as the frequency position (SS REF ) and the index (SSB_index) of the SSB.
  • the convolutional neural network can output more than one (n>1) optimal beams according to the predicted probability. Therefore, in an example, according to the order of the probability of each beam from high to low, the low-frequency base station can determine the time-frequency resource information containing the corresponding n SSBs and send it to the UE, that is, one of the n SSBs Have different priorities.
  • the UE can directly detect the corresponding SSB and estimate its signal quality (such as RSRP). If the SSB meets the threshold requirement, it will continue to decode the SSB to try to access the candidate high frequency
  • the base station is as described above with reference to FIG. 13. If the SSB does not meet the threshold requirement, the UE can continue to detect other SSBs with lower priority on the same frequency location. If no SSB that meets the requirements is found, the search range is expanded to other candidate high-frequency base stations or other possible frequency locations. Through this priority search method, the search range of the UE can be further reduced in the time domain, which improves the efficiency of initial access.
  • the low-frequency base station may also notify the corresponding candidate high-frequency base station of the predicted beam and the identification code of the UE through the interface between the base stations (for example, the Xn interface) (S53).
  • the identification code of a UE refers to identification information that uniquely identifies the UE, such as an International Mobile Subscriber Identity (IMSI).
  • IMSI International Mobile Subscriber Identity
  • the low-frequency base station can learn the user's identification code. Therefore, when the UE accesses the candidate high-frequency base station, the high-frequency base station can use the predicted optimal beam to communicate with the UE. For example, in the uplink random access process as shown in FIG. 14, the high-frequency base station can use the optimal beam to transmit MSG-1 and/or MSG-4, etc.
  • the high-frequency base station can further refine the beam through beam training. Specifically, the high-frequency base station preferentially scans the multiple narrow beams contained in the beam, receives the signal quality (such as RSRP) of each narrow beam reported by the UE, and determines the narrow beam with the highest quality as the one used for subsequent data transmission. Beam. If the received signal quality of these narrow beams does not meet the requirements, continue to try beam directions adjacent to the optimal beam until the signal quality meets the requirements or all beam directions have been traversed. Therefore, using the optimal beam predicted by the neural network can narrow the range of the scanning beam and effectively reduce the overhead of beam training.
  • the signal quality such as RSRP
  • the low-frequency base station can send the beam IDs (for example, the identifiers of the corresponding reference signals) sorted from high to low with the predicted probability to the candidate high-frequency base stations.
  • the high-frequency base station can scan these beams in sequence to improve the efficiency of beam training.
  • the low-frequency base station uses the convolutional neural network to predict the candidate high-frequency base station + path loss (or RSRP) and the candidate high-frequency base station + the optimal beam
  • path loss or RSRP
  • the low-frequency base station can simultaneously predict the candidate high-frequency base station, Path loss (or RSRP) and optimal beam.
  • the process of initial access of high-frequency communication can be understood in conjunction with FIG. 15A or 15B and FIG. 16, and the description will not be repeated here.
  • Predictive models such as convolutional neural networks are deep learning models that can learn from themselves.
  • the low-frequency base station can collect a large amount of training data from the UE and the high-frequency base station to determine or update the parameters of the convolutional neural network.
  • Fig. 17 shows a communication flow chart for collecting training data of a convolutional neural network.
  • the UE that accesses the high-frequency communication link can periodically report the identification information of the high-frequency base station with which it communicates as a model output of the low-frequency base station training convolutional neural network.
  • the convolutional neural network of the first example shown in FIG. 9 in order to ensure the accuracy of the prediction of the convolutional neural network, it is expected that the high-frequency base station reported by the UE has the smallest path loss.
  • the convolutional neural network of the second example shown in FIG. 12 it is expected that the high-frequency base station reported by the UE has the largest RSRP.
  • the UE also periodically sends a low-frequency reference signal to the low-frequency base station, and the low-frequency base station obtains a low-frequency CSI matrix based on the reference signal, which is used as a model input for training a convolutional neural network.
  • the UE can also periodically estimate the path loss (or RSRP) on the current high-frequency communication link and report it to the low-frequency base station as a branch network for training the convolutional neural network (branch 1 in Figure 9 or 12). ) Model output.
  • path loss or RSRP
  • the high-frequency base station that performs high-frequency communication with the UE may periodically notify the low-frequency base station of the identification code of the UE communicating with it and the beam used for the UE through the interface between the base stations (for example, the Xn interface), as a training convolution
  • the model output of the branch network of the neural network (branch 2 in Fig. 9 or 12).
  • the low-frequency base station will pair the training data collected from the UE and the high-frequency base station according to the UE’s identification code, such as the CSI matrix associated with the same UE, the high-frequency base station ID, the path loss value (or RSRP value), and the value used by the high-frequency base station. Beam etc.
  • the position data and transmission power of all candidate high-frequency base stations can also be model input data of the corresponding convolutional neural network, and this data can be acquired and stored in advance.
  • the real data collected by the low-frequency base station is used to learn the parameters of the neural network, such as the convolution kernel of each convolutional layer, so that the convolutional neural network can obtain the predictive ability.
  • a convolutional neural network is trained on a large amount of data and then deployed to a low-frequency base station for actual prediction.
  • the online learning strategy can also be applied to the update of the convolutional neural network, that is, after the deployment of the convolutional neural network, the low-frequency base station still collects training data in real time and trains the prediction model. Online learning strategies allow predictive models to adapt to changes in the communication environment.
  • the training scheme can include the following steps:
  • the neural network takes accurate prediction of the high-frequency base station ID as the optimization goal and trains all Output shared network structure parameters to fully extract the basic features common to all outputs. After the pre-training, each output will further learn the branch network based on the basic features that have been extracted.
  • the training data collected from one or more low-frequency base stations can be used to determine the parameters of the convolutional neural network, and the preliminarily trained convolutional neural network can be applied to more Multiple low-frequency base stations.
  • the convolutional neural networks used in all low-frequency base stations can be initialized in advance with training data collected from some low-frequency base stations.
  • each low-frequency base station can use the position data and/or transmission power of the high-frequency base station in its respective coverage area and the acquired low-frequency CSI matrix as model input, and continuously optimize the parameters of the neural network through online learning.
  • Online learning mainly learns environmental information such as obstructions and reflectors around the current low-frequency base station, so as to improve the adaptability of the parameters to the communication environment.
  • Low frequency base station coverage radius 500 Number of low frequency base station antennas n 64 Number of low frequency UE antenna m 4 Number of low-frequency UE sub-carriers k 288 Number of millimeter wave base stations b 5 Beamforming codebook size for millimeter wave base stations 4 SNR/dB 5 ⁇ 25, randomly generated
  • the millimeter wave path loss model is:
  • d represents the distance between the UE and the millimeter wave base station
  • f c represents the millimeter wave center frequency
  • f c 28 GHz
  • ⁇ s represents shadow fading
  • its distribution obeys a Gaussian distribution with a mean of 0 and a variance of 4.
  • FIG. 18 shows an example of the convolutional neural network of the first example
  • FIG. 19 shows an example of the convolutional neural network of the second example.
  • the input of the convolutional neural network in Figure 18 is: Is the low-frequency CSI matrix, k is the number of low-frequency UE sub-carriers, m is the number of low-frequency UE antennas, and n is the number of low-frequency base station antennas.
  • B is the position of the candidate millimeter wave base station. The positions of the millimeter wave base station expressed in rectangular coordinates are successively spliced into a vector input with a length of 2b, and b is the number of candidate base stations; the vector input is regarded as a two-dimensional with a dimension of 1.
  • Matrix input i.e., ).
  • the input of the convolutional neural network in Figure 19 is: Is the low-frequency CSI matrix, k is the number of low-frequency UE sub-carriers, m is the number of low-frequency UE antennas, and n is the number of low-frequency base station antennas.
  • Table 3 The meaning of symbols or numbers in convolutional neural networks
  • the convolutional neural network in FIG. 18 In addition to outputting the ID of the millimeter wave base station with the smallest path loss, the convolutional neural network in FIG. 18 also outputs the corresponding path loss value and the beam of the millimeter wave base station.
  • the convolutional neural network in Figure 19 In addition to outputting the ID of the millimeter wave base station with the largest RSRP, the convolutional neural network in Figure 19 also outputs the corresponding RSRP value and the beam of the millimeter wave base station.
  • 20A-20C show schematic diagrams of the effect of using the prediction result of the convolutional neural network in FIG. 18 to assist high-frequency communication access.
  • Figure 20A shows the proportion of millimeter wave base stations with the smallest actual path loss in the top X positions with the highest predicted probability in the predicted results. It can be seen from the figure that in 67.1% of the samples, the millimeter-wave base station with the smallest path loss can be accurately predicted as a candidate high-frequency base station. At the same time, in more than 90% of the samples, the millimeter-wave base station with the smallest path loss can actually be among the two high-frequency base stations with the highest predicted probability.
  • FIG. 20B shows the cumulative distribution function of the absolute error between the predicted millimeter wave path loss and the actual path loss. It can be seen from the figure that more than 70% of the predictions have an absolute error within 5dB, and about 92.5% of the predictions have an absolute error within 10dB. At the same time, the average absolute error is 5.1dB. This shows that it is feasible to predict the minimum path loss between the user and the surrounding millimeter wave base station based on low-frequency CSI.
  • Figure 20C shows the ratio of the actual optimal beam in the top X with the highest predicted probability in the prediction result.
  • the beam actually used can be accurately predicted as the optimal beam of the high-frequency base station; at the same time, the beam actually used is in the top two positions with the highest predicted probability by about 75% of the predicted results.
  • the simulation results show that preferential selection of the optimal beam obtained by prediction can effectively reduce the overhead of beam search.
  • FIG. 21A is a block diagram illustrating the electronic device 100 according to the present disclosure.
  • the electronic device 100 may be a low-frequency base station or a component thereof.
  • the electronic device 100 includes a processing circuit 101.
  • the processing circuit 101 at least includes a CSI matrix acquisition unit 102, a candidate high-frequency base station determination unit 103, an access assistance information determination unit 104, and an access assistance information sending unit 105.
  • the processing circuit 101 may be configured to execute the communication method shown in FIG. 21B.
  • the processing circuit 101 may refer to various implementations of a digital circuit system, an analog circuit system, or a mixed signal (combination of analog signal and digital signal) circuit system that performs functions in a low-frequency base station.
  • the CSI matrix acquisition unit 102 of the processing circuit 101 is configured to acquire the low-frequency CSI matrix based on the reference signal received from the UE via the low-frequency communication, that is, perform step S101 in FIG. 21B.
  • the UE may send reference signals, such as CSI-RS or SRS, to the low-frequency base station when it needs to access the high-frequency base station.
  • the candidate high-frequency base station determination unit 103 is configured to use a neural network to determine candidate high-frequency base stations suitable for high-frequency communication with the UE from a plurality of high-frequency base stations based on the CSI matrix acquired by the CSI matrix acquisition unit 102, that is, perform FIG. 21B Step S102 in.
  • the neural network may be a convolutional neural network as shown in FIG. 9 or FIG. 12.
  • the determined candidate high-frequency base station may be a high-frequency base station predicted to have the smallest path loss for high-frequency communication with the UE, or may be a high-frequency base station predicted to have the highest received power at the UE for high-frequency communication with the UE .
  • the access assistance information determining unit 104 is configured to determine the access assistance information associated with the candidate high-frequency base station, that is, perform step S103 in FIG. 21B.
  • the access assistance information may include the frequency position of the SSB broadcast by the candidate high-frequency base station or the GSCN indicating the frequency position of the SSB.
  • the access assistance information may also include the path loss value or RSRP value corresponding to the candidate high-frequency base station predicted by the branch network of the neural network.
  • the access assistance information can also include the frequency position of the SSB corresponding to the optimal beam (or GSCN indicating the frequency position) and index.
  • the access assistance information sending unit 105 is configured to send the access assistance information determined by the access assistance information determination unit 104 to the UE, that is, perform step S104 in FIG. 21B. These access assistance information will help the UE to decide whether to access the high-frequency base station, and can narrow the scope of the UE to search for the SSB, thereby improving the efficiency and quality of the initial access of the high-frequency communication.
  • the electronic device 100 may also include a communication unit 106.
  • the communication unit 106 may be configured to communicate with the UE under the control of the processing circuit 101.
  • the communication unit 106 may be implemented as a transceiver, including communication components such as an antenna array and/or a radio frequency link.
  • the communication unit 106 is drawn with a dashed line because it can also be located outside the electronic device 100.
  • the electronic device 100 may also include a memory 107.
  • the memory 107 can store various data and instructions, such as programs and data used for the operation of the electronic device 100, various data generated by the processing circuit 101, various control signaling or service data sent or received by the communication unit 106, and the like.
  • the memory 107 is drawn with a dashed line because it can also be located inside the processing circuit 101 or outside the electronic device 100.
  • FIG. 22A is a block diagram illustrating an electronic device 200 according to the present disclosure.
  • the electronic device 200 may be a UE or a component thereof.
  • the electronic device 200 includes a processing circuit 201.
  • the processing circuit 201 at least includes a reference signal sending unit 202, an access auxiliary information receiving unit 203, and an access unit 204.
  • the processing circuit 201 may be configured to execute the communication method shown in FIG. 22B.
  • the processing circuit 201 may refer to various implementations of a digital circuit system, an analog circuit system, or a mixed signal (combination of analog signal and digital signal) circuit system that performs a function in the UE.
  • the reference signal sending unit 202 may be configured to send a reference signal to the low-frequency base station via the low-frequency link for the low-frequency base station to obtain the CSI matrix, that is, perform step S201 in FIG. 22B.
  • the reference signal sent by the reference signal sending unit 202 may be a low-frequency reference signal such as CSI-RS or SRS.
  • the access assistance information receiving unit 203 may be configured to receive the access assistance information determined by the low-frequency base station and associated with the candidate high-frequency base station, that is, perform step S202 in FIG. 22B.
  • the candidate high-frequency base station is determined by the low-frequency base station inputting the acquired CSI matrix into the convolutional neural network as shown in FIG. 9 or 12.
  • the candidate high-frequency base station is predicted to be a high-frequency base station suitable for high-frequency communication with the UE, for example, the high-frequency communication with the UE has the smallest path loss, or the high-frequency communication with the UE has the maximum received power at the UE.
  • the access assistance information received by the access assistance information receiving unit 203 may include the frequency position and/or index of the SSB broadcast by the candidate high frequency base station.
  • the access unit 204 may be configured to access the candidate high-frequency base station based on the received access assistance information, that is, perform step S203 in FIG. 22B.
  • the access unit 204 may search for one or more SSBs sent by the candidate high-frequency base station in the time-frequency resources indicated by the access assistance information, and obtain downlink synchronization and uplink synchronization by decoding the SSB.
  • the initial access using the access assistance information has higher efficiency.
  • the electronic device 200 may further include a communication unit 206.
  • the communication unit 206 may be configured to communicate with the base station under the control of the processing circuit 201.
  • the communication unit 206 may be implemented as a transmitter or a transceiver, including communication components such as an antenna array and/or a radio frequency link.
  • the communication unit 206 is drawn with a dashed line because it can also be located outside the electronic device 200.
  • the electronic device 200 may also include a memory 207.
  • the memory 207 may store various data and instructions, programs and data for the operation of the electronic device 200, various data generated by the processing circuit 201, data to be transmitted by the communication unit 207, and the like.
  • the memory 207 is drawn with a dashed line because it can also be located inside the processing circuit 201 or outside the electronic device 200.
  • FIG. 23A is a block diagram illustrating an electronic device 300 according to the present disclosure.
  • the electronic device 300 may be a high-frequency base station or a component thereof.
  • the electronic device 300 includes a processing circuit 301.
  • the processing circuit 301 at least includes a receiving unit 302 and a communication establishing unit 303.
  • the processing circuit 301 may be configured to execute the communication method shown in FIG. 23B.
  • the processing circuit 301 may refer to various implementations of a digital circuit system, an analog circuit system, or a mixed signal (combination of analog signal and digital signal) circuit system that performs functions in the base station equipment.
  • the receiving unit 302 may be configured to receive the identification code of the UE and the information about the beams available for the UE of the high-frequency base station from the low-frequency base station, that is, perform step S301 in FIG. 23B.
  • the beam is determined by the low-frequency base station by inputting the CSI matrix obtained based on the reference signal sent by the UE via the low-frequency link into the neural network.
  • the neural network may be a convolutional neural network as shown in FIG. 9 or 12, which includes a branch network for predicting the optimal beam of a high-frequency base station.
  • the communication establishing unit 303 may be configured to use the beam to establish high frequency communication with the UE, that is, perform step S302 in FIG. 23B. For example, when the UE initially accesses the high-frequency base station, the communication establishment unit 303 of the high-frequency base station may use the beam recommended by the low-frequency base station to send signaling to the UE. In addition, the communication establishment unit 303 may also refine the beam through beam training.
  • the electronic device 300 may further include a communication unit 306.
  • the communication unit 306 may be configured to communicate with the base station under the control of the processing circuit 301.
  • the communication unit 306 may be implemented as a transmitter or transceiver, including communication components such as an antenna array and/or a radio frequency link.
  • the communication unit 306 is drawn with a dashed line because it can also be located outside the electronic device 300.
  • the electronic device 300 may further include a memory 307.
  • the memory 307 may store various data and instructions, programs and data for the operation of the electronic device 300, various data generated by the processing circuit 301, data to be transmitted by the communication unit 307, and the like.
  • the memory 307 is drawn with a dashed line because it can also be located inside the processing circuit 301 or outside the electronic device 300.
  • each of the above-mentioned units may be implemented as an independent physical entity, or may also be implemented by a single entity (for example, a processor (CPU or DSP, etc.), an integrated circuit, etc.).
  • the processing circuits 101, 201, or 301 described in the above embodiments may include, for example, circuits such as integrated circuits (ICs), application specific integrated circuits (ASICs), parts or circuits of individual processor cores, and entire processors.
  • a core a separate processor, a programmable hardware device such as a Field Programmable Array (FPGA), and/or a system including multiple processors.
  • the memory 107, 207, or 307 may be a volatile memory and/or a non-volatile memory.
  • the memory 107, 207, or 307 may include, but is not limited to, random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), read only memory (ROM), flash memory.
  • each of the above-mentioned units may be implemented as an independent physical entity, or may also be implemented by a single entity (for example, a processor (CPU or DSP, etc.), an integrated circuit, etc.).
  • An electronic device for a low-frequency base station comprising: a processing circuit configured to: acquire a channel state information (CSI) matrix based on a reference signal received from a user equipment via low-frequency communication; based on the CSI matrix, use The neural network determines from a plurality of high-frequency base stations a candidate high-frequency base station suitable for high-frequency communication with the user equipment; determines access assistance information associated with the candidate high-frequency base station; and sends the access assistance information To the user equipment.
  • CSI channel state information
  • the processing circuit is further configured to: input the position data of the CSI matrix and the plurality of high-frequency base stations into the neural network to determine the The high-frequency base station with the smallest path loss for high-frequency communication of the user equipment is used as a candidate high-frequency base station.
  • processing circuit is further configured to: input the CSI matrix and the position data and transmission power of the multiple high-frequency base stations into the neural network, The high-frequency base station with the largest received power for high-frequency communication at the user equipment is determined as a candidate high-frequency base station.
  • the processing circuit is further configured to: use the neural network to determine the beam of the candidate high-frequency base station for the user equipment; The identification code of and the information of the beam are notified to the candidate high-frequency base station.
  • the access assistance information includes the frequency position of the synchronization signal/physical broadcast channel block (SSB) of the candidate high-frequency base station.
  • SSB synchronization signal/physical broadcast channel block
  • the access assistance information includes the frequency position and index of the synchronization signal/physical broadcast channel block (SSB) of the candidate high-frequency base station.
  • SSB synchronization signal/physical broadcast channel block
  • the electronic device wherein the processing circuit is further configured to: receive from a user equipment that performs high-frequency communication with any one of the plurality of high-frequency base stations via low-frequency communication The sent reference signal and the identification information of any high-frequency base station; by using the CSI matrix obtained based on the reference signal and the position data of the any high-frequency base station as input and using the any high-frequency base station The identification information of is used as output, and the parameters of the neural network are updated.
  • the electronic device wherein the processing circuit is further configured to: receive from a user equipment that performs high-frequency communication with any one of the plurality of high-frequency base stations.
  • the electronic device wherein the processing circuit is further configured to: receive an identification of the user equipment from a high-frequency base station that performs high-frequency communication with the user equipment among the plurality of high-frequency base stations Code and information about the beam used for the high-frequency communication; by using the information about the beam used for the high-frequency communication as an output, the parameters of the branch network of the neural network are updated.
  • An electronic device for user equipment comprising: a processing circuit configured to: send a reference signal to a low-frequency base station via low-frequency communication for the low-frequency base station to obtain a channel state information (CSI) matrix; and receive a channel state information (CSI) matrix determined by the low-frequency base station Access assistance information associated with a candidate high-frequency base station, where the candidate high-frequency base station is a high-frequency base station that is determined by the low-frequency base station based on the CSI matrix using a neural network and is suitable for high-frequency communication with the user equipment; and Using the access assistance information, access the candidate high-frequency base station.
  • CSI channel state information
  • the candidate high-frequency base station is a high-frequency base station predicted by the neural network to have the highest received power at the user equipment for high-frequency communication with the user equipment.
  • the processing circuit is further configured to: receive a path loss value corresponding to a candidate high-frequency base station from a low-frequency base station; and when the path loss value is lower than a predetermined threshold Next, turn on the high-frequency communication module to access candidate high-frequency base stations.
  • the electronic device wherein the processing circuit is further configured to: receive a received power value corresponding to a candidate high-frequency base station from a low-frequency base station; and when the received power value exceeds a predetermined threshold , Turn on the high-frequency communication module to access candidate high-frequency base stations.
  • the access assistance information includes the frequency position of the synchronization signal/physical broadcast channel block (SSB) of the candidate high-frequency base station.
  • SSB synchronization signal/physical broadcast channel block
  • the access assistance information further includes the index of the SSB of the candidate high-frequency base station.
  • the electronic device according to 19) or 20), wherein the processing circuit is further configured to adjust the predetermined threshold according to the current power of the user equipment, connection preference, and transmission success rate of high-frequency communication.
  • An electronic device for a high-frequency base station comprising: a processing circuit configured to: receive an identification code of a user equipment and information about a beam of the high-frequency base station that can be used for the user equipment from the low-frequency base station , Wherein the beam is determined by the low-frequency base station by inputting the CSI matrix obtained based on the reference signal sent by the user equipment via the low-frequency link into the neural network; Frequency communication.
  • a method of training a neural network comprising: receiving a reference signal sent via low-frequency communication from a user equipment; acquiring a channel state information (CSI) matrix based on the reference signal; The identification information of the high-frequency base station; by using the CSI matrix as an input and the identification information of the high-frequency base station as an output for deep learning, the parameters of the neural network are determined.
  • CSI channel state information
  • a communication method comprising: acquiring a channel state information (CSI) matrix based on a reference signal received from a user equipment via low-frequency communication; based on the CSI matrix, using a neural network to determine a suitable signal from a plurality of high-frequency base stations A candidate high-frequency base station for high-frequency communication with the user equipment; determining access assistance information associated with the candidate high-frequency base station; and sending the access assistance information to the user equipment.
  • CSI channel state information
  • a communication method comprising: sending a reference signal to a low-frequency base station via low-frequency communication for the low-frequency base station to obtain a channel state information (CSI) matrix; receiving access assistance information associated with a candidate high-frequency base station determined by the low-frequency base station Wherein the candidate high-frequency base station is a high-frequency base station suitable for high-frequency communication with the user equipment determined by the low-frequency base station based on the CSI matrix using a neural network; and using the access auxiliary information to access the Candidate high frequency base station.
  • CSI channel state information
  • a communication method comprising: receiving an identification code of a user equipment from a low-frequency base station and information about a beam of the high-frequency base station that can be used for the user equipment, wherein the beam is the low-frequency base station through
  • the CSI matrix obtained by the user equipment via the reference signal sent by the low frequency link is input into the neural network and determined; and the high frequency communication with the user equipment is established by using the beam.
  • a non-transitory computer-readable storage medium storing executable instructions that, when executed, implement the method described in any one of 27) to 33).
  • the electronic devices 100, 300 may be implemented as various base stations or installed in base stations, and the electronic device 200 may be implemented as various user equipments or installed in various user equipments.
  • the communication method according to the embodiments of the present disclosure may be implemented by various base stations or user equipment; the methods and operations according to the embodiments of the present disclosure may be embodied as computer-executable instructions, stored in a non-transitory computer-readable storage medium, and It can be executed by various base stations or user equipment to realize one or more of the above-mentioned functions.
  • the technology according to the embodiments of the present disclosure can be made into various computer program products, which can be used in various base stations or user equipment to implement one or more functions described above.
  • the base station mentioned in this disclosure can be implemented as any type of base station, preferably, such as macro gNB and ng-eNB defined in the 5G NR standard of 3GPP.
  • the gNB may be a gNB covering a cell smaller than a macro cell, such as pico gNB, micro gNB, and home (femto) gNB.
  • the base station may be implemented as any other type of base station, such as NodeB, eNodeB, and base transceiver station (BTS).
  • the base station may also include: a main body configured to control wireless communication, and one or more remote wireless headends (RRH), wireless relay stations, drone towers, control nodes in automated factories, etc., arranged in different places from the main body.
  • RRH remote wireless headends
  • the user equipment may be implemented as a mobile terminal (such as a smart phone, a tablet personal computer (PC), a notebook PC, a portable game terminal, a portable/dongle type mobile router, and a digital camera) or a vehicle-mounted terminal (such as a car navigation device).
  • the user equipment can also be implemented as a terminal (also referred to as a machine type communication (MTC) terminal) that performs machine-to-machine (M2M) communication, a drone, sensors and actuators in an automated factory, and so on.
  • the user equipment may be a wireless communication module (such as an integrated circuit module including a single chip) installed on each of the above-mentioned terminals.
  • the first application example of the base station is the first application example of the base station
  • FIG. 24 is a block diagram showing a first example of a schematic configuration of a base station to which the technology of the present disclosure can be applied.
  • the base station can be implemented as gNB 1400.
  • the gNB 1400 includes multiple antennas 1410 and base station equipment 1420.
  • the base station device 1420 and each antenna 1410 may be connected to each other via an RF cable.
  • the gNB 1400 (or base station device 1420) here may correspond to the above-mentioned electronic device 100 or 300.
  • the antenna 1410 includes a plurality of antenna elements.
  • the antenna 1410 may be arranged in an antenna array matrix, for example, and used for the base station device 1420 to transmit and receive wireless signals.
  • multiple antennas 1410 may be compatible with multiple frequency bands used by gNB 1400.
  • the base station device 1420 includes a controller 1421, a memory 1422, a network interface 1423, and a wireless communication interface 1425.
  • the controller 1421 may be, for example, a CPU or a DSP, and operates various functions of higher layers of the base station device 1420.
  • the controller 1421 may include the processing circuit 101 or 301 described above to execute the communication method described in FIG. 21B or 23B, or to control various components of the base station device 1420.
  • the controller 1421 generates a data packet based on the data in the signal processed by the wireless communication interface 1425, and transmits the generated packet via the network interface 1423.
  • the controller 1421 may bundle data from multiple baseband processors to generate a bundled packet, and transfer the generated bundled packet.
  • the controller 1421 may have a logic function for performing control such as radio resource control, radio bearer control, mobility management, admission control, and scheduling. This control can be performed in conjunction with nearby gNB or core network nodes.
  • the memory 1422 includes RAM and ROM, and stores programs executed by the controller 1421 and various types of control data (such as a terminal list, transmission power data, and scheduling data).
  • the network interface 1423 is a communication interface for connecting the base station device 1420 to a core network 1424 (for example, a 5G core network).
  • the controller 1421 may communicate with the core network node or another gNB via the network interface 1423.
  • the gNB 1400 and the core network node or other gNB can be connected to each other through a logical interface (such as an NG interface and an Xn interface).
  • the network interface 1423 may also be a wired communication interface or a wireless communication interface for a wireless backhaul line. If the network interface 1423 is a wireless communication interface, the network interface 1423 can use a higher frequency band for wireless communication than the frequency band used by the wireless communication interface 1425.
  • the wireless communication interface 1425 supports any cellular communication scheme (such as 5G NR), and provides a wireless connection to a terminal located in the cell of the gNB 1400 via the antenna 1410.
  • the wireless communication interface 1425 may generally include a baseband (BB) processor 1426 and an RF circuit 1427, for example.
  • the BB processor 1426 can perform, for example, encoding/decoding, modulation/demodulation, and multiplexing/demultiplexing, and perform various types of signals of various layers (such as physical layer, MAC layer, RLC layer, PDCP layer, SDAP layer) deal with.
  • the BB processor 1426 may have a part or all of the above-mentioned logical functions.
  • the BB processor 1426 may be a memory storing a communication control program, or a module including a processor and related circuits configured to execute the program.
  • the update program can change the function of the BB processor 1426.
  • the module may be a card or a blade inserted into the slot of the base station device 1420. Alternatively, the module can also be a chip mounted on a card or blade.
  • the RF circuit 1427 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via the antenna 1410.
  • FIG. 24 shows an example in which one RF circuit 1427 is connected to one antenna 1410, the present disclosure is not limited to this illustration, but one RF circuit 1427 can connect multiple antennas 1410 at the same time.
  • the wireless communication interface 1425 may include a plurality of BB processors 1426.
  • multiple BB processors 1426 may be compatible with multiple frequency bands used by gNB 1400.
  • the wireless communication interface 1425 may include a plurality of RF circuits 1427.
  • multiple RF circuits 1427 may be compatible with multiple antenna elements.
  • FIG. 24 shows an example in which the wireless communication interface 1425 includes a plurality of BB processors 1426 and a plurality of RF circuits 1427, the wireless communication interface 1425 may also include a single BB processor 1426 or a single RF circuit 1427.
  • the gNB 1400 shown in FIG. 24 one or more units included in the processing circuit 101 described with reference to FIG. 21A (for example, the access assistance information sending unit 105) or one of the processing circuits 301 described with reference to FIG. 23A Or multiple units (for example, the receiving unit 302) may be implemented in the wireless communication interface 825. Alternatively, at least a part of these components may be implemented in the controller 821.
  • the gNB 1400 includes a part (for example, the BB processor 1426) or the whole of the wireless communication interface 1425, and/or a module including the controller 1421, and one or more components may be implemented in the module.
  • the module may store a program for allowing the processor to function as one or more components (in other words, a program for allowing the processor to perform operations of one or more components), and may execute the program.
  • a program for allowing the processor to function as one or more components may be installed in the gNB 1400, and the wireless communication interface 1425 (for example, the BB processor 1426) and/or the controller 1421 may execute this program.
  • gNB 1400, base station equipment 1420, or modules may be provided, and a program for allowing the processor to function as one or more components may be provided.
  • a readable medium in which the program is recorded may be provided.
  • FIG. 25 is a block diagram showing a second example of a schematic configuration of a base station to which the technology of the present disclosure can be applied.
  • the base station is shown as gNB 1530.
  • the gNB 1530 includes multiple antennas 1540, base station equipment 1550, and RRH 1560.
  • the RRH 1560 and each antenna 1540 may be connected to each other via an RF cable.
  • the base station device 1550 and the RRH 1560 may be connected to each other via a high-speed line such as an optical fiber cable.
  • the gNB 1530 (or base station device 1550) herein may correspond to the above-mentioned electronic device 100 or 300.
  • the antenna 1540 includes a plurality of antenna elements.
  • the antenna 1540 may be arranged in an antenna array matrix, for example, and used for the base station device 1550 to transmit and receive wireless signals.
  • multiple antennas 1540 may be compatible with multiple frequency bands used by gNB 1530.
  • the base station equipment 1550 includes a controller 1551, a memory 1552, a network interface 1553, a wireless communication interface 1555, and a connection interface 1557.
  • the controller 1551, the memory 1552, and the network interface 1553 are the same as the controller 1421, the memory 1422, and the network interface 1423 described with reference to FIG. 24.
  • the wireless communication interface 1555 supports any cellular communication scheme (such as 5G NR), and provides wireless communication to a terminal located in a sector corresponding to the RRH 1560 via the RRH 1560 and the antenna 1540.
  • the wireless communication interface 1555 may generally include, for example, a BB processor 1556.
  • the BB processor 1556 is the same as the BB processor 1426 described with reference to FIG. 24 except that the BB processor 1556 is connected to the RF circuit 1564 of the RRH 1560 via the connection interface 1557.
  • the wireless communication interface 1555 may include a plurality of BB processors 1556.
  • multiple BB processors 1556 may be compatible with multiple frequency bands used by gNB 1530.
  • FIG. 25 shows an example in which the wireless communication interface 1555 includes a plurality of BB processors 1556, the wireless communication interface 1555 may also include a single BB processor 1556.
  • connection interface 1557 is an interface for connecting the base station device 1550 (wireless communication interface 1555) to the RRH 1560.
  • the connection interface 1557 may also be a communication module used to connect the base station device 1550 (wireless communication interface 1555) to the communication in the above-mentioned high-speed line of the RRH 1560.
  • the RRH 1560 includes a connection interface 1561 and a wireless communication interface 1563.
  • connection interface 1561 is an interface for connecting the RRH 1560 (wireless communication interface 1563) to the base station device 1550.
  • the connection interface 1561 may also be a communication module used for communication in the above-mentioned high-speed line.
  • the wireless communication interface 1563 transmits and receives wireless signals via the antenna 1540.
  • the wireless communication interface 1563 may generally include an RF circuit 1564, for example.
  • the RF circuit 1564 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via the antenna 1540.
  • FIG. 25 shows an example in which one RF circuit 1564 is connected to one antenna 1540, the present disclosure is not limited to this illustration, but one RF circuit 1564 can connect multiple antennas 1540 at the same time.
  • the wireless communication interface 1563 may include a plurality of RF circuits 1564.
  • multiple RF circuits 1564 may support multiple antenna elements.
  • FIG. 25 shows an example in which the wireless communication interface 1563 includes a plurality of RF circuits 1564, the wireless communication interface 1563 may also include a single RF circuit 1564.
  • the gNB 1500 shown in FIG. 25 one or more units included in the processing circuit 101 described with reference to FIG. 21A (for example, the access assistance information sending unit 105) or one of the units included in the processing circuit 301 described with reference to FIG. 23A Or multiple units (for example, the receiving unit 302) may be implemented in the wireless communication interface 1525. Alternatively, at least a part of these components may be implemented in the controller 1521.
  • the gNB 1500 includes a part (for example, the BB processor 1526) or the whole of the wireless communication interface 1525, and/or a module including the controller 1521, and one or more components may be implemented in the module.
  • the module may store a program for allowing the processor to function as one or more components (in other words, a program for allowing the processor to perform operations of one or more components), and may execute the program.
  • a program for allowing the processor to function as one or more components may be installed in the gNB 1500, and the wireless communication interface 1525 (for example, the BB processor 1526) and/or the controller 1521 may execute this program.
  • gNB 1500, base station equipment 1520, or modules may be provided, and a program for allowing the processor to function as one or more components may be provided.
  • a readable medium in which the program is recorded may be provided.
  • the first application example of user equipment is the first application example of user equipment
  • FIG. 26 is a block diagram showing an example of a schematic configuration of a smart phone 1600 to which the technology of the present disclosure can be applied.
  • the smart phone 1600 may be implemented as the electronic device 200 described in this disclosure.
  • the smart phone 1600 includes a processor 1601, a memory 1602, a storage device 1603, an external connection interface 1604, a camera device 1606, a sensor 1607, a microphone 1608, an input device 1609, a display device 1610, a speaker 1611, a wireless communication interface 1612, one or more An antenna switch 1615, one or more antennas 1616, a bus 1617, a battery 1618, and an auxiliary controller 1619.
  • the processor 1601 may be, for example, a CPU or a system on a chip (SoC), and controls the functions of the application layer and other layers of the smart phone 1600.
  • the processor 1601 may include or serve as the processing circuit 201 described with reference to FIG. 22A.
  • the memory 1602 includes RAM and ROM, and stores data and programs executed by the processor 1601 to implement the communication method described with reference to FIG. 22B.
  • the storage device 1603 may include a storage medium such as a semiconductor memory and a hard disk.
  • the external connection interface 1604 is an interface for connecting an external device such as a memory card and a universal serial bus (USB) device to the smart phone 1600.
  • USB universal serial bus
  • the imaging device 1606 includes an image sensor such as a charge coupled device (CCD) and a complementary metal oxide semiconductor (CMOS), and generates a captured image.
  • the sensor 1607 may include a group of sensors, such as a measurement sensor, a gyroscope sensor, a geomagnetic sensor, and an acceleration sensor.
  • the microphone 1608 converts the sound input to the smart phone 1600 into an audio signal.
  • the input device 1609 includes, for example, a touch sensor, a keypad, a keyboard, a button, or a switch configured to detect a touch on the screen of the display device 1610, and receives an operation or information input from the user.
  • the display device 1610 includes a screen such as a liquid crystal display (LCD) and an organic light emitting diode (OLED) display, and displays an output image of the smart phone 1600.
  • the speaker 1611 converts the audio signal output from the smart phone 1600 into sound.
  • the wireless communication interface 1612 supports any cellular communication scheme (such as 4G LTE or 5G NR, etc.), and performs wireless communication.
  • the wireless communication interface 1612 may generally include, for example, a BB processor 1613 and an RF circuit 1614.
  • the BB processor 1613 may perform, for example, encoding/decoding, modulation/demodulation, and multiplexing/demultiplexing, and perform various types of signal processing for wireless communication.
  • the RF circuit 1614 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via the antenna 1616.
  • the wireless communication interface 1612 may be a chip module on which the BB processor 1613 and the RF circuit 1614 are integrated. As shown in FIG.
  • the wireless communication interface 1612 may include a plurality of BB processors 1613 and a plurality of RF circuits 1614.
  • FIG. 26 shows an example in which the wireless communication interface 1612 includes a plurality of BB processors 1613 and a plurality of RF circuits 1614, the wireless communication interface 1612 may also include a single BB processor 1613 or a single RF circuit 1614.
  • the wireless communication interface 1612 may support another type of wireless communication scheme, such as a short-range wireless communication scheme, a near field communication scheme, and a wireless local area network (LAN) scheme.
  • the wireless communication interface 1612 may include a BB processor 1613 and an RF circuit 1614 for each wireless communication scheme.
  • Each of the antenna switches 1615 switches the connection destination of the antenna 1616 among a plurality of circuits included in the wireless communication interface 1612 (e.g., circuits for different wireless communication schemes).
  • the antenna 1616 includes a plurality of antenna elements.
  • the antenna 1616 may be arranged in an antenna array matrix, for example, and used for the wireless communication interface 1612 to transmit and receive wireless signals.
  • the smart phone 1600 may include one or more antenna panels (not shown).
  • the smart phone 1600 may include an antenna 1616 for each wireless communication scheme.
  • the antenna switch 1615 may be omitted from the configuration of the smart phone 1600.
  • the bus 1617 connects the processor 1601, the memory 1602, the storage device 1603, the external connection interface 1604, the camera device 1606, the sensor 1607, the microphone 1608, the input device 1609, the display device 1610, the speaker 1611, the wireless communication interface 1612, and the auxiliary controller 1619 to each other. connection.
  • the battery 1618 supplies power to each block of the smart phone 1600 shown in FIG. 26 via a feeder line, and the feeder line is partially shown as a dashed line in the figure.
  • the auxiliary controller 1619 operates the minimum necessary functions of the smartphone 1600 in the sleep mode, for example.
  • one or more components included in the processing circuit 201 described with reference to FIG. 22A may be implemented in wireless communication.
  • at least a part of these components may be implemented in the processor 1601 or the auxiliary controller 1619.
  • the smart phone 1600 includes a part (for example, the BB processor 1613) or the whole of the wireless communication interface 1612, and/or a module including the processor 1601 and/or the auxiliary controller 1619, and one or more components may be Implemented in this module.
  • the module may store a program that allows processing to function as one or more components (in other words, a program for allowing the processor to perform operations of one or more components), and may execute the program.
  • a program for allowing the processor to function as one or more components may be installed in the smart phone 1600, and the wireless communication interface 1612 (for example, the BB processor 1613), the processor 1601, and/or the auxiliary The controller 1619 can execute this program.
  • a device including one or more components a smart phone 1600 or a module may be provided, and a program for allowing a processor to function as one or more components may be provided.
  • a readable medium in which the program is recorded may be provided.
  • FIG. 27 is a block diagram showing an example of a schematic configuration of a car navigation device 1720 to which the technology of the present disclosure can be applied.
  • the car navigation device 1720 may be implemented as the electronic device 200 described with reference to FIG. 22A.
  • the car navigation device 1720 includes a processor 1721, a memory 1722, a global positioning system (GPS) module 1724, a sensor 1725, a data interface 1726, a content player 1727, a storage medium interface 1728, an input device 1729, a display device 1730, a speaker 1731, a wireless A communication interface 1733, one or more antenna switches 1736, one or more antennas 1737, and a battery 1738.
  • the car navigation device 1720 may be implemented as the UE described in this disclosure.
  • the processor 1721 may be, for example, a CPU or SoC, and controls the navigation function of the car navigation device 1720 and other functions.
  • the memory 1722 includes RAM and ROM, and stores data and programs executed by the processor 1721.
  • the GPS module 1724 uses GPS signals received from GPS satellites to measure the position of the car navigation device 1720 (such as latitude, longitude, and altitude).
  • the sensor 1725 may include a group of sensors, such as a gyroscope sensor, a geomagnetic sensor, and an air pressure sensor.
  • the data interface 1726 is connected to, for example, an in-vehicle network 1741 via a terminal not shown, and acquires data (such as vehicle speed data) generated by the vehicle.
  • the content player 1727 reproduces content stored in a storage medium (such as CD and DVD), which is inserted into the storage medium interface 1728.
  • the input device 1729 includes, for example, a touch sensor, a button, or a switch configured to detect a touch on the screen of the display device 1730, and receives an operation or information input from the user.
  • the display device 1730 includes a screen such as an LCD or OLED display, and displays an image of a navigation function or reproduced content.
  • the speaker 1731 outputs the sound of the navigation function or the reproduced content.
  • the wireless communication interface 1733 supports any cellular communication scheme (such as 4G LTE or 5G NR), and performs wireless communication.
  • the wireless communication interface 1733 may generally include, for example, a BB processor 1734 and an RF circuit 1735.
  • the BB processor 1734 may perform, for example, encoding/decoding, modulation/demodulation, and multiplexing/demultiplexing, and perform various types of signal processing for wireless communication.
  • the RF circuit 1735 may include, for example, a mixer, a filter, and an amplifier, and transmit and receive wireless signals via the antenna 1737.
  • the wireless communication interface 1733 can also be a chip module on which the BB processor 1734 and the RF circuit 1735 are integrated. As shown in FIG.
  • the wireless communication interface 1733 may include a plurality of BB processors 1734 and a plurality of RF circuits 1735.
  • FIG. 27 shows an example in which the wireless communication interface 1733 includes a plurality of BB processors 1734 and a plurality of RF circuits 1735, the wireless communication interface 1733 may also include a single BB processor 1734 or a single RF circuit 1735.
  • the wireless communication interface 1733 may support another type of wireless communication scheme, such as a short-range wireless communication scheme, a near field communication scheme, and a wireless LAN scheme.
  • the wireless communication interface 1733 may include a BB processor 1734 and an RF circuit 1735 for each wireless communication scheme.
  • Each of the antenna switches 1736 switches the connection destination of the antenna 1737 among a plurality of circuits included in the wireless communication interface 1733, such as circuits for different wireless communication schemes.
  • the antenna 1737 includes a plurality of antenna elements.
  • the antenna 1737 may be arranged in an antenna array matrix, for example, and used for the wireless communication interface 1733 to transmit and receive wireless signals.
  • the car navigation device 1720 may include an antenna 1737 for each wireless communication scheme.
  • the antenna switch 1736 may be omitted from the configuration of the car navigation device 1720.
  • the battery 1738 supplies power to each block of the car navigation device 1720 shown in FIG. 27 via a feeder line, and the feeder line is partially shown as a dashed line in the figure.
  • the battery 1738 accumulates electric power supplied from the vehicle.
  • the car navigation device 1720 shown in FIG. 27 one or more components included in the processing circuit 201 described with reference to FIG. Communication interface 1733. Alternatively, at least a part of these components may be implemented in the processor 1721.
  • the car navigation device 1720 includes a part (for example, the BB processor 1734) or the whole of the wireless communication interface 1733, and/or a module including the processor 1721, and one or more components may be implemented in the module.
  • the module may store a program that allows processing to function as one or more components (in other words, a program for allowing the processor to perform operations of one or more components), and may execute the program.
  • a program for allowing the processor to function as one or more components may be installed in the car navigation device 1720, and the wireless communication interface 1733 (for example, the BB processor 1734) and/or the processor 1721 may Execute the procedure.
  • a device including one or more components a car navigation device 1720 or a module may be provided, and a program for allowing the processor to function as one or more components may be provided.
  • a readable medium in which the program is recorded may be provided.
  • the technology of the present disclosure may also be implemented as an in-vehicle system (or vehicle) 1740 including one or more blocks of a car navigation device 1720, an in-vehicle network 1741, and a vehicle module 1742.
  • vehicle module 1742 generates vehicle data (such as vehicle speed, engine speed, and failure information), and outputs the generated data to the vehicle network 1741.
  • a plurality of functions included in one unit in the above embodiments may be realized by separate devices.
  • the multiple functions implemented by multiple units in the above embodiments may be implemented by separate devices, respectively.
  • one of the above functions can be implemented by multiple units. Needless to say, such a configuration is included in the technical scope of the present disclosure.
  • the steps described in the flowchart include not only processing performed in time series in the described order, but also processing performed in parallel or individually rather than necessarily in time series.
  • the order can be changed appropriately.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Computer Security & Cryptography (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

本公开涉及无线通信系统中的电子设备、通信方法和存储介质。一种用于低频基站的电子设备，包括处理电路，处理电路被配置为：基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；确定与候选高频基站相关联的接入辅助信息；以及将所述接入辅助信息发送给所述用户设备。

Description

无线通信系统中的电子设备、通信方法和存储介质 技术领域
本公开涉及无线通信系统中的电子设备、通信方法和存储介质，更具体地，本公开涉及在高低频混合网络架构中利用低频通信辅助毫米波通信接入的电子设备、通信方法和存储介质。
背景技术
随着无线通信用户的不断增多及用户需求和业务量的不断增加，移动通信的数据量呈现爆发式增长。当前的移动通信受限于微波频段的带宽，已不能适应用户的业务需求。为此，开发利用更高频段的宽带移动通信成为热点趋势。毫米波(Millimeter wave)可以极大地丰富可用频谱资源，这意味着更宽的带宽、更快的传输速率，此外，根据天线理论，毫米波通信所使用的天线尺寸也在毫米量级，从而在小范围空间中可以放置成百甚至上千根毫米波天线，有利于大规模多输入多输出(Massive MIMO)技术在实际系统中的应用。因此，毫米波技术成为5G无线通信系统中的关键技术之一。
另一方面，毫米波通信存在覆盖范围过小、功耗过大的缺陷。利用波束赋形技术可以形成具有指向性的空间波束，以在特定空间方向聚集能量对抗信道路径衰落，从而扩大覆盖范围。此外，还可以利用高低频混合部署的拓扑结构，通过低频基站(例如，传统的LTE基站)作为锚点实现大范围的稳定覆盖以及为用户设备(UE)提供低速率传输服务，而当UE具有高速率数据传输的需求时才考虑切换至高频基站(例如，毫米波基站)。这种高低频混合部署可以结合低频通信和高频通信这两者的优点。
然而，在决策接入毫米波基站时，由于UE与各个毫米波基站之间的无线信道质量是未知的，UE无法知道哪个毫米波基站最适合接入。如果依次执行对于各个毫米波基站的搜索，则接入的效率将是低下的。甚至作为最差的情况，如果任何毫米波基站的信道质量都无法满足通信要求，则贸然开启毫米波通信模块将会导致功耗浪费。
因此，在应用例如毫米波的无线通信系统中，存在对于高频通信接入方法的需求。
发明内容
针对上面提到的问题及其它问题，本公开的各个方面提供了适于在高低频混合网络架构中高效地接入诸如毫米波基站之类的高频基站的解决方案。
在下文中给出了关于本公开的简要概述，以便提供关于本公开的一些方面的基本理解。但是，应当理解，这个概述并不是关于本公开的穷举性概述。它并不是意图用来确定本公开的关键性部分或重要部分，也不是意图用来限定本公开的范围。其目的仅仅是以简化的形式给出关于本公开的某些概念，以此作为稍后给出的更详细描述的前序。
根据本公开的一个方面，提供了一种用于低频基站的电子设备，包括处理电路，该处理电路被配置为：基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；确定与候选高频基站相关联的接入辅助信息；以及将所述接入辅助信息发送给所述用户设备。
根据本公开的一个方面，提供了一种用于用户设备的电子设备，包括处理电路，该处理电路被配置为：经由低频链路向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及利用所述接入辅助信息，接入所述候选高频基站。
根据本公开的一个方面，提供了一种用于高频基站的电子设备，包括处理电路，该处理电路被配置为：从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束的信息，其中所述波束是所述低频基站通过将基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；利用所述波束建立与所述用户设备的高频通信。
根据本公开的一个方面，提供了一种训练神经网络的方法，包括：从用户设备 接收经由低频通信发送的参考信号；基于所述参考信号获取信道状态信息(CSI)矩阵；从用户设备接收与之进行高频通信的高频基站的标识信息；通过使用所述CSI矩阵作为输入以及使用所述高频基站的标识信息作为输出进行深度学习，确定所述神经网络的参数。
根据本公开的一个方面，提供了一种用于低频基站的通信方法：基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；确定与候选高频基站相关联的接入辅助信息；以及将所述接入辅助信息发送给所述用户设备。
根据本公开的一个方面，提供了一种用于用户设备的通信方法：经由低频通信向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及利用所述接入辅助信息，接入所述候选高频基站。
根据本公开的一个方面，提供了一种用于高频基站的通信方法：从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束的信息，其中所述波束是所述低频基站通过将基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；利用所述波束建立与所述用户设备的高频通信。
根据本公开的另一个方面，提供了一种存储有可执行指令的非暂时性计算机可读存储介质，所述可执行指令当被执行时实现如上所述的任一个方法。
附图说明
本公开可以通过参考下文中结合附图所给出的详细描述而得到更好的理解，其中在所有附图中使用了相同或相似的附图标记来表示相同或者相似的要素。所有附图连同下面的详细说明一起包含在本说明书中并形成说明书的一部分，用来进一步举例说明本公开的实施例和解释本公开的原理和优点。其中：
图1是例示了NR通信系统的体系架构的简化示图；
图2A和2B分别例示了用户平面和控制平面的NR无线电协议架构；
图3是例示了高低频混合网络架构的示意图；
图4例示了5G NR中使用的帧结构；
图5是例示了波束与同步信号之间的关联的示意图；
图6例示了5G NR中的同步信号块(SSB)的时频结构；
图7例示了角度域CSI的幅度与角度之间的关系；
图8例示了UE与基站之间的路径损耗与距离的关系；
图9例示了根据本公开的卷积神经网络(CNN)的一个示例；
图10例示了利用卷积核进行的卷积操作；
图11例示了池化(pooling)处理的示例；
图12例示了根据本公开的卷积神经网络(CNN)的另一示例；
图13例示了根据本公开的高频通信接入的通信流程的一个示例；
图14例示了初始接入的通信流程图；
图15A-15B例示了根据本公开的高频通信接入的通信流程的另一示例；
图16例示了根据本公开的高频通信接入的通信流程的另一示例；
图17例示了根据本公开的用于收集训练数据的通信流程；
图18和图19例示了根据本公开的CNN的实例；
图20A-20C例示了根据本公开的高频通信接入的效果仿真图；
图21A-21B例示了根据本公开的用于低频基站的电子设备及其通信方法；
图22A-22B例示了根据本公开的用于UE的电子设备及其通信方法；
图23A-23B例示了根据本公开的用于高频基站的电子设备及其通信方法；
图24例示了根据本公开的基站的示意性配置的第一示例；
图25例示了根据本公开的基站的示意性配置的第二示例；
图26例示了根据本公开的智能电话的示意性配置示例；
图27例示了根据本公开的汽车导航设备的示意性配置示例。
通过参照附图阅读以下详细描述，本公开的特征和方面将得到清楚的理解。
具体实施方式
在下文中将参照附图来详细描述本公开的各种示例性实施例。为了清楚和简明起见，在本说明书中并未描述实施例的所有特征。然而应注意，在实现本公开的实施例时可以根据具体需求做出很多特定于实现方式的设置，以便例如符合与设备及业务相关的那些限制条件，并且这些限制条件可能会随着实现方式的不同而有所改变。此外，还应该了解，虽然开发工作有可能是较复杂和费事的，但对得益于本公开内容的本领域技术人员来说，这种开发公开仅仅是例行的任务。
此外，还应注意，为了避免因不必要的细节而模糊了本公开，在有些附图中仅仅示出了与至少根据本公开的技术内容密切相关的处理步骤和/或设备结构，而在另一些附图中，为了便于本公开的更好理解，额外示出了现有的处理步骤和/或设备结构。
将参照附图来详细描述根据本公开的示例性实施例和应用实例。以下示例性实施例的描述仅仅是说明性的，不意在作为对本公开及其应用的任何限制。
出于方便解释的目的，下面将在5G NR的背景下描述本公开的各个方面。但是应注意，这不是对本公开的应用范围的限制，本公开的一个或多个方面还可以被应用于诸如4G LTE/LTE-A的现有无线通信系统或者未来发展的各种无线通信系统。下面的描述中提及的架构、实体、功能、过程等可以在NR或其它的通信标准中找到对应。
【概述】
图1是示出了5G NR通信系统的体系架构的简化示图。如图1中所示，在网络控制侧，NR通信系统的无线接入网(NG-RAN)节点包括gNB和ng-eNB，其中gNB是在5G NR通信标准中新定义的节点，其经由NG接口连接到5G核心网(5GC)，并且 提供与终端设备(也可称为“用户设备”，下文中简称为“UE”)终接的NR用户平面和控制平面协议；ng-eNB是为了与4G LTE通信系统兼容而定义的节点，其可以是LTE无线接入网的演进型节点B(eNB)的升级，经由NG接口连接设备到5G核心网，并且提供与UE终接的演进通用陆地无线接入(E-UTRA)用户平面和控制平面协议。在NG-RAN节点(例如，gNB、ng-eNB)之间具有Xn接口，以便于节点之间的相互通信。下文中将gNB和ng-eNB统称为“基站”。
但是应注意，本公开中所使用的术语“基站”不仅限于上面这两种节点，而是无线通信系统中的控制设备的示例，具有其通常含义的全部广度。例如，除了5G通信标准中规定的gNB和ng-eNB之外，取决于本公开的技术方案被应用的场景，“基站”例如还可以是LTE或LTE-A通信系统中的eNB、远程无线电头端(RRH)、无线接入点、中继节点、无人机控制塔台、自动化工厂中的控制节点或者执行类似控制功能的通信装置或其元件。后面的章节将详细描述基站的应用示例。
另外，本公开中所使用的术语“UE”具有其通常含义的全部广度，包括与基站通信的各种终端设备或车载设备。作为例子，UE例如可以是移动电话、膝上型电脑、平板电脑、车载通信设备、无人机、自动化工厂中的传感器和执行器等之类的终端设备或其元件。后面的章节将详细描述UE的应用示例。
接下来结合图2A和2B来描述用于图1中的基站和UE的NR无线电协议架构。图2A示出了用于UE和基站的用户平面的无线电协议栈，图2B示出了用于UE和基站的控制平面的无线电协议栈。
无线电协议栈的层1(L1)是最低层，也被称为物理层。L1实现各种物理层信号处理以提供信号的透明传输功能。例如，在发送数据时，L1对来自MAC层的用户数据进行诸如循环冗余校验(CRC)、信道编码、速率匹配、加扰、调制、预编码、资源映射等一系列物理层处理，以便映射到传输信道，反之在接收数据时可以执行一系列逆处理。
层2(L2)在物理层之上并且负责管理UE与基站之间的无线链路。在用户平面中，L2包括介质接入控制(MAC)子层、无线电链路控制(RLC)子层、分组数据汇聚协议(PDCP)子层、以及业务数据适配协议(SDAP)子层。另外，在控制平面中， L2包括MAC子层、RLC子层、PDCP子层。这些子层的关系在于：物理层为MAC子层提供传输信道，诸如物理上行共享信道(PUSCH)、物理上行控制信道(PUCCH)、物理随机接入信道(PRACH)、物理下行共享信道(PDSCH)、物理下行控制信道(PDCCH)、物理广播信道(PBCH)；MAC子层为RLC子层提供逻辑信道；RLC子层为PDCP子层提供RLC信道，PDCP子层为SDAP子层提供无线电承载。
在控制平面中，UE和基站中还包括层3(L3)中的无线电资源控制(RRC)子层。RRC子层负责获得无线电资源(即，无线电承载)以及负责使用RRC信令来配置各下层。另外，UE中的非接入层(NAS)控制协议执行例如认证、移动性管理、安全控制等功能。
在5G NR中，下行和上行传输都被组织成帧。图4示出了5G通信系统中的帧结构的示图。作为与LTE/LTE-A兼容的固定构架，NR中的帧同样具有10ms的长度，包括2个长度为5ms的半帧，并且进一步包括10个相等大小的子帧，每个子帧为1ms。不同于LTE/LTE-A，NR中的帧结构具有取决于子载波间隔的灵活构架。每个子帧具有可配置的
Figure PCTCN2020138206-appb-000001
个时隙，例如1、2、4、8、16。每个时隙也具有可配置的
Figure PCTCN2020138206-appb-000002
个OFDM符号，对于正常的循环前缀，每个时隙包含14个连贯的OFDM符号，而对于延长的循环前缀，每个时隙包括12个连贯的OFDM符号。在频域维度上，每个时隙包括若干个资源块，每个资源块包含频域中的例如12个连贯子载波。由此，可使用资源网格来表示时隙中的资源元素(RE)，如图4中所示。
5G NR通信系统考虑了三大应用场景：增强型移动宽带(eMBB)、海量机器类通信(mMTC)、超可靠低延时通信(URLLC)，分别具有更宽的带宽(例如，大于1Gbps)、更多的用户接入(每平方公里100万个连接)、更低的时延(小于1毫秒)等特点，而这些都有赖于丰富的频谱资源。NR可以使用的频谱可分为两个部分：FR1，大约450MHz～6GHz，也称为sub-6GHz频段；FR2，大约24GHz～52GHz。其中FR2中的电磁波波长基本上是毫米级别的，即，属于所谓的毫米波。
在NR通信系统中，基站和UE具有支持大规模MIMO的许多天线，例如几十根、几百根、甚至上千根，并且可以通过调整天线的参数来造成某些角度的无线电信号的相长干涉以及另一些角度的无线电信号的相消干涉，从而形成较窄范围内的波束以在特定的方向上提供较强的功率覆盖，这个过程也称为波束赋形。基站和UE可以使用方 向不同的多个波束来实现小区覆盖。利用大规模MIMO和波束赋形技术，可以克服毫米波的信道路径衰落过大的缺陷。
此外，低频和高频的混合组网是在实际的通信系统中克服毫米波的一些缺陷的可行选择。如本公开使用的，“低频”是指比毫米波频段低的通信频段，诸如LTE或LTE-A通信系统使用的频段、NR通信系统使用的FR1频段(sub-6GHz频段)等。低频段的优点是频率低，绕射能力强，覆盖效果好，因此可以作为基础覆盖频段，以实现5G网络的快速部署。“高频”是指毫米波频段或附近范围的通信频段，诸如NR通信系统使用的FR2频段。高频段的优点是超大带宽，频谱干净，干扰较小，因此可以作为容量补充频段，以支持高速率应用。
图3示出了高低频混合网络架构的示意图。如图3中所示，通信网络包括宏蜂窝和微蜂窝/微微蜂窝等的两层以上的混合网络。宏蜂窝采用低频段的频率部署，由低频基站提供大范围的稳定覆盖。在低频基站的覆盖范围内，可以存在多个(如图3中有5个，但数量不限于此)高频基站，每个高频基站采用高频段的频率部署，实现小范围的热点增强。
在此网络架构下的UE可以支持双连接(dual connectivity)，即，在大多数情况下，控制信号以及低速率数据传输服务通过与低频基站的低频链路完成，只有当UE有高速率数据传输的需求时才会考虑切换为与高频基站(例如，毫米波基站)连接，这是因为由于功耗和稳定性的限制，UE一直使用毫米波进行通信是困难的，尤其当UE是资源受限的设备时。
当UE决策是否开启高频通信模块时，各个高频基站的信道状况对于UE而言是未知的。UE不知道哪个高频基站最适合为其提供服务，也不知道高频基站的信道质量是否满足传输要求。UE将在所有频率范围内依次检测各个高频基站的小区可能驻留的频点，即，UE“盲搜索”可用的小区并尝试接入。
相比于4G LTE的基于广播的机制，5G NR的初始接入采用波束管理机制。具体而言，如图5中所示，gNB按规律的周期T广播同步信号突发(SS Burst)，SS突发包括一个或多个同步信号/物理广播信道块(SSB)，每个SSB对应于一个具有不同方向的波束。由此，gNB可以周期性地利用SS突发用波束扫描所有预定方向。典型地，一个SS 突发可以在5ms时间窗口(半帧)内发送完，以例如20ms的周期重复。
图6示出了NR通信系统中的SSB的时频结构。SSB由主同步信号(PSS)、辅同步信号(SSS)和PBCH共同构成。如图6中所示，在时域中，每个SSB占用4个连续的OFDM符号，在频域中，每个SSB包含240个连续的子载波，其中PSS和SSS占用1个OFDM符号和127个子载波，PBCH跨越3个OFDM符号和240个子载波，但是在一个OFDM符号(图6中的OFDM符号2)中间留有嵌入SSS的部分。SS突发内的每个SSB具有相应的索引(SSB_index)，因此，SSB_index也可以用于标识相应的波束。
对于如图3中的多个高频基站来说，每个高频基站发送的SSB可以具有相互不同的频率位置SS REF，由相应的全局同步信道编号(GSCN)指示。所有频率范围的SS REF和GSCN如下表1所示：
表1：SS REF和GSCN之间的关系
Figure PCTCN2020138206-appb-000003
按照传统的5G NR初始接入机制，UE在其公共陆地移动网络(PLMN)所用的频率范围内盲检测所有可能的SSB频率位置，当SSB的信号质量(例如，参考信号接收功率(RSRP))符合要求时，UE尝试接入发送该SSB的高频基站的小区。然而，UE接入的不一定是能够为其提供最佳高频通信的基站。检测所有高频基站发送的SSB将带来非常大的开销，就效率而言是不可取的。
本公开考虑使用维持网络连接的低频通信来提供有利于高频通信接入的信息，以提高与高频基站的初始接入的效率和质量。
在如图3所示的高低频混合网络中，低频基站可以充当锚点，实现UE与无线通 信网络的稳定连接。低频基站例如可以是4G通信系统的eNB或5G通信系统的ng-eNB等等。对于低频基站与UE之间的低频链路，可以利用低频参考信号来评估信道状况，例如，低频基站可以发送例如信道状态信息参考信号(CSI-RS)之类的下行参考信号，UE通过测量参考信号获得下行低频信道的信道状态信息并将其反馈给低频基站，或者UE可以发送例如探测参考信号(SRS)或CSI-RS之类的上行参考信号，低频基站通过测量参考信号获得上行低频信道的信道状态信息。特别地，在使用时分双工的情况下，上行信道和下行信道的信道状态信息可以从单个方向上的测量获得。
通过测量参考信号获得的CSI可以是与基站的天线数、UE的天线数、子载波数相关联的三维复数矩阵，也可以称为空间域CSI。CSI矩阵中的元素描述了从UE的各天线到低频基站的各天线之间的无线传输路径的状况，例如信号散射、衰落等信息。
本公开的发明人注意到，除了关于信道状况的信息以外，CSI矩阵实际上隐含了UE的位置信息。为了更清楚地说明这一点，图7示意性地示出了当将空间域CSI变换为角度域CSI时三个UE的CSI的幅度与角度之间的关系，其中横轴是0～π的角度范围，纵轴是CSI的幅度值。从图7中可以明显看出，CSI在某个方向上具有显著的峰值，在存在视距(LOS)路径的环境中，这个方向往往是LOS方向。在MIMO中，天线响应向量为信道到达角(AoA)、信道离开角(AoD)的函数，因此CSI包含AoA、AoD等角度信息。此外，图8示出了纵轴为路径损耗、横轴为UE和基站之间的距离的散点图。从图8可以看出，CSI的幅度衰减与UE和基站之间的距离相关，距离越大，CSI的幅度趋于越小，这表明CSI包含距离信息。
考虑到在极坐标系中角度和距离可以表示位置，低频CSI包含了UE的位置信息。但是，CSI矩阵中隐藏的UE位置信息难以通过传统的方法来获取，因为低频天线数目较少限制了传统角度估计方法的性能；同时，从图8也可以看出，低频CSI和毫米波路径衰减之间的关系难以用显式表达式直接表达。
根据本公开的实施例，基于深度学习的预测模型被引入以提取CSI和UE位置之间隐藏的复杂关系，并进而预测辅助UE决策高频通信接入的带外信息。作为示例，卷积神经网络(CNN)是一种得到广泛应用的深度学习模型，包含卷积计算和深度结构，具有极强的非线性拟合能力。借助于例如卷积神经网络的非线性拟合能力，当将经由低频通信获取的CSI矩阵输入到卷积神经网络中时，可以挖掘隐藏在低频CSI矩阵中 的关于UE位置的信息。卷积神经网络具有自学习能力，可以通过真实数据的训练来确定神经网络的参数，而无需复杂的手动参数设计。
预测模型得到的带外信息包括关于最适合为UE提供服务的高频基站的信息、关于UE与高频基站的信道状况的信息、关于高频基站与UE通信使用的最优波束的信息，等等，这些信息有利于UE的高频通信接入，在本公开中也称为“接入辅助信息”。
本公开了提供了基于深度学习的预测模型的不同示例。下面主要以卷积神经网络为例进行描述，然而可以想到，本公开可以不限于卷积神经网络，而是可以使用其他类型的神经网络或者任何合适的深度学习模型，只要能够基于相同的输入产生期望的输出即可。
【卷积神经网络的第一示例】
图9例示了根据第一示例的卷积神经网络的构造示意图。通常，CNN由卷积层、激活函数、全连接层、池化层和批归一化层五个部分组成。
根据第一示例，卷积神经网络的输入数据是低频基站基于上行参考信号获取的低频CSI矩阵和所有备选高频基站的位置数据。CSI矩阵是三维的复数矩阵。高频基站的位置数据可以表示为二维的矩阵。一般而言，高频基站的位置是固定不变的，因此高频基站的位置数据可以事先确定并存储在低频基站的存储器中。当新增或去除高频基站时，低频基站中存储的位置数据可以相应地更新。高频基站的位置数据的表示方法可以有多种，例如以低频基站为参考的相对位置、绝对的地理坐标等等，不同表示的位置数据本质上仅涉及线性变换，没有本质区别。
批归一化(BN)层：批归一化层将输入的数据矩阵归一化为均值为0、方差为1的标准分布，从而加速神经网络的收敛。CNN可以包括一个或多个批归一化层。例如，批归一化层可以用于对输入到CNN中的输入数据进行批归一化，然后再输入到卷积层，或者可以用于对神经网络的中间数据进行批归一化。
卷积层：如图9中所示，在卷积层中，参数可自学习的滤波器(即卷积核)和数据矩阵进行卷积，以提取输入数据中隐藏的特征。考虑到卷积核的尺寸往往远小于数据矩阵，卷积核在数据矩阵上移动以遍历数据矩阵，移动的距离被称为步长。此外，为了同卷积核的移动相匹配，数据矩阵可能进行边缘扩展(即边缘填充)。不同参数的 卷积核被用于从数据矩阵中提取不同的特征，其对应的卷积后的输出被称为特征通道。为了提取更丰富的特征，随着网络层数的加深，特征通道数目逐渐增加。图10中例示了用不同的卷积核对同一数据矩阵进行卷积的示意图，其中卷积的步长为2。
激活函数：卷积层的输出往往在被输入到下一层之前通过激活函数。激活函数通常为非线性函数，因此激活函数能为CNN引入非线性拟合能力。深度学习可以表现出高性能，正是因为通过用多层结构重复非线性变换可以获得高非线性。如果没有激活函数负责非线性变化并且网络仅包括线性变换，那么无论层数多少都仅存在等效的单层线性变换，并且多次结构是无用的。显然，随着层数的增加，深度学习表现出更强的非线性和更高的性能。
池化层：池化层对输入矩阵进行降采样，以降低神经网络的数据量和运算量。池化操作包含最大值池化和平均值池化。图11分别示出了两种池化处理的示意图。如图11中所示，最大值池化保留数据矩阵的最大值，而平均值池化保留数据矩阵的平均值。不同池化层获得的特征向量可以组合成一个特征向量，以便于后续网络结构预测输出。
全连接层：在全连接层中，输入的特征向量经过线性拟合得到输出。全连接层能控制输出尺寸，因此全连接层通常被用于实现从提取的数据特征到输出的尺寸变换。
卷积神经网络可以输出最适合为UE提供高频通信服务的高频基站(下文中也称为“候选高频基站”)。这意味着在UE当前的位置，该高频基站有可能提供信道状况最好的高频链路连接。候选高频基站的输出形式可以是其标识信息，例如用于唯一地标识高频基站的标识符——高频基站ID，包括但不限于例如eNodeB ID、gNBID(其与物理层小区ID(PCI)一起构成唯一地标识小区的NR小区ID(NCI)中)或者更简单的基站编号，只要低频基站能够将高频基站ID与相应的高频基站一一关联即可。
考虑到预测候选高频基站是从数量有限的备选高频基站中选取一个，因此图9中的卷积神经网络的分类器与备选高频基站对应。
为实现多分类预测，从低频CSI矩阵和高频基站位置中所提取的特征通过全连接层转化为尺寸为备选高频基站数量的输出，随后通过例如Softmax激活函数
Figure PCTCN2020138206-appb-000004
Softmax激活函数将全连接层输出归一化为概率。在硬判决下，卷积神经网络选取概率最大的分类器的类别作为预测得到的输出结果。该输出的表达形式提供了预测的高频基站优先程度的排序，即概率越大，认为用户更可能优先接入该高频基站。卷积神经网络可以输出概率最大的仅一个高频基站，或者可以按概率排序的多个高频基站。
对于第一示例的卷积神经网络，候选高频基站的预测标准可以是UE与高频基站之间的高频链路的路径损耗。换句话说，期望的是卷积神经网络输出的候选高频基站与UE的无线信道具有最小路径损耗。事实上，卷积神经网络的预测标准取决于训练数据的获取原则，不同的训练数据可能导致输出不同意义上的候选高频基站。
如图9中所示，卷积神经网络还可以输出其他接入辅助信息。
例如，卷积神经网络的一个分支网络(分支1)可以输出与候选高频基站对应的路径损耗值。路径损耗值的预测有助于估计UE与候选高频基站之间的高频通信链路的信道状况是否满足连接要求。如果路径损耗值指示无线信道不满足要求，则UE可以暂不开启高频通信模块，而是等待更好的时机。
考虑到路径损耗是连续值，因此预测路径损耗值可以被建模为回归问题，即，全连接层将所提取的特征转化为标量，该标量直接作为路径损耗值输出。为减小路径损耗值的动态范围，神经网络预测的路径损耗值可以以dB表示。
又例如，卷积神经网络的一个分支网络(分支2)可以输出候选高频基站用于为UE提供高频通信服务的最优波束。一般而言，对于UE当前所处的位置，最优波束的AOD和AOA可以最接近信道方向，例如LOS方向。低频基站可以将预测得到的最优波束通知给候选高频基站，由此高频基站可以使用这个波束建立与UE的高频通信，或者对用于UE的波束进行进一步的精细化调整。
由于高频基站的波束赋形码本大小有限，卷积神经网络预测最优波束实质上是从高频基站可用的有限个波束中选取一个，因此该类问题也可以建模为多分类问题，分类器数量为备选波束的数量，分类器类型未高频基站的有限个波束。波束可以通过相关联的参考信号(例如SSB)来索引。
此外，损失函数被用于衡量预测结果和目标输出之间的差异。为了降低预测误差(即损失)，梯度反向传播算法被用于更新卷积神经网络的参数。作为示例，关于候选 高频基站或最优波束的多分类预测可以采用交叉熵损失函数，关于路径损耗值的回归预测可以采用Smooth-L1损失函数。在神经网络同时进行多个预测时，可以采用线性加权来涉及整体损失函数Loss，即：
Loss＝μ BSloss BS+μ beamloss beam+μ pathloss path
其中loss BS、loss beam、loss path分别表示关于候选高频基站、最优波束、路径损耗值这三个预测目标的损失函数，μ BS、μ beam、μ path分别表示相应的线性权重系数。
【卷积神经网络的第二示例】
在上面描述的第一示例中，作为预测模型的卷积神经网络用于预测路径损耗最小的高频基站作为适于为UE提供高频通信服务的候选高频基站。然而，在实际情况中，各个高频基站的发射功率往往有差异，具有最小路径损耗的高频基站不一定在UE处的信号接收功率最大，而UE通常通过比较接收信号功率来选择接入的基站，这可能导致预测的候选高频基站与实际的候选高频基站存在偏差。
因此，第二示例的卷积神经网络考虑了这种发射功率差异，候选高频基站的预测标准可以是同步信号在UE处的接收功率(RSRP)。换句话说，期望的是预测模型输出的候选高频基站发射的同步信号在被UE接收具有最好的信号质量。
图12例示了作为根据第二实施例的预测模型的示例的卷积神经网络的构造示意图。卷积神经网络包括卷积层、激活函数、全连接层、池化层和批归一化层五个部分。下面着重描述与图9中所示的卷积神经网络的区别之处。
如图12中所示，除了低频CSI矩阵和高频基站的位置数据以外，卷积神经网络的输入还包括各个高频基站的发射功率。这里，发射功率可以是高频基站广播SSB信号的功率。每个高频基站的位置数据可以与发射功率共同作为高频基站信息输入到神经网络中。一般而言，高频基站的发射功率不经常变化，因此可以与位置数据一样预先存储在低频基站处。
图12中的卷积神经网络输出在RSRP意义上的候选高频基站ID。可选地，卷积神经网络还包括输出对应的RSRP值的分支网络(分支1)和候选高频基站的最优波束的分支网络(分支2)。
RSRP值的预测有助于估计UE与候选高频基站之间的高频通信链路的信道状况是 否满足连接要求。如果RSRP值指示无线信道不满足要求，则UE可以暂不开启高频通信模块，而是等待更好的时机。
类似于第一实施例中的路径损耗值，考虑到RSRP是连续值，因此预测RSRP同样可以被建模为回归问题，即，全连接层将所提取的特征转化为标量，该标量直接作为RSRP值输出。
对于损失函数，关于候选高频基站或最优波束的多分类预测可以采用交叉熵损失函数，关于RSRP值的回归预测可以采用Smooth-L1损失函数。在神经网络同时进行多个预测时，可以采用线性加权来涉及整体损失函数Loss，即：
Loss＝μ BSloss BS+μ beamloss beam+μ RSRPloss RSRP
其中loss BS、loss beam、loss RSRP分别表示关于候选高频基站、最优波束、RSRP值这三个预测目标的损失函数，μ BS、μ beam、μ RSRP分别表示相应的线性权重系数。
【高频通信初始接入流程】
下面结合附图描述利用基于深度学习的预测模型的预测结果辅助高频通信初始接入的流程。
图13是示出了高频通信初始接入的通信流程的一个示例。图13中所示的通信流程可以从UE向低频基站发送上行参考信号(S1)开始。上行参考信号可以是例如SRS、CSI-RS等，并且经由低频链路发送到低频基站。
UE发送上行参考信号的行为可以在UE有高速率数据传输需求的情况下发生。当UE需要连接到高频基站时，UE可以向低频基站发送高频链路接入请求(图13中未示出)，以请求低频基站提供有利于接入高频链路的信息。作为一个示例，UE可以利用预先配置的通信资源(例如，时频资源块)分开发送或一起发送上行参考信号与高频链路接入请求。作为另一个示例，UE可以先向低频基站发送高频链路接入请求，响应于该请求，低频基站可以为UE分配通信资源(例如通过DCI)，由此UE可以利用被分配的通信资源发送上行参考信号。
响应于接收到上行参考信号，低频基站可以获取CSI矩阵(S2)。低频基站可以对通过其多个天线接收的参考信号进行测量，并基于测量值估计CSI矩阵。CSI矩阵可以使用传统的估计方法来获取，这里不再详述。所估计的CSI矩阵可以是如上所述的 三维矩阵的形式。
然后，低频基站利用获取的CSI矩阵来预测对于UE来说最优选的高频基站(S3)。具体而言，低频基站将CSI矩阵作为输入数据输入到已训练好的预测模型中，例如图9或12中所示的卷积神经网络。对于如图9中所示的第一示例的卷积神经网络，输入数据还包括所有可用的高频基站的位置数据，这些位置数据可以事先存储在低频基站处。而对于图12中所示的第二示例的卷积神经网络，输入数据还包括所有可用的高频基站的位置数据和发射功率，这些位置数据和发射功率可以事先存储在低频基站处。卷积神经网络可以确定适于为UE提供高频通信服务的候选高频基站作为输出，例如，取决于低频基站利用的预测模型，候选高频基站可以是与UE的高频通信链路具有最小路径损耗的高频基站，或者在UE处具有最大接收信号功率的高频基站。
低频基站可以确定与候选高频基站相关联的接入辅助信息(S4)。尽管直接向UE通知候选高频基站的ID是一种可行的方式，但是优选地，低频基站可以确定在UE接入候选高频基站时更方便使用的信息，这有利于兼容传统的NR初始接入过程。在一个示例中，低频基站确定的接入辅助信息包括有利于UE识别候选高频基站的同步信号的信息，例如候选高频基站广播的SSB的频率位置SS REF。SSB的频率位置SS REF也可以由GSCN来指示。
低频基站可以将确定的接入辅助信息(例如GSCN)发送给UE(S5)。UE将可以利用这样的接入辅助信息接入候选高频基站(S6)。特别地，在UE的高频通信模块(例如毫米波通信模块)平时处于关闭的情况下，UE可以在接收到与候选高频基站相关联的接入辅助信息的触发下开启高频通信模块。
下面结合图14来描述UE与基站的初始接入过程。UE在开机或要切换到特定基站时首先需要进行小区搜索，小区搜索就是UE获取与小区的时间和频率同步并且检测该小区的物理层小区ID的过程。
在101处，UE通过接收SSB来执行小区搜索。根据本公开的第一实施例，UE可以直接搜索在接入辅助信息中指示的频率位置处的同步信号，在这个频率位置处，UE可以接收到候选高频基站周期性地发送的SS突发，包括分别与不同波束对应的一个或多个SSB。相比于传统的频域盲搜索，直接定位候选高频基站的SSB显然更高效。
当候选高频基站广播的SS突发包括两个或更多个SSB的情况下，UE可以检测各个SSB的信号质量(例如RSRP)是否满足阈值要求，或者检测具有最好信号质量的SSB。UE可以对满足阈值要求的SSB进行解码，以便同步到下行链路定时，例如，通过以下步骤：
1)检测并解码SSB中的PSS，获得传输组内
Figure PCTCN2020138206-appb-000005
2)基于SSS与PSS之间的相对时域位置，检测并解码SSS，获得传输组
Figure PCTCN2020138206-appb-000006
并根据
Figure PCTCN2020138206-appb-000007
获得对应小区的物理层小区ID(PCI)，其中
Figure PCTCN2020138206-appb-000008
3)此外，通过解码PSS/SSS，还可以得到符号的同步，从而间接得到SSB的子载波间隔(SCS)和SSB的绝对频点(absoluteFrequencySSB)；
4)获得PCI之后，就可以确定PBCH的DMRS的位置，例如，DMRS的位置偏移量为PCI mod 4；
5)通过解调PBCH的DMRS和有效载荷，可以得到SSB索引(i SSB)以及半帧信息(n hf)，UE可以获得10ms帧同步。
在获得下行链路小区同步之后，UE可以在下行链路帧中的适当位置接收小区系统信息，诸如主信息块(MIB)和各种系统信息块(SIB)。系统信息可以由基站通过广播用的信道(例如广播信道PBCH、共享信道PDSCH等)周期广播，并且可以包括UE要接入基站所必需的信息，如随机接入相关的信息。
之后，为了获得上行链路小区同步，UE需要进行随机接入过程。如图14中所示，在102处，UE可以通过向候选高频基站发送随机接入前导码(例如包括在MSG-1中)来向基站通知自己的接入行为。随机接入前导码的发送使基站能够估计终端设备的上行链路定时提前(Timing Advance)。在103处，基站可以通过向UE发送随机接入响应(例如包括在MSG-2中)来向UE通知上述定时提前。UE可以通过该定时提前实现上行链路小区同步。随机接入响应中还可以包括上行链路资源的信息，UE可以在操作104中使用该上行链路资源。对于竞争型的随机接入过程，在104处，UE可以通过上述调度的上行链路资源发送UE标识符以及可能的其他信息(例如包括在MSG-3中)。基站可以通过UE标识符确定竞争解决结果。在105处，基站可以告知UE该竞争解决 结果(例如包括在MSG-4中)。此时，如果竞争成功，则UE成功接入基站，该随机接入过程结束；否则，UE需重复操作102至105的随机接入过程。
随着UE与候选高频基站之间的随机接入过程完成，UE建立了与候选高频基站的高频通信，并且可以进行后续的通信。
当在图13的S3中，神经网络输出不止一个候选高频基站的情况下，低频基站可以分别确定与各个高频基站相关联的接入辅助信息(S4)，并将其发送给UE(S5)。UE可以按照这些候选高频基站的优先级依次搜索对应的SS突发。如果具有最高优先级(即，概率最大)的候选高频基站的SS突发中的SSB均不满足阈值要求，UE可以搜索具有第二高优先级的候选高频基站的SS突发，以此类推。如果没有找到满足要求的SSB，则本次初始接入失败。
下面结合图15A和15B来描述根据本公开的高频通信接入的通信流程的另一示例。将主要描述与图13中的通信流程的区别之处，其余相同的部分将不再重复描述。
如图15A或15B中所示，在S31中，低频基站利用卷积神经网络除了预测候选高频基站之外，还可以预测与候选高频基站对应的路径损耗值或RSRP值，这取决于低频基站使用的是第一示例还是第二示例的卷积神经网络。所预测的路径损耗值或RSRP值可以用于判断所预测的候选高频基站是否值得接入。S31中使用的卷积神经网络可以包括用于预测路径损耗值或RSRP值的分支网络(图9或12中的分支1)。
在一个示例中，如图15A中所示，低频基站可以将卷积神经网络输出的路径损耗值(或RSRP值)与预定阈值相比较(S40)。在路径损耗值低于预定阈值(或者RSRP值超过预定阈值)的情况下，低频基站可以认为所预测的优选高频基站能够提供符合要求的高频通信链路，并在S4中确定相关联的接入辅助信息，以促进UE的初始接入。UE响应于接收到接入辅助信息而开启高频通信模块并开始初始接入过程。可替代地，低频基站可以生成开启指示，并将其与接入辅助信息一起发送给UE(如图15A的S5的括号所示)，以指示UE开启高频通信模式并开始初始接入过程。
相反，在路径损耗值超过预定阈值(或者RSRP值低于预定阈值)的情况下，低频基站可以认为所预测的候选高频基站可能无法提供符合要求的高频通信链路，将不进行后面的步骤。此时，低频基站可以改为向UE发送当前不存在适合接入的高频基站 的指示。
在另一个示例中，如图15B中所示，低频基站可以将确定的接入辅助信息与路径损耗值(或RSRP值)一起发送给UE(S51)。UE可以将接收的路径损耗值(或RSRP值)与预定阈值相比较，在路径损耗值低于预定阈值(或者RSRP值超过预定阈值)的情况下，UE可以决策开启高频通信模式，并通过初始接入过程连接到候选高频基站。相反，在路径损耗值超过预定阈值(或者RSRP值低于预定阈值)的情况下，UE可以决策当前不存在适合接入的高频基站，并在后续的时机尝试接入，例如重复图15B中的步骤S1至S6。
应理解的是，上面提到的用于与路径损耗值比较的预定阈值可以是可调整的参数。低频基站或UE使用的预定阈值可以取决于各种因素，诸如：UE的当前电量，UE的剩余电量越多，预定阈值可以越高，从而增加高频接入的机会；UE的连接偏好，例如UE或其使用者可以设置为尽可能接入无线信道状况好(即，路径损耗小)的高频基站；高频通信的传输成功率，例如基于上一次高频接入的结果或者高频通信的数据传输成功率来动态地调整预定阈值；业务紧急程度，例如如果业务要求立刻进行高频通信，则UE可以临时调高预定阈值；等等。类似地，用于与RSRP值比较的预定阈值也可以是可调整的参数，低频基站或UE可以根据例如UE的当前电量、连接偏好、高频通信的传输成功率、业务紧急程度等来调整该预定阈值。应理解，关于路径损耗的预定阈值和关于RSRP的预定阈值是不同的阈值。
下面结合图16来描述根据本公开的高频通信接入的通信流程的另一示例。将主要描述与图13中的通信流程的区别之处，其余相同的部分将不再重复描述。
如图16中所示，在S32处，低频基站可以利用卷积神经网络来预测候选高频基站对于该UE而言的最优波束。最优波束可以对应于候选高频基站发送的SS突发中的SSB。此时，低频基站在S4中确定的接入辅助信息可以包括关于该SSB的时频资源的信息，例如SSB的频率位置(SS REF)和索引(SSB_index)。
如上面所提到的，卷积神经网络可以按预测概率高低输出不止一个(n>1)最优波束。因此，在一个示例中，按照各个波束的概率从高到低的顺序，低频基站可以确定包含对应的n个SSB的时频资源信息并将其发送给UE，也就是说，这n个SSB之间 具有不同的优先级。
当UE接收到这样的接入辅助信息时，UE可以直接检测对应的SSB，并估计其信号质量(例如RSRP)，如果该SSB满足阈值要求，则继续解码该SSB以尝试接入到候选高频基站，如上面参照图13所述。如果该SSB不满足阈值要求，UE可以继续检测相同频率位置上的具有较低优先级的其它SSB。如果没有找到满足要求的SSB，则将搜索范围扩大到其它候选高频基站或其它可能的频率位置。通过这种具有优先级的搜索方式，UE的搜索范围可以在时域上进一步缩小，提高了初始接入的效率。
可选地，低频基站还可以通过基站之间的接口(例如Xn接口)将预测的波束与UE的标识码通知给相应的候选高频基站(S53)。如本公开使用的，UE的标识码是指唯一地标识UE的标识信息，诸如国际移动用户标识码(IMSI)，当UE接入低频基站时，低频基站即可获知该用户的标识码。由此，当该UE接入候选高频基站时，高频基站可以使用预测的最优波束与UE通信。例如，在如图14中所示的上行随机接入过程中，高频基站可以使用最优波束发送MSG-1和/或MSG-4等。
此外，根据本公开的实施例，在得到UE的标识码和最优波束后，高频基站还可以通过波束训练来实现波束的进一步精细化。具体而言，高频基站优先扫描该波束内包含的多个窄波束，接收UE上报的关于各窄波束的信号质量(例如RSRP)，并确定具有最高质量的窄波束作为用于后续数据传输的波束。如果这些窄波束的接收信号质量不满足要求，则继续尝试与最优波束相邻的波束方向，直到信号质量满足要求或已遍历所有波束方向。因此，利用神经网络预测的最优波束，可以缩小扫描波束的范围，有效地降低了波束训练的开销。
当卷积神经网络输出不止一个(n>1)最优波束时，低频基站可以将预测概率从高到低排序的波束ID(例如，对应的参考信号的标识符)发送给候选高频基站，高频基站可以按顺序依次扫描这些波束，以提高波束训练的效率。
以上虽然分开描述了低频基站利用卷积神经网络预测候选高频基站+路径损耗(或RSRP)与候选高频基站+最优波束的情况，但是应理解，低频基站可以同时预测候选高频基站、路径损耗(或RSRP)与最优波束，此时，高频通信初始接入的流程可以结合图15A或15B和图16来理解，这里不再重复描述。
下面描述根据本公开的预测模型的学习方案。
诸如卷积神经网络的预测模型是可以自学习的深度学习模型。低频基站可以从UE和高频基站收集大量的训练数据来确定或更新卷积神经网络的参数。
图17示出了用于收集卷积神经网络的训练数据的通信流程图。如图17中所示，接入高频通信链路的UE可以周期性上报与之通信的高频基站的标识信息，作为低频基站训练卷积神经网络的模型输出。对于如图9中所示的第一示例的卷积神经网络，为了保证卷积神经网络预测的准确度，期望UE上报的高频基站具有最小的路径损耗。而对于图12中所示的第二示例的卷积神经网络，期望UE上报的高频基站具有最大的RSRP。
该UE还周期性经由向低频基站发送低频参考信号，低频基站基于参考信号获取低频CSI矩阵，作为训练卷积神经网络的模型输入。
可选地，UE还可以周期性地估计当前高频通信链路上的路径损耗(或RSRP)，并上报给低频基站，作为训练卷积神经网络的分支网络(图9或12中的分支1)的模型输出。
可选地，与UE进行高频通信的高频基站可以通过基站间的接口(例如Xn接口)周期性向低频基站通知与之通信的UE的标识码和用于该UE的波束，作为训练卷积神经网络的分支网络(图9或12中的分支2)的模型输出。
低频基站根据UE的标识码将从UE和高频基站收集到的训练数据配对，例如与同一UE相关联的CSI矩阵、高频基站ID、路径损耗值(或RSRP值)、高频基站使用的波束等。此外，所有备选高频基站的位置数据和发射功率可以也是相应卷积神经网络的模型输入数据，这种数据可以被预先获取和存储。
由低频基站收集的真实数据被用于学习神经网络的参数，诸如各卷积层的卷积核等，以使得卷积神经网络获得预测能力。通常，卷积神经网络通过大量数据的训练后被部署到低频基站用于实际预测。此外，在线学习策略也可应用于卷积神经网络的更新，即，卷积神经网络在部署之后，低频基站仍然实时地收集训练数据并训练预测模型。在线学习策略允许预测模型适应通信环境的变化。
对于如图9或12中所示的包括分支网络的卷积神经网络，训练方案可以包括以 下步骤：
1、利用低频CSI矩阵和高频基站的位置数据作为模型输入、高频基站ID作为模型输出，预训练卷积神经网络的主体网络结构的参数；
2、利用相应的模型输出训练各分支网络的参数，例如，利用路径损耗值训练图9中的分支1的参数，利用RSRP值训练图12中的分支1的参数，利用高频基站的波束训练分支2的参数。
由于准确地预测候选高频基站ID是预测对应的路径损耗(或RSRP)和高频基站最优波束的基础，因此在预训练阶段，神经网络以准确预测高频基站ID为优化目标，训练所有输出共享的网络结构参数，以充分提取所有输出共同需要的基本特征。在预训练结束后，各个输出根据已经提取得到的基本特征对分支网络进行进一步学习。
在实际部署如上面所述的卷积神经网络时，可以先利用从一个或多个低频基站收集的训练数据来确定卷积神经网络的参数，并将初步训练好的卷积神经网络应用于更多个低频基站。换句话说，所有低频基站中应用的卷积神经网络可以预先利用从部分低频基站收集的训练数据来初始化。然后，每个低频基站可以使用各自覆盖范围内的高频基站的位置数据和/或发射功率以及获取的低频CSI矩阵作为模型输入，通过在线学习不断地持续优化神经网络的参数。在线学习主要学习当前的低频基站周围的遮挡物、反射体等环境信息，从而提高参数对于所处通信环境的适应性。
接下来，将描述根据本公开的卷积神经网络的实例以及基于卷积神经网络的预测结果辅助高频通信接入的仿真效果。
下面描述的仿真基于如图3中所示的低频与毫米波混合网络。具体的设置如表2所示。
表2：仿真具体设置
低频基站覆盖半径(米) 500
低频基站天线数n 64
低频UE天线数m 4
低频UE子载波数k 288
毫米波基站数b 5
毫米波基站的波束赋形码本大小 4
SNR/dB 5～25，随机生成
毫米波路径损耗模型为：
PL[dB]＝32.4+20log 10d+20log 10f c+σ s
其中，其中d表示UE和毫米波基站之间的距离，f c表示毫米波中心频率，f c＝28GHz，σ s表示阴影衰落，其分布服从均值为0、方差为4的高斯分布。上述仿真设置用于生成卷积神经网络的输入数据集。
图18示出了第一示例的卷积神经网络的实例，图19示出第二示例的卷积神经网络的实例。
图18中的卷积神经网络的输入为：
Figure PCTCN2020138206-appb-000009
为低频CSI矩阵，k为低频UE子载波数，m为低频UE天线数，n为低频基站天线数。B为备选毫米波基站位置，以直角坐标表示的毫米波基站位置依次拼接形成为长度为2b的向量输入，b为备选基站数目；该向量输入被视为一个维度尺寸为1的二维矩阵输入(即，
Figure PCTCN2020138206-appb-000010
)。
图19中的卷积神经网络的输入为：
Figure PCTCN2020138206-appb-000011
为低频CSI矩阵，k为低频UE子载波数，m为低频UE天线数，n为低频基站天线数。B’为备选毫米波基站位置和发射功率，被构造成尺寸为
Figure PCTCN2020138206-appb-000012
的矩阵，其中b ij,i＝1,2,3分别表示第j个毫米波基站的二维横坐标、二维纵坐标和发射功率。
图18和19中的符号或数字代表的含义如下面的表3中所示：
表3：卷积神经网络中的符号或数字的含义
Figure PCTCN2020138206-appb-000013
Figure PCTCN2020138206-appb-000014
图18中的卷积神经网络除了输出具有最小路径损耗的毫米波基站ID，还输出相应的路径损耗值和毫米波基站的波束。
图19中的卷积神经网络除了输出具有最大RSRP的毫米波基站ID，还输出相应的RSRP值和毫米波基站的波束。
图20A-20C示出了利用图18中的卷积神经网络的预测结果辅助高频通信接入的 效果示意图。
图20A展示了实际路径损耗最小的毫米波基站在预测结果中排在预测概率最高的前X位的比例。从图中可以看到，在67.1％的样本中，实际具有最小路径损耗的毫米波基站能够被准确地预测为候选高频基站。同时，在超过90％的样本中，实际具有最小路径损耗的毫米波基站能在预测得到的概率最高的两个高频基站中。
图20B示出了预测得到的毫米波路径损耗与实际路径损耗的绝对误差的累计分布函数。从图中可以看到，超过70％的预测绝对误差在5dB以内，约92.5％的预测绝对误差在10dB以内。同时，平均绝对误差为5.1dB。这表明根据低频CSI预测用户与周围毫米波基站的最小路径损耗具有可行性。
图20C展示了实际最优波束在预测结果中排在预测概率最高的前X位的比例。在44.5％的样本中，实际使用的波束能够被准确地预测为高频基站的最优波束；同时，实际使用的波束在预测结果中以约75％的比例位于预测概率最高的前两位。仿真结果表明，优先选择预测得到的最优波束能有效地降低波束搜索的开销。
【电子设备与通信方法】
接下来描述根据本公开的电子设备和通信方法。
图21A是例示了根据本公开的电子设备100的框图。电子设备100可以是低频基站或其部件。
如图21A中所示，电子设备100包括处理电路101。处理电路101至少包括CSI矩阵获取单元102、候选高频基站确定单元103、接入辅助信息确定单元104和接入辅助信息发送单元105。处理电路101可被配置为执行图21B中所示的通信方法。处理电路101可以指在低频基站中执行功能的数字电路系统、模拟电路系统或混合信号(模拟信号和数字信号的组合)电路系统的各种实现。
处理电路101的CSI矩阵获取单元102被配置为基于经由低频通信从UE接收的参考信号，获取低频CSI矩阵，即执行图21B中的步骤S101。UE可以在需要接入到高频基站时向低频基站发送参考信号，例如CSI-RS或SRS。
候选高频基站确定单元103被配置为基于CSI矩阵获取单元102获取的CSI矩阵，利用神经网络从多个高频基站中确定适于与UE进行高频通信的候选高频基站， 即执行图21B中的步骤S102。神经网络可以是如图9或图12中所示的卷积神经网络。所确定的候选高频基站可以是被预测为与UE的高频通信具有最小路径损耗的高频基站，或者可以是被预测为与UE的高频通信在UE处具有最大接收功率的高频基站。
接入辅助信息确定单元104被配置为确定与候选高频基站相关联的接入辅助信息，即执行图21B中的步骤S103。接入辅助信息可以包括候选高频基站广播的SSB所在的频率位置或指示SSB的频率位置的GSCN。此外，接入辅助信息还可以包括利用神经网络的分支网络预测的与候选高频基站对应的路径损耗值或RSRP值。在可以利用神经网络的分支网络预测候选高频基站应使用的最优波束的情况下，接入辅助信息还可以包括与最优波束对应的SSB所在的频率位置(或指示频率位置的GSCN)以及索引。
接入辅助信息发送单元105被配置为将接入辅助信息确定单元104确定的接入辅助信息发送给UE，即执行图21B中的步骤S104。这些接入辅助信息将帮助UE决策是否接入高频基站，并且可以缩小UE搜索SSB的范围，从而提高高频通信初始接入的效率和质量。
电子设备100还可以包括通信单元106。通信单元106可以被配置为在处理电路101的控制下与UE进行通信。在一个示例中，通信单元106可以被实现为收发机，包括天线阵列和/或射频链路等通信部件。通信单元106用虚线绘出，因为它还可以位于电子设备100外。
电子设备100还可以包括存储器107。存储器107可以存储各种数据和指令，例如用于电子设备100操作的程序和数据、由处理电路101产生的各种数据、由通信单元106发送或接收的各种控制信令或业务数据等。存储器107用虚线绘出，因为它还可以位于处理电路101内或者位于电子设备100外。
图22A是例示了根据本公开的电子设备200的框图。电子设备200可以是UE或其部件。
如图22A中所示，电子设备200包括处理电路201。处理电路201至少包括参考信号发送单元202、接入辅助信息接收单元203和接入单元204。处理电路201可被配置为执行图22B中所示的通信方法。处理电路201可以指在UE中执行功能的数字电路 系统、模拟电路系统或混合信号(模拟信号和数字信号的组合)电路系统的各种实现。
参考信号发送单元202可以被配置为经由低频链路向低频基站发送参考信号以供低频基站获取CSI矩阵，即执行图22B中的步骤S201。由参考信号发送单元202发送的参考信号可以是例如CSI-RS或SRS的低频参考信号。
接入辅助信息接收单元203可以被配置为接收由低频基站确定的与候选高频基站相关联的接入辅助信息，即执行图22B中的步骤S202。候选高频基站是低频基站将所获取的CSI矩阵输入到如图9或12中所示的卷积神经网络确定的。候选高频基站被预测为适于与UE进行高频通信的高频基站，例如与UE的高频通信具有最小路径损耗，或者与UE的高频通信在UE处具有最大接收功率。由接入辅助信息接收单元203接收的接入辅助信息可以包括候选高频基站广播的SSB的频率位置和/或索引。
接入单元204可以被配置为基于所接收的接入辅助信息，接入候选高频基站，即执行图22B中的步骤S203。接入单元204可以在接入辅助信息所指示的时频资源中搜索候选高频基站发送的一个或多个SSB，并通过解码SSB来获得下行链路同步和上行链路同步。相比于UE在所属PLMN的所有同步信道频率位置和所有时域位置上盲搜索，利用接入辅助信息的初始接入具有更高的效率。
电子设备200还可以包括通信单元206。通信单元206可以被配置为在处理电路201的控制下与基站进行通信。在一个示例中，通信单元206可以被实现为发射机或收发机，包括天线阵列和/或射频链路等通信部件。通信单元206用虚线绘出，因为它还可以位于电子设备200外。
电子设备200还可以包括存储器207。存储器207可以存储各种数据和指令、用于电子设备200操作的程序和数据、由处理电路201产生的各种数据、将由通信单元207发送的数据等。存储器207用虚线绘出，因为它还可以位于处理电路201内或者位于电子设备200外。
图23A是例示了根据本公开的电子设备300的框图。电子设备300可以是高频基站或其部件。
如图23A中所示，电子设备300包括处理电路301。处理电路301至少包括接收单元302和通信建立单元303。处理电路301可被配置为执行图23B中所示的通信方法。处理电路301可以指在基站设备中执行功能的数字电路系统、模拟电路系统或混 合信号(模拟信号和数字信号的组合)电路系统的各种实现。
接收单元302可以被配置为从低频基站接收UE的标识码和关于高频基站的可用于该UE的波束的信息，即执行图23B中的步骤S301。其中所述波束是低频基站通过将基于UE经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的。神经网络可以是如图9或12中所示的卷积神经网络，其包括用于预测高频基站的最优波束的分支网络。
通信建立单元303可以被配置为利用所述波束建立与UE的高频通信，即执行图23B中的步骤S302。例如，在UE初始接入该高频基站时，高频基站的通信建立单元303可以使用低频基站推荐的波束向UE发送信令。此外，通信建立单元303还可以通过波束训练对波束进行精细化。
电子设备300还可以包括通信单元306。通信单元306可以被配置为在处理电路301的控制下与基站进行通信。在一个示例中，通信单元306可以被实现为发射机或收发机，包括天线阵列和/或射频链路等通信部件。通信单元306用虚线绘出，因为它还可以位于电子设备300外。
电子设备300还可以包括存储器307。存储器307可以存储各种数据和指令、用于电子设备300操作的程序和数据、由处理电路301产生的各种数据、将由通信单元307发送的数据等。存储器307用虚线绘出，因为它还可以位于处理电路301内或者位于电子设备300外。
上面已经详细描述了本公开的实施例的各个方面，但是应注意，上面为了描述了所示出的天线阵列的结构、布置、类型、数量等，端口，参考信号，通信设备，通信方法等等，都不是为了将本公开的方面限制到这些具体的示例。
应当理解，上述各实施例中描述的电子设备100、200、300的各个单元仅是根据其所实现的具体功能划分的逻辑模块，而不是用于限制具体的实现方式。在实际实现时，上述各单元可被实现为独立的物理实体，或者也可以由单个实体(例如，处理器(CPU或DSP等)、集成电路等)来实现。
应当理解，上面各实施例中描述的处理电路101、201或301可以包括例如诸如集成电路(IC)、专用集成电路(ASIC)之类的电路、单独处理器核心的部分或电路、整个处理器核心、单独的处理器、诸如现场可编程们阵列(FPGA)的可编程硬件设备、 和/或包括多个处理器的系统。存储器107、207或307可以是易失性存储器和/或非易失性存储器。例如，存储器107、207或307可以包括但不限于随机存储存储器(RAM)、动态随机存储存储器(DRAM)、静态随机存取存储器(SRAM)、只读存储器(ROM)、闪存存储器。
应当理解，上述各实施例中描述的电子设备100、200或300的各个单元仅是根据其所实现的具体功能划分的逻辑模块，而不是用于限制具体的实现方式。在实际实现时，上述各单元可被实现为独立的物理实体，或者也可以由单个实体(例如，处理器(CPU或DSP等)、集成电路等)来实现。
【本公开的示例性实现】
根据本公开的实施例，可以想到各种实现本公开的概念的实现方式，包括但不限于：
1)、一种用于低频基站的电子设备，包括：处理电路，被配置为：基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；确定与候选高频基站相关联的接入辅助信息；以及将所述接入辅助信息发送给所述用户设备。
2)、如1)所述的电子设备，其中所述处理电路还被配置为：通过将所述CSI矩阵与所述多个高频基站的位置数据输入到所述神经网络中，确定与所述用户设备的高频通信具有最小路径损耗的高频基站作为候选高频基站。
3)、如2)所述的电子设备，其中所述处理电路还被配置为：将与候选高频基站对应的路径损耗值发送给所述用户设备。
4)、如2)所述的电子设备，其中所述处理电路还被配置为：在与候选高频基站对应的路径损耗值低于预定阈值的情况下，指示所述用户设备开启高频通信模块。
5)、如1)所述的电子设备，其中所述处理电路还被配置为：通过将所述CSI矩阵与所述多个高频基站的位置数据和发射功率输入到所述神经网络中，确定在所述用户设备处高频通信具有最大接收功率的高频基站作为候选高频基站。
6)、如5)所述的电子设备，其中所述处理电路还被配置为：将与候选高频基 站对应的接收功率值发送给所述用户设备。
7)、如5)所述的电子设备，其中所述处理电路还被配置为：在与候选高频基站对应的接收功率值超过预定阈值的情况下，指示所述用户设备开启高频通信模块。
8)、如1)所述的电子设备，其中，其中所述处理电路还被配置为：利用所述神经网络，确定候选高频基站的用于所述用户设备的波束；将所述用户设备的标识码和所述波束的信息通知给候选高频基站。
9)、如8)所述的电子设备，其中关于所述波束的信息是与所述波束对应的同步信号/物理广播信道块(SSB)的索引。
10)、如1)所述的电子设备，其中所述接入辅助信息包括候选高频基站的同步信号/物理广播信道块(SSB)的频率位置。
11)、如8)所述的电子设备，其中所述接入辅助信息包括候选高频基站的同步信号/物理广播信道块(SSB)的频率位置和索引。
12)、如1)所述的电子设备，其中所述处理电路还被配置为：从与所述多个高频基站中的任一高频基站进行高频通信的用户设备，接收经由低频通信发送的参考信号、以及所述任一高频基站的标识信息；通过使用基于所述参考信号获取的CSI矩阵和所述任一高频基站的位置数据作为输入以及使用所述任一高频基站的标识信息作为输出，更新所述神经网络的参数。
13)、如2)所述的电子设备，其中所述处理电路还被配置为：从与所述多个高频基站中的任一高频基站进行高频通信的用户设备，接收由该用户设备估计的高频通信的路径损耗值；通过使用所述路径损耗值作为输出，更新所述神经网络的分支网络的参数。
14)、如2)所述的电子设备，其中所述处理电路还被配置为：从所述多个高频基站中与用户设备进行高频通信的高频基站，接收所述用户设备的标识码和关于用于所述高频通信的波束的信息；通过使用所述关于用于所述高频通信的波束的信息作为输出，更新所述神经网络的分支网络的参数。
15)、如1)所述的电子设备，其中所述低频通信工作在LTE频段、LTE-A频段或sub-6GHz频段，并且其中，所述高频通信工作在毫米波频段。
16)、一种用于用户设备的电子设备，包括：处理电路，被配置为：经由低频通信向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及利用所述接入辅助信息，接入所述候选高频基站。
17)、如16)所述的电子设备，其中候选高频基站是由所述神经网络预测的与所述用户设备的高频通信的路径损耗最小的高频基站。
18)、如16)所述的电子设备，其中候选高频基站是由所述神经网络预测的与所述用户设备的高频通信在所述用户设备处的接收功率最大的高频基站。
19)、如17)所述的电子设备，其中所述处理电路还被配置为：从低频基站接收与候选高频基站对应的路径损耗值；以及在所述路径损耗值低于预定阈值的情况下，开启高频通信模块以接入候选高频基站。
20)、如18)所述的电子设备，其中所述处理电路还被配置为：从低频基站接收与候选高频基站对应的接收功率值；以及在所述接收功率值超过预定阈值的情况下，开启高频通信模块以接入候选高频基站。
21)、如16)所述的电子设备，其中所述接入辅助信息包括候选高频基站的同步信号/物理广播信道块(SSB)的频率位置。
22)、如21)所述的电子设备，其中所述接入辅助信息还包括候选高频基站的SSB的索引。
23)、如19)或20)所述的电子设备，其中所述处理电路还被配置为：根据用户设备的当前电量、连接偏好、高频通信的传输成功率来调整所述预定阈值。
24)、一种用于高频基站的电子设备，包括：处理电路，被配置为：从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束的信息，其中所述波束是所述低频基站通过将基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；利用所述波束建立与所述用户设备的高频通信。
25)、如24)所述的电子设备，其中所述处理电路还被配置为：基于所述波束， 通过波束扫描来确定用于与所述用户设备的高频通信的更窄波束。
26)、如25)所述的电子设备，其中所述波束包括具有不同优先级的多个波束，并且所述处理电路还被配置为：基于所述多个波束的优先级来扫描波束。
27)、一种训练神经网络的方法，包括：从用户设备接收经由低频通信发送的参考信号；基于所述参考信号获取信道状态信息(CSI)矩阵；从用户设备接收与之进行高频通信的高频基站的标识信息；通过使用所述CSI矩阵作为输入以及使用所述高频基站的标识信息作为输出进行深度学习，确定所述神经网络的参数。
28)、如27)所述的方法，还包括：从用户设备接收由该用户设备估计的高频通信的路径损耗值；通过使用所述路径损耗值作为输出，确定所述神经网络的分支网络的参数。
29)、如27)所述的方法，还包括：从用户设备接收由该用户设备测量的对于高频通信的接收功率值；通过使用所述接收功率值作为输出，确定所述神经网络的分支网络的参数。
30)、如27)所述的方法，还包括：从与用户设备进行高频通信的高频基站，接收所述用户设备的标识码和关于用于所述高频通信的波束的信息；通过使用关于用于所述高频通信的波束的信息作为输出，确定所述神经网络的分支网络的参数。
31)、一种通信方法，包括：基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；确定与候选高频基站相关联的接入辅助信息；以及将所述接入辅助信息发送给所述用户设备。
32)、一种通信方法，包括：经由低频通信向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及利用所述接入辅助信息，接入所述候选高频基站。
33)、一种通信方法，包括：从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束的信息，其中所述波束是所述低频基站通过将 基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；利用所述波束建立与所述用户设备的高频通信。
34)、一种存储有可执行指令的非暂时性计算机可读存储介质，所述可执行指令当被执行时实现如27)至33)中任一项所述的方法。
【本公开的应用实例】
本公开中描述的技术能够应用于各种产品。
例如，根据本公开的实施例的电子设备100、300可以被实现为各种基站或者安装在基站中，电子设备200可以被实现为各种用户设备或被安装在各种用户设备中。
根据本公开的实施例的通信方法可以由各种基站或用户设备实现；根据本公开的实施例的方法和操作可以体现为计算机可执行指令，存储在非暂时性计算机可读存储介质中，并可以由各种基站或用户设备执行以实现上面所述的一个或多个功能。
根据本公开的实施例的技术可以制成各个计算机程序产品，被用于各种基站或用户设备以实现上面所述的一个或多个功能。
本公开中所说的基站可以被实现为任何类型的基站，优选地，诸如3GPP的5G NR标准中定义的宏gNB和ng-eNB。gNB可以是覆盖比宏小区小的小区的gNB，诸如微微gNB、微gNB和家庭(毫微微)gNB。代替地，基站可以被实现为任何其他类型的基站，诸如NodeB、eNodeB和基站收发台(BTS)。基站还可以包括：被配置为控制无线通信的主体以及设置在与主体不同的地方的一个或多个远程无线头端(RRH)、无线中继站、无人机塔台、自动化工厂中的控制节点等。
用户设备可以被实现为移动终端(诸如智能电话、平板个人计算机(PC)、笔记本式PC、便携式游戏终端、便携式/加密狗型移动路由器和数字摄像装置)或者车载终端(诸如汽车导航设备)。用户设备还可以被实现为执行机器对机器(M2M)通信的终端(也称为机器类型通信(MTC)终端)、无人机、自动化工厂中的传感器和执行器等。此外，用户设备可以为安装在上述终端中的每个终端上的无线通信模块(诸如包括单个晶片的集成电路模块)。
基站的第一应用示例
图24是示出可以应用本公开内容的技术的基站的示意性配置的第一示例的框图。在图24中，基站可以实现为gNB 1400。gNB 1400包括多个天线1410以及基站设备1420。基站设备1420和每个天线1410可以经由RF线缆彼此连接。在一种实现方式中，此处的gNB 1400(或基站设备1420)可以对应于上述电子设备100或300。
天线1410包括多个天线元件。天线1410例如可以被布置成天线阵列矩阵，并且用于基站设备1420发送和接收无线信号。例如，多个天线1410可以与gNB 1400使用的多个频段兼容。
基站设备1420包括控制器1421、存储器1422、网络接口1423以及无线通信接口1425。
控制器1421可以为例如CPU或DSP，并且操作基站设备1420的较高层的各种功能。例如，控制器1421可以包括上面所述的处理电路101或301，以执行图21B或23B中描述的通信方法，或者控制基站设备1420的各个部件。例如，控制器1421根据由无线通信接口1425处理的信号中的数据来生成数据分组，并经由网络接口1423来传递所生成的分组。控制器1421可以对来自多个基带处理器的数据进行捆绑以生成捆绑分组，并传递所生成的捆绑分组。控制器1421可以具有执行如下控制的逻辑功能：该控制诸如为无线资源控制、无线承载控制、移动性管理、接纳控制和调度。该控制可以结合附近的gNB或核心网节点来执行。存储器1422包括RAM和ROM，并且存储由控制器1421执行的程序和各种类型的控制数据(诸如终端列表、传输功率数据以及调度数据)。
网络接口1423为用于将基站设备1420连接至核心网1424(例如，5G核心网)的通信接口。控制器1421可以经由网络接口1423而与核心网节点或另外的gNB进行通信。在此情况下，gNB 1400与核心网节点或其他gNB可以通过逻辑接口(诸如NG接口和Xn接口)而彼此连接。网络接口1423还可以为有线通信接口或用于无线回程线路的无线通信接口。如果网络接口1423为无线通信接口，则与由无线通信接口1425使用的频段相比，网络接口1423可以使用较高频段用于无线通信。
无线通信接口1425支持任何蜂窝通信方案(诸如5G NR)，并且经由天线1410来提供到位于gNB 1400的小区中的终端的无线连接。无线通信接口1425通常可以包括 例如基带(BB)处理器1426和RF电路1427。BB处理器1426可以执行例如编码/解码、调制/解调以及复用/解复用，并且执行各层(例如物理层、MAC层、RLC层、PDCP层、SDAP层)的各种类型的信号处理。代替控制器1421，BB处理器1426可以具有上述逻辑功能的一部分或全部。BB处理器1426可以为存储通信控制程序的存储器，或者为包括被配置为执行程序的处理器和相关电路的模块。更新程序可以使BB处理器1426的功能改变。该模块可以为插入到基站设备1420的槽中的卡或刀片。可替代地，该模块也可以为安装在卡或刀片上的芯片。同时，RF电路1427可以包括例如混频器、滤波器和放大器，并且经由天线1410来传送和接收无线信号。虽然图24示出一个RF电路1427与一根天线1410连接的示例，但是本公开并不限于该图示，而是一个RF电路1427可以同时连接多根天线1410。
如图24所示，无线通信接口1425可以包括多个BB处理器1426。例如，多个BB处理器1426可以与gNB 1400使用的多个频段兼容。如图24所示，无线通信接口1425可以包括多个RF电路1427。例如，多个RF电路1427可以与多个天线元件兼容。虽然图24示出其中无线通信接口1425包括多个BB处理器1426和多个RF电路1427的示例，但是无线通信接口1425也可以包括单个BB处理器1426或单个RF电路1427。
在图24中示出的gNB 1400中，参照图21A描述的处理电路101中包括的一个或多个单元(例如接入辅助信息发送单元105)或参照图23A描述的处理电路301中包括的一个或多个单元(例如接收单元302)可被实现在无线通信接口825中。可替代地，这些组件中的至少一部分可被实现在控制器821中。例如，gNB 1400包含无线通信接口1425的一部分(例如，BB处理器1426)或者整体，和/或包括控制器1421的模块，并且一个或多个组件可被实现在模块中。在这种情况下，模块可以存储用于允许处理器起一个或多个组件的作用的程序(换言之，用于允许处理器执行一个或多个组件的操作的程序)，并且可以执行该程序。作为另一个示例，用于允许处理器起一个或多个组件的作用的程序可被安装在gNB 1400中，并且无线通信接口1425(例如，BB处理器1426)和/或控制器1421可以执行该程序。如上所述，作为包括一个或多个组件的装置，gNB 1400、基站设备1420或模块可被提供，并且用于允许处理器起一个或多个组件的作用的程序可被提供。另外，将程序记录在其中的可读介质可被提供。
基站的第二应用示例
图25是示出可以应用本公开的技术的基站的示意性配置的第二示例的框图。在图25中，基站被示出为gNB 1530。gNB 1530包括多个天线1540、基站设备1550和RRH 1560。RRH 1560和每个天线1540可以经由RF线缆而彼此连接。基站设备1550和RRH 1560可以经由诸如光纤线缆的高速线路而彼此连接。在一种实现方式中，此处的gNB 1530(或基站设备1550)可以对应于上述电子设备100或300。
天线1540包括多个天线元件。天线1540例如可以被布置成天线阵列矩阵，并且用于基站设备1550发送和接收无线信号。例如，多个天线1540可以与gNB 1530使用的多个频段兼容。
基站设备1550包括控制器1551、存储器1552、网络接口1553、无线通信接口1555以及连接接口1557。控制器1551、存储器1552和网络接口1553与参照图24描述的控制器1421、存储器1422和网络接口1423相同。
无线通信接口1555支持任何蜂窝通信方案(诸如5G NR)，并且经由RRH 1560和天线1540来提供到位于与RRH 1560对应的扇区中的终端的无线通信。无线通信接口1555通常可以包括例如BB处理器1556。除了BB处理器1556经由连接接口1557连接到RRH 1560的RF电路1564之外，BB处理器1556与参照图24描述的BB处理器1426相同。如图25所示，无线通信接口1555可以包括多个BB处理器1556。例如，多个BB处理器1556可以与gNB 1530使用的多个频段兼容。虽然图25示出其中无线通信接口1555包括多个BB处理器1556的示例，但是无线通信接口1555也可以包括单个BB处理器1556。
连接接口1557为用于将基站设备1550(无线通信接口1555)连接至RRH 1560的接口。连接接口1557还可以为用于将基站设备1550(无线通信接口1555)连接至RRH 1560的上述高速线路中的通信的通信模块。
RRH 1560包括连接接口1561和无线通信接口1563。
连接接口1561为用于将RRH 1560(无线通信接口1563)连接至基站设备1550的接口。连接接口1561还可以为用于上述高速线路中的通信的通信模块。
无线通信接口1563经由天线1540来传送和接收无线信号。无线通信接口1563 通常可以包括例如RF电路1564。RF电路1564可以包括例如混频器、滤波器和放大器，并且经由天线1540来传送和接收无线信号。虽然图25示出一个RF电路1564与一根天线1540连接的示例，但是本公开并不限于该图示，而是一个RF电路1564可以同时连接多根天线1540。
如图25所示，无线通信接口1563可以包括多个RF电路1564。例如，多个RF电路1564可以支持多个天线元件。虽然图25示出其中无线通信接口1563包括多个RF电路1564的示例，但是无线通信接口1563也可以包括单个RF电路1564。
在图25中示出的gNB 1500中，参照图21A描述的处理电路101中包括的一个或多个单元(例如接入辅助信息发送单元105)或参照图23A描述的处理电路301中包括的一个或多个单元(例如接收单元302)可被实现在无线通信接口1525中。可替代地，这些组件中的至少一部分可被实现在控制器1521中。例如，gNB 1500包含无线通信接口1525的一部分(例如，BB处理器1526)或者整体，和/或包括控制器1521的模块，并且一个或多个组件可被实现在模块中。在这种情况下，模块可以存储用于允许处理器起一个或多个组件的作用的程序(换言之，用于允许处理器执行一个或多个组件的操作的程序)，并且可以执行该程序。作为另一个示例，用于允许处理器起一个或多个组件的作用的程序可被安装在gNB 1500中，并且无线通信接口1525(例如，BB处理器1526)和/或控制器1521可以执行该程序。如上所述，作为包括一个或多个组件的装置，gNB 1500、基站设备1520或模块可被提供，并且用于允许处理器起一个或多个组件的作用的程序可被提供。另外，将程序记录在其中的可读介质可被提供。
用户设备的第一应用示例
图26是示出可以应用本公开内容的技术的智能电话1600的示意性配置的示例的框图。在一个示例中，智能电话1600可以被实现为本公开中描述的电子设备200。
智能电话1600包括处理器1601、存储器1602、存储装置1603、外部连接接口1604、摄像装置1606、传感器1607、麦克风1608、输入装置1609、显示装置1610、扬声器1611、无线通信接口1612、一个或多个天线开关1615、一个或多个天线1616、总线1617、电池1618以及辅助控制器1619。
处理器1601可以为例如CPU或片上系统(SoC)，并且控制智能电话1600的应用层和另外层的功能。处理器1601可以包括或充当参照图22A描述的处理电路201。存储器1602包括RAM和ROM，并且存储数据和由处理器1601执行的程序，以实现参照图22B所述的通信方法。存储装置1603可以包括存储介质，诸如半导体存储器和硬盘。外部连接接口1604为用于将外部装置(诸如存储卡和通用串行总线(USB)装置)连接至智能电话1600的接口。
摄像装置1606包括图像传感器(诸如电荷耦合器件(CCD)和互补金属氧化物半导体(CMOS))，并且生成捕获图像。传感器1607可以包括一组传感器，诸如测量传感器、陀螺仪传感器、地磁传感器和加速度传感器。麦克风1608将输入到智能电话1600的声音转换为音频信号。输入装置1609包括例如被配置为检测显示装置1610的屏幕上的触摸的触摸传感器、小键盘、键盘、按钮或开关，并且接收从用户输入的操作或信息。显示装置1610包括屏幕(诸如液晶显示器(LCD)和有机发光二极管(OLED)显示器)，并且显示智能电话1600的输出图像。扬声器1611将从智能电话1600输出的音频信号转换为声音。
无线通信接口1612支持任何蜂窝通信方案(诸如4G LTE或5G NR等等)，并且执行无线通信。无线通信接口1612通常可以包括例如BB处理器1613和RF电路1614。BB处理器1613可以执行例如编码/解码、调制/解调以及复用/解复用，并且执行用于无线通信的各种类型的信号处理。同时，RF电路1614可以包括例如混频器、滤波器和放大器，并且经由天线1616来传送和接收无线信号。无线通信接口1612可以为其上集成有BB处理器1613和RF电路1614的一个芯片模块。如图26所示，无线通信接口1612可以包括多个BB处理器1613和多个RF电路1614。虽然图26示出其中无线通信接口1612包括多个BB处理器1613和多个RF电路1614的示例，但是无线通信接口1612也可以包括单个BB处理器1613或单个RF电路1614。
此外，除了蜂窝通信方案之外，无线通信接口1612可以支持另外类型的无线通信方案，诸如短距离无线通信方案、近场通信方案和无线局域网(LAN)方案。在此情况下，无线通信接口1612可以包括针对每种无线通信方案的BB处理器1613和RF电路1614。
天线开关1615中的每一个在包括在无线通信接口1612中的多个电路(例如用于 不同的无线通信方案的电路)之间切换天线1616的连接目的地。
天线1616包括多个天线元件。天线1616例如可以被布置成天线阵列矩阵，并且用于无线通信接口1612传送和接收无线信号。智能电话1600可以包括一个或多个天线面板(未示出)。
此外，智能电话1600可以包括针对每种无线通信方案的天线1616。在此情况下，天线开关1615可以从智能电话1600的配置中省略。
总线1617将处理器1601、存储器1602、存储装置1603、外部连接接口1604、摄像装置1606、传感器1607、麦克风1608、输入装置1609、显示装置1610、扬声器1611、无线通信接口1612以及辅助控制器1619彼此连接。电池1618经由馈线向图26所示的智能电话1600的各个块提供电力，馈线在图中被部分地示为虚线。辅助控制器1619例如在睡眠模式下操作智能电话1600的最小必需功能。
在图26中示出的智能电话1600中，参照图22A描述的处理电路201中包括的一个或多个组件(例如参考信号发送单元202、接入辅助信息接收单元203)可被实现在无线通信接口1612中。可替代地，这些组件中的至少一部分可被实现在处理器1601或者辅助控制器1619中。作为一个示例，智能电话1600包含无线通信接口1612的一部分(例如，BB处理器1613)或者整体，和/或包括处理器1601和/或辅助控制器1619的模块，并且一个或多个组件可被实现在该模块中。在这种情况下，该模块可以存储允许处理起一个或多个组件的作用的程序(换言之，用于允许处理器执行一个或多个组件的操作的程序)，并且可以执行该程序。作为另一个示例，用于允许处理器起一个或多个组件的作用的程序可被安装在智能电话1600中，并且无线通信接口1612(例如，BB处理器1613)、处理器1601和/或辅助控制器1619可以执行该程序。如上所述，作为包括一个或多个组件的装置，智能电话1600或者模块可被提供，并且用于允许处理器起一个或多个组件的作用的程序可被提供。另外，将程序记录在其中的可读介质可被提供。
用户设备的第二应用示例
图27是示出可以应用本公开的技术的汽车导航设备1720的示意性配置的示例的框图。汽车导航设备1720可以被实现为参照图22A描述的电子设备200。汽车导航设 备1720包括处理器1721、存储器1722、全球定位系统(GPS)模块1724、传感器1725、数据接口1726、内容播放器1727、存储介质接口1728、输入装置1729、显示装置1730、扬声器1731、无线通信接口1733、一个或多个天线开关1736、一个或多个天线1737以及电池1738。在一个示例中，汽车导航设备1720可以被实现为本公开中描述的UE。
处理器1721可以为例如CPU或SoC，并且控制汽车导航设备1720的导航功能和另外的功能。存储器1722包括RAM和ROM，并且存储数据和由处理器1721执行的程序。
GPS模块1724使用从GPS卫星接收的GPS信号来测量汽车导航设备1720的位置(诸如纬度、经度和高度)。传感器1725可以包括一组传感器，诸如陀螺仪传感器、地磁传感器和空气压力传感器。数据接口1726经由未示出的终端而连接到例如车载网络1741，并且获取由车辆生成的数据(诸如车速数据)。
内容播放器1727再现存储在存储介质(诸如CD和DVD)中的内容，该存储介质被插入到存储介质接口1728中。输入装置1729包括例如被配置为检测显示装置1730的屏幕上的触摸的触摸传感器、按钮或开关，并且接收从用户输入的操作或信息。显示装置1730包括诸如LCD或OLED显示器的屏幕，并且显示导航功能的图像或再现的内容。扬声器1731输出导航功能的声音或再现的内容。
无线通信接口1733支持任何蜂窝通信方案(诸如4G LTE或5G NR)，并且执行无线通信。无线通信接口1733通常可以包括例如BB处理器1734和RF电路1735。BB处理器1734可以执行例如编码/解码、调制/解调以及复用/解复用，并且执行用于无线通信的各种类型的信号处理。同时，RF电路1735可以包括例如混频器、滤波器和放大器，并且经由天线1737来传送和接收无线信号。无线通信接口1733还可以为其上集成有BB处理器1734和RF电路1735的一个芯片模块。如图27所示，无线通信接口1733可以包括多个BB处理器1734和多个RF电路1735。虽然图27示出其中无线通信接口1733包括多个BB处理器1734和多个RF电路1735的示例，但是无线通信接口1733也可以包括单个BB处理器1734或单个RF电路1735。
此外，除了蜂窝通信方案之外，无线通信接口1733可以支持另外类型的无线通 信方案，诸如短距离无线通信方案、近场通信方案和无线LAN方案。在此情况下，针对每种无线通信方案，无线通信接口1733可以包括BB处理器1734和RF电路1735。
天线开关1736中的每一个在包括在无线通信接口1733中的多个电路(诸如用于不同的无线通信方案的电路)之间切换天线1737的连接目的地。
天线1737包括多个天线元件。天线1737例如可以被布置成天线阵列矩阵，并且用于无线通信接口1733传送和接收无线信号。
此外，汽车导航设备1720可以包括针对每种无线通信方案的天线1737。在此情况下，天线开关1736可以从汽车导航设备1720的配置中省略。
电池1738经由馈线向图27所示的汽车导航设备1720的各个块提供电力，馈线在图中被部分地示为虚线。电池1738累积从车辆提供的电力。
在图27中示出的汽车导航装置1720中，参照图22A描述的处理电路201中包括的一个或多个组件(例如参考信号发送单元202、接入辅助信息接收单元203)可被实现在无线通信接口1733中。可替代地，这些组件中的至少一部分可被实现在处理器1721中。作为一个示例，汽车导航装置1720包含无线通信接口1733的一部分(例如，BB处理器1734)或者整体，和/或包括处理器1721的模块，并且一个或多个组件可被实现在该模块中。在这种情况下，该模块可以存储允许处理起一个或多个组件的作用的程序(换言之，用于允许处理器执行一个或多个组件的操作的程序)，并且可以执行该程序。作为另一个示例，用于允许处理器起一个或多个组件的作用的程序可被安装在汽车导航装置1720中，并且无线通信接口1733(例如，BB处理器1734)和/或处理器1721可以执行该程序。如上所述，作为包括一个或多个组件的装置，汽车导航装置1720或者模块可被提供，并且用于允许处理器起一个或多个组件的作用的程序可被提供。另外，将程序记录在其中的可读介质可被提供。
本公开的技术也可以被实现为包括汽车导航设备1720、车载网络1741以及车辆模块1742中的一个或多个块的车载系统(或车辆)1740。车辆模块1742生成车辆数据(诸如车速、发动机速度和故障信息)，并且将所生成的数据输出至车载网络1741。
以上参照附图描述了本公开的示例性实施例，但是本公开当然不限于以上示例。本领域技术人员可在所附权利要求的范围内得到各种变更和修改，并且应理解这些变 更和修改自然将落入本公开的技术范围内。
例如，在以上实施例中包括在一个单元中的多个功能可以由分开的装置来实现。替选地，在以上实施例中由多个单元实现的多个功能可分别由分开的装置来实现。另外，以上功能之一可由多个单元来实现。无需说，这样的配置包括在本公开的技术范围内。
在该说明书中，流程图中所描述的步骤不仅包括以所述顺序按时间序列执行的处理，而且包括并行地或单独地而不是必须按时间序列执行的处理。此外，甚至在按时间序列处理的步骤中，无需说，也可以适当地改变该顺序。
虽然已经详细说明了本公开及其优点，但是应当理解在不脱离由所附的权利要求所限定的本公开的精神和范围的情况下可以进行各种改变、替代和变换。而且，本公开实施例的术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、物品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、物品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、物品或者设备中还存在另外的相同要素。

Claims (34)

  1. 一种用于低频基站的电子设备，包括：
    处理电路，被配置为：
    基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；
    基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；
    确定与候选高频基站相关联的接入辅助信息；以及
    将所述接入辅助信息发送给所述用户设备。
  2. 如权利要求1所述的电子设备，其中所述处理电路还被配置为：
    通过将所述CSI矩阵与所述多个高频基站的位置数据输入到所述神经网络中，确定与所述用户设备的高频通信具有最小路径损耗的高频基站作为所述候选高频基站。
  3. 如权利要求2所述的电子设备，其中所述处理电路还被配置为：
    将与所述候选高频基站对应的路径损耗值发送给所述用户设备。
  4. 如权利要求2所述的电子设备，其中所述处理电路还被配置为：
    在与所述候选高频基站对应的路径损耗值低于预定阈值的情况下，指示所述用户设备开启高频通信模块。
  5. 如权利要求1所述的电子设备，其中所述处理电路还被配置为：
    通过将所述CSI矩阵与所述多个高频基站的位置数据和发射功率输入到所述神经网络中，确定在所述用户设备处高频通信具有最大接收功率的高频基站作为所述候选高频基站。
  6. 如权利要求5所述的电子设备，其中所述处理电路还被配置为：
    将与所述候选高频基站对应的接收功率值发送给所述用户设备。
  7. 如权利要求5所述的电子设备，其中所述处理电路还被配置为：
    在与所述候选高频基站对应的接收功率值超过预定阈值的情况下，指示所述用户设备开启高频通信模块。
  8. 如权利要求1所述的电子设备，其中所述处理电路还被配置为：
    利用所述神经网络，确定所述候选高频基站的用于所述用户设备的波束；
    将所述用户设备的标识码和所述波束的信息通知给所述候选高频基站。
  9. 如权利要求8所述的电子设备，其中关于所述波束的信息是与所述波束对应的同步信号/物理广播信道块(SSB)的索引。
  10. 如权利要求1所述的电子设备，其中所述接入辅助信息包括所述候选高频基站的同步信号/物理广播信道块(SSB)的频率位置。
  11. 如权利要求8所述的电子设备，其中所述接入辅助信息包括所述候选高频基站的同步信号/物理广播信道块(SSB)的频率位置和索引。
  12. 如权利要求1所述的电子设备，其中所述处理电路还被配置为：
    从与所述多个高频基站中的任一高频基站进行高频通信的用户设备，接收经由低频通信发送的参考信号、以及所述任一高频基站的标识信息；
    通过使用基于所述参考信号获取的CSI矩阵和所述任一高频基站的位置数据作为输入以及使用所述任一高频基站的标识信息作为输出，更新所述神经网络的参数。
  13. 如权利要求2所述的电子设备，其中所述处理电路还被配置为：
    从与所述多个高频基站中的任一高频基站进行高频通信的用户设备，接收由该用户设备估计的高频通信的路径损耗值；
    通过使用所述路径损耗值作为输出，更新所述神经网络的分支网络的参数。
  14. 如权利要求2所述的电子设备，其中所述处理电路还被配置为：
    从所述多个高频基站中与用户设备进行高频通信的高频基站，接收所述用户设备的标识码和关于用于所述高频通信的波束的信息；
    通过使用所述关于用于所述高频通信的波束的信息作为输出，更新所述神经网络的分支网络的参数。
  15. 如权利要求1所述的电子设备，其中所述低频通信工作在LTE频段、LTE-A频段或sub-6GHz频段，并且其中，所述高频通信工作在毫米波频段。
  16. 一种用于用户设备的电子设备，包括：
    处理电路，被配置为：
    经由低频通信向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；
    接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中所述候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及
    利用所述接入辅助信息，接入所述候选高频基站。
  17. 如权利要求16所述的电子设备，其中所述候选高频基站是由所述神经网络预测的与所述用户设备的高频通信的路径损耗最小的高频基站。
  18. 如权利要求16所述的电子设备，其中所述候选高频基站是由所述神经网络预测的与所述用户设备的高频通信在所述用户设备处的接收功率最大的高频基站。
  19. 如权利要求17所述的电子设备，其中所述处理电路还被配置为：
    从低频基站接收与所述候选高频基站对应的路径损耗值；以及
    在所述路径损耗值低于预定阈值的情况下，开启高频通信模块以接入所述候选高频基站。
  20. 如权利要求18所述的电子设备，其中所述处理电路还被配置为：
    从低频基站接收与所述候选高频基站对应的接收功率值；以及
    在所述接收功率值超过预定阈值的情况下，开启高频通信模块以接入所述候选高频基站。
  21. 如权利要求16所述的电子设备，其中所述接入辅助信息包括所述候选高频基站的同步信号/物理广播信道块(SSB)的频率位置。
  22. 如权利要求21所述的电子设备，其中所述接入辅助信息还包括所述候选高频基站的SSB的索引。
  23. 如权利要求19或20所述的电子设备，其中所述处理电路还被配置为：
    根据用户设备的当前电量、连接偏好、高频通信的传输成功率来调整所述预定阈值。
  24. 一种用于高频基站的电子设备，包括：
    处理电路，被配置为：
    从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束的信息，其中所述波束是所述低频基站通过将基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；
    利用所述波束建立与所述用户设备的高频通信。
  25. 如权利要求24所述的电子设备，其中所述处理电路还被配置为：
    基于所述波束，通过波束扫描来确定用于与所述用户设备的高频通信的更窄波束。
  26. 如权利要求25所述的电子设备，其中所述波束包括具有不同优先级的多个波束，并且所述处理电路还被配置为：
    基于所述多个波束的优先级来扫描波束。
  27. 一种训练神经网络的方法，包括：
    从用户设备接收经由低频通信发送的参考信号；
    基于所述参考信号获取信道状态信息(CSI)矩阵；
    从用户设备接收与之进行高频通信的高频基站的标识信息；
    通过使用所述CSI矩阵作为输入以及使用所述高频基站的标识信息作为输出进行深度学习，确定所述神经网络的参数。
  28. 如权利要求27所述的方法，还包括：
    从用户设备接收由该用户设备估计的高频通信的路径损耗值；
    通过使用所述路径损耗值作为输出，确定所述神经网络的分支网络的参数。
  29. 如权利要求27所述的方法，还包括：
    从用户设备接收由该用户设备测量的对于高频通信的接收功率值；
    通过使用所述接收功率值作为输出，确定所述神经网络的分支网络的参数。
  30. 如权利要求27所述的方法，还包括：
    从与用户设备进行高频通信的高频基站，接收所述用户设备的标识码和关于用于所述高频通信的波束的信息；
    通过使用关于用于所述高频通信的波束的信息作为输出，确定所述神经网络的分支网络的参数。
  31. 一种通信方法，包括：
    基于经由低频通信从用户设备接收的参考信号，获取信道状态信息(CSI)矩阵；
    基于所述CSI矩阵，利用神经网络从多个高频基站中确定适于与所述用户设备进行高频通信的候选高频基站；
    确定与候选高频基站相关联的接入辅助信息；以及
    将所述接入辅助信息发送给所述用户设备。
  32. 一种通信方法，包括：
    经由低频通信向低频基站发送参考信号以供低频基站获取信道状态信息(CSI)矩阵；
    接收由低频基站确定的与候选高频基站相关联的接入辅助信息，其中所述候选高频基站是所述低频基站基于所述CSI矩阵利用神经网络确定的适于与所述用户设备进行高频通信的高频基站；以及
    利用所述接入辅助信息，接入所述候选高频基站
  33. 一种通信方法，包括：
    从低频基站接收用户设备的标识码和关于所述高频基站的可用于所述用户设备的波束 的信息，其中所述波束是所述低频基站通过将基于所述用户设备经由低频链路发送的参考信号获取的CSI矩阵输入到神经网络中而确定的；
    利用所述波束建立与所述用户设备的高频通信。
  34. 一种存储有可执行指令的非暂时性计算机可读存储介质，所述可执行指令当被执行时实现如权利要求27-33中任一项所述的方法。
PCT/CN2020/138206 2019-12-26 2020-12-22 无线通信系统中的电子设备、通信方法和存储介质 WO2021129591A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911362252.8A CN113055981A (zh) 2019-12-26 2019-12-26 无线通信系统中的电子设备、通信方法和存储介质
CN201911362252.8 2019-12-26

Publications (1)

Publication Number Publication Date
WO2021129591A1 true WO2021129591A1 (zh) 2021-07-01

Family

ID=76505273

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/138206 WO2021129591A1 (zh) 2019-12-26 2020-12-22 无线通信系统中的电子设备、通信方法和存储介质

Country Status (2)

Country Link
CN (1) CN113055981A (zh)
WO (1) WO2021129591A1 (zh)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113852583A (zh) * 2021-09-18 2021-12-28 江苏亨鑫众联通信技术有限公司 一种解调参考信号动态配置方法
CN114364007A (zh) * 2022-01-10 2022-04-15 西南科技大学 低轨道卫星与无人机蜂窝融合网络的子载波功率控制方法
CN115802480A (zh) * 2022-10-17 2023-03-14 武汉大学 基于5g多波束下行信号的指纹定位方法及系统
WO2024036631A1 (zh) * 2022-08-19 2024-02-22 北京小米移动软件有限公司 一种信息反馈方法、装置、设备及存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113727397A (zh) * 2021-08-25 2021-11-30 深圳国人无线通信有限公司 高低频网络系统协同通信的方法
CN116193585A (zh) * 2021-11-24 2023-05-30 华为技术有限公司 一种通信方法及装置
CN114978842B (zh) * 2022-05-19 2023-05-02 西华大学 一种基于神经网络的二阶段ofdm系统的定时同步方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150257073A1 (en) * 2014-03-10 2015-09-10 Samsung Electronics Co., Ltd. Apparatus and method for determining beam in wireless communication system
CN107306162A (zh) * 2016-04-22 2017-10-31 香港城市大学 多小区及多用户毫米波蜂窝网络中的干扰管理方法
CN107733501A (zh) * 2016-08-10 2018-02-23 中兴通讯股份有限公司 波束管理方法及装置
CN108566621A (zh) * 2018-04-23 2018-09-21 电子科技大学 一种毫米波蜂窝系统小区切换判决方法
CN110063069A (zh) * 2016-12-20 2019-07-26 瑞典爱立信有限公司 多载波网络中的切换过程
CN110401964A (zh) * 2019-08-06 2019-11-01 北京邮电大学 一种面向用户为中心网络基于深度学习的功率控制方法

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20150257073A1 (en) * 2014-03-10 2015-09-10 Samsung Electronics Co., Ltd. Apparatus and method for determining beam in wireless communication system
CN107306162A (zh) * 2016-04-22 2017-10-31 香港城市大学 多小区及多用户毫米波蜂窝网络中的干扰管理方法
CN107733501A (zh) * 2016-08-10 2018-02-23 中兴通讯股份有限公司 波束管理方法及装置
CN110063069A (zh) * 2016-12-20 2019-07-26 瑞典爱立信有限公司 多载波网络中的切换过程
CN108566621A (zh) * 2018-04-23 2018-09-21 电子科技大学 一种毫米波蜂窝系统小区切换判决方法
CN110401964A (zh) * 2019-08-06 2019-11-01 北京邮电大学 一种面向用户为中心网络基于深度学习的功率控制方法

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113852583A (zh) * 2021-09-18 2021-12-28 江苏亨鑫众联通信技术有限公司 一种解调参考信号动态配置方法
CN113852583B (zh) * 2021-09-18 2024-01-26 江苏亨鑫科技有限公司 一种解调参考信号动态配置方法
CN114364007A (zh) * 2022-01-10 2022-04-15 西南科技大学 低轨道卫星与无人机蜂窝融合网络的子载波功率控制方法
WO2024036631A1 (zh) * 2022-08-19 2024-02-22 北京小米移动软件有限公司 一种信息反馈方法、装置、设备及存储介质
CN115802480A (zh) * 2022-10-17 2023-03-14 武汉大学 基于5g多波束下行信号的指纹定位方法及系统
CN115802480B (zh) * 2022-10-17 2024-03-08 武汉大学 基于5g多波束下行信号的指纹定位方法及系统

Also Published As

Publication number Publication date
CN113055981A (zh) 2021-06-29

Similar Documents

Publication Publication Date Title
WO2021129591A1 (zh) 无线通信系统中的电子设备、通信方法和存储介质
CN111052630B (zh) 毫米波系统中的波束选择
US20240137078A1 (en) Electronic device, method and storage medium for wireless communication system
WO2019129006A1 (zh) 用于无线通信系统的电子设备、方法、装置和存储介质
JP2022091924A (ja) 通信システム、基地局装置および通信端末装置
US11824607B2 (en) Electronic device, method and storage medium for wireless communication system
WO2018095305A1 (zh) 一种波束训练方法及装置
CN114982190A (zh) 载波聚集的波束相关性
EP4307594A1 (en) Electronic device, communication method, and storage medium
US11616628B2 (en) Beam search pilots for paging channel communications
US20220182847A1 (en) Base station device, communication method and storage medium
CN113424628A (zh) 蜂窝系统中的多用户协调传输
CN111345094A (zh) 用于上行链路调度的方法和装置
CN110603823B (zh) 用于无线通信系统的电子设备、方法和存储介质
US20220286214A1 (en) Electronic device, wireless communication method and computer-readable storage medium
CN109075833B (zh) 通信控制装置、终端装置、方法和程序
WO2023092360A1 (zh) 通信系统中的终端以及基站
WO2023279226A1 (en) Channel estimation based beam determination in holographic multiple-in multiple-out system
JP7149324B2 (ja) 端末及び基地局装置
WO2019109483A1 (zh) 一种通信的方法及装置
WO2019109484A1 (zh) 一种通信的方法及装置
CN117397175A (zh) 用于无线通信的电子设备和方法、计算机可读存储介质

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: 20907872

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20907872

Country of ref document: EP

Kind code of ref document: A1