Beam selection method, storage medium, computer device and computer program product

Abstract

The invention discloses a beam selection method, a storage medium, computer equipment and a computer program product. The method comprises the steps that an initial training sample is acquired, and the initial training sample comprises candidate detection beams in a beam space and reference signal received power (RSRP) true values corresponding to the candidate detection beams respectively; and performing iterative training on the L beam prediction networks according to the initial training sample, and selecting L detection beams from the candidate detection beams in a one-to-one correspondence manner based on the trained L beam prediction networks, the L detection beams forming a fixed codebook of a beam space, and the fixed codebook being used for predicting a communication beam selected between the base station and the mobile terminal. According to the invention, the technical problem that an accurate beam prediction result is difficult to obtain by using a uniform codebook in the prior art in the field of airspace beam prediction is solved.

Description

Technical Field

[0001] The present invention relates to the field of communications, and in particular to a beam selection method, a storage medium, a computer device, and a computer program product. Background Art

[0002] In 5G and 6G wireless communications, millimeter wave communications emerge as a solution to provide abundant available spectrum, meeting the critical need for explosively growing data traffic. However, data transmission in the millimeter wave band has limited coverage due to high path loss. Millimeter wave systems use large-scale antenna arrays to send directional beams to overcome the huge free space path loss of millimeter wave signals and reduce spatial interference.

[0003] In the related art, the communication beam can be selected by aligning / tracking with the test beam. The test beam alignment / tracking is to predict the reference signal received power (RSRP) of the entire beam space or the beam with the maximum RSRP based on several sensing beams or even without any beam sensing. Among them, the spatial beam prediction predicts the RSRP of the entire beam domain through the RSRP feedback of the detection beam codebook. However, the detection beam codebook used in the related art is an artificially designed uniform codebook, and its codewords are uniform in the two-dimensional beam domain. In actual specific scenarios, the RSRP distribution of the beam domain channel is non-uniform, so the prediction performance of the spatial beam is insufficient.

[0004] To address the above-mentioned problems, no effective solution has been proposed yet. Summary of the invention

[0005] The embodiments of the present invention provide a beam selection method, a storage medium, a computer device, and a computer program product to at least solve the technical problem that the uniform codebook used in the prior art in the field of spatial domain beam prediction is difficult to obtain accurate beam prediction results.

[0006] According to one aspect of an embodiment of the present invention, a beam selection method is provided, comprising: obtaining an initial training sample, wherein the initial training sample comprises candidate sounding beams in a beam space and true values ​​of reference signal received power (RSRP) corresponding to each of the candidate sounding beams; iteratively training L beam prediction networks according to the initial training sample, and selecting L sounding beams one by one from the candidate sounding beams based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between a base station and a mobile terminal, and the process of training an i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam comprises the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting the first i-1 sounding beams from the candidate sounding beams except the first i-1 sounding beams; A training sounding beam; marking the first i-1 sounding beams and the training sounding beam as known beams, and obtaining an i-th generation training sample according to the RSRP true values ​​corresponding to the known beams and the known beams; using the i-th generation training sample to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP predicted values ​​of all sounding beams included in the beam space according to the RSRP true values ​​of the known beams, wherein the RSRP true values ​​of the known beams are the training targets; using the trained i-th beam prediction network to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are the other beams in the candidate sounding beams except the first i-1 sounding beams, and the corresponding performance index of the i-th sounding beam is better than that of other sounding beams in the i-th generation candidate sounding beams.

[0007] Optionally, the trained i-th beam prediction network is used to select the i-th detection beam from the i-th generation candidate detection beams, including: inputting the i-th generation candidate detection beams and the corresponding RSRP true values ​​into the i-th beam prediction network in sequence to obtain multiple groups of prediction results corresponding to the i-th generation candidate detection beams; calculating the performance indicators corresponding to the multiple groups of prediction results respectively according to the multiple groups of prediction results; and determining the detection beam with the best performance indicator of the corresponding prediction result among the i-th generation candidate detection beams as the i-th detection beam.

[0008] Optionally, the number of the training detection beam is one.

[0009] Optionally, obtaining the initial training sample includes: randomly selecting a part of the detection beams from all the detection beams included in the beam space as the candidate detection beams; obtaining the RSRP true value corresponding to the candidate detection beam, and generating the initial training sample based on the candidate detection beam and the RSRP true value.

[0010] Optionally, the performance indicator includes any one of the following: reference signal reception power difference, mean square error, and TOP-K accuracy.

[0011] Optionally, the method also includes: sending the fixed codebook to the base station, wherein the base station is used to send the base station detection beam to the mobile terminal according to the fixed codebook, and predict the RSRP prediction value of each of the full detection beams included in the beam space according to the RSRP measurement feedback value returned by the mobile terminal in response to the base station detection beam, and the base station is also used to select the communication beam according to the RSRP prediction value of each of the full detection beams and use the communication beam to communicate with the mobile terminal.

[0012] According to another aspect of an embodiment of the present invention, a beam selection method is further provided, including: obtaining a fixed codebook corresponding to a beam space, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained on L beam prediction networks according to initial training samples, and selected one-to-one from candidate sounding beams based on the trained L beam prediction networks, the initial training samples include the candidate sounding beams in the beam space and the true value of the reference signal received power RSRP corresponding to each of the candidate sounding beams, and the i-th beam in the L beam prediction networks is selected. The process of training a beam prediction network and selecting a corresponding i-th sounding beam comprises the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; marking the first i-1 sounding beams and the training sounding beams as known beams, and obtaining an i-th generation training sample according to the known beams and the true values ​​of RSRP corresponding to the known beams; using the i-th generation training sample to train the i-th beam prediction network; The trained i-th beam prediction network is trained to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP predicted values ​​of all the sounding beams included in the beam space according to the RSRP true values ​​of the known beams, wherein the RSRP true values ​​of the known beams are the training targets; the trained i-th beam prediction network is used to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are the other beams in the candidate sounding beams except the first i-1 sounding beams, and the i-th sounding beam is selected from the first i-1 generation candidate sounding beams. The corresponding performance indicator of the beam is better than other sounding beams in the i-th generation candidate sounding beams; a base station sounding beam is selected from the fixed codebook and sent to the mobile terminal; an RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam is received; according to the base station sounding beam and the RSRP measurement feedback value, the RSRP prediction value of each of the full sounding beams included in the beam space is predicted; according to the RSRP prediction value of each of the full sounding beams, a communication beam is selected from the full sounding beams, and the communication beam is used to communicate with the mobile terminal.

[0013] According to another aspect of an embodiment of the present invention, a non-volatile storage medium is further provided, wherein the non-volatile storage medium includes a stored program, wherein when the program is running, the device where the non-volatile storage medium is located is controlled to execute any one of the above-mentioned beam selection methods.

[0014] According to another aspect of an embodiment of the present invention, a computer device is provided, comprising a memory and a processor, wherein the memory is used to store programs, and the processor is used to run the programs stored in the memory, wherein when the program is run, any one of the above-mentioned beam selection methods is executed.

[0015] According to another aspect of the embodiments of the present application, a computer program product is also provided, including a computer program, and when the computer program is executed by a processor, the beam selection method in each embodiment of the present application is implemented.

[0016] According to another aspect of the embodiments of the present application, a computer program is further provided. When the computer program is executed by a processor, the beam selection method in each embodiment of the present application is implemented.

[0017] In an embodiment of the present invention, an initial training sample is obtained, wherein the initial training sample includes candidate sounding beams in the beam space and true values ​​of reference signal received power RSRP corresponding to each of the candidate sounding beams; according to the initial training sample, L beam prediction networks are iteratively trained, and L sounding beams are selected one-to-one from the candidate sounding beams based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal, thereby achieving the purpose of generating a more reasonable fixed codebook of the beam space, thereby achieving the technical effect of improving the rationality of the fixed codebook, and further solving the technical problem that the uniform codebook used in the prior art in the field of spatial beam prediction is difficult to obtain accurate beam prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0019] Figure 1 A hardware structure block diagram of a computer terminal for implementing a beam selection method is shown;

[0020] Figure 2 is a flowchart of a beam selection method 1 provided according to an embodiment of the present invention;

[0021] Figure 3 is a schematic diagram of a uniform codebook provided according to the related art;

[0022] Figure 4 is a schematic flow chart of a second beam selection method according to an embodiment of the present invention;

[0023] Figure 5is a schematic diagram of a simulated communication scenario according to an optional embodiment of the present invention;

[0024] Figure 6 This is a schematic diagram of the prediction accuracy comparison using MSE, Top-K accuracy, and RSRP difference as performance indicators;

[0025] Figure 7 This is a schematic diagram comparing the MSE and RSRP difference indicators of the beam prediction network using MSE, Top-K accuracy, and RSRP difference as performance indicators;

[0026] Figure 8 is a structural block diagram of a beam selection device 1 provided according to an embodiment of the present invention;

[0027] Fig. 9 It is a structural block diagram of a second beam selection device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0028] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0029] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0030] According to an embodiment of the present invention, a method embodiment of beam selection is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] The method embodiments provided in the embodiments of the present application can be executed in a mobile terminal, a computer terminal or a similar computing device. Figure 1 FIG. 1 shows a hardware structure block diagram of a computer terminal for implementing a beam selection method. Figure 1 As shown, the computer terminal 10 may include one or more processors (processors 102a, 102b, ..., 102n are used as examples in the figure) (the processors may include but are not limited to processing devices such as microprocessors MCU or programmable logic devices FPGA), and a memory 104 for storing data. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the BUS bus), a network interface, a power supply, and / or a camera. It will be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above electronic device. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0032] It should be noted that the one or more processors and / or other data processing circuits described above may generally be referred to herein as "data processing circuits". The data processing circuits may be embodied in whole or in part as software, hardware, firmware, or any other combination thereof. In addition, the data processing circuit may be a single independent processing module, or may be incorporated in whole or in part into any of the other components in the computer terminal 10. As described in the embodiments of the present application, the data processing circuit acts as a processor control (e.g., selection of a variable resistor terminal path connected to an interface).

[0033] The memory 104 can be used to store software programs and modules of application software, such as program instructions / data storage devices corresponding to the beam selection method in the embodiment of the present invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, the beam selection method of the above-mentioned application is realized. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor, and these remote memories may be connected to the computer terminal 10 via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0034] The display may be, for example, a touch screen liquid crystal display (LCD) that enables a user to interact with a user interface of the computer terminal 10 .

[0035] Millimeter wave wireless communications are typically characterized by multipath propagation in rural and urban scenarios, which is caused by interactions (reflection, diffraction, transmission, scattering) with stationary obstacles (hills, buildings, towers) and moving objects (cars, pedestrians). Beamforming is challenging for mmWave scenarios such as UMa. Environmental factors, such as wind flow, precipitation, moving vehicles, and pedestrians, can cause significant and dramatic fluctuations in received power. Since the transceiver units of the mmWave base station (BS) are mounted on facilities such as poles, brackets, or streetlights, vibrations and movements can cause unacceptable outage probabilities if beam alignment / tracking is not performed frequently. Frequent beam alignment / tracking can improve link stability, but it adds a lot of beam training overhead. Therefore, it is very necessary to keep a low-cost mmWave beam alignment / tracking solution while maintaining link stability.

[0036] The present invention proposes a performance indicator driven beam selection method, which can be implemented based on a greedy selection algorithm. In the solution of the present invention, a millimeter wave MISO communication system consisting of a single base station BS and a single mobile terminal MU can be considered. Optionally, the beam selection method can be extended to scenarios of multiple BSs and multiple MUs through time division multiplexing. The BS is equipped with a large-scale planar antenna array connected to a radio frequency link, and the UE is equipped with an omnidirectional antenna.

[0037] Figure 2 is a flow chart of a beam selection method 1 according to an embodiment of the present invention. Figure 2 As shown, the method comprises the following steps:

[0038] Step S202: Acquire initial training samples, wherein the initial training samples include candidate sounding beams in the beam space and true values ​​of reference signal received powers (RSRP) corresponding to the candidate sounding beams.

[0039] Optionally, the initial training sample may include multiple candidate detection beams in the beam space and the true RSRP value corresponding to each beam. For example, the number of candidate detection beams is 128, and the initial training sample may include 128 candidate detection beams and the true RSRP value of each of the 128 candidate detection beams. The initial training sample may come from the training sample set.

[0040] As an optional embodiment, the initial training sample can be generated by the following steps: randomly selecting a part of the detection beams in the total detection beams included in the beam space as candidate detection beams; obtaining the RSRP true value corresponding to the candidate detection beam, and generating the initial training sample based on the candidate detection beam and the RSRP true value.

[0041] In this optional embodiment, the candidate detection beams may be part of the full detection beams included in the beam space, and may be part of the beams randomly selected from the full detection beams as candidate detection beams. For example, the full detection beams have 128 beams, and the candidate detection beams are 40 beams selected from the 128 beams. Since these 40 beams are randomly selected, they can represent the overall characteristics of the full detection beams to a certain extent. Each beam prediction network obtained by training with 40 candidate detection beams can better predict the RSRP distribution of the entire beam space. At the same time, using candidate detection beams to train the beam prediction network can greatly save training performance overhead and improve network training speed while ensuring the accuracy of the trained network. It can be understood that the closer the number of candidate detection beams is to the number of full detection beams, the higher the prediction accuracy of the trained beam prediction network, the fewer the number of candidate detection beams, the faster the training of the beam prediction network, and the lower the overhead.

[0042] Step S204, iteratively train the L beam prediction networks according to the initial training samples, and select L sounding beams one by one from the candidate sounding beams based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal.

[0043] Among them, the number of L can be freely set according to demand, and the size of L depends on the codebook size of the fixed codebook. For example, L can be 8. According to the number of L, L beam prediction networks are pre-set, and each beam prediction network is used to select a detection beam until L detection beams are selected by L beam prediction networks to obtain a fixed codebook. In the process of iterative training of the L beam prediction networks, the first beam prediction network can be used to predict the first detection beam, and then the prediction result of the first beam prediction network is used to train the second beam prediction network, and then the second beam prediction network predicts the second detection beam, and iterative training is performed in this way until the L beam prediction networks are trained respectively and predict a total of L detection beams respectively.

[0044] Through the above method, L sounding beams can be selected in steps, and each sounding beam can be the optimal solution selected by the network trained in each step. Therefore, the final fixed codebook can perform better in the RSRP prediction work in the full beam space. Figure 3 is a schematic diagram of a uniform codebook provided according to the related art, such as Figure 3As shown in the figure, the beams in the uniform codebook are evenly distributed in the beam space and cannot adapt to different environments and scenarios. The fixed codebook selected by the above method is not evenly distributed in the beam space, but is selected for better prediction performance. Therefore, it is more suitable for the base station to find the best communication path in different scenarios and environments. The direction of a beam can be described by two angles, namely and The two angular domains are discretized according to the number of antennas in the transmit array (16×8 in this article), and the full spatial beam domain is also discretized into the corresponding 128. Therefore, the horizontal axis AOD (azimuth of departure) index is the index after the beam angle φ is discretized, and the vertical axis ZOD (zenith of departure) is the index after the beam angle θ is discretized.

[0045] In order to further illustrate how to iteratively train the network based on the greedy algorithm and how to use the trained network to select L detection beams, step S204 is described below using any one of the L beam prediction networks as an example. In step S204, the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th detection beam includes the following steps, where 1≤i≤L:

[0046] Step S2041, determining the first i-1 sounding beams selected by the first i-1 beam prediction networks among the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams.

[0047] As an optional embodiment, the number of the training sounding beams may be one, that is, when each beam prediction network is trained, based on the first i-1 sounding beams predicted by the first i-1 beam prediction networks, an additional training sounding beam is selected for the training task of the i-th beam prediction network. Optionally, the training sounding beam may be randomly selected.

[0048] It should be noted that when i=1, that is, for the first beam prediction network among the L beam prediction networks, since no previous beam prediction network has selected a detection beam, for the first beam prediction network, the first i-1 detection beams are an empty set, and the training detection beam is any one of the candidate detection beams.

[0049] Step S2042: Mark the first i-1 sounding beams and the training sounding beams as known beams, and obtain the i-th generation training samples according to the known beams and the true values ​​of RSRP corresponding to the known beams.

[0050] Step S2043, using the i-th generation training samples to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction value of all detection beams included in the beam space according to the RSRP true value of the known beam, wherein the RSRP true value of the known beam is the training target.

[0051] For the i-th beam prediction network, the known beam is the input condition during training, and the RSRP prediction value of the full detection beam is the output result of the beam prediction network. The i-th beam prediction network is used to predict the RSRP value of the known beam based on the RSRP value of the known beam, and to tune the parameters of the network itself according to the deviation between the RSRP prediction value of the known beam and the RSRP true value of the known beam, until the parameters of the i-th beam prediction network converge and complete the training. In the actual training process, all detection beams can be uniformly input into the i-th beam network, where the detection beams marked as known beams can be directly input into the i-th beam prediction network during training, and the RSRP true values ​​of the detection beams other than the known beams in the full detection beam can be masked with a mask and then input into the beam prediction network. It should be noted that when i is 1, since no detection beam is selected by the previous beam prediction network, the first i-1 detection beams are an empty set, and the known beam only includes any one beam from the candidate detection beams. The first generation of training samples for training the first beam prediction network can be obtained based on this beam and its corresponding true RSRP value, and the first generation of training samples is used to train the first beam prediction network to obtain the trained first beam prediction network.

[0052] Step S2044: Use the trained i-th beam prediction network to select the i-th detection beam from the i-th generation candidate detection beams, wherein the i-th generation candidate detection beams are the other beams in the candidate detection beams except the first i-1 detection beams, and the corresponding performance index of the i-th detection beam is better than other detection beams in the i-th generation candidate detection beams.

[0053] As an optional embodiment, the i-th detection beam can be selected from the i-th generation candidate detection beams in the following manner: the i-th generation candidate detection beams and the corresponding RSRP true values ​​are input into the i-th beam prediction network in sequence to obtain multiple groups of prediction results corresponding to the i-th generation candidate detection beams; based on the multiple groups of prediction results, performance indicators corresponding to the multiple groups of prediction results are calculated; and the detection beam with the best performance indicator corresponding to the prediction result in the i-th generation candidate detection beams is determined as the i-th detection beam.

[0054] Among them, the trained beam prediction network can predict the RSRP value of each beam in the full beam space based on the input beam and the RSRP value of the beam. Therefore, it can be determined that after each beam in the i-th generation candidate detection beam is input into the trained i-th beam prediction network, the network gives the full beam space prediction results for each input beam. Based on the full beam space prediction result, the performance index corresponding to the prediction result can be determined, and the performance index can characterize the spatial channel performance under the corresponding full beam space prediction result.

[0055] As an optional embodiment, the performance indicator may include any one of the following: reference signal reception power difference, mean square error, TOP-K accuracy.

[0056] The mean square error (MSE) is used to evaluate the RSRP average prediction mean square error in beam space, expressed in (dBm 2 ) is the unit, that is Among them, E x,s,1 Characterize the mean square error MSE, N s Characterize the number of samples, x i represents the true value of RSRP of the ith beam, Represents the RSRP prediction value of the ith beam. This is an indicator for evaluating the overall prediction error.

[0057] The TOP-K accuracy rate represents the proportion of the best beam included in the top K predicted beams (sorted by predicted RSRP), where K∈{1,3,5}.

[0058] The reference signal received power difference (RSRP difference) indicates the predicted optimal beam and the corresponding ground-truth optimal beam i * The absolute RSRP difference between Among them, E x,s,2 Characterizes the reference signal received power difference RSRP difference, represents the predicted optimal beam The true value of RSRP, represents the ground-truth optimal beam i * The true value of RSRP.

[0059] In the above steps, by obtaining the initial training samples, wherein the initial training samples include the candidate detection beams in the beam space and the true value of the reference signal received power RSRP corresponding to each of the candidate detection beams; according to the initial training samples, the L beam prediction networks are iteratively trained, and based on the trained L beam prediction networks, L detection beams are selected one by one from the candidate detection beams, wherein the L detection beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal, so as to achieve the purpose of generating a more reasonable fixed codebook of the beam space, thereby achieving the technical effect of improving the rationality of the fixed codebook, and further solving the technical problem that the uniform codebook used in the prior art in the field of spatial domain beam prediction is difficult to obtain accurate beam prediction results. Compared with the uniform detection codebook, the beam prediction result of the fixed codebook designed using the performance indicator drive can obtain better prediction performance (such as MSE, Top-K accuracy, RSRP difference).

[0060] As an optional embodiment, the above-mentioned beam selection method also includes: sending a fixed code book to the base station, wherein the base station is used to send the base station detection beam to the mobile terminal according to the fixed code book, and predict the RSRP prediction value of each of the full detection beams included in the beam space according to the RSRP measurement feedback value returned by the mobile terminal in response to the base station detection beam, and the base station is also used to select the communication beam according to the RSRP prediction value of each of the full detection beams and use the communication beam to communicate with the mobile terminal.

[0061] Optionally, the above-mentioned beam selection method 1 can be applied to a server other than a base station or an edge computing device that provides computing services. The server calculates a fixed codebook and sends the fixed codebook to the base station. The base station communicates with the mobile terminal based on the received fixed codebook.

[0062] Figure 4 is a flow chart of a second beam selection method according to an embodiment of the present invention. The second beam selection method can be applied to a base station. The base station obtains a fixed codebook generated by the first beam selection method and communicates with a mobile terminal. Figure 4 As shown, the method comprises the following steps:

[0063] Step 402, obtaining a fixed codebook corresponding to the beam space, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained for L beam prediction networks according to initial training samples, and selected one by one from the candidate sounding beams based on the trained L beam prediction networks, the initial training samples include the candidate sounding beams in the beam space and the true values ​​of the reference signal received power RSRP corresponding to each of the candidate sounding beams, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; the first i-1 sounding beams and the training detection beams are marked as known beams, and the i-th generation training samples are obtained according to the known beams and the true RSRP values ​​corresponding to the known beams respectively; the i-th generation training samples are used to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction values ​​of all detection beams included in the beam space according to the true RSRP values ​​of the known beams, wherein the true RSRP values ​​of the known beams are the training targets; the i-th beam prediction network after training is used to select the i-th detection beam from the i-th generation candidate detection beams, wherein the i-th generation candidate detection beams are the other beams in the candidate detection beams except the first i-1 detection beams, and the corresponding performance indicators of the i-th detection beam are better than those of other detection beams in the i-th generation candidate detection beams.

[0064] Step 404: Select a base station sounding beam from the fixed codebook and send it to the mobile terminal.

[0065] Step 406: Receive the RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam.

[0066] Step 408: According to the base station sounding beam and the RSRP measurement feedback value, the RSRP prediction value of each of the full sounding beams included in the beam space is predicted.

[0067] Step 410, selecting a communication beam from all the sounding beams according to the RSRP prediction values ​​of each of the all the sounding beams, and using the communication beam to communicate with the mobile terminal.

[0068] As an optional embodiment, step S408 can be implemented based on a fixed codebook network. The fixed codebook network can be a network that predicts the full beam space RSRP according to a uniform codebook in the related art, or a network that has the same structure as the beam prediction network in the beam training method 1 but has different parameters. In summary, the fixed codebook network can use the optimal beam set (fixed codebook) selected by the beam prediction network to predict the full beam space RSRP and obtain the RSRP distribution prediction result of the full beam space.

[0069] In combination with the above-mentioned beam selection method 1 and beam selection method 2, the present application provides an embodiment of generating a fixed codebook and using the fixed codebook for communication. In this embodiment, L networks with the same structure can be generated in the offline training stage. Among them, the network = means that i-1 sounding beams have been selected so far, and the network is used to select the i-th sounding beam, which means the i-th beam prediction network. When all L sounding beams are selected, the selected L sounding beams are stored as a fixed codebook. The next problem is simplified to a fixed codebook prediction problem: using the network structure and For a network with the same structure, the RSRP measurement values ​​of L beams are used to predict the RSRP value of the entire beam space. The training data of the beam prediction network can be collected offline in the form of Where N s is the number of samples, parameter set Θ g The update is iteratively performed via mini-batch gradient descent (MBGD) until convergence.

[0070] In this embodiment, the process of iteratively training the L beam prediction networks and using the trained L beam prediction networks to predict the L detection beams can be represented as follows:

[0071] The input of the training process includes a sample data set D and the maximum number of sounding beams L. The sample data set D can be an initial training sample. The output of the training process includes the sounding beam fixed codebook BeamList selected by the base station BS and the fixed codebook network {g fix}.

[0072]

[0073]

[0074] In the above process, psd batch Represents a batch of training samples when the network is trained in batches, psd mask represents the i-th generation training sample when training the i-th beam prediction network, psd mask [l,pl The l in ] indicates psd batch The lth sample (the lth row) in l Indicates that this row of samples randomly selects the pth l beams are trained (the corresponding positions of the unselected beams are set to 0), P i represents the non-candidate beam set of the i-th iteration, and the candidate beam set Q i Corresponding (Q i ∪P i is the full spatial beam index set), q best Indicates the optimal beam selected in this iteration. Steps 1 to 10 are the process of training L beam prediction networks, and steps 11 to 14 are the process of training the fixed codebook network.

[0075] In the online inference phase, the base station selects a set of sounding beams to be sent in the fixed codebook based on the given maximum number of sounding beams L and receives the corresponding RSRP feedback from the mobile terminal. Then, the RSRP in the full beam space is predicted based on the feedback. Then, the BS selects the optimal data transmission beam based on the predicted RSRP in the full beam space. As a communication beam.

[0076] network and The structure is consistent with that of , and optionally, its specific structure can be as shown in Table 1.

[0077] Table 1 Topology of beam prediction network and fixed codebook network

[0078]

[0079] The size of the network data input can be [batch_size, 128], which means that batch_size samples of data are sent to the network at one time, and the dimension of each sample is 128, that is, the full beam space contains detection beams from 128 directions. When inputting data to the i-th beam prediction network, if a beam is marked as a known beam, the value of the beam position in the input data is the true measurement value of RSRP; if the beam is not marked as a known beam, the value of the beam position is 0.

[0080] The above embodiments can be further described and explained in conjunction with simulation. The simulation configuration is as follows: the millimeter wave channel is established as a map-based hybrid model, which complies with the 3GPP 38.901 channel model. Among them, the deterministic ray tracing model is established using Feko Winprop software; the random scattering of moving objects is to verify the generalization ability of the proposed solution. The configurations of the base station BS and the mobile terminal MU are shown in Table 2. The communication scenarios are as follows: Figure 5 shown. Figure 5Medium 210×130m 2 The millimeter wave network geometric layout scenario is representative of the suburban scenario, where the BS is located on the top of the building or arranged along the street, and its height is between 15 and 40m. The height of the building or tree is also marked. The outdoor positioning data is collected uniformly. In this simulation embodiment, although the proposed positioning method is performed in three-dimensional space, only the plane mean square error is considered as the performance indicator. The simulation platform can be: Python 3.11.5, Torch 2.1.1+cu121, CPU Intel i7-10700K GPU NVIDIAGeForceGTX 1070Ti.

[0081] Table 2 Simulation configuration

[0082]

[0083] During the simulation, a sounding beam is sent at the beginning of each time slot, and each sounding beam occupies one symbol resource. S and T C are the duration of a symbol and a slot respectively.

[0084] In this simulation process, the detection beam is uniformly selected in the beam space as a uniform fixed codebook (such as Figure 3 The prediction results using the uniform fixed codebook method are used as the comparison benchmark. Using multiple performance indicators (MSE, TOP-K accuracy, RSRP difference) as the beam selection indicators, the final results are compared with the performance baseline obtained using the uniform fixed codebook. The comparison results are shown in Figure 6 and Figure 7 shown.

[0085] Figure 6 This is a diagram showing the comparison of prediction accuracy using MSE, Top-K accuracy, and RSRP difference as performance indicators. Figure 7 The MSE, Top-K accuracy, and RSRP difference are used as performance indicators for the beam prediction network. The beam set selected by the baseline (index, such as Figure 3 As shown in the figure, the beams are uniformly distributed in the beam space.

[0086] like Figure 6 and Figure 7As shown in the figure, compared with the baseline (a fixed codebook with beams evenly distributed in the entire space), the Top-K accuracy of the prediction results of the fixed codebook network designed using the beam selection algorithm provided by the present application is higher than the baseline. Among them, the overall accuracy performance of the network driven by the Top-3 accuracy is better than that of the network driven by the Top-1 and Top-5 accuracy. Although the accuracy of the network driven by MSE is higher than the baseline, it is slightly lower than other networks.

[0087] In terms of the MSE of the prediction results, the network driven directly by MSE achieves the best performance among all driving strategies. The MSE of the network driven by RSRP difference is even slightly higher than the baseline. In terms of the RSRP difference of the prediction results, the network driven directly by RSRP difference itself still performs best.

[0088] The MSE indicator and Top-K accuracy are correlated. When the MSE is as low as a certain level, the RSRP difference and Top-K accuracy can reach a relatively good state. However, the network will select the detection beam with the highest RSRP estimation value as the communication beam, and the Top-K accuracy and RSRP difference are also based on this mechanism. It can be imagined that the change trend of the RSRP distribution estimated by a network is consistent with that of the actual RSRP distribution, but the specific RSRP value is not completely close to the actual value. In this case, the prediction result of the model has better RSRP difference and Top-K accuracy indicators, but not necessarily lower MSE. For example, a network with RSRP difference as the driving indicator has excellent performance in Top-K accuracy and RSRP difference between the predicted result and the actual result.

[0089] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0090] Through the description of the above implementation methods, those skilled in the art can clearly understand that the beam selection method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, a computer, a server, or a network device, etc.) to execute the methods of various embodiments of the present invention.

[0091] According to an embodiment of the present invention, a beam selection device for implementing the above-mentioned beam selection method 1 is also provided. Figure 8 is a structural block diagram of a beam selection device according to an embodiment of the present invention. Figure 8 As shown, the first beam selection device includes: a first acquisition module 82 and a first selection module 84. The first beam selection device is described below.

[0092] A first acquisition module 82 is used to acquire an initial training sample, wherein the initial training sample includes a candidate sounding beam in the beam space and a true value of a reference signal received power RSRP corresponding to each of the candidate sounding beams;

[0093] The first selection module 84 is used to iteratively train the L beam prediction networks according to the initial training samples, and select L sounding beams from the candidate sounding beams one by one based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal. The selection module includes:

[0094] A determination submodule 841 is used to determine the first i-1 sounding beams selected by the first i-1 beam prediction networks among the L beam prediction networks, and to select a training sounding beam from the candidate sounding beams except the first i-1 sounding beams;

[0095] The marking submodule 842 is used to mark the first i-1 sounding beams and the training sounding beam as known beams, and obtain the i-th generation training samples according to the known beams and the RSRP true values ​​corresponding to the known beams;

[0096] A training submodule 843 is used to train the i-th beam prediction network using the i-th generation training samples to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction value of all the detection beams included in the beam space according to the RSRP true value of the known beam, wherein the RSRP true value of the known beam is the training target;

[0097] The selection submodule 844 is used to select the i-th detection beam from the i-th generation candidate detection beams using the trained i-th beam prediction network, wherein the i-th generation candidate detection beams are the other beams in the candidate detection beams except the first i-1 detection beams, and the corresponding performance index of the i-th detection beam is better than other detection beams in the i-th generation candidate detection beams.

[0098] It should be noted that the first acquisition module 82 and the first selection module 84 correspond to steps S202 to S204 in the embodiment, and the examples and application scenarios implemented by the two modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment. It should be noted that the above modules, as part of the device, can be run in the computer terminal 10 provided in the embodiment.

[0099] According to an embodiment of the present invention, a beam selection device for implementing the above-mentioned beam selection method 2 is also provided. Fig. 9 is a structural block diagram of a second beam selection device according to an embodiment of the present invention. Fig. 9 As shown, the second beam selection device includes: a second acquisition module 92, a sending module 94, a receiving module 96, a prediction module 98 and a second selection module 100. The second beam selection device is described below.

[0100] The second acquisition module 92 is used to obtain a fixed codebook corresponding to the beam space, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained for L beam prediction networks according to initial training samples, and are selected one by one from the candidate sounding beams based on the trained L beam prediction networks, the initial training samples include the candidate sounding beams in the beam space and the true values ​​of the reference signal received power RSRP corresponding to each of the candidate sounding beams, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; the first i-1 sounding beams are selected from the first i-1 sounding beams; The known beam and the training detection beam are marked as known beams, and the i-th generation training samples are obtained according to the known beams and the RSRP true values ​​corresponding to the known beams respectively; the i-th generation training samples are used to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction values ​​of all detection beams included in the beam space according to the RSRP true values ​​of the known beams, wherein the RSRP true values ​​of the known beams are the training targets; the i-th beam prediction network after training is used to select the i-th detection beam from the i-th generation candidate detection beams, wherein the i-th generation candidate detection beams are the other beams in the candidate detection beams except the first i-1 detection beams, and the corresponding performance index of the i-th detection beam is better than that of other detection beams in the i-th generation candidate detection beams;

[0101] A sending module 94, configured to select a base station sounding beam from a fixed codebook and send the beam to the mobile terminal;

[0102] A receiving module 96, configured to receive an RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam;

[0103] A prediction module 98, configured to predict RSRP prediction values ​​of all the sounding beams included in the beam space according to the sounding beams of the base station and the RSRP measurement feedback value;

[0104] The second selection module 100 is used to select a communication beam from all the sounding beams according to the RSRP prediction values ​​of each of the all the sounding beams, and use the communication beam to communicate with the mobile terminal.

[0105] It should be noted that the second acquisition module 92, the sending module 94, the receiving module 96, the prediction module 98 and the second selection module 100 correspond to steps S402 to S410 in the embodiment, and the examples and application scenarios implemented by the multiple modules and the corresponding steps are the same, but are not limited to the contents disclosed in the above embodiment. It should be noted that the above modules as part of the device can be run in the computer terminal 10 provided in the embodiment.

[0106] An embodiment of the present invention may provide a computer device. Optionally, in this embodiment, the computer device may be located in at least one network device among multiple network devices of a computer network. The computer device includes a memory and a processor.

[0107] Among them, the memory can be used to store software programs and modules, such as program instructions / modules corresponding to the beam selection method and device in the embodiments of the present invention. The processor executes various functional applications and data processing by running the software programs and modules stored in the memory, that is, realizing the above-mentioned beam selection method. The memory may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory may further include a memory remotely arranged relative to the processor, and these remote memories can be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0108] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtaining an initial training sample, wherein the initial training sample includes candidate sounding beams in the beam space and true values ​​of reference signal received power RSRP corresponding to each of the candidate sounding beams; iteratively training L beam prediction networks according to the initial training samples, and selecting L sounding beams from the candidate sounding beams one by one based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the i-1st sounding beams selected by the i-1st beam prediction networks in the L beam prediction networks, and removing the i-1st sounding beams from the candidate sounding beams; a training sounding beam is selected from the first i-1 sounding beams and the training sounding beams; the first i-1 sounding beams and the training sounding beams are marked as known beams, and the i-th generation training samples are obtained according to the known beams and the true RSRP values ​​corresponding to the known beams; the i-th generation training samples are used to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction values ​​of all sounding beams included in the beam space according to the true RSRP values ​​of the known beams, wherein the true RSRP values ​​of the known beams are the training targets; the i-th beam prediction network is used to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are the other beams in the candidate sounding beams except the first i-1 sounding beams, and the corresponding performance index of the i-th sounding beam is better than that of other sounding beams in the i-th generation candidate sounding beams.

[0109] The processor can call the information and application stored in the memory through the transmission device to perform the following steps: obtain a fixed codebook corresponding to the beam space, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained for L beam prediction networks based on initial training samples, and selected one by one from the candidate sounding beams based on the trained L beam prediction networks, the initial training samples include the candidate sounding beams in the beam space and the true values ​​of the reference signal received power RSRP corresponding to each of the candidate sounding beams, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; marking the first i-1 sounding beams and the training sounding beams as known beams, and obtaining the i-th generation training samples according to the known beams and the true values ​​of the RSRP corresponding to each of the known beams; using the i-th generation training samples The i-th beam prediction network is trained to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction values ​​of all the sounding beams included in the beam space according to the RSRP true values ​​of the known beams, wherein the RSRP true values ​​of the known beams are the training targets; the i-th beam prediction network after the training is used to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are the other beams in the candidate sounding beams except the first i-1 sounding beams, The corresponding performance index of the i-th sounding beam is better than other sounding beams in the i-th generation candidate sounding beams; a base station sounding beam is selected from a fixed codebook and sent to a mobile terminal; an RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam is received; based on the base station sounding beam and the RSRP measurement feedback value, the RSRP prediction values ​​of all the sounding beams included in the beam space are predicted; based on the RSRP prediction values ​​of all the sounding beams, a communication beam is selected from the full sounding beams, and the communication beam is used to communicate with the mobile terminal.

[0110] A person of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing the hardware related to the terminal device through a program, and the program can be stored in a non-volatile storage medium, and the storage medium may include: a flash drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0111] The embodiment of the present invention further provides a non-volatile storage medium. Optionally, in this embodiment, the non-volatile storage medium can be used to store the program code executed by the beam selection method provided in the embodiment.

[0112] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a computer terminal group in a computer network, or in any mobile terminal in a mobile terminal group.

[0113] Optionally, in this embodiment, the non-volatile storage medium is configured to store a program code for executing the following steps: obtaining an initial training sample, wherein the initial training sample includes candidate sounding beams in the beam space and true values ​​of reference signal received power RSRP corresponding to each of the candidate sounding beams; iteratively training L beam prediction networks according to the initial training sample, and selecting L sounding beams one by one from the candidate sounding beams based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and removing the first i-1 sounding beams from the candidate sounding beams; -1 sounding beams; the first i-1 sounding beams and the training sounding beams are marked as known beams, and the i-th generation training samples are obtained according to the known beams and the true RSRP values ​​corresponding to the known beams; the i-th generation training samples are used to train the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP predicted values ​​of all sounding beams included in the beam space according to the true RSRP values ​​of the known beams, wherein the true RSRP values ​​of the known beams are the training targets; the i-th beam prediction network is used to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are the other beams in the candidate sounding beams except the first i-1 sounding beams, and the corresponding performance indicators of the i-th sounding beam are better than those of other sounding beams in the i-th generation candidate sounding beams.

[0114] Optionally, in this embodiment, the non-volatile storage medium is configured to store a program code for executing the following steps: obtaining a fixed codebook corresponding to the beam space, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained on the L beam prediction networks according to the initial training samples, and selected one by one from the candidate sounding beams based on the trained L beam prediction networks, and the initial training samples include the candidate sounding beams in the beam space and the true value of the reference signal received power RSRP corresponding to each of the candidate sounding beams , the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: determining the first i-1 sounding beams selected by the first i-1 beam prediction networks in the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; marking the first i-1 sounding beams and the training sounding beams as known beams, and obtaining the i-th generation training samples according to the RSRP true values ​​corresponding to the known beams and the known beams; using the i-th generation training The sample trains the i-th beam prediction network to obtain the trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction value of all the detection beams included in the beam space according to the RSRP true value of the known beam, wherein the RSRP true value of the known beam is the training target; the trained i-th beam prediction network is used to select the i-th detection beam from the i-th generation candidate detection beams, wherein the i-th generation candidate detection beams are the other beams in the candidate detection beams except the first i-1 detection beams. , the corresponding performance index of the i-th sounding beam is better than other sounding beams in the i-th generation candidate sounding beams; select a base station sounding beam from a fixed codebook and send it to a mobile terminal; receive an RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam; predict the RSRP prediction values ​​of all the sounding beams included in the beam space according to the base station sounding beam and the RSRP measurement feedback value; select a communication beam from the full sounding beams according to the RSRP prediction values ​​of all the sounding beams, and use the communication beam to communicate with the mobile terminal.

[0115] The embodiment of the present application further provides a computer program product. Optionally, in this embodiment, the computer program product may include a computer program, and when the computer program is executed by a processor, the beam selection method provided in the embodiment is implemented.

[0116] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages or disadvantages of the embodiments.

[0117] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0118] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0119] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed over multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0120] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0121] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0122] The above are only preferred embodiments of the present invention. It should be pointed out that, for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A beam selection method, characterized in that: include: Acquire an initial training sample, wherein the initial training sample includes candidate sounding beams in the beam space and true values ​​of reference signal received powers (RSRP) corresponding to the candidate sounding beams; According to the initial training samples, L beam prediction networks are iteratively trained, and L sounding beams are selected one-to-one from the candidate sounding beams based on the trained L beam prediction networks, wherein the L sounding beams constitute a fixed codebook of the beam space, and the fixed codebook is used to predict the communication beam selected between the base station and the mobile terminal, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam comprises the following steps: Determining first i-1 sounding beams selected by first i-1 beam prediction networks among the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; Mark the first i-1 sounding beams and the training sounding beam as known beams, and obtain an i-th generation training sample according to the known beams and the true RSRP values ​​corresponding to the known beams; Using the i-th generation training samples to train the i-th beam prediction network to obtain a trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction value of the full amount of detection beams included in the beam space according to the RSRP true value of the known beam, wherein the RSRP true value of the known beam is the training target; The trained i-th beam prediction network is used to select the i-th sounding beam from the i-th generation candidate sounding beams, wherein the i-th generation candidate sounding beams are other beams in the candidate sounding beams except the first i-1 sounding beams, and the corresponding performance index of the i-th sounding beam is better than other sounding beams in the i-th generation candidate sounding beams.

2. The method according to claim 1, characterized in that: Selecting the i-th sounding beam from the i-th generation candidate sounding beams using the trained i-th beam prediction network includes: Inputting the i-th generation candidate sounding beams and the corresponding RSRP true values ​​into the i-th beam prediction network in sequence, respectively, to obtain multiple groups of prediction results corresponding to the i-th generation candidate sounding beams; According to the multiple groups of prediction results, calculating the performance indicators respectively corresponding to the multiple groups of prediction results; The sounding beam with the best performance indicator corresponding to the prediction result among the i-th generation candidate sounding beams is determined as the i-th sounding beam.

3. The method according to claim 1, characterized in that The number of the training detection beam is one.

4. The method according to claim 1, characterized in that The obtaining of the initial training sample comprises: Randomly selecting a portion of the probing beams from all the probing beams included in the beam space as the candidate probing beams; Acquire an RSRP true value corresponding to the candidate sounding beam, and generate the initial training sample based on the candidate sounding beam and the RSRP true value.

5. The method according to claim 1, characterized in that The performance indicator includes any one of the following: reference signal reception power difference, mean square error, and TOP-K accuracy.

6. The method according to any one of claims 1 to 5, characterized in that The method further comprises: The fixed codebook is sent to the base station, wherein the base station is used to send a base station sounding beam to the mobile terminal according to the fixed codebook, and predict the RSRP prediction value of each of the full sounding beams included in the beam space according to the RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam, and the base station is also used to select the communication beam according to the RSRP prediction value of each of the full sounding beams and use the communication beam to communicate with the mobile terminal.

7. A beam selection method, characterized in that: include: A fixed codebook corresponding to the beam space is obtained, wherein the fixed codebook is composed of L sounding beams, the L sounding beams are iteratively trained on L beam prediction networks according to initial training samples, and are selected one-to-one from candidate sounding beams based on the trained L beam prediction networks, the initial training samples include the candidate sounding beams in the beam space and the true value of the reference signal received power RSRP corresponding to each of the candidate sounding beams, and the process of training the i-th beam prediction network in the L beam prediction networks and selecting the corresponding i-th sounding beam includes the following steps: Determining first i-1 sounding beams selected by first i-1 beam prediction networks among the L beam prediction networks, and selecting a training sounding beam from the candidate sounding beams except the first i-1 sounding beams; Mark the first i-1 sounding beams and the training sounding beam as known beams, and obtain an i-th generation training sample according to the known beams and the true RSRP values ​​corresponding to the known beams; Using the i-th generation training samples to train the i-th beam prediction network to obtain a trained i-th beam prediction network, wherein the i-th beam prediction network is used to predict the RSRP prediction value of the full amount of detection beams included in the beam space according to the RSRP true value of the known beam, wherein the RSRP true value of the known beam is the training target; Selecting the i-th sounding beam from the i-th generation candidate sounding beams using the trained i-th beam prediction network, wherein the i-th generation candidate sounding beams are other beams in the candidate sounding beams except the first i-1 sounding beams, and the corresponding performance index of the i-th sounding beam is better than other sounding beams in the i-th generation candidate sounding beams; Selecting a base station sounding beam from the fixed codebook and sending it to the mobile terminal; Receiving an RSRP measurement feedback value returned by the mobile terminal in response to the base station sounding beam; Predicting RSRP prediction values ​​of all the sounding beams included in the beam space according to the base station sounding beam and the RSRP measurement feedback value; A communication beam is selected from the full sounding beams according to the respective RSRP prediction values ​​of the full sounding beams, and the communication beam is used to communicate with the mobile terminal.

8. A non-volatile storage medium, characterized in that: The non-volatile storage medium includes a stored program, wherein when the program is executed, the device where the non-volatile storage medium is located is controlled to execute the beam selection method according to any one of claims 1 to 7.

9. A computer device, characterized in that: The computer device comprises a memory and a processor, the memory is used to store a program, and the processor is used to run the program stored in the memory, wherein the program executes the beam selection method according to any one of claims 1 to 7 when running.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the beam selection method according to any one of claims 1 to 7 are implemented.

Images