A distributed sensing-assisted beam selection method for 5G AeroMACS

Abstract

The present invention relates to a distributed sensing-assisted beam selection method for 5G AeroMACS, and belongs to the field of wireless communication technology. The present invention provides a distributed sensing-assisted beam selection method for 5G AeroMACS for high-dynamic terminals in airport scenarios, such as airborne terminals, vehicle-mounted terminals, etc., and realizes fast beam codeword selection in airport scenarios where aircraft and ground vehicles move at high speed by extracting terminal multimodal features. More specifically, the multimodal features of the position, speed and posture of the high-dynamic terminal are extracted through an attention mechanism, and multi-base station collaborative training beam selection is realized through federated learning. A beamforming codebook is used to improve the operation efficiency, enhance the performance of the communication system, and solve the problems of insufficient timeliness of airport communication and large beam overhead in the prior art.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communications, and specifically relates to a distributed sensing-assisted beam selection method for 5G AeroMACS, aiming at the application requirements of real-time beamforming for terminals such as aircraft and support vehicles at airports. Background Art

[0002] Beam selection is one of the key technologies in multi-antenna systems. It significantly improves the reliability of communication systems by establishing highly directional links between users and base stations. In airport systems, since terminals such as aircraft and support vehicles are highly mobile, the terminals and base stations need to re-perform fast beamforming after the terminal position changes to ensure the stability of the directional link. However, the difficulty of fast beam selection is how to calculate the beamforming matrix in real time. In scenarios with high terminal dynamics, it is necessary to iteratively solve the transmit and receive ends to calculate the beamforming matrix, which requires a lot of system overhead. In addition, it is also difficult to obtain accurate channel state information in practice to perform beam selection.

[0003] Although there are studies on the beam selection problem, the applicable technical solutions do not meet the requirements of beam selection in high-dynamic scenarios. The main problems of related technical research are reflected in the following three aspects: On the one hand, the existing beam selection technology assumes that the terminal remains stationary for a period of time, which cannot meet the communication needs of high-dynamic scenarios. On the other hand, the existing beamforming algorithm has problems such as high complexity and poor real-time performance. In addition, the existing beam selection technology does not fully utilize the characteristics of airports with multiple base stations and cannot fully utilize distributed base station resources. Summary of the invention

[0004] In view of the above problems, the present invention provides a distributed perception-assisted beam selection method for 5G AeroMACS for high-dynamic terminals in airport scenarios, such as airborne terminals and vehicle-mounted terminals. By extracting terminal multimodal features, fast beam codeword selection is achieved in airport scenarios where aircraft and ground vehicles move at high speed. More specifically, the multimodal features of the position, speed and posture of the high-dynamic terminal are extracted through the attention mechanism, and multi-base station collaborative training beam selection is achieved through federated learning. The beamforming codebook is used to improve the computing efficiency, enhance the performance of the communication system, and solve the problems of insufficient timeliness of airport communications and large beam overhead in the prior art.

[0005] The present invention provides a distributed sensing-assisted beam selection method for 5G AeroMACS, and the specific steps are as follows:

[0006] Step 1. Establish an airport communication beam selection model; based on the airport communication beam selection model, construct a model that maximizes the accuracy of the network's prediction of the codeword label corresponding to the actual optimal beam;

[0007] Step 2. Obtain high dynamic terminal data;

[0008] Step 3. Based on the high dynamic terminal data obtained in step 2 and the accuracy model of maximizing the network's prediction of the codeword label corresponding to the actual optimal beam in step 1, a distributed sensing-assisted MIMO base station beam selection model is trained;

[0009] Step 4. Use the distributed sensing-assisted MIMO base station beam selection model obtained in step 3 to perform beam selection on the beam of the selected high dynamic terminal to the base station, and obtain the optimal communication transmission rate corresponding to the beam codeword label with the maximum weight.

[0010] Optionally, the airport communication beam selection model in step 1 includes a channel matrix under a single subcarrier sent by a high dynamic terminal to a single base station and a signal model received by the base station.

[0011] Optionally, the specific steps of constructing a model for maximizing the accuracy of the network prediction of the codeword label corresponding to the actual optimal beam based on the airport communication beam selection model in step 1 are:

[0012] Construct the optimization problem of communication transmission rate under the airport communication beam selection model;

[0013] The optimization problem of communication transmission rate is transformed into the problem of finding the actual optimal beam corresponding to the codeword label. Under the airport communication beam selection model, a theoretical optimal beam corresponding to the codeword label estimation model is constructed.

[0014] Based on the theoretical best beam corresponding codeword label estimation, a model is constructed to maximize the accuracy of the network's prediction of the actual best beam corresponding codeword label.

[0015] Optionally, the communication transmission rate is expressed as:

[0016]

[0017] in, Indicates The communication transmission rate of each time block; Indicates A high-dynamic terminal sends a channel matrix under a single subcarrier to a single base station in a time block; Indicates The codeword selected by the time block; For the Signals from highly dynamic terminals within a time block; represents the noise variance; Indicates the communication bandwidth.

[0018] Optionally, the high-dynamic terminal data includes terminal information of the position, speed and posture of the high-dynamic terminal relative to the base station.

[0019] Optionally, the high-dynamic terminal data also includes the high-dynamic terminal performing beam scanning on all base stations according to the code book, and obtaining the received signal strength corresponding to the code word label selected by the high-dynamic terminal in each time block collected by each base station, and the actual optimal beam corresponding code word label used by the high-dynamic terminal when the signal strength is maximum.

[0020] Compared with the prior art, the present invention has at least the following beneficial effects:

[0021] (1) The beam selection method of the present invention adopts a distributed multi-base station collaboration method to perform beam selection with low memory consumption, which is suitable for real-time beam selection scenarios;

[0022] (2) The distributed sensing-assisted MIMO base station beam selection method proposed by the present invention has low computational complexity and is suitable for airport scenarios with high dynamic terminals such as aircraft and support vehicles;

[0023] (3) The beam selection method of the present invention is combined with the distributed sensing-assisted MIMO base station beam selection algorithm to train each node simultaneously, distributing the computational workload to each base station, thereby improving the learning efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings are only for the purpose of illustrating particular embodiments and are not to be construed as limiting the invention.

[0025] Figure 1 A schematic diagram of a high-dynamic terminal data acquisition coordinate system for the beam selection method of the present invention;

[0026] Figure 2 A schematic diagram of a distributed sensing-assisted MIMO base station beam selection model of the beam selection method of the present invention;

[0027] Figure 3 A training flow chart of a distributed sensing-assisted MIMO base station beam selection model of the beam selection method of the present invention;

[0028] Figure 4 The figure is a flow chart of the beam selection method of the present invention. DETAILED DESCRIPTION

[0029] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments can be combined with each other. In addition, the present invention can also be implemented in other ways different from those described herein, and therefore, the protection scope of the present invention is not limited by the specific embodiments disclosed below.

[0030] A specific embodiment of the present invention, as Figure 1-Figure 4 , discloses a distributed sensing-assisted beam selection method for 5G AeroMACS, the specific steps are as follows:

[0031] Step 1. Establish an airport communication beam selection model; based on the airport communication beam selection model, construct a model that maximizes the accuracy of the network's prediction of the codeword label corresponding to the actual optimal beam.

[0032] The present invention considers the scenario of a single high-dynamic terminal and multiple base stations. Antenna array, Global Position System (GPS), Inertial Navigation System (INS), each base station (BS) is equipped with A uniform planar antenna array, Indicates the number of rows of the base station array antenna; Indicates the number of columns of the base station array antenna.

[0033] Step 11. The airport communication beam selection model includes a channel matrix under which a high dynamic terminal sends a single subcarrier to a single base station and a signal model received by the base station;

[0034] Furthermore, a channel matrix is ​​constructed when a high-dynamic terminal sends a single subcarrier to a single base station. The expression is:

[0035] (1)

[0036] in, Indicates The high-dynamic terminal sends a channel matrix under a single subcarrier to a single base station in a time block. , represents the complex domain, Indicates the number of antennas equipped with a single base station; Indicates the number of antennas equipped by the high-dynamic terminal; and Respectively represent The complex gain of the LoS path and the NLoS path for each time block; and Respectively represent Antenna steering vectors for base station and high dynamic terminal LoS paths in time blocks; and Respectively represent Antenna steering vectors for base stations and highly dynamic terminals NLoS paths in time blocks; and Respectively represent The azimuth and elevation angles of the base station arrival angle under the LoS path in each time block; and Respectively represent The azimuth and elevation angles of the departure angle of the high-dynamic terminal under the LoS path in each time block; and Respectively represent The azimuth and elevation angles of the base station arrival angle under the NLoS path in each time block; and Respectively represent The azimuth and elevation angles of the departure angle of the highly dynamic terminal under the NLoS path in each time block; is the Rice factor, which is the power ratio between the LoS path and the NLoS path; Indicates the number of NLoS paths; LoS path refers to line-of-sight wireless transmission path (Line of Sight), and NLoS path refers to non-line-of-sight wireless transmission path.

[0037] Furthermore, a signal model received by the base station is constructed, and the expression is:

[0038]

[0039] in, Indicates that the base station is The signal received within a time block; Indicates A high-dynamic terminal sends a channel matrix under a single subcarrier to a single base station in a time block; Indicates The codeword selected by the time block; For the The signal from the high-dynamic terminal in a time block, For the The mean in the time block is 0 and the variance is Gaussian white noise.

[0040] Further, It's a codebook The codeword labels in .

[0041] Step 12. Construct a communication transmission rate model under the airport communication beam selection model, and the expression is:

[0042]

[0043] in, Indicates The communication transmission rate of each time block; Indicates the communication bandwidth.

[0044] Step 13. Under the airport communication beam selection model, construct the optimization problem of communication transmission rate, which is expressed as:

[0045]

[0046] in, For the code book.

[0047] It can be understood that when the communication transmission rate is the highest, the beam is the optimal beam selected.

[0048] Step 14. The optimization problem of the communication transmission rate in step 13 is transformed into the problem of finding the actual optimal beam corresponding to the codeword label. Under the airport communication beam selection model, a theoretical optimal beam corresponding to the codeword label estimation model is constructed, and the expression is:

[0049]

[0050] in, Indicates The codeword label corresponding to the theoretical best beam predicted within a time block, ; The representative parameters are In the network, the terminal information such as the position, speed and posture of the highly dynamic terminal is Function of selecting beam; Indicates The terminal information such as the position, speed and posture of a highly dynamic terminal in a time block.

[0051] It can be understood that the network includes base stations, highly dynamic terminals and a server side.

[0052] Step 15. Based on the theoretical best beam corresponding codeword label estimation in step 14, a model is constructed to maximize the accuracy of the network's prediction of the actual best beam corresponding codeword label, which is expressed as:

[0053]

[0054] in, Indicates that through Terminal information for time blocks A function that maximizes the accuracy of selecting the actual best beam corresponding to the codeword label; Indicates the first Time block terminal information The codeword label corresponding to the predicted theoretical optimal beam is is the codeword corresponding to the actual optimal beam probability; Represents the total prediction time.

[0055] Step 2. Obtain high dynamic terminal data;

[0056] Furthermore, high dynamic terminals include aircraft and support vehicles.

[0057] Step 21. High dynamic terminal according to the codebook Perform beam scanning on all base stations to obtain the high-dynamic terminal first Time-selected codeword labels The corresponding received signal strength; record the actual optimal beam corresponding codeword label used by the high dynamic terminal when the signal strength is maximum .

[0058] Step 22. Collect terminal information such as the position, speed, and posture of the high-dynamic terminal relative to the base station to obtain a high-dynamic terminal beam selection data set.

[0059] For example, Figure 1 As shown, in the geographic coordinate system, the origin is the base station center, The axis points due east. The axis points due north. Axis axis, The vertical axis of the axis points to the opposite direction of the center of the earth. High-dynamic terminals use GPS, INS and other devices to record in time blocks The GPS position coordinates of the internal terminal relative to the base station 、INS device position coordinates ,speed and posture Information, including , , The high dynamic terminal measured by GPS is axis, axis, The position coordinates of the axis, , , The high dynamic terminal measured by INS is axis, axis, The position coordinates of the axis, , , They are respectively high dynamic terminals in axis, axis, The speed in the axis direction, , , They are the heading angle, pitch angle, and roll angle of the high dynamic terminal respectively.

[0060] Combining the above data into The high dynamic terminal beam selection data set collected in a time block is expressed as:

[0061]

[0062] Step 23. Obtain high dynamic terminal data according to the high dynamic terminal beam selection data set, expressed as:

[0063] .

[0064] Step 3. Based on the high dynamic terminal data obtained in step 2 and the accuracy model of maximizing the network's prediction of the codeword label corresponding to the actual optimal beam in step 1, a distributed sensing-assisted MIMO base station beam selection model is trained.

[0065] The present invention distributes the task of training the distributed sensing-assisted MIMO base station beam selection model to multiple base stations by introducing a federated learning algorithm. Multiple base stations collaborate to extract multimodal data features of the position, speed, and posture of high-dynamic terminals such as aircraft and support vehicles, which can make full use of data and computing resources and improve the operating efficiency of the algorithm.

[0066] Specifically, Figure 2 As shown, the base station includes a base station operation unit, which is composed of a Transformer neural network with multiple encoder layers. Composition: The distributed sensing-assisted MIMO base station beam selection model is embedded in the server side, including a server-side computing unit, which contains a classification network composed of a multi-layer perceptron (MLP) , and Respectively The Transformer neural network of the base station and the network parameters of the server, where: , is the total number of base stations.

[0067] No. Transformer neural network for each base station It consists of 1 embedding layer, 1 position encoding layer, 2 encoder layers, 2 linear layers and 1 dropout layer; an encoder layer consists of 1 multi-head attention layer, 2 residual connections and layer normalization operations (Add&Norm) and 1 feed forward layer; the classification network on the server side Consists of 1 Linear layer.

[0068] Based on the highly dynamic terminal data obtained in step 2 and the model that maximizes the accuracy of the network's prediction of the codeword label corresponding to the actual optimal beam, the distributed sensing-assisted MIMO base station beam selection model is trained. The specific steps are as follows:

[0069] Step 31. Based on the high-dynamic terminal data obtained in step 2, obtain the training set of each base station and initialize the Transformer neural network parameters of each base station and the network parameters on the server side , the number of layers of the Transformer neural network of the base station , total number of training steps and the network learning rate .

[0070] Step 32. Each base station selects a batch of data from the training set and inputs it into its Transformer neural network to extract the multimodal features of the position, speed and posture of the high-dynamic terminal, and uploads it to the server.

[0071] Step 33. The server inputs the received multimodal features into the classification network to obtain the optimal beam corresponding codeword label prediction results for each base station, and transmits them back to the corresponding base station.

[0072] Step 34. Based on the prediction results of the codeword labels corresponding to the optimal beams of each base station, each base station obtains the corresponding beam selection loss according to the Transformer neural network beam selection loss function, and uploads the beam selection loss to the server.

[0073] No. Transformer neural network beam selection loss for each base station The expression is:

[0074]

[0075] in, For the The optimal beam corresponding to the codeword label prediction result of each base station; For the The actual optimal beam corresponding to each base station codeword label; For the The total number of base station samples, For the The probability that the optimal beam corresponding to the codeword label of each time block is correct.

[0076] Step 35. The server aggregates the beam selection losses of all base stations into a global beam selection loss and transmits the global beam selection loss back to the base station; back gradient propagation is used to update the Transformer neural network of the base station and the network parameters of the server.

[0077] Furthermore, the expression of the global beam selection loss is: (9).

[0078] Furthermore, the expression of the network parameters of the base station's Transformer neural network and the server's network is updated by reverse gradient propagation:

[0079]

[0080] (10)

[0081] in, Represents the global beam selection loss value; represents the network learning rate; Indicates the next generation Transformer neural network parameters of each base station; Indicates the current generation Base station network parameters; Indicates the next generation server-side network parameters; Indicates the network parameters of the current generation server.

[0082] Step 36. Determine whether the global beam selection loss converges. If so, end the training and save the network parameters to obtain the distributed sensing-assisted MIMO base station beam selection model. Otherwise, repeat steps 32-35 using the updated base station's Transformer neural network and the server-side network parameters.

[0083] It can be understood that the global beam selection loss is the convergence of The Transformer neural network beam selection loss of the base station is the smallest, and the actual communication transmission rate achieved by using this codeword label is Best.

[0084] Step 4. Use the distributed sensing-assisted MIMO base station beam selection model obtained in step 3 to perform beam selection on the beam of the selected high dynamic terminal to the base station, and obtain the optimal communication transmission rate corresponding to the beam codeword label with the maximum weight.

[0085] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A distributed sensing-assisted beam selection method for 5G AeroMACS, characterized in that: The specific steps are as follows: Step 1. Establish an airport communication beam selection model; based on the airport communication beam selection model, construct a model that maximizes the accuracy of the network's prediction of the codeword label corresponding to the actual optimal beam; Step 11. The airport communication beam selection model includes a channel matrix under which a high dynamic terminal sends a single subcarrier to a single base station and a signal model received by the base station; Step 12. Construct a communication transmission rate model under the airport communication beam selection model; Step 13. Under the airport communication beam selection model, construct the optimization problem of communication transmission rate; Step 14. The optimization problem of the communication transmission rate in step 13 is transformed into the problem of finding the actual optimal beam corresponding codeword label, and a theoretical optimal beam corresponding codeword label estimation model is constructed under the airport communication beam selection model; Step 15. Based on the theoretical best beam corresponding codeword label estimation in step 14, a model is constructed to maximize the accuracy of the network's prediction of the actual best beam corresponding codeword label; Step 2. Obtain high dynamic terminal data; Step 21. The high-dynamic terminal performs beam scanning on all base stations according to the codebook, obtains the received signal strength corresponding to the codeword label selected by the high-dynamic terminal at each time collected by each base station; records the actual optimal beam corresponding to the codeword label used by the high-dynamic terminal when the signal strength is the maximum; Step 22: Collect terminal information of the position, speed and posture of the high-dynamic terminal relative to the base station to obtain a high-dynamic terminal beam selection data set; Step 23. Obtain high dynamic terminal data according to the high dynamic terminal beam selection data set; Step 3. Based on the high dynamic terminal data obtained in step 2 and the accuracy model of maximizing the network's prediction of the codeword label corresponding to the actual optimal beam in step 1, a distributed sensing-assisted MIMO base station beam selection model is trained; Step 4. Use the distributed sensing-assisted MIMO base station beam selection model obtained in step 3 to select the beam of the selected high-dynamic terminal to the base station, and obtain the optimal communication transmission rate corresponding to the beam codeword label with the maximum weight, which is expressed as: in, Indicates The communication transmission rate of each time block; Indicates A high-dynamic terminal sends a channel matrix under a single subcarrier to a single base station in a time block; Indicates The codeword selected by the time block; For the Signals from highly dynamic terminals within a time block; represents the noise variance; Indicates the communication bandwidth.

Images