Lightweight beam forming method based on unsupervised MLP

Abstract

The invention discloses a lightweight beam forming method based on unsupervised MLP, which belongs to the field of wireless communication, and comprises the following steps: constructing a communication signal-to-noise ratio and a sensing signal-to-noise ratio of an echo signal of an ISAC signal at a target according to the ISAC signal sent by a base station; constructing a target optimization problem of the sensing signal-to-noise ratio according to the ISAC signal and the communication signal-to-noise ratio; solving the target optimization problem to obtain a target beam former; inputting channel parameters in the communication signal-to-noise ratio and the sensing signal-to-noise ratio into the trained lightweight network based on the multilayer sensor to obtain a phase control vector; obtaining a target phase shift matrix according to the phase control vector; and obtaining the maximum sensing signal-to-noise ratio of the echo signal according to the target beam former and the target phase shift matrix. According to the method, the communication and sensing performance can be effectively balanced, the calculation complexity is relatively low, the performance is relatively good, and the sensing signal-to-noise ratio of the echo signal is improved.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication, specifically relating to a lightweight beamforming method based on unsupervised MLP. Background Technology

[0002] Integrated Sensing and Communications (ISAC) is widely regarded as a key technology in next-generation wireless networks. This technology significantly improves overall system performance by sharing hardware and spectrum resources and collaboratively designing waveform and signal processing algorithms. In recent years, numerous studies have focused on combining advanced technologies such as Reconfigurable Intelligent Surface (RIS), terahertz communication, and multiple-input multiple-output (MIMO) with ISAC to further optimize its performance. In particular, this paper focuses on a communication-centric RIS-assisted ISAC system, where precise configuration of the RIS unit phase enables efficient beamforming and effectively improves the quality of the received signal.

[0003] Currently, existing solutions to various beamforming optimization problems all have certain limitations. On the one hand, model-driven optimization methods typically rely on convex approximations to transform the original problem into a computationally feasible suboptimal solution; however, such methods are difficult to adapt to dynamically changing communication environments. On the other hand, while neural network-based methods possess strong modeling capabilities, they are often accompanied by high computational complexity, and the design of their loss functions still faces many challenges, such as overly complex network structures and difficulties in effectively constructing loss functions. Summary of the Invention

[0004] To address the aforementioned problems in the prior art, this invention provides a lightweight beamforming method based on unsupervised MLP.

[0005] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a lightweight beamforming method based on unsupervised MLP, the method comprising: The communication signal-to-noise ratio and the sensing signal-to-noise ratio are constructed based on the ISAC signals transmitted by the base station; Based on the ISAC signal and the communication signal-to-noise ratio, a target optimization problem for the perceived signal-to-noise ratio is constructed; wherein, the target optimization problem is used to optimize the beamformer and the phase shift matrix to maximize the perceived signal-to-noise ratio of the echo signal of the ISAC signal at the target; Solving the objective optimization problem yields the target beamformer; The channel parameters in the communication signal-to-noise ratio and the sensing signal-to-noise ratio are input into the trained lightweight network based on a multilayer perceptron to obtain the phase control vector; wherein, the lightweight network based on a multilayer perceptron is trained according to a preset loss function; The target phase shift matrix is ​​obtained based on the phase control vector; The maximum perceived signal-to-noise ratio of the echo signal is obtained based on the target beamformer and the target phase shift matrix.

[0006] Optionally, the communication signal-to-noise ratio is expressed as follows: ; in, This indicates the signal-to-noise ratio of the communication. This represents the channel through which ISAC signals propagate from the base station to the RIS and then to the user. Indicates beamformer, Indicates communication noise power. This indicates the transpose operation.

[0007] Optionally, the perceived signal-to-noise ratio is expressed as follows: ; in, This represents the perceived signal-to-noise ratio. Indicates a two-way sensing channel. This represents the perceived noise power.

[0008] Optionally, the objective optimization problem is expressed as follows: ; in, Indicates the communication threshold. This indicates the upper limit of the transmission power.

[0009] Optionally, the target beamformer is represented as follows: ; in, This refers to the target beamformer. This represents the channel through which the signal is propagated from the base station to the RIS and then to the target. Indicates modulo, Indicates and Direction vectors in the same direction Indicates and Orthogonal unit vectors, and located at and Within Zhang Cheng's plane, Indicates assignment to The range, Indicates assignment to The range.

[0010] Optionally, the lightweight network based on a multilayer perceptron includes a flattened layer, a fully connected layer, a ReLU activation function layer, and a linear layer linked together in sequence.

[0011] Optionally, the preset loss function is expressed as follows: ; in, This represents the preset loss function. Indicates the batch size during the training phase. Indicates the number of components on the RIS. Indicates the first The batch was processed after the first Communication channels for each RIS element Indicates the first The batch was processed after the [number]th [period]. Sensing channels of RIS components Indicates the first Each batch of bidirectional sensing channels, This indicates taking the complex conjugate of each element of the vector. For balance coefficient, This indicates taking the Frobenius norm.

[0012] Optionally, obtaining the target phase shift matrix based on the phase control vector includes: The transpose of the phase control vector is obtained from the phase control vector. The target phase shift matrix is ​​obtained by diagonalizing the phase control vector and the transpose vector.

[0013] The technical solutions provided by the embodiments of the present invention may include the following beneficial effects: In the above technical solutions, this invention proposes a loss function based on channel alignment for unsupervised learning, which can effectively balance communication and sensing performance; and the lightweight network based on multilayer perceptron proposed in this invention has both low computational complexity and good performance; this invention constructs an objective optimization problem to maximize the sensing signal-to-noise ratio under the constraint of communication signal-to-noise ratio, and solves this problem by optimizing the phase shift matrix of the beamformer to improve the sensing signal-to-noise ratio of the echo signal.

[0014] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an overall architecture provided by an embodiment of the present invention; Figure 2This is a flowchart of a lightweight beamforming method based on unsupervised MLP provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of a lightweight network based on a multilayer perceptron provided in an embodiment of the present invention; Figure 4 This is a schematic diagram illustrating the convergence performance of the sensing signal-to-noise ratio under different parameters according to an embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the communication signal-to-noise ratio convergence performance of the present invention under different parameters, provided by an embodiment of the present invention; Figure 6 This is a schematic diagram showing a comparison between the present invention and IBF, provided by an embodiment of the present invention; Figure 7 This is another schematic diagram showing the comparison results between the present invention and IBF provided by the embodiments of the present invention. Detailed Implementation

[0016] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0017] Figure 1 This is a schematic diagram of an overall architecture provided by an embodiment of the present invention, such as... Figure 1 As shown, this invention is based on an RIS-assisted ISAC system for analysis, wherein the base station (BS) is equipped with Each antenna can simultaneously sense targets and serve a single-antenna user; the RIS has Several components are used to adjust the channel environment. Specifically, the BS transmits the ISAC signal. ,in, It is a unity-power baseband signal, and Indicates beamformer, Represents the field of complex numbers. and The noise representing the communication and sensing channels, respectively, both follow a Gaussian distribution. and The power of communication and sensing noise are respectively and Based on this, a detailed analysis of the ISAC signal is conducted: ISAC signals pass through the channel The signal propagates from the BS to the RIS, and after phase tuning at the RIS, it passes through the channel. The signal is propagated to the user, therefore the signal received by the user can be represented as: ; in, This indicates that the ISAC signal originates from the BS-RIS user's channel, therefore the signal received by the user can be represented as: ; Simultaneously, the ISAC signal passes through the channel. The target is sensed, and the echo signal from the target is reflected at the RIS and received by the BS. The sensed signal received by the base station can be represented as: ; in, This indicates the channel from which the ISAC signal originates in the BS-RIS-target channel; therefore, the signal received by the user can be represented as: .

[0018] Figure 2 This is a flowchart of a lightweight beamforming method based on unsupervised MLP provided by an embodiment of the present invention, as shown below. Figure 2 As shown, the method may include the following steps: S201. Construct the communication signal-to-noise ratio and the perceived signal-to-noise ratio of the echo signal of the ISAC signal at the target based on the ISAC signal sent by the base station.

[0019] Optionally, the communication signal-to-noise ratio is expressed as follows: ; in, Indicates the signal-to-noise ratio in communication. This represents the channel through which ISAC signals propagate from the base station to the RIS and then to the user. Indicates beamformer, Indicates communication noise power. Indicates the transpose operation; The perceived signal-to-noise ratio is expressed as follows: ; in, This represents the perceived signal-to-noise ratio. Indicates a two-way sensing channel. This represents the perceived noise power.

[0020] S202. Construct a target optimization problem for the perceived signal-to-noise ratio based on the ISAC signal and the communication signal-to-noise ratio; wherein, the target optimization problem is used to optimize the beamformer and the phase shift matrix to maximize the perceived signal-to-noise ratio.

[0021] Understandably, the design goal is to jointly optimize the beamformer. Phase shift matrix of RIS To maximize the perceived signal-to-noise ratio of the echo signal. The corresponding initial optimization problem can be formulated as: ; The first constraint is achieved by limiting the communication threshold. To ensure communication quality; the second constraint limits the upper limit of transmit power. The first constraint ensures that the beamforming design is within the maximum transmit power budget of the base station; the second constraint controls the modulus of each non-zero element on the diagonal of the phase shift matrix to 1, which means that each element can only change the phase of the incident wave, but not its amplitude.

[0022] Since the initial optimization problem described above is highly nonconvex, an alternating optimization method can be used to solve it. This method involves updating one variable while keeping another variable constant and iterating alternately. In this case, channel correlation... It becomes crucial.

[0023] The alignment degree between the beamformer and the sensing channel directly determines the strength of the channel correlation. Strong channel correlation enables effective power reuse for communication, while weak correlation leads to substandard communication performance. In such cases, adjustments to the beamformer are necessary. Optimization is performed to place it within the extended channel subspace, thereby improving communication performance. Therefore, another variable, the phase shift matrix, can be fixed, transforming the problem into an objective optimization problem, as follows: ; in, Indicates the communication threshold. This indicates the upper limit of the transmission power.

[0024] S203. Solve the target optimization problem to obtain the target beamformer.

[0025] It is understandable that the optimal solution to this objective optimization problem, i.e., the target beamformer, can be expressed as follows: ; in, Indicates the target beamformer. This represents the channel through which the signal is propagated from the base station to the RIS and then to the target. Indicates modulo, Indicates and Direction vectors in the same direction , Indicates and Orthogonal unit vectors and located and Within Zhang Cheng's plane, Indicates assignment to The range is intended to precisely satisfy... This constraint, , Indicates assignment to The range, The goal is to maximize It is not difficult to see that the optimal solution to the objective optimization problem always belongs to... and In the linear subspace spanned by the beamformer, strong channel correlation enables effective power reuse for communication when the alignment between the beamformer and the sensing channel is high, while weak correlation leads to unsatisfactory communication performance. In this case, the beamformer needs to be optimized to place it within the extended channel subspace, thereby improving communication performance.

[0026] S204. Input the channel parameters from the communication signal-to-noise ratio and the sensing signal-to-noise ratio into the trained lightweight network based on the multilayer perceptron to obtain the phase control vector; wherein, the lightweight network based on the multilayer perceptron is trained according to a preset unsupervised learning loss function.

[0027] Understandable, Figure 3 This is a schematic diagram of a lightweight network based on a multilayer perceptron provided in an embodiment of the present invention, as shown below. Figure 3 As shown, the lightweight network based on the multilayer perceptron includes a flattened layer, a fully connected layer, a ReLU activation function layer, and a linear layer connected in sequence.

[0028] Specifically, using and The communication channel gain and the sensing channel gain can be obtained, as shown below:

[0029] ; in, For communication channel gain, Indicates the perceived channel gain. , Indicates the characteristics of the communication channel. This indicates the characteristics of the sensing channel.

[0030] A lightweight network based on a multilayer perceptron was designed, taking 4×N×N multichannel data as input. Traditional neural networks process real numbers, but... and They are usually complex numbers. Therefore, their real and imaginary parts are extracted separately and used as channel parameters in the communication signal-to-noise ratio and the sensing signal-to-noise ratio, denoted as . , , , ,in, This indicates an operation that extracts the real part of the data. This indicates the operation of taking the imaginary part of the data. Let be the real part of the communication channel characteristics. Let the imaginary part of the communication channel characteristics be denoted as . Let be the real part of the sensing channel characteristics. This represents the imaginary part of the sensing channel characteristics. Subsequently, Dimensions expanded , Let the numbers be in the real number field, and concatenate them along the extended dimensions. Therefore, the input to the constructed lightweight multilayer perceptron-based network is of dimension [dimensionality missing]. 3D tensors.

[0031] For example, in this invention, the number of components on the RIS is N=32, so the input is first flattened into a one-dimensional vector of length 4096 through a flattening layer; then it passes through two hidden fully connected layers (both with a dimension of 512), with a ReLU activation function inserted in between to achieve a non-linear transformation; finally, the output passes through a last linear layer and is mapped to a phase vector of dimension N. To satisfy the complex representation of the output and the unit modulus constraint of the phase angle, the final step of the network performs cosine and sine function transformations on the output to obtain the real and imaginary parts, respectively, constructing a complex form phase control vector: ; However, the sensing channel experiences double fading during bidirectional propagation, while the communication channel only experiences single fading. To address this issue, a regularization term is introduced into the loss function to balance the correlation between the two channels and the fading problem of the sensing channel. The pre-defined unsupervised learning loss function is expressed as follows: ; in, This indicates the predefined unsupervised learning loss function. Indicates the batch size during the training phase. Indicates the number of components on the RIS. Indicates the first The batch was processed after the [number]th [period]. Communication channels for each RIS element Indicates the first The batch was processed after the [number]th [period]. Sensing channels of RIS components Indicates the first Each batch of bidirectional sensing channels, This indicates taking the complex conjugate of each element of the vector. For balance coefficient, This indicates taking the Frobenius norm.

[0032] S205. Obtain the target phase shift matrix based on the phase control vector.

[0033] Optionally, S205 may include: The transpose of the phase control vector is obtained from the phase control vector. The target phase shift matrix is ​​obtained by diagonalizing the phase control vector and the transpose vector.

[0034] Specifically, the obtained phase control vector and transpose vector are diagonalized to obtain the phase shift matrix of RIS. This is compared with the channel matrix from BS to RIS. Multiplying these yields the equivalent channel after RIS phase adjustment. This equivalent channel is then multiplied by the RIS-to-user channel. and RIS to sensing channel The product yields the phase-adjusted cascaded communication and sensing channels. Based on this, the corresponding channel gain can be calculated, and a loss function can be designed to drive the next round of optimization iterations.

[0035] S206. Obtain the maximum perceived signal-to-noise ratio of the echo signal based on the target beamformer and the target phase shift matrix.

[0036] In one embodiment, to make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the operating results.

[0037] Model parameters: 8 base station antennas, 32 RIS components, 8dBm transmit power, and -20dBm noise power; training set of 5000 parameters, test set of 100 parameters; Batch size of network architecture is set to 200, Lr = 0.0001.

[0038] First, using SNRc and SNRs as performance metrics, we scanned different mini-batch sizes and learning rates. Figure 4 This is a schematic diagram illustrating the convergence performance of the sensing signal-to-noise ratio under different parameters according to an embodiment of the present invention, as shown below. Figure 4 As shown, it can be observed that the signal-to-noise ratio (SNR) performance curves all show an increasing trend with the increase of epochs, indicating that the present invention effectively improves the sensing performance in ISAC signals. When the learning rate is set to 0.0001 and the mini-batch size is set to 200, the best performance curve is observed. Furthermore, as the mini-batch size increases, the convergence speed slows down and the convergence value decreases. When the mini-batch size is set to 200, reducing the learning rate to a certain extent leads to improved performance. Figure 5 This is a schematic diagram illustrating the communication signal-to-noise ratio convergence performance of the present invention under different parameters, as provided in an embodiment of the present invention. Figure 5As shown, at the beginning of the training phase, a significant decrease in SNRc can be observed. This is because communication performance is affected to some extent in order to achieve optimal listening performance. The goal of this invention is to minimize communication loss while realizing the sensing function. Furthermore, the SNRc curve changes in accordance with SNRs as the parameter settings change. However, it is worth noting that when the learning rate is reduced to 0.00001, the final convergence values ​​of SNRc and SNRs are the worst among all parameter combinations, indicating that reducing the learning rate does not guarantee improved performance.

[0039] Based on the above experimental results, the performance of IBF (ISAC BeamForming neural network) and the present invention was further compared in terms of runtime, signal-to-noise ratio (SNR), and inference time. First, the training and inference times of the two were compared. Figure 6 This is a schematic diagram illustrating the comparison between the present invention and IBF, provided by an embodiment of the present invention. Figure 6 As shown, compared to IBF, the time required by this invention is reduced by 30%, demonstrating that this invention has lower computational complexity than IBF. Furthermore, Figure 7 This is another comparative diagram of the present invention and IBF provided by an embodiment of the present invention, as shown in the figure. Figure 7 As shown, this invention also compares the performance of the two in terms of SNRs and SNRc. It can be seen that compared with IBF, the convergence performance of this invention is improved by 0.7dB and the communication performance is improved by 2dB. The superior performance of this invention stems from its loss design, which effectively decouples the beamforming process from the influence of signal amplitude fluctuations, thereby leading to more stable and effective optimization.

[0040] In the above technical solutions, this invention proposes a loss function based on channel alignment for unsupervised learning, which can effectively balance communication and sensing performance; and the lightweight network based on multilayer perceptron proposed in this invention has both low computational complexity and good performance; this invention constructs an objective optimization problem to maximize the sensing signal-to-noise ratio under the constraint of communication signal-to-noise ratio, and solves this problem by optimizing the phase shift matrix of the beamformer to improve the sensing signal-to-noise ratio of the echo signal.

[0041] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0042] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art will understand and implement other variations of the disclosed embodiments by reviewing the accompanying drawings and the disclosure in carrying out the claimed invention. In the description of the invention, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0043] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the concept of the present invention, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A lightweight beamforming method based on unsupervised MLP, characterized in that, The method includes: The communication signal-to-noise ratio and the perceived signal-to-noise ratio of the echo signal of the ISAC signal at the target are constructed based on the ISAC signal sent by the base station. Based on the ISAC signal and the communication signal-to-noise ratio, a target optimization problem for the perceived signal-to-noise ratio is constructed; wherein, the target optimization problem is used to optimize the beamformer and the phase shift matrix to maximize the perceived signal-to-noise ratio; Solving the objective optimization problem yields the target beamformer; The channel parameters in the communication signal-to-noise ratio and the sensing signal-to-noise ratio are input into a trained lightweight network based on a multilayer perceptron to obtain a phase control vector; wherein, the lightweight network based on a multilayer perceptron is trained according to a preset unsupervised learning loss function; The target phase shift matrix is ​​obtained based on the phase control vector; The maximum perceived signal-to-noise ratio of the echo signal is obtained based on the target beamformer and the target phase shift matrix.

2. The lightweight beamforming method based on unsupervised MLP according to claim 1, characterized in that, The communication signal-to-noise ratio is expressed as follows: ； in, This indicates the signal-to-noise ratio of the communication. This represents the channel through which ISAC signals propagate from the base station to the RIS and then to the user. Indicates beamformer, Indicates communication noise power. This indicates the transpose operation.

3. The lightweight beamforming method based on unsupervised MLP according to claim 2, characterized in that, The perceived signal-to-noise ratio is expressed as follows: ； in, This represents the perceived signal-to-noise ratio. Indicates a two-way sensing channel. This represents the perceived noise power.

4. The lightweight beamforming method based on unsupervised MLP according to claim 3, characterized in that, The objective optimization problem is expressed as follows: ； in, Indicates the communication threshold. This indicates the upper limit of the transmission power.

5. The lightweight beamforming method based on unsupervised MLP according to claim 4, characterized in that, The target beamformer is represented as follows: ； in, This refers to the target beamformer. This represents the channel through which the signal is propagated from the base station to the RIS and then to the target. Indicates modulo, Indicates and Direction vectors in the same direction Indicates and Orthogonal unit vectors, and located at and Within Zhang Cheng's plane, Indicates assignment to The range, Indicates assignment to The range.

6. The lightweight beamforming method based on unsupervised MLP according to claim 1, characterized in that, The lightweight network based on a multilayer perceptron includes a flattened layer, a fully connected layer, a ReLU activation function layer, and a linear layer linked together in sequence.

7. The lightweight beamforming method based on unsupervised MLP according to claim 1, characterized in that, The preset unsupervised learning loss function is expressed as follows: ； in, This represents the preset unsupervised learning loss function. Indicates the batch size during the training phase. Indicates the number of components on the RIS. Indicates the first The batch was processed after the [number]th [period]. Communication channels for each RIS element Indicates the first The batch was processed after the [number]th [period]. Sensing channels of RIS components Indicates the first Each batch of bidirectional sensing channels, This indicates taking the complex conjugate of each element of the vector. For balance coefficient, This indicates taking the Frobenius norm.

8. The lightweight beamforming method based on unsupervised MLP according to claim 1, characterized in that, The step of obtaining the target phase shift matrix based on the phase control vector includes: The transpose of the phase control vector is obtained from the phase control vector. The target phase shift matrix is ​​obtained by diagonalizing the phase control vector and the transpose vector.

Images