Cloud wireless access network millimeter wave signal optimization method based on machine learning

Abstract

The invention discloses a cloud wireless access network millimeter wave signal optimization method based on machine learning, which adopts a double-ring OEO to replace a traditional radio frequency signal source, improves phase noise and improves signal stability. Meanwhile, the gain behavior of the millimeter wave signal is modeled by using a BiLSTM network, and high-precision power prediction is realized by combining the DML chirp effect and the influence of optical fiber dispersion. Besides, the BiLSTM model is optimized through the ACO algorithm, the optimization weight can be flexibly adjusted according to different communication scene requirements, and millimeter wave signals meeting the requirements are generated.

Description

Technical Field

[0001] The present invention belongs to the technical field of wireless communication, and particularly relates to a method for optimizing millimeter-wave signals in a cloud radio access network based on machine learning. Background Art

[0002] Cloud Radio Access Network (C-RAN) is an innovative wireless access network architecture that enables resource sharing and dynamic allocation through centralized baseband processing units (BBUs), significantly improving network capacity and reducing operating costs. However, traditional radio frequency (RF) signal sources are limited in signal transmission quality due to their inherent phase noise problems. Phase noise directly affects the stability and communication quality of signals. Especially in high-frequency communications, it can cause signal distortion and interference, thereby affecting the performance of the entire communication system.

[0003] In addition, the chirp effect of directly modulated lasers (DMLs) and the impact of fiber dispersion on millimeter-wave signals. The chirp effect of DMLs causes instantaneous frequency changes in the modulated signal, and fiber dispersion further amplifies this frequency change, resulting in signal gain behavior. This gain behavior has a significant impact on the transmission quality of millimeter-wave signals, especially in long-distance fiber optic transmissions.

[0004] However, existing research usually only focuses on ideal gain behavior without considering various influences of equipment and environment in experiments. In addition, millimeter-wave signal quality optimization strategies for remote C-RAN transmitters based on DMLs have not been widely reported. Therefore, existing methods still have limitations in practical applications, especially in terms of insufficient adaptability in complex environments and high-dynamic conditions. Summary of the Invention

[0005] The present invention proposes a method for optimizing millimeter-wave signals in a cloud radio access network based on machine learning to solve the problems existing in the above-mentioned prior art.

[0006] To achieve the above object, the present invention provides a method for optimizing millimeter-wave signals in a cloud radio access network based on machine learning, including the following steps:

[0007] Generate millimeter-wave signals with high side-mode suppression ratio and low phase noise through a dual-loop optoelectronic oscillator;

[0008] Input input features related to the characteristics of millimeter-wave signals into a pre-trained bidirectional long short-term memory network model to obtain the power index of the millimeter-wave signals;

[0009] Based on the ant colony optimization algorithm, optimize the power index by optimizing the objective function, and adjust the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain optimized millimeter-wave signals.

[0010] Preferably, the calculation expression for the effective mode spacing of the double-ring optoelectronic oscillator is as follows:

[0011]

[0012] In the formula, l1 and l2 are the lengths of the two single-mode fiber loops respectively, c is the speed of light, and nf is the refractive index of the optical fiber.

[0013] Preferably, the input features include the optical fiber link length, the amplitude and frequency of the modulated intermediate frequency signal; the power metrics include the carrier power, the upper and lower sideband powers, and the second harmonic distortion power.

[0014] Preferably, the calculation expression for predicting the power metric value by the pre-trained bidirectional long short-term memory network model is as follows:

[0015]

[0016] In the formula, w out is the weight matrix of the output layer, b out is the bias matrix of the output layer, is the bidirectional hidden state.

[0017] Preferably, the optimization objective function is as follows:

[0018]

[0019] In the formula, α i represents the optimization weight, represents the carrier power, represents the lower sideband power of the IF component, represents the upper sideband power of the IF component, represents the lower sideband power of the second harmonic distortion, represents the upper sideband power of HD2.

[0020] Preferably, the transfer probability expression for an ant to transfer from node i to node j in the ant colony optimization algorithm is as follows:

[0021]

[0022] In the formula, τ ij is the pheromone concentration from node i to node j, η ij is the heuristic information, which is related to the heuristic solution of the problem. α and β respectively control the importance of the pheromone and the heuristic information. allowed is the set of nodes that an ant can transfer to when it is at node i.

[0023] The present invention also provides a millimeter-wave signal optimization system for cloud radio access network based on machine learning, including:

[0024] A signal generation module, configured to generate a millimeter-wave signal with a high side mode suppression ratio and low phase noise through a dual-loop optoelectronic oscillator;

[0025] A prediction module, configured to input input features related to the characteristics of the millimeter-wave signal into a pre-trained bidirectional long short-term memory network model to predict the power index of the millimeter-wave signal;

[0026] A parameter optimization module, configured to optimize the power index based on the ant colony optimization algorithm by optimizing the objective function, and adjust the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain an optimized millimeter-wave signal.

[0027] The present invention also provides a computer device, including a memory, a processor, and a computer program stored on the memory, and the processor executes the computer program to implement the steps of the method.

[0028] The present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method are implemented.

[0029] The present invention also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method are implemented.

[0030] Compared with the prior art, the present invention has the following advantages and technical effects:

[0031] The present invention discloses a method for optimizing millimeter-wave signals in a cloud radio access network based on machine learning, including the following steps: generating a millimeter-wave signal with a high side mode suppression ratio and low phase noise through a dual-loop optoelectronic oscillator; inputting input features related to the characteristics of the millimeter-wave signal into a pre-trained bidirectional long short-term memory network model to obtain the power index of the millimeter-wave signal; based on the ant colony optimization algorithm, optimizing the power index by optimizing the objective function, and adjusting the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain an optimized millimeter-wave signal. The present invention uses a dual-loop OEO to replace the traditional radio frequency signal source, improves the phase noise, and enhances the signal stability. At the same time, the BiLSTM network is used to model the gain behavior of the millimeter-wave signal, and combined with the influence of the DML chirping effect and fiber dispersion, high-precision power prediction is achieved. In addition, by optimizing the BiLSTM model through the ACO algorithm, the optimization weights can be flexibly adjusted according to different communication scenario requirements to generate millimeter-wave signals that meet the requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0033] Figure 1 is the method flow chart of the embodiment of the present invention;

[0034] Figure 2 is the microwave phase noise comparison chart of the dual-ring OEO and RF source in the embodiment of the present invention;

[0035] Figure 3 is the training flow chart of the millimeter-wave signal gain behavior model based on BiLSTM in the embodiment of the present invention;

[0036] Figure 4 is the fitting effect diagram of the BiLSTM model in the embodiment of the present invention, where (a) is the fitting effect diagram of the carrier power, (b) is the fitting effect diagram of the lower sideband power of the intermediate frequency component, (c) is the fitting effect diagram of the upper sideband power of the intermediate frequency component, (d) is the fitting effect diagram of the lower sideband power of HD2, and (e) is the fitting effect diagram of the upper sideband power of HD2;

[0037] Figure 5 is the input parameter optimization flow chart of the remote C-RAN transmitter based on the ACO algorithm in the embodiment of the present invention;

[0038] Figure 6 is the comparison chart of the optimization curves of the ACO algorithm and the Adam algorithm in the embodiment of the present invention;

[0039] Figure 7 is the comparison chart of the optimized spectrum and the actual spectrum under different SMF link lengths in the embodiment of the present invention, where (a) is the 5 km link length, (b) is the 10 km link length, (c) is the 15 km link length, (d) is the 20 km link length, (e) is the 25 km link length, (f) is the 30 km link length, and (g) is the 35 km link length. Detailed implementation manners

[0040] It should be noted that, without conflict, the embodiments in the present application and the features in the embodiments may be combined with each other. The present application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0041] It should be noted that the steps shown in the flow charts of the drawings may be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flow charts, in some cases, the steps shown or described may be executed in a different order than here.

[0042] Embodiment 1

[0043] As Figure 1 shown, in this embodiment, a millimeter-wave signal optimization method for a cloud radio access network based on machine learning is provided, including the following steps:

[0044] Generating millimeter-wave signals with high side-mode suppression ratio and low phase noise through a dual-ring optoelectronic oscillator;

[0045] Inputting input features related to the characteristics of millimeter-wave signals into a pre-trained bidirectional long short-term memory network model to obtain the power index of the millimeter-wave signals;

[0046] Based on the ant colony optimization algorithm, optimizing the power index by optimizing the objective function, and adjusting the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain optimized millimeter-wave signals.

[0047] Specifically, it is divided into the following steps:

[0048] Step 1: Design of a remote C-RAN transmitter based on a dual-ring OEO.

[0049] The core of the dual-ring OEO of the present invention lies in generating microwave signals with high SMSR and low phase noise through two independent feedback loops. Compared with the traditional single-ring OEO, the dual-ring OEO utilizes the Vernier effect to significantly increase the effective mode spacing, thereby improving the single-mode selectivity and side-mode suppression ability. Specifically, the effective mode spacing of the dual-ring OEO can be calculated by the following formula:

[0050]

[0051] where l1 and l2 are the lengths of the two single-mode fiber loops respectively. By selecting l1 = 2 km and l2 = 170 m, the dual-ring OEO can effectively suppress side modes and improve signal stability without increasing the filter bandwidth. In addition, compared with the traditional RF source, the dual-ring OEO achieves a 27 dBc / Hz phase noise reduction at a 10 kHz frequency offset. The microwave phase noise comparison between the dual-ring OEO and the RF source is as Figure 2 shown.

[0052] Step 2: Modeling the gain behavior of millimeter-wave signals based on BiLSTM.

[0053] The training process of the millimeter-wave signal gain behavior model based on BiLSTM is as Figure 3As shown, the input features of the BiLSTM network structure of the present invention include the optical fiber link length, the amplitude and frequency of the modulated intermediate frequency (IF) signal. These features are processed by the BiLSTM network, and the output is the power metrics of the millimeter-wave signal, including the carrier power, the upper and lower sideband powers, and the HD2 power. Through the bidirectional structure and the LSTM gating mechanism, the BiLSTM network can effectively capture the long-range dependencies in the sequence data, thereby more accurately predicting the gain behavior of the millimeter-wave signal. The modeling process of the gain behavior is mainly based on experimental data, and the BiLSTM network is trained to predict the power metrics of the millimeter-wave signal.

[0054] First, define the input features and output features. The input features include the single-mode optical fiber link length (l SMF ), the amplitude (A IF ) of the modulated intermediate frequency (IF) signal, and the frequency (F IF ) of the modulated IF signal. These input features are organized into the input matrix x(n) = [l SMF (n), A IF (n), F IF (n)] T , where n is the sample index of the training set. The selection of these input features is based on their influence on signal transmission and gain behavior.

[0055] The output features include the carrier power the lower sideband power of the IF component the upper sideband power of the IF component the lower sideband power of the second harmonic distortion (HD2) and the upper sideband power of HD2 These output features are organized into the output matrix to characterize the spectral characteristics of the millimeter-wave signal.

[0056] The training data set contains 10,500 sample groups, which are composed of different combinations of SMF link lengths (5 km to 35 km), IF signal amplitudes (210 mVpp to 500 mVpp), and IF signal frequencies (11 MHz to 60 MHz). The sample groups are divided into a training set, a validation set, and a test set in a ratio of 7:2:1. The goal of the training process is to minimize the mean square error (MSE) between the predicted output and the actual output by adjusting the model parameters.

[0057] During the training process, the BiLSTM network, leveraging its bidirectional structure and LSTM gating mechanism, can process both the forward and backward information of sequential data simultaneously. Specifically, the forward LSTM processes the sequence from the first time step to the last time step; while the backward LSTM processes the sequence from the last time step to the first time step. At each time step, the hidden states of the forward and backward LSTMs are concatenated together to form the final bidirectional hidden state where h t and h' t represent the hidden states of the forward and backward LSTMs respectively. The network has two hidden layers, each containing 15 neurons, and the output layer of the network contains 5 neurons, corresponding to 5 output features. The predicted power level is calculated as follows:

[0058]

[0059] where, w out is the weight matrix of the output layer, and b out is the bias matrix of the output layer.

[0060] The Adam optimizer is used in the training process, and the activation function is the hyperbolic tangent function (Tanh). The upper limit of the training iteration is set to 10,000 times. In each iteration, the model calculates the hidden states of the forward and backward LSTMs, concatenates these hidden states to form the bidirectional hidden state, and then calculates the loss function (i.e., the MSE between the predicted output and the actual output). Next, the model calculates the gradients of the loss function with respect to the model parameters and updates the model parameters according to the rules of the Adam optimizer.

[0061] The fitting effects of the BiLSTM model on the carrier power, the lower sideband power of the intermediate frequency component, the upper sideband power of the intermediate frequency component, the lower sideband power of HD2, and the upper sideband power of HD2 are as Figure 4 shown.

[0062] Step 3: Design of the millimeter-wave signal optimization algorithm based on ACO.

[0063] The present invention combines the ACO algorithm with the BiLSTM model to jointly optimize the SNR and distortion level of millimeter-wave signals. The ACO algorithm simulates the process of ants releasing pheromones to mark paths to find the optimal combination of input parameters, thereby achieving the optimization of the millimeter-wave signal quality.

[0064] The optimization process of the input parameters of the remote C-RAN transmitter based on the ACO algorithm is as Figure 5 shown. The optimization process of the ACO algorithm first needs to initialize relevant parameters, specifically including: the number of ants N a= 50, the maximum number of iterations T = 100, the pheromone evaporation rate ρ = 0.5, the pheromone importance factor α = 1, the heuristic information importance factor β = 5, the total pheromone Q = 100, and the initial pheromone concentration and initialize the optimal solution to be empty, and the optimal cost C b is initialized to infinity. In addition, define the optimization objective function O[·] to evaluate the optimization degree of the carrier power, IF component power, and HD2 power of the millimeter-wave signal, and its form is:

[0065]

[0066] where α1, α2, and α3 are optimization weights used to adjust the optimization priorities of different power metrics. In each iteration, each ant starts building a solution from a random starting node. The transition probability of an ant moving from node i to node j is calculated by the following formula:

[0067]

[0068] where τ ij is the pheromone concentration from node i to node j, η ij is the heuristic information, which is usually related to the heuristic solution of the problem. α and β control the importance of the pheromone and the heuristic information respectively, and allowed is the set of nodes that an ant can transfer to when at node i.

[0069] The ant gradually builds the solution S k according to the above probability formula and calculates the cost C k of this solution. When all ants have completed the path construction, update the pheromone concentration on the path according to the cost of the solution constructed by each ant. The pheromone update rule is:

[0070] τ ij (t + 1) = (1 - ρ)·τ ij (t) + Δτ ij (5)

[0071] where ρ is the pheromone evaporation rate, and Δτ ij is the total pheromone released by all ants on the path from i to j, and its calculation formula is:

[0072]

[0073] where L k is the cost of the solution constructed by the k-th ant, and Q is the total pheromone. In each iteration, compare the costs C k of the solutions constructed by all ants and update the current optimal solution and the optimal cost C b If the cost C of the solution of a certain ant k is less than the current optimal cost C b , then update the optimal solution:

[0074] If C k < C b , then

[0075] The ACO algorithm gradually approaches the optimal solution by repeating the above processes of path selection, pheromone update, and optimization result update. When the preset maximum number of iterations T is reached or the optimal solution that meets the conditions is found, the algorithm terminates. Finally, the ACO algorithm outputs the optimal parameter combination These parameters are used as the input of the BiLSTM model to predict the power index of the millimeter-wave signal and are applied in the actual remote C-RAN transmitter to optimize the quality of the millimeter-wave signal. The optimized millimeter-wave signal shows good consistency between the fitting spectrum line and the actual spectrum, and the mean absolute error (MAE) is as low as 0.043 dB, verifying the effectiveness and accuracy of the proposed optimization strategy.

[0076] In this embodiment, to verify the performance of the ant colony optimization (ACO) algorithm, we compared it with the commonly used Adam optimization algorithm. Figure 6 Shows the comparison of the performance curves of the two algorithms during the optimization process. It can be seen from the figure that the ACO algorithm shows more superior performance during the entire optimization process.

[0077] The ACO algorithm utilizes the behavior of simulating ants searching for food paths and realizes the global search of the solution space through the mechanism of pheromone release and evaporation. This method is particularly good at avoiding being trapped in local optimal solutions and can effectively find the global optimal solution or an approximate optimal solution. In this embodiment, the ACO algorithm shows a faster convergence speed and higher optimization accuracy, indicating its significant advantages in dealing with complex optimization problems.

[0078] In contrast, although the Adam algorithm performs well in many machine learning tasks and combines the adaptive estimation of gradient descent and the momentum method, in the specific optimization problem of this embodiment, the performance of the ACO algorithm is better. This may be because the global search feature of the ACO algorithm is more suitable for solving the complexity of millimeter-wave signal optimization in the present invention. Especially when facing multiple local optimal solutions, the ACO algorithm can more effectively explore the solution space and thus find a better solution.

[0079] Therefore, the experimental results prove that the ACO algorithm in the millimeter-wave signal optimization method of the present invention can not only effectively improve the signal-to-noise ratio (SNR) of the signal and reduce distortion, but also provide better optimization performance than traditional gradient descent algorithms such as Adam. This discovery provides a new and effective tool for the optimization of millimeter-wave signals in future wireless communication networks, which helps to improve the overall performance and reliability of communication systems.

[0080] As Figure 7 shown, in this embodiment, the optimized millimeter-wave signal spectrum line and the actual spectrum are compared and analyzed under different single-mode fiber (SMF) link lengths, including 5 km, 10 km, 15 km, 20 km, 25 km, 30 km, and 35 km. Through these comparisons, it can be clearly seen that the millimeter-wave signal optimization method based on the bidirectional long short-term memory network (BiLSTM) and ant colony optimization (ACO) algorithm proposed in this application exhibits excellent performance under each tested fiber link length. There is a high consistency between the optimized signal spectrum line and the actual spectrum, and the mean absolute error (MAE) reaches an excellent level as low as 0.043 dB, which verifies the effectiveness and accuracy of the proposed optimization strategy under different link lengths.

[0081] In addition, this method not only performs well in short-distance transmission, but also maintains its optimization effect in fiber optic transmission up to 35 km, demonstrating its excellent adaptability and flexibility to the requirements of different communication scenarios. By adjusting the optimization weights in the BiLSTM model and ACO algorithm, this method can flexibly generate millimeter-wave signals that meet different signal-to-noise ratio (SNR) and distortion requirements. This flexibility and optimization ability provide a solid theoretical basis and practical application solution for the design and optimization of C-RAN transmitters, indicating that the method of the present invention has broad application prospects in the development of low-cost, high-efficiency, and high-capacity wireless communication networks, and can provide an efficient, flexible, and reliable signal optimization solution for future wireless communication networks.

[0082] The contributions of the present invention mainly include the following aspects:

[0083] (1) The present invention uses a dual-loop OEO to replace the traditional RF signal source to generate microwave signals with SMSR. Due to the inherent limitations of its electronic components, the traditional RF signal source usually generates relatively high phase noise, which directly affects the signal stability and communication quality, especially more significantly in high-frequency communication and high-precision applications. In contrast, OEO converts light into microwaves through feedback oscillation and uses high-quality-factor optical fibers to achieve low phase noise, thus maintaining the signal stability and quality in high-frequency communication.

[0084] (2) In a remote C-RAN transmitter, the chirp effect of the DML and fiber dispersion have important effects on the gain behavior of millimeter-wave signals. The chirp effect of the DML causes the instantaneous frequency of the modulated signal to change, while fiber dispersion amplifies this frequency change, which is manifested as a gain behavior in millimeter-wave signals. To accurately model this complex dynamic process, the present invention proposes a gain behavior model based on BiLSTM. BiLSTM can effectively solve the gradient problem and capture long-distance dependencies by combining forward and backward information to process sequential data. The present invention uses the dataset collected through experiments to train the BiLSTM model so that it can accurately predict key indicators such as the carrier power, upper and lower sideband powers, and second harmonic distortion (HD2) power of millimeter-wave signals. Compared with traditional feedforward neural networks (FFNNs), time-delay neural networks (TDNNs), recurrent neural networks (RNNs), and unidirectional long short-term memory networks (LSTMs), the BiLSTM model exhibits better performance on the training set, validation set, and test set, and has stronger generalization ability and robustness.

[0085] (3) To further optimize the SNR and distortion level of millimeter-wave signals, the present invention combines the ACO algorithm with the BiLSTM model. Compared with traditional gradient descent algorithms (such as Adam), the ACO algorithm has a powerful global search ability by simulating the behavior of ants releasing pheromones and can effectively avoid local optimal solutions. In addition, the present invention designs an optimization function that can flexibly generate millimeter-wave signals that meet requirements according to the needs of different communication scenarios by adjusting the optimization weights. The optimization function comprehensively considers indicators such as carrier power, upper and lower sideband powers, and second harmonic distortion power, and can flexibly control the optimization result by adjusting the optimization weights to meet the requirements of SNR and distortion in different communication scenarios.

[0086] (4) By combining the BiLSTM model and the ACO algorithm, the present invention can not only accurately predict the gain behavior of millimeter-wave signals, but also flexibly control the optimization result by adjusting the optimization weights according to the needs of different communication scenarios. This joint optimization strategy provides theoretical guidance for the design and optimization of C-RAN transmitters and has broad application prospects, especially in the development of low-cost, high-efficiency, and high-capacity wireless communication networks. By adjusting the optimization weights, millimeter-wave signals that meet different SNR and distortion requirements can be flexibly generated, thus providing an efficient, flexible, and reliable solution for future wireless communication networks.

[0087] This embodiment also provides a cloud radio access network millimeter-wave signal optimization system based on machine learning, including:

[0088] A signal generation module for generating millimeter-wave signals with high side-mode suppression ratio and low phase noise through a dual-loop optoelectronic oscillator;

[0089] A prediction module, configured to input input features related to millimeter-wave signal characteristics into a pre-trained bidirectional long short-term memory network model to predict the power index of the millimeter-wave signal;

[0090] A parameter optimization module, configured to optimize the power index based on the ant colony optimization algorithm by optimizing the objective function, and adjust the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain an optimized millimeter-wave signal.

[0091] This embodiment also provides a computer device, including a memory, a processor, and a computer program stored on the memory, where the processor executes the computer program to implement the steps of the method.

[0092] This embodiment also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method are implemented.

[0093] This embodiment also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method are implemented.

[0094] The above is only a preferred specific embodiment of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A millimeter-wave signal optimization method for cloud radio access network based on machine learning, characterized in that, It includes the following steps: Generate a millimeter-wave signal with a high side-mode suppression ratio and low phase noise through a dual-ring optoelectronic oscillator; Input the input features related to the characteristics of the millimeter-wave signal into a pre-trained bidirectional long short-term memory network model to obtain the power index of the millimeter-wave signal; Based on the ant colony optimization algorithm, optimize the power index through the optimization objective function, and adjust the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain an optimized millimeter-wave signal.

2. The method according to claim 1, characterized in that, The calculation expression for the effective mode spacing of the dual-ring optoelectronic oscillator is: where l1 and l2 are the lengths of the two single-mode fiber loops respectively, c is the speed of light, and n f is the refractive index of the optical fiber.

3. The method according to claim 1, wherein The input features include the optical fiber link length, the amplitude and frequency of the modulated intermediate-frequency signal; the power index includes the carrier power, the upper and lower sideband powers, and the second harmonic distortion power.

4. The method according to claim 1, wherein The calculation expression for predicting the power index value by the pre-trained bidirectional long short-term memory network model is: where, w out is the weight matrix of the output layer, b out is the bias matrix of the output layer, is the bidirectional hidden state.

5. The method according to claim 1, wherein The optimization objective function is as follows: where α i represents the optimization weight, represents the carrier power, represents the lower sideband power of the IF component, represents the upper sideband power of the IF component, represents the lower sideband power of the second harmonic distortion, represents the upper sideband power of HD2.

6. The method according to claim 1, characterized in that, The transfer probability expression for an ant to transfer from node i to node j in the ant colony optimization algorithm is: where τ ij is the pheromone concentration from node i to node j, η ij is the heuristic information, which is related to the heuristic solution of the problem. α and β respectively control the importance of pheromone and heuristic information, and allowed is the set of nodes that the ant can transfer to when at node i.

7. A millimeter-wave signal optimization system for cloud radio access network based on machine learning, characterized in that, It includes: A signal generation module for generating a millimeter-wave signal with a high side-mode suppression ratio and low phase noise through a dual-ring optoelectronic oscillator; A prediction module for inputting the input features related to the characteristics of the millimeter-wave signal into a pre-trained bidirectional long short-term memory network model to predict the power index of the millimeter-wave signal; A parameter optimization module for optimizing the power index through the optimization objective function based on the ant colony optimization algorithm, and adjusting the input parameters of the remote cloud radio access network transmitter through the optimized power index to obtain an optimized millimeter-wave signal.

8. A computer device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1-6.

Images