A cloud radio access network modeling method based on machine learning

Abstract

This invention relates to a machine learning-based modeling method for cloud wireless access networks, comprising: acquiring a remote cloud wireless access network transmitter; improving the remote cloud wireless access network transmitter by replacing the traditional microwave source with a dual-ring opto-oscillator to suppress side modes through single-mode oscillation; and introducing a directly modulated laser after an arbitrary waveform generator to utilize the chirp effect and fiber dispersion to generate power gain; and modeling the millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter based on a lightweight gradient booster model to match the dimensions of input and output features. This invention uses a dual-ring opto-oscillator to replace the traditional microwave source, reducing phase noise; utilizes the synergistic gain behavior of the directly modulated laser chirp and fiber dispersion to improve the transmission gain of the millimeter-wave signal; and models the spectral characteristics of the millimeter-wave signal using a lightweight gradient booster model to achieve spectrum prediction.

Description

Technical Field

[0001] This invention relates to the field of resource allocation technology, and in particular to a cloud wireless access network modeling method based on machine learning. Background Technology

[0002] Cloud-based wireless access networks (BRANs) achieve flexible resource allocation and efficient management by separating centralized baseband processing units from distributed remote wireless heads, making them one of the core architectures of next-generation communication systems. In BRANs, directly modulated lasers are widely used in optical fronthaul links due to their low cost and ease of deployment; while optical wireless technology transmits radio frequency signals through optical fibers, supporting high-speed communication in the millimeter-wave band.

[0003] However, performance optimization of cloud wireless access network systems faces two key challenges: on the one hand, traditional microwave sources suffer from high phase noise, which limits the stability of millimeter-wave signals; on the other hand, the synergistic effect of chirp in directly modulated lasers and fiber dispersion produces complex gain behavior, accompanied by high-order harmonic distortion, and traditional analytical models struggle to accurately capture the interaction of these nonlinear, dynamic characteristics and dispersion effects.

[0004] Specifically, the existing modeling methods have the following limitations: (1) The analytical models proposed by the existing technical solutions rely on complex mathematical derivations, which are insufficient in modeling the coupling effect of nonlinearity of direct modulation lasers, dynamic characteristics of photoelectric oscillators and dispersion effects, and have high computational complexity; (2) When dealing with the problem of mismatch between the dimensions of input and output, traditional machine learning models are prone to underfitting or unstable prediction, and have long training time and limited generalization ability. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention aims to provide a machine learning-based cloud wireless access network modeling method. This method utilizes a data-driven modeling framework based on a lightweight gradient booster, combined with dual-ring opto-oscillator technology, to achieve efficient and high-precision modeling of the spectral characteristics of millimeter-wave signals in a remote cloud wireless access network architecture. The dual-ring opto-oscillator achieves low phase noise and high side-mode rejection ratio through a phase-matching mechanism with dual fiber loops. The lightweight gradient booster, through automatic learning of feature interactions, resolves the input-output dimension mismatch problem while preserving complete spectral information, thus improving modeling efficiency and accuracy.

[0006] To achieve the above objectives, the present invention provides the following solution:

[0007] A machine learning-based cloud wireless access network modeling method includes:

[0008] Obtain a remote cloud wireless access network transmitter and improve the remote cloud wireless access network transmitter:

[0009] A dual-ring opto-oscillator is used to replace the traditional microwave source of the remote cloud wireless access network transmitter. This is used to suppress side modes through single-mode oscillation, and a direct-modulation laser is introduced after the arbitrary waveform generator to generate power gain by utilizing the chirp effect and fiber dispersion to achieve synergistic gain.

[0010] The millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter is modeled based on a lightweight gradient booster model, matching the dimensions of input and output features.

[0011] Optionally, suppressing side modes through the single-mode oscillation includes:

[0012] The dual-ring optoelectronic oscillator uses a long fiber loop and a short fiber loop for single-mode oscillation and suppresses side modes:

[0013]

[0014] Among them, f osc τ is the oscillation frequency, k and m are integers, and τ is the oscillation frequency. l For long loop delay, τ s This is for short loop delay.

[0015] Optionally, the optical field expression of the directly modulated laser is:

[0016]

[0017] Where, m a ω is the amplitude modulation index. IF ω is the angular frequency of the intermediate frequency signal. o The laser carrier frequency angular frequency, E represents the maximum phase shift caused by intermediate frequency modulation. DML (t) represents the optical field of the directly modulated laser, and t represents time.

[0018] Optionally, the Mach-Zehnder modulator of the improved remote cloud wireless access network transmitter operates in carrier suppression mode to enhance sideband power, and the optical signal is transmitted through single-mode fiber, with loss compensated by erbium-doped fiber amplifier, and the optical signal is converted into millimeter-wave electrical signal by photodetector at the remote wireless head.

[0019] The electric field expression of the photodetector at the remote wireless head is:

[0020]

[0021] Among them, P o For laser output power, The DC bias phase shift of the Mach-Zehnder modulator is given, β is the fiber dispersion parameter, and L is the length of the single-mode fiber (m). RF m is the radio frequency modulation index.+ m - These are the upper and lower sideband modulation indices, E R (t) represents the electric field of the photodetector at the remote wireless head, e is the natural constant, and ω IF ω is the angular modulation frequency of the intermediate frequency signal. osc ω is the angular modulation frequency of the microwave signal.

[0022] Optionally, the upper and lower sideband modulation indices include:

[0023]

[0024] Where α is the linewidth enhancement factor, κ is the adiabatic chirp parameter, and m a The amplitude modulation index j is an imaginary unit.

[0025] Optionally, the dimensions for matching the input features with the output features include:

[0026] Collect input and output features, and construct a training set using the input and output features;

[0027] Training the lightweight gradient boosting machine model using the training set includes:

[0028] The initial predicted value of each output dimension is used as a constant to minimize the loss function, and the first and second derivatives of the loss function are calculated using each training sample in the training set.

[0029] The first derivative is used as the gradient, and the first target number of high gradient sample sets are retained under the absolute value of the gradient. The second target number of low gradient sample sets are randomly selected from the remaining samples, and the gradient of the low gradient samples is further compensated.

[0030] Histogram aggregation and feature bundling are used to accelerate splitting and construct a decision tree. During the construction of the decision tree, splitting is stopped when the splitting gain is lower than the first target value or the sum of the second derivatives of the child nodes is less than the second target value, indicating that the training is complete, thus obtaining the trained lightweight gradient boosting machine model.

[0031] The millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter is modeled using a trained lightweight gradient booster model, matching the input and output feature dimensions.

[0032] Optionally, the loss function includes:

[0033]

[0034] Where l(·) is the mean squared error loss function, y ij For real spectrum labels, This is the model's predicted output, where u is the number of spectrum sampling points.

[0035] Optionally, calculating the first and second derivatives of the loss function includes:

[0036]

[0037] Among them, g i h is the first derivative. i It is the second derivative. Let y be the predicted value for the (m-1)th iteration. i For real labels, This is the sign for a partial derivative.

[0038] Optionally, the splitting gain includes:

[0039]

[0040] in, For split gain, I L I R Let I be the sample set of the left and right child nodes after the split, I be the sample set of the current node, and λ be the L2 regularization coefficient.

[0041] The beneficial effects of this invention are as follows:

[0042] This invention employs a dual-ring optoelectronic oscillator to replace the traditional microwave source, significantly reducing phase noise. The dual-ring optoelectronic oscillator achieves single-mode oscillation and suppresses side modes through a phase-matching mechanism using two fiber loops of varying lengths. Compared to a traditional microwave source, the dual-ring optoelectronic oscillator achieves a 27 dBc / Hz reduction in phase noise at a 10 kHz frequency offset.

[0043] This invention utilizes the synergistic gain behavior of direct-modulated laser chirp and fiber dispersion to improve millimeter-wave signal transmission performance. The synergistic effect of the chirp effect of the direct-modulated laser and the fiber dispersion produces a significant power gain.

[0044] This invention proposes a millimeter-wave signal spectrum modeling framework based on a lightweight gradient booster, addressing the input-output dimension mismatch problem. The lightweight gradient booster automatically learns the complex interaction between intermediate frequency (IF), IF amplitude, and single-mode fiber length and spectral profile through techniques such as leaf-wise growth of gradient boosting decision trees, histogram-accelerated splitting, and gradient-guided sampling. It preserves complete spectral information without requiring manual feature expansion. Compared to traditional models, this framework significantly improves training efficiency while maintaining high accuracy.

[0045] The lightweight gradient booster model of this invention exhibits excellent accuracy, robustness, and efficiency. The average absolute error of the lightweight gradient booster model for the intermediate frequency (IF) sideband does not exceed 1.70 dB, and the average absolute error for the second harmonic distortion (HHDI) sideband does not exceed 0.81 dB, outperforming models such as feedforward neural networks, recurrent neural networks, and long short-term memory networks. Furthermore, the average absolute error fluctuates minimally when the IF frequency, amplitude, and single-mode fiber length change. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0047] Figure 1 This is a schematic diagram of a cloud wireless access network modeling method based on machine learning according to an embodiment of the present invention;

[0048] Figure 2 A comparison of the phase noise of the dual-ring optoelectronic oscillator in this embodiment of the invention with that of a conventional microwave source;

[0049] Figure 3 The following are comparisons of side-mode suppression between a dual-loop photoelectric oscillator and a single-loop photoelectric oscillator in embodiments of the present invention: (a) a comparison of side-mode suppression between the 2km long loop in the dual-loop photoelectric oscillator and the single-loop photoelectric oscillator; (b) a comparison of side-mode suppression between the 170m short loop in the dual-loop photoelectric oscillator and the single-loop photoelectric oscillator.

[0050] Figure 4 A comparison of the spectral curves of millimeter-wave signals modulated with a 60MHz intermediate frequency signal under 5km and 35km single-mode fiber links in embodiments of the present invention;

[0051] Figure 5 This is the training structure of the lightweight gradient boosting machine model in an embodiment of the present invention;

[0052] Figure 6 A comparison of convergence performance during model training and validation in this embodiment of the invention; (a) convergence curve of the training set, (b) convergence curve of the validation set;

[0053] Figure 7 In this embodiment of the invention, at an intermediate frequency of 60MHz and a voltage of 440mV... pp Comparison of predicted and actual spectrum curves under mid-frequency amplitude and 35km single-mode fiber link length conditions;

[0054] Figure 8This is a comparison of the mean absolute error of the upper sideband spectrum curves on the test set in this embodiment of the invention;

[0055] Figure 9 This is a comparison of the computation time for the model training and testing processes in an embodiment of the present invention. Detailed Implementation

[0056] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0058] like Figure 1 As shown, this embodiment discloses a cloud wireless access network modeling method based on machine learning, including: acquiring a remote cloud wireless access network transmitter; improving the remote cloud wireless access network transmitter by replacing the traditional microwave source of the remote cloud wireless access network transmitter with a dual-ring opto-oscillator to suppress side modes through single-mode oscillation; and introducing a direct-modulation laser after the arbitrary waveform generator to generate power gain by utilizing the chirp effect and fiber dispersion to achieve synergistic gain. The millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter is modeled based on a lightweight gradient booster model to match the dimensions of input and output features.

[0059] Figure 1This diagram illustrates a remote cloud wireless access network architecture assisted by a directly modulated laser and optoelectronic oscillator. The network consists of: CO (Central Office), ODN (Optical Distribution Network), RRH (Remote Radio Head), DML (Directly Modulated Laser), AWG (Arbitrary Waveform Generator), PC (Polarization Controller), MZM (Mach-Zehnder Modulator), OC1 (First Optical Coupler), OC2 (Second Optical Coupler), OC3 (Third Optical Coupler), SMF (Single-Mode Fiber), PD1 (First Photodetector), PD2 (Second Photodetector), PD3 (Third Photodetector), LNA (Low-Noise Amplifier), BPF (Bandpass Filter), PA1 (First Power Amplifier), PA2 (Second Power Amplifier), PNA (Phase Noise Analyzer), EDFA (Erbium-Doped Fiber Amplifier), OSA (Optical Spectrum Analyzer), and SA (Spectrum Analyzer). The optical distribution network transmits signals from the central office to end users. The arbitrary waveform generator generates intermediate frequency (IF) waveforms, which are then used to modulate the directly modulated laser. The polarization controller adjusts the polarization state of the optical signal to ensure stability and compatibility throughout the transmission process. The bias voltage of the Mach-Zehnder modulator is controlled to achieve carrier suppression, thereby obtaining a frequency-doubled radio frequency signal at the third photodetector. The first and second photodetectors convert the recirculating optical signal into an electrical signal. A low-noise amplifier amplifies the signal and minimizes noise introduction to maintain signal quality. A bandpass filter ensures signal integrity by filtering out unwanted frequency components. The first power amplifier further provides gain to the oscillation loop. A phase noise analyzer is used to observe phase noise. An erbium-doped fiber amplifier amplifies the optical signal to compensate for any losses incurred during fiber transmission. The third photodetector in the remote RF head performs photoelectric signal conversion for spectral analysis. The second power amplifier further amplifies the millimeter-wave signal to meet the requirements of transmission distance and signal strength. A spectrum analyzer is used to analyze the spectral characteristics of the millimeter-wave signal, facilitating monitoring and optimization of signal transmission quality. A spectrum analyzer is used to observe the spectrum and carrier suppression status.

[0060] Furthermore, the suppression of side modes through single-mode oscillation includes: the dual-ring optoelectronic oscillator uses a long fiber loop and a short fiber loop for single-mode oscillation and suppresses side modes.

[0061] Specifically, the core of the remote cloud wireless access network transmitter architecture of the present invention includes a central office, an optical distribution network, and a remote wireless head. The key components and designs are as follows:

[0062] Dual-ring optoelectronic oscillator: Single-mode oscillation is achieved using a long fiber loop and a short fiber loop, as shown in the following formula:

[0063]

[0064] Among them, f osc τ is the oscillation frequency, k and m are integers, and τ is the oscillation frequency.l For long loop delay, τ s The short loop determines the mode spacing, increasing it to simplify mode competition; the long loop dominates phase noise suppression, utilizing the high Q-value of the fiber to reduce phase noise. Experiments show that this structure reduces phase noise by 27 dBc / Hz at a 10 kHz frequency offset compared to a traditional microwave source, and improves side-mode suppression by 40.83 dB and 12.66 dB compared to long-loop and short-loop opto-oscillators, respectively.

[0065] Furthermore, the improved remote cloud wireless access network transmitter's Mach-Zehnder modulator operates in carrier suppression mode, enhancing sideband power. The optical signal is transmitted via single-mode fiber, with losses compensated by an erbium-doped fiber amplifier, and the optical signal is converted into a millimeter-wave electrical signal by a photodetector at the remote wireless head.

[0066] Specifically, the signal generation and transmission link is as follows: an arbitrary waveform generator produces an intermediate frequency signal, which drives a direct-modulation laser to generate an optical modulation signal. The optical field expression of the direct-modulation laser is:

[0067]

[0068] Where, m a ω is the amplitude modulation index. IF ω is the angular frequency of the intermediate frequency signal. o The laser carrier frequency angular frequency, This represents the maximum phase shift caused by intermediate frequency modulation. The Mach-Zehnder modulator operates in carrier-suppressed mode to enhance sideband power; the optical signal is transmitted through single-mode fiber, with losses compensated by an erbium-doped fiber amplifier; a photodetector at the remote wireless head converts the optical signal into a millimeter-wave electrical signal. The electric field expression of the photodetector in the remote architecture is:

[0069]

[0070] Among them, P o For laser output power, β is the DC bias phase shift of the Mach-Zehnder modulator (set to π / 2 for carrier suppression), β is the fiber dispersion parameter, and L is the length of the single-mode fiber (m). RF m is the radio frequency modulation index. + m - These are the upper and lower sideband modulation indices, respectively, and their expressions are:

[0071]

[0072] Where α is the linewidth enhancement factor and κ is the adiabatic chirp parameter. The signal is amplified by a power amplifier and then monitored by a spectrum analyzer.

[0073] Further, matching the input and output feature dimensions includes: collecting input and output features, and constructing a training set using the input and output features; training the lightweight gradient booster model using the training set includes: using the initial predicted value of each output dimension as a constant to minimize the loss function, and calculating the first and second derivatives of the loss function using each training sample in the training set; using the first derivative as the gradient, and retaining the first target number of high gradient samples under the absolute value of the gradient, randomly selecting the second target number of low gradient samples from the remaining samples, and further compensating the gradient of the low gradient samples; using histogram aggregation and feature bundling to accelerate splitting and constructing a decision tree; during the construction of the decision tree, when the splitting gain is lower than the first target value or the sum of the second derivatives of the child nodes is less than the second target value, the splitting is stopped, resulting in the trained lightweight gradient booster model; using the trained lightweight gradient booster model to model the millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter, matching the input and output feature dimensions.

[0074] Specifically, millimeter-wave signal spectrum modeling based on lightweight gradient boosters, such as... Figure 5 As shown:

[0075] This invention employs a lightweight gradient booster model to model the spectral profile of millimeter-wave signals. The specific process is as follows:

[0076] Input and output characteristics definition: Input characteristics: Intermediate frequency signal frequency 11-60MHz, intermediate frequency signal amplitude 210-500mV pp Single-mode fiber optic links are 5-35 km long and organized as an input matrix:

[0077] x i =[F i A i ,L i ] T i = 1, 2, ..., n (6)

[0078] Among them, F i Let A be the intermediate frequency of the i-th sample group. i Let L be the mid-frequency amplitude of the i-th sample group. i Let be the fiber length of the i-th sample group.

[0079] Output characteristics: The millimeter-wave spectrum profile acquired by the spectrum analyzer, including sampling points of the intermediate frequency upper and lower sidebands, second harmonic distortion upper and lower sidebands, and background noise, is organized into an output matrix:

[0080] y ij =[s i1 ,s i2 ,…,s ij ] Ti=1,2,…,n; j=1,2,…,u (7)

[0081] Among them, s ij Let be the j-th spectral sampling point of the i-th sample group, and u be the total number of sampling points.

[0082] Dataset construction: 10,500 samples were collected by a spectrum analyzer and divided into training set, validation set and test set in a ratio of 7:2:1.

[0083] Lightweight Gradient Boosting Machine Model Training:

[0084] Initialization: For each output dimension, the initial prediction value is a constant that minimizes the loss function, which uses the mean squared error.

[0085]

[0086] The formula for calculating the initial predicted value is:

[0087]

[0088] Where γ is the constant to be optimized, and l(·) is the mean square error loss function mentioned above.

[0089] Hyperparameter settings: Boost iteration count M = 100, maximum leaf node count K = 20, learning rate η = 0.1, gradient-guided sampling ratios a = 0.2, b = 0.1, L2 regularization coefficient λ = 0.5, minimum split gain γ min =0.1, minimum Hessian and H min =1.0.

[0090] Improve and iterate:

[0091] Gradient and Hessian calculations: For each sample, calculate the first derivative (gradient) and second derivative (Hessian) of the loss function:

[0092]

[0093] in, This is the predicted value for the (m-1)th iteration.

[0094] Sample Sampling: Sample selection is achieved through a gradient-guided sampling strategy: A set of high-gradient samples A with the highest absolute gradient value a|D| is retained, and a set of low-gradient samples R with the lowest absolute gradient value b|D| is randomly selected from the remaining samples. The gradients of these low-gradient samples are then compensated.

[0095]

[0096] Among them, g r Let g' be the scaled gradient value of sample r. rThe original gradient value of sample r. The compensation coefficient is S, and the final iterative sample set is S. m =a∪R, where r is a sample from a random subset. Let a be a random subset, a be the proportion of high-gradient samples retained, b be the sampling proportion of the random subset, and a be the high-gradient sample subset.

[0097] Decision tree construction: Histogram aggregation and feature bundling are used to accelerate splitting. The formula for calculating the histogram of the feature bundle is:

[0098] hist b =∑ f∈b hist f (13)

[0099] Among them, hist f Let f be the histogram of a single feature f, and b be the feature bundle. The formula for calculating the split gain is:

[0100]

[0101] Where I is the sample set of the current node, I L I R This is the sample set of the left and right child nodes after the split.

[0102] Leaf node value calculation: when the splitting gain is lower than γ min Or child nodes Hessian and less than H min When the split stops, the leaf node value v is calculated using the following formula:

[0103]

[0104] Prediction Update: After each iteration, the formula for updating the predicted value is:

[0105]

[0106] Among them, v (m) (x i ) represents the m-th decision tree for sample x i The assigned leaf node values, where η is the learning rate.

[0107] Model performance validation:

[0108] (1) Accuracy verification: On the test set, the average absolute error of the lightweight gradient booster to the mid-frequency sideband is 0.98-1.30dB and the average absolute error to the second harmonic distortion sideband is 0.75-1.55dB, which is significantly better than models such as feedforward neural networks.

[0109] (2) Robustness verification: When the intermediate frequency, amplitude and fiber length change, the average absolute error fluctuation of the lightweight gradient booster is 0.32dB, 0.29dB and 0.80dB respectively, which is better than the comparison model.

[0110] (3) Efficiency verification: The training time is 26.09 seconds, which is much faster than bidirectional long short-term memory networks, extreme gradient boosting and other models; the test time is 447 milliseconds, which meets the actual deployment requirements.

[0111] This invention uses a dual-ring optoelectronic oscillator to replace the traditional microwave source, significantly reducing phase noise. The dual-ring optoelectronic oscillator achieves single-mode oscillation and suppresses side modes through a phase matching mechanism of two fiber loops of different lengths.

[0112] like Figure 2 As shown, compared with traditional microwave sources, the dual-ring opto-oscillator achieves a phase noise reduction of 27 dBc / Hz at a frequency offset of 10 kHz.

[0113] This invention significantly outperforms single-loop structures in terms of side-mode suppression performance, such as... Figure 3 As shown in (a)-(b), the side-mode rejection ratio is improved by 40.83 dB compared to the single-loop opto-oscillator in the 2km long-loop configuration; and by 12.66 dB in the 170m short-loop configuration.

[0114] This invention utilizes the synergistic gain behavior of direct-modulated laser chirp and fiber dispersion to improve millimeter-wave signal transmission performance. The synergistic effect of the chirp effect of the direct-modulated laser and the fiber dispersion produces a significant power gain. For example... Figure 4 As shown, the millimeter-wave signal gain under 35km single-mode fiber transmission is increased by 19dB compared to 5km, which is better than the attenuation characteristics of the local architecture.

[0115] This invention proposes Figure 5 The millimeter-wave signal spectrum modeling framework based on a lightweight gradient booster, as shown, addresses the input-output dimension mismatch problem. The lightweight gradient booster automatically learns the complex interaction between intermediate frequency (IF), IF amplitude, and single-mode fiber length and spectral profile through techniques such as leaf-wise growth of gradient boosting decision trees, histogram-accelerated splitting, and gradient-guided sampling. It preserves complete spectral information without requiring manual feature expansion. Compared to traditional models, this framework significantly improves training efficiency while maintaining high accuracy.

[0116] The lightweight gradient booster model used in this invention exhibits excellent and stable convergence performance during training. For example... Figure 6As shown in (a)-(b), the model loss decreases rapidly and stabilizes with increasing iterations on both the training and validation sets, without overfitting, demonstrating the effectiveness of the training strategy and the model's generalization ability.

[0117] This invention has high-precision prediction capabilities, such as... Figure 7 As shown, under the conditions of 60MHz intermediate frequency, 440mVpp intermediate frequency amplitude and 35km single-mode fiber link length, the spectrum curve predicted by the lightweight gradient booster model is in high agreement with the actual measured spectrum curve, and the effect is significantly better than that of the feedforward neural network.

[0118] The lightweight gradient booster model of this invention exhibits excellent accuracy, robustness, and efficiency. For example... Figure 8 As shown, the model's average absolute error for the intermediate frequency sideband does not exceed 1.70 dB, and its average absolute error for the second harmonic distortion sideband does not exceed 0.81 dB, which is better than models such as feedforward neural networks, recurrent neural networks, and long short-term memory networks. The average absolute error fluctuates little when the intermediate frequency, amplitude, and single-mode fiber length change.

[0119] The lightweight gradient boosting machine modeling framework proposed in this invention has advantages in terms of efficiency, such as... Figure 9 As shown, the model training time is only 26.09 seconds and the testing time is only 447 milliseconds. Its overall time consumption is lower than that of other traditional machine learning models and neural network models, which meets the requirements for modeling efficiency in practical engineering applications.

[0120] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for modeling a cloud radio access network based on machine learning, characterized in that, include: Obtain a remote cloud wireless access network transmitter and improve the remote cloud wireless access network transmitter: A dual-ring opto-oscillator is used to replace the traditional microwave source of the remote cloud wireless access network transmitter. This is used to suppress side modes through single-mode oscillation, and a direct-modulation laser is introduced after the arbitrary waveform generator to generate power gain by utilizing the chirp effect and fiber dispersion to achieve synergistic gain. The millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter is modeled based on a lightweight gradient booster model, matching the dimensions of input and output features. The improved remote cloud wireless access network transmitter's Mach-Zehnder modulator operates in carrier suppression mode, enhancing sideband power. The optical signal is transmitted via single-mode fiber, with losses compensated by an erbium-doped fiber amplifier, and the optical signal is converted into a millimeter-wave electrical signal by a photodetector at the remote wireless head. The electric field expression of the photodetector at the remote wireless head is: in, For laser output power, DC bias phase shift for Mach-Zehnder modulators For fiber dispersion parameters, L This refers to the length of a single-mode fiber. The radio frequency modulation index. , These are the upper and lower sideband modulation indices, respectively. The electric field of the photodetector at the remote wireless head. It is a natural constant. This is the angular modulation frequency of the intermediate frequency signal. The angular modulation frequency of the microwave signal; The dimensions for matching the input features and output features include: Collect input and output features, and construct a training set using the input and output features; Training the lightweight gradient boosting machine model using the training set includes: The initial predicted value of each output dimension is used as a constant to minimize the loss function, and the first and second derivatives of the loss function are calculated using each training sample in the training set. The first derivative is used as the gradient, and the first target number of high gradient sample sets are retained under the absolute value of the gradient. The second target number of low gradient sample sets are randomly selected from the remaining samples, and the gradient of the low gradient samples is further compensated. Histogram aggregation and feature bundling are used to accelerate splitting and construct a decision tree. During the construction of the decision tree, splitting is stopped when the splitting gain is lower than the first target value or the sum of the second derivatives of the child nodes is less than the second target value, indicating that the training is complete, thus obtaining the trained lightweight gradient boosting machine model. The millimeter-wave signal spectrum of the improved remote cloud wireless access network transmitter is modeled using a trained lightweight gradient booster model, matching the input and output feature dimensions.

2. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, Suppressing side modes through the single-mode oscillation includes: The dual-ring optoelectronic oscillator uses a long fiber loop and a short fiber loop for single-mode oscillation and suppresses side modes: in, The oscillation frequency is... k , m It is an integer. For long loop delay, This is for short loop delay.

3. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, The optical field expression of the directly modulated laser is: in, For amplitude modulation index, The angular frequency of the intermediate frequency signal. The laser carrier frequency angular frequency, The maximum phase shift caused by intermediate frequency modulation. To directly modulate the optical field of the laser, For time.

4. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, The upper and lower sideband modulation indices include: in, Linewidth enhancement factor For adiabatic chirp parameters, For amplitude index It is the imaginary unit.

5. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, The loss function includes: in, Let the mean squared error loss function be . These are real spectrum labels. For the model's predicted output, This represents the number of spectrum sampling points.

6. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, Calculating the first and second derivatives of the loss function includes: in, The first derivative, It is the second derivative. For the first The predicted value of the next iteration. For real labels, This is the sign for a partial derivative.

7. The cloud wireless access network modeling method based on machine learning according to claim 1, characterized in that, The splitting gain includes: in, For split gain, , This is the sample set of the left and right child nodes after the split. For the current node's sample set, is the L2 regularization coefficient.

Images