5G signal module network performance adaptive optimization method based on dynamic link quality evaluation

Abstract

The invention relates to the technical field of adaptive optimization, and discloses a 5G signal module network performance adaptive optimization method based on dynamic link quality assessment, comprising the following steps: step S1, constructing a multi-dimensional link quality index model, collecting channel state information, end-to-end delay, packet loss rate and instantaneous rate change value of the 5G module in different network environments in real time; s2, reinforcement learning strategy generation: inputting the link quality matrix into a deep Q network, and taking a network access mode, a spectrum selection strategy and a physical resource block allocation proportion as an action space; and S3, dynamic network performance adjustment and optimization. A deep Q network is utilized to generate a reinforcement learning strategy, network parameters are dynamically adjusted according to a real-time link quality evaluation result, and the self-adaptive adjustment mechanism can enable the network to quickly respond and optimize the performance in different environments.

Description

Technical Field

[0001] The present invention relates to the field of adaptive optimization technology, and specifically to a 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation. Background Art

[0002] With the widespread application of 5G technology in areas such as augmented reality, industrial IoT, and autonomous driving, users are increasingly demanding high network reliability, low latency, and adaptability to dynamic environments. Traditional network optimization relies on fixed thresholds, such as the Qin / Qout thresholds of radio link monitoring (RLM), or periodic manual adjustments to parameters such as base station antenna tilt and power control. These methods are unable to perceive real-time channel state changes, leading to problems such as handover delays and spikes in bit error rates in dynamic environments such as high-speed mobile and densely populated urban areas.

[0003] After searching, the Chinese patent number CN118102318B discloses a data transmission system based on 5G technology, which relates to the field of wireless communication network technology, including: a dynamic spectrum resource allocation module, a data reordering and encryption module, and an intelligent transmission control center; the dynamic spectrum resource allocation module adopts a spectrum dynamic allocation algorithm optimized for 5G networks. The algorithm can adaptively manage and allocate spectrum resources based on the real-time load of the network, user needs and spectrum usage status. The data reordering and encryption module can intelligently adjust the processing order of the data packet according to its priority, size and destination. In addition, the module also adopts adaptive encryption technology. The intelligent transmission control center is the core of the system and is responsible for comprehensively evaluating the operating results of the dynamic spectrum resource allocation module and the data reordering and encryption module. Based on the results of these comprehensive evaluations, the intelligent transmission control center can take targeted adjustments, significantly improving the performance and user experience of the 5G network.

[0004] The above-mentioned data transmission system mainly includes a dynamic spectrum resource allocation module, a data reordering and encryption module, and an intelligent transmission control center. Although the system has certain optimizations in spectrum resource management and data packet processing, the core module of this patent relies on static formulas to calculate network status scores. Among them, ULD, SPD and DMD are all based on macro network indicators, such as spectrum utilization rate and service type weight calculation. However, these indicators do not include real-time link quality parameters such as channel state information, delay, packet loss rate, rate change, etc., which makes the system unable to capture dynamic changes in the physical layer or signal level, making the system performance limited in complex network environments and unable to effectively optimize throughput and connection stability. Based on this, the present invention designs a 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation to solve the above problems. Summary of the Invention

[0005] The purpose of the present invention is to provide a 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation, which solves the problem in the background technology that the dynamic changes of the physical layer or signal layer cannot be captured.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] A method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation includes the following steps:

[0008] Step S1: Build a multi-dimensional link quality indicator model to collect channel state information, end-to-end delay, packet loss rate, and instantaneous rate change values ​​of the 5G module in real time under different network environments, and generate a dynamic link quality matrix: Qlink = [CSI, Delay, Loss, ΔRate], with a sampling period of ≤ 100ms.

[0009] Step S2: Reinforcement learning strategy generation. The link quality matrix is ​​input into the deep Q network, with the network access mode, spectrum selection strategy, and physical resource block allocation ratio as the action space. The optimal parameter configuration solution is output through iterative optimization of the reward function: R = 0.7 × Throughput - 0.3 × Delay.

[0010] Step S3, dynamic network performance tuning, dynamically adjusts the following parameters according to the generated solution: switching network access mode; selecting Sub-6GHz or millimeter wave spectrum; adjusting the modulation and coding scheme level and PRB allocation ratio.

[0011] Preferably, the action space definition satisfies: the spectrum selection strategy includes Sub-6GHz and millimeter wave; the physical resource block allocation ratio ranges from 10% to 90%; when the interference-to-noise ratio is detected to be less than 10dB, the millimeter wave spectrum switching action is forcibly triggered.

[0012] Preferably, the dynamic adjustment includes the following rules: Rule S1, when the packet loss rate is greater than 5% and the delay is greater than 20ms, increase the physical resource block allocation ratio by 15% and improve the modulation and coding scheme level; Rule S2, when the instantaneous rate change value fluctuates by more than 30%, switch to the independent networking mode; Rule S3, in an environment where the mobile speed is greater than or equal to 60km / h, give priority to millimeter wave spectrum and increase the physical resource block allocation ratio; Rule S4, in an environment where the multipath fading is greater than or equal to 15dB, reduce the modulation and coding scheme level and switch to the independent networking mode.

[0013] Preferably, the training process of the deep Q network includes: step s1, using a generative adversarial network to simulate a high-interference scenario and inject adversarial training data; step s2, building a dual-network architecture, including a main network and a target network, and synchronizing the main network parameters to the target network every 1000 iterations; step s3, setting the experience replay pool capacity to 10,000 groups of state and action transfer samples, and extracting 128 groups of samples in batches for training each time.

[0014] Preferably, the generation of the dynamic link quality matrix in step S1 includes data preprocessing: step s1, performing Kalman filtering smoothing on the channel state information; step s2, using sliding window mean filtering on the end-to-end delay data; step s3, performing a first-order difference calculation ΔRate=Ratet-Ratet-1 on the instantaneous rate change value.

[0015] Preferably, the reward function further introduces a packet loss rate penalty factor: the updated reward function is: R=0.6×Throughput-0.25×Delay-0.15×Loss.

[0016] Preferably, the modulation and coding scheme level is adjusted in step S3 using a gradient strategy: when the channel quality indicator CQI increases by ≥2 levels, the modulation and coding scheme level is increased by 1 level; when the CQI decreases by ≥3 levels for three consecutive sampling periods, the modulation and coding scheme level is decreased by 2 levels.

[0017] Preferably, the generation of adversarial training data in step s1 includes: simulating a high-scattering environment with a multipath delay spread of ≥50ns; injecting pulse noise interference to cause a sudden drop in the signal-to-noise ratio of ≥15dB; and constructing a high-speed mobile scenario with a Doppler frequency shift of ≥1.2kHz.

[0018] Preferably, the system also includes a hardware execution module: the data acquisition module is directly connected to the 5G module baseband processor to obtain physical layer measurement reports in real time; the decision engine is deployed in the edge computing unit of the module, and the response time is ≤5ms; the parameter configuration instructions are directly written into the RF front-end controller through the module driver layer API.

[0019] Preferably, the method further includes an optimization effect verification step: step s1, comparing the key performance indicators of throughput, delay and packet loss rate before and after optimization in a simulated high interference, high-speed movement and multipath fading scenario; step s2, using statistical methods to analyze the verification results to ensure that the optimization effect meets the preset performance improvement target; step s3, recording and saving various data in the verification process for subsequent strategy iteration and optimization.

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

[0021] 1. The present invention uses a deep Q network to generate reinforcement learning strategies and dynamically adjusts network parameters based on real-time link quality assessment results. This adaptive adjustment mechanism enables the network to respond quickly and optimize performance in different environments.

[0022] 2. This invention builds a multidimensional link quality indicator model that collects key parameters such as channel state information, end-to-end delay, packet loss rate, and instantaneous rate change in real time to generate a dynamic link quality matrix. These parameters can comprehensively reflect the actual network operation status, providing accurate data support for subsequent optimization strategies. The dynamic link quality matrix can capture dynamic changes at the physical layer or signal level, thereby more effectively optimizing the network.

[0023] 3. By introducing a reward function and continuously iterating and optimizing during reinforcement learning, the present invention can output an optimal parameter configuration solution that can balance throughput and latency and achieve efficient utilization of network resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is the overall flow chart of the 5G signal module network performance adaptive optimization method;

[0025] Figure 2 A flow chart for generating a dynamic link quality matrix in step S1;

[0026] Figure 3 Flowchart for reinforcement learning strategy generation and training in step S2;

[0027] Figure 4 This is a flowchart of the dynamic network performance tuning rules in step S3. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0029] Example 1;

[0030] See also Figure 1-Figure 4In an embodiment of the present invention, a method for adaptively optimizing network performance of a 5G signal module based on dynamic link quality evaluation is characterized by comprising the following steps: Step S1, constructing a multi-dimensional link quality indicator model, collecting channel state information, end-to-end delay, packet loss rate, and instantaneous rate change values ​​of the 5G module in different network environments in real time, and generating a dynamic link quality matrix: Qlink = [CSI, Delay, Loss, ΔRate], with a sampling period of ≤ 100ms;

[0031] Among them, Throughput is defined as the amount of data successfully transmitted per second (unit: Mbps), Delay is defined as the time difference between the sending and receiving of a data packet (unit: ms), and Loss is defined as the proportion of lost data packets to the total sent packets (unit: %). △Rate is calculated by first-order difference: △Rate = Rate_t - Rate_{t-1}, where Rate_t is the current sampling period rate.

[0032] Step S2: Reinforcement learning strategy generation. The link quality matrix is ​​input into the deep Q network, with the network access mode, spectrum selection strategy, and physical resource block allocation ratio as the action space. The optimal parameter configuration solution is output through iterative optimization of the reward function: R = 0.7 × Throughput - 0.3 × Delay.

[0033] Step S3, dynamic network performance tuning, dynamically adjusts the following parameters according to the generated solution: switching network access mode; selecting Sub-6GHz or millimeter wave spectrum; adjusting the modulation and coding scheme level and PRB allocation ratio.

[0034] The action space definition satisfies the following requirements: spectrum selection strategies include Sub-6GHz and millimeter wave; the physical resource block allocation ratio ranges from 10% to 90%; and when the interference-to-noise ratio (INR) is detected to be less than 10dB, a millimeter wave spectrum switch is forcibly triggered, specifically implemented by the DQN output layer action selection module. Dynamic adjustment includes the following rules: Rule S1: When the packet loss rate is greater than 5% and the latency is greater than 20ms, the physical resource block allocation ratio is increased by 15% and the modulation and coding scheme level is increased; Rule S2: When the instantaneous rate change value fluctuates by more than 30%, switch to standalone networking mode; Rule S3: In environments with a mobile speed greater than or equal to 60km / h, prioritize millimeter wave spectrum and increase the physical resource block allocation ratio; Rule S4: In environments with multipath fading greater than or equal to 15dB, reduce the modulation and coding scheme level and switch to standalone networking mode. The training process of the deep Q network includes: step s1, using a generative adversarial network to simulate high-interference scenarios and inject adversarial training data; step s2, building a dual-network architecture, including a main network and a target network, and synchronizing the main network parameters to the target network every 1000 iterations; step s3, setting the experience replay pool capacity to 10,000 sets of state and action transition samples, and extracting 128 sets of samples in batches for training each time.

[0035] This embodiment performs the following steps to work:

[0036] Step S1: Construct a multi-dimensional link quality indicator model, collect CSI, Delay, Loss, and ΔRate in real time, and generate a dynamic link quality matrix Qlink. Smooth the CSI data using a Kalman filter (with parameters Q = 0.1, R = 1.0), reducing noise fluctuations by more than 40%. Use a sliding window mean filter (window size = 5) to stabilize the delay data. ΔRate calculation explicitly uses first-order differences.

[0037] Step S2: Qlink is fed into the DQN. The action space includes network access mode, spectrum selection, and PRB allocation ratio. The reward function R = 0.6 × Throughput - 0.25 × Delay - 0.15 × Loss balances throughput, delay, and packet loss rate. During DQN training, GAN-generated data is directly injected into the experience replay pool to simulate high-interference scenarios. An iteration is defined as a network parameter update loop, with the target network synchronized every 1000 iterations. Random sampling is used for sample extraction, and batches of 128 groups are used for Q-value loss calculation.

[0038] Step S3: Dynamically tune parameters, such as rules S1-S4.

[0039] The working principle of this embodiment of the present invention is to build a multi-dimensional link quality indicator model to accurately collect channel state information (CSI), end-to-end delay, packet loss rate, and instantaneous rate change values ​​of the 5G module in real time under different network environments, and generate a dynamic link quality matrix Qlink. This process uses a short sampling period (≤100ms) to ensure the real-time and accuracy of the data.

[0040] The collected link quality matrix is ​​then fed into the Deep Q Network, initiating the reinforcement learning strategy generation process. The Deep Q Network constructs an action space based on network access methods, spectrum selection strategies, and physical resource block allocation ratios. It then performs iterative optimization using a set reward function, R = 0.7 × Throughput - 0.3 × Delay. Through a continuous process of trial and error and learning, the Deep Q Network gradually discovers the optimal parameter configuration.

[0041] Based on this optimal parameter configuration, dynamic network performance tuning is then implemented. Specifically, network access methods can be flexibly switched, selecting the most appropriate connection mode based on network conditions. Spectrum selection can be switched between Sub-6 GHz and millimeter wave spectrum to accommodate transmission requirements in different scenarios. Furthermore, the modulation and coding scheme level and physical resource block allocation ratio can be adjusted to achieve refined allocation of network resources. The action space definition clearly defines the spectrum selection range and physical resource block allocation ratio, and sets a mandatory trigger for millimeter wave spectrum switching when the interference-to-noise ratio is less than 10 dB, further enhancing the network's ability to cope with complex interference environments. Dynamic adjustment rules are equally sophisticated and targeted. For example, when packet loss and latency exceed specified limits, the physical resource block allocation ratio is increased and the modulation and coding scheme level is upgraded. Parameter adjustments are also implemented for situations with large fluctuations in instantaneous rate changes, varying mobile speeds, and multipath fading. These rules work together to ensure comprehensive network performance stability and optimization. The training process of the Deep Q Network has also been carefully designed. By using a generative adversarial network to simulate high-interference scenarios and inject adversarial training data, the network can adapt to harsh environments in advance; a dual-network architecture is constructed and a parameter synchronization mechanism is reasonably set to improve the stability and efficiency of training; an experience replay pool is used to store state and action transition samples, and samples are batch-extracted for training to fully tap the value of the data and promote the continuous improvement of the network's learning and optimization capabilities.

[0042] Example 2;

[0043] See also Figure 1-Figure 4In this embodiment of the present invention, the generation of the dynamic link quality matrix in step S1 includes data preprocessing: Step s1, Kalman filtering is performed on the channel state information. The Kalman filter parameters are set as follows: process noise covariance Q = 0.1, measurement noise covariance R = 1.0, and initial state estimate X0 = 0. After processing, the CSI signal-to-noise ratio fluctuation is reduced by more than 40%, improving signal stability. Step s2, sliding window mean filtering is performed on the end-to-end delay data. Step s3, first-order difference calculation of the instantaneous rate change is performed: ΔRate = Ratet - Ratet - 1. The reward function further introduces a packet loss rate penalty factor: the updated reward function is: R = 0.6 × Throughput - 0.25 × Delay - 0.15 × Loss.

[0044] The adjustment of the modulation and coding scheme level in step S3 adopts a gradient strategy: when the channel quality indicator CQI increases by ≥2 levels, the modulation and coding scheme level is increased by 1 level; when the CQI decreases by ≥3 levels for three consecutive sampling periods, the modulation and coding scheme level is decreased by 2 levels. The generation of adversarial training data in step s1 includes: simulating a high-scattering environment with a multipath delay spread of ≥50ns; injecting pulse noise interference to cause a sudden drop in the signal-to-noise ratio of ≥15dB; and constructing a high-speed mobile scenario with a Doppler frequency shift of ≥1.2kHz. It also includes a hardware execution module: the data acquisition module is directly connected to the 5G module baseband processor to obtain physical layer measurement reports in real time; the decision engine is deployed in the edge computing unit of the module with a response time of ≤5ms; and the parameter configuration instructions are directly written to the RF front-end controller through the module driver layer API.

[0045] The method also includes an optimization effect verification step: step s1, comparing key performance indicators of throughput, delay, and packet loss rate before and after optimization in a simulated high-interference, high-speed movement, and multipath fading scenario; step s2, using statistical methods to analyze the verification results to ensure that the optimization effect meets the preset performance improvement target; step s3, recording and saving various data during the verification process for subsequent strategy iteration and optimization.

[0046] The working principle of the embodiment of the present invention is as follows: in step S1, a data preprocessing step is incorporated into the generation of the dynamic link quality matrix. First, a Kalman filter is used to smooth the channel state information to remove noise interference in the data, making the CSI data more stable and reliable. Then, a sliding window mean filter is used to smooth the end-to-end delay data, effectively filtering out short-term fluctuations and presenting a stable trend of delay. Then, a first-order difference calculation is performed on the instantaneous rate change value to accurately obtain ΔRate, so as to more clearly capture the dynamic change characteristics of the rate.

[0047] The reward function further introduces a packet loss penalty factor, updating it to R = 0.6 × Throughput - 0.25 × Delay - 0.15 × Loss. This comprehensively evaluates network performance from three perspectives: throughput, delay, and packet loss, making the optimization more comprehensive and tailored to actual needs. In step S3, the modulation and coding scheme level is adjusted using a gradient strategy, closely following the changes in the channel quality indicator (CQI). The modulation and coding scheme level is increased when the CQI improves by ≥2 levels, and decreased when the CQI decreases by ≥3 levels for three consecutive sampling periods. This precise and dynamic approach accurately matches changing channel conditions, achieving a balance between transmission efficiency and reliability. To generate adversarial training data in step S1, a high-scattering environment with multipath delay spread ≥50ns is simulated to create a complex signal propagation scenario. Impulse noise is injected to cause a sudden drop in the signal-to-noise ratio by ≥15dB, testing the network's ability to cope with sudden interference.

[0048] At the hardware level, the data acquisition module is directly connected to the 5G module baseband processor, establishing an efficient data path and obtaining physical layer measurement reports in real time, ensuring the timeliness and accuracy of link quality data. The decision engine is deployed in the module's edge computing unit, and its ultra-short response time of ≤5ms enables the optimization strategy to quickly respond to network changes. Parameter configuration instructions are directly written to the RF front-end controller through the module driver layer API, and instructions are transmitted quickly and accurately. In addition, the method also covers the optimization effect verification step. First, in typical complex scenarios such as simulated high interference, high-speed movement, and multipath fading, key performance indicators such as throughput, latency, and packet loss rate are compared before and after optimization. Statistical methods are then used to conduct an in-depth analysis of the verification results to determine whether the preset performance improvement goals have been achieved. At the same time, the verification process data is recorded and saved.

[0049] Example 3;

[0050] See also Figure 1-Figure 4 , providing a specific example. On the automated assembly line of a certain automobile manufacturing plant, the 5GRedCap module model RG520N is installed on the AGV transport cart to perform material transportation tasks. This scenario has significant dynamic interference sources. The AGV cart moves at a speed of 60 kilometers per hour during the acceleration phase of the linear track. The reflection of the metal shelf causes multipath fading of up to 18 decibels. At the same time, 10 millisecond pulse noise caused by electromagnetic interference from the welding machine periodically occurs.

[0051] The system initializes with an 80-millisecond sampling period for the link quality matrix, collecting data directly through the baseband processor. The initial action space is configured to use the 3.5 GHz spectrum in the sub-6 GHz band, with a 40% physical resource block allocation ratio and an 18-level modulation and coding scheme corresponding to 256QAM modulation. The reinforcement learning model uses a dual-network architecture, with the primary network updating 100 times per second and the target network synchronizing parameters every 1,000 iterations.

[0052] During the optimization process, the channel state information collected in real time during the t-th sampling period showed a signal-to-noise ratio (SNR) of 8.2 dB, a channel quality indicator (CQI) of level 14, an end-to-end delay of 28 milliseconds, and a packet loss rate of 6.3%. The instantaneous rate change, calculated by calculating the difference between the current rate of 350 Mbps and the previous cycle's 265 Mbps, was 85 Mbps. During the data preprocessing phase, a Kalman filter was applied to the channel state information, taking as input the raw SNR fluctuation sequence of 7.1 dB, 7.9 dB, 8.2 dB, 6.5 dB, and 8.2 dB, and outputting a smoothed value of 7.8 dB.

[0053] The Deep Q Network's decision-making phase uses the link quality matrix as input to the model for action space analysis. It predicts that switching to the 28 GHz millimeter wave band will increase throughput by 42%, and the incremental reward for increasing the physical resource block allocation from 40% to 55% is 1.3. Using the updated reward function, the reward is calculated by multiplying 0.6 by the throughput of 350 megabits per second, minus 0.25 by the latency of 29.2 milliseconds, and minus 0.15 by the packet loss rate of 6.3%, resulting in a reward of 201.755.

[0054] Verification of the optimization results showed significant performance improvements at the same location, with throughput increasing from 350 Mbps to 502 Mbps, a 43.4% increase; end-to-end latency decreasing from 29.2 milliseconds to 16.5 milliseconds, a 43.5% decrease; packet loss rate decreasing from 6.3% to 1.8%, a 71.4% improvement; and signal-to-interference-plus-noise ratio increasing from 7.8 dB to 14.2 dB. Statistically validated using a paired t-test, the p-value for throughput improvement at a 95% confidence interval was 0.0032, significantly below the 0.05 threshold; and the packet loss rate fluctuation coefficient decreased from 0.35 to 0.12, improving stability by 65.7%.

[0055] This embodiment has been verified through actual testing of the RG520N module. The embedded hardware achieves millisecond-level response for direct data acquisition by the baseband processor and decision-making by the edge computing unit. Millimeter wave switching takes only 4.7 milliseconds, and the error in adjusting the physical resource block allocation ratio is less than plus or minus 2%. In the scenario of counteracting impulse noise, the signal-to-interference-plus-noise ratio is stably maintained at above 14 decibels.

[0056] Working Principle: First, a multi-dimensional link quality indicator model is constructed. Channel state information, end-to-end delay, packet loss rate, and instantaneous rate change are collected in real time to generate a dynamic link quality matrix. After data preprocessing, this matrix accurately reflects network conditions. Next, the link quality matrix is ​​input into a Deep Q Network. Using network access mode, spectrum selection strategy, and physical resource block allocation ratio as the action space, an iterative optimization reward function is used to output the optimal parameter configuration. Network parameters are then dynamically adjusted based on this configuration, such as switching network access mode, selecting spectrum, adjusting modulation and coding scheme levels, and adjusting PRB allocation ratios. These adjustment rules work together to ensure stable and optimized network performance. During Deep Q Network training, a generative adversarial network is used to inject adversarial training data, simulating high-interference scenarios. This dual-network architecture is constructed and utilizes an experience replay pool to enhance training effectiveness. At the hardware level, the data acquisition module is directly connected to the 5G module baseband processor, and the decision engine is deployed in the edge computing unit for rapid response. Parameter configuration commands are written to the RF front-end controller via the module driver API for precise execution. Finally, through the optimization verification step, performance indicators before and after optimization are compared and analyzed, and the data is recorded for subsequent strategy iteration and optimization.

[0057] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and alterations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation, characterized in that: The following steps are involved: Step S1: Build a multi-dimensional link quality indicator model to collect channel state information, end-to-end delay, packet loss rate, and instantaneous rate change values ​​of the 5G module in different network environments in real time to generate a dynamic link quality matrix: Qlink=[CSI,Delay,Loss,ΔRate], sampling period ≤ 100ms; Step S2: Reinforcement learning strategy generation. The link quality matrix is ​​input into the deep Q network, and the network access mode, spectrum selection strategy, and physical resource block allocation ratio are used as the action space. The optimal parameter configuration solution is output through iterative optimization of the reward function. The reward function is: R=0.7×Throughput-0.3×Delay; Step S3: Dynamic network performance tuning. According to the generated solution, dynamically adjust the following parameters: Switch network access mode; Choose Sub-6GHz or mmWave spectrum; Adjust the modulation and coding scheme level and PRB allocation ratio.

2. A 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation according to claim 1, characterized in that: The action space definition satisfies: the spectrum selection strategy includes Sub-6GHz and millimeter wave; the physical resource block allocation ratio ranges from 10% to 90%; when the interference-to-noise ratio is detected to be less than 10dB, the millimeter wave spectrum switching action is forcibly triggered.

3. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: The dynamic adjustment includes the following rules: Rule S1: When the packet loss rate is greater than 5% and the delay is greater than 20ms, increase the physical resource block allocation ratio by 15% and improve the modulation and coding scheme level; Rule S2: When the instantaneous rate change value fluctuates by more than 30%, it switches to the independent networking mode. Rule S3: In an environment where the mobile speed is greater than or equal to 60 km / h, millimeter wave spectrum is preferred and the proportion of physical resource blocks allocated is increased; Rule S4: When the multipath fading is greater than or equal to 15 dB, reduce the modulation and coding scheme level and switch to the independent networking mode.

4. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: The training process of the deep Q network includes: Step s1: Generative adversarial networks are used to simulate high-interference scenarios and inject adversarial training data. Step s2: construct a dual network architecture, including a main network and a target network. The main network parameters are synchronized to the target network every 1000 iterations. In step s3, the experience replay pool capacity is set to 10,000 sets of state and action transition samples, and 128 sets of samples are extracted for training each time.

5. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: The generation of the dynamic link quality matrix in step S1 includes data preprocessing: Step s1, performing Kalman filtering smoothing on the channel state information; Step s2, applying sliding window mean filtering to the end-to-end delay data; Step s3, perform first-order difference calculation on the instantaneous rate change value ΔRate=Rate t -Rate t-1 .

6. A 5G signal module network performance adaptive optimization method based on dynamic link quality evaluation according to claim 2, characterized in that: The reward function further introduces a packet loss rate penalty factor: The updated reward function is: R = 0.6×Throughput-0.25×Delay-0.15×Loss.

7. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: The modulation and coding scheme level is adjusted in step S3 using a gradient strategy: When the channel quality indicator CQI increases by ≥2 levels, the modulation and coding scheme level is increased by 1 level; When the CQI drops by ≥3 levels for three consecutive sampling periods, the modulation and coding scheme level is reduced by 2 levels.

8. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 4, characterized in that: The generation of adversarial training data in step s1 includes: simulating a high scattering environment with a multipath delay spread of ≥50ns; Injecting pulse noise interference causes the signal-to-noise ratio to drop by ≥15dB; Construct a high-speed movement scenario with a Doppler shift ≥ 1.2kHz.

9. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: It also includes a hardware execution module: the data acquisition module is directly connected to the 5G module baseband processor to obtain physical layer measurement reports in real time; the decision engine is deployed in the module's edge computing unit, with a response time of ≤5ms; parameter configuration instructions are directly written to the RF front-end controller through the module driver layer API.

10. The method for adaptively optimizing 5G signal module network performance based on dynamic link quality evaluation according to claim 1, characterized in that: The method further comprises the step of optimizing the effect: Step s1: Compare the key performance indicators of throughput, delay, and packet loss rate before and after optimization in a simulated high-interference, high-speed mobility, and multipath fading scenario; Step s2: Analyze the verification results using statistical methods to ensure that the optimization effect meets the preset performance improvement goals; Step s3: record and save various data during the verification process for subsequent strategy iteration and optimization.

Images