Energy-saving and self-adaptive DRX control method based on 6G multi-connection network
By introducing a three-layer architecture and hierarchical reinforcement learning method in 6G networks, the global coordination problem across layers and nodes in DRX optimization is solved, achieving efficient energy consumption management and quality of service assurance in multi-connection scenarios, and improving the system's operating efficiency and adaptability.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-31
- Publication Date
- 2026-04-07
AI Technical Summary
Existing DRX optimization methods in 6G networks lack cross-layer and cross-node global coordination in multi-connection scenarios. They suffer from insufficient local optimization, high model complexity, large training overhead, and difficulty in efficient deployment in real-time scenarios, making it difficult to guarantee a balance between global energy efficiency and service quality.
A three-layer architecture consisting of a terminal layer, a micro base station layer, and a macro base station layer is adopted. Through hierarchical reinforcement learning and graph neural networks, a policy model with continuous learning and iterative update capabilities is constructed to achieve cross-layer collaborative control. The terminal layer is responsible for state perception and reporting, the micro base station layer performs local scheduling, and the macro base station layer performs global coordination. The action decision is made by combining a multi-agent deep deterministic policy gradient algorithm and a deep Q-network.
It significantly improves system operating efficiency and real-time response capabilities, enables intelligent energy consumption control under dynamic changes in user scale and business load, ensures service quality and seamless switching experience, and reduces system operating costs.
Smart Images
Figure CN121815380A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of wireless communication technology, and particularly relates to an energy-saving and adaptive DRX control method based on a 6G multi-connection network. Background Technology
[0002] In 6G networks, Discontinuous Reception (DRX) is a crucial energy-saving mechanism for user terminals. By periodically entering and exiting sleep mode in idle mode, DRX can effectively reduce energy consumption while maintaining service quality. Its basic components include: an "active period" for monitoring and controlling the channel, an active period for data processing and retransmission, and a timer that determines the receive window. DRX can switch between short and long periods; the former ensures response speed, while the latter maximizes energy saving, thus achieving a balance between energy efficiency and service quality.
[0003] Early research primarily employed static parameter configuration and threshold control methods. For example, Jia-Ming Liang et al. reduced redundant wake-ups in multicast scenarios by using predefined scheduling windows; M. Sajid Mushtaq et al. extended the traditional three-state model, introducing new state and transition rules; and Dharmaraja Selvamuth et al. combined energy harvesting mechanisms with DRX. These methods achieved some success in energy saving, but they generally relied on fixed thresholds and predefined rules, lacking the ability to adapt to complex dynamic network environments.
[0004] With the diversification of network environments and service requirements, research has begun to explore dynamic optimization methods. Some works utilize traffic prediction algorithms to achieve adaptive adjustment of DRX cycles, while others introduce adaptive packet queuing techniques or hybrid static and dynamic methods to improve flexibility. Furthermore, researchers have attempted to model DRX using stochastic Markov chains, semi-Markov processes, and Markov decision processes to achieve a better trade-off between latency and energy consumption. These methods have propelled the development of DRX from static configuration to intelligent scheduling, but their optimization remains limited to a single level, lacking global coordination across levels and nodes.
[0005] In recent years, machine learning and deep learning methods have been introduced into DRX optimization. For example, methods based on Q-learning and deep Q-networks (DQN) can model the relationship between network state and DRX actions, reinforcement learning methods support online real-time parameter adjustment, and multi-agent reinforcement learning combined with attention mechanisms can achieve collaborative optimization among multiple devices. In addition, some studies have combined new technologies such as network slicing, beam search, and beam tracking to further improve energy efficiency. These AI-driven methods have shown strong adaptability, but still have significant limitations: First, most studies focus on local optimization of single devices or single base stations, lacking coordination and scheduling between macro base stations, micro base stations, and user terminals; second, the model complexity and training overhead are large, making it difficult to deploy efficiently in real-time scenarios; third, feature modeling often only focuses on a single dimension of spatial topology or time series, failing to fully capture the spatiotemporal coupling characteristics of network state; fourth, in ultra-dense networks and multi-connection scenarios, existing methods are still insufficient in ensuring a balance between global energy efficiency and service quality. Summary of the Invention
[0006] In view of the above-mentioned deficiencies of the prior art, this invention proposes an energy-saving and adaptive DRX control method based on a 6G multi-connection network. The technical solution designed in this invention includes the following steps: S1: Each terminal collects its own operating status and service association information and reports it to the micro base station it is connected to; the operating status includes DRX mode, active status, dormant status and data transmission requirements; the service association information includes the multi-connection association relationship between the terminal and the micro base station; S2: Each micro base station collects its own local operating status and aggregates the operating status and service-related information reported by all terminals connected to it to generate a micro base station local status vector. S3: Each micro base station generates and executes local DRX control actions based on its local state vector and global guidance parameters from the macro base station through the first strategy model; S4: The macro base station aggregates the information uploaded by all micro base stations, generates a global state vector, and generates a global DRX control strategy based on the global state vector through the second strategy model. The global DRX control strategy includes global guidance parameters sent to each micro base station. S5: Collect performance metrics after the network executes local DRX control actions and global DRX control strategies, calculate the global reward value, and update the first strategy model and the second strategy model based on the global reward value.
[0007] Preferably, S1 further includes: Based on the operating status and service association information reported by each terminal, the DRX distribution characteristics of the terminal are calculated, and a multi-connection mapping relationship table between the terminal and the micro base station is constructed. The DRX distribution characteristics include the proportion of active terminals, the proportion of DRX long-cycle mode, and the proportion of dormant terminals.
[0008] Preferably, S2 further includes: When generating the local state vector of the micro base station, the operating status and service-related information reported by the terminal are weighted and aggregated according to the preset contribution weight; The local operating status includes the DRX mode configured locally by the base station, the current service load, and the real-time energy consumption level.
[0009] Preferably, S3 further includes: Construct a micro base station topology graph, where nodes are micro base stations, and the weight of the edges between nodes is positively correlated with the number of terminals shared by the two base stations; A spatial modeling method is used to aggregate the states of neighboring nodes of each micro base station to generate a local spatial feature vector for each micro base station. When each micro base station generates a local DRX control action, its input information also includes the local spatial feature vector.
[0010] Preferably, S3 further includes: After generating local DRX control actions, the selection probability of DRX control actions is adjusted according to the consistency constraints of shared terminals. For terminals that are connected to multiple micro base stations at the same time, the most active priority rule is adopted, which includes the terminal selecting the mode with the most active DRX status among the connected micro base stations as its own operating mode.
[0011] Preferably, S3 further includes: When the network state changes abruptly, a fast adaptive mechanism is activated to update the parameters of the first strategy model in real time.
[0012] Preferably, S4 includes: The global DRX control strategy also includes the DRX control mode configuration of the macro base station itself; The global guidance parameters include the DRX long-cycle ratio and the sleep ratio; When the network environment changes significantly, a rapid adaptation process is triggered to update the parameters of the second strategy model.
[0013] Preferably, S5 includes: The step of updating the first strategy model and the second strategy model based on the global reward value includes decomposing the global reward value according to the micro base station dimension to obtain the local reward value of each micro base station. During the decomposition process, the energy consumption performance, latency control effect, throughput contribution and multi-connection consistency constraint satisfaction of each micro base station are taken into account. Each micro base station updates its first strategy model based on its local reward value; The macro base station updates the second strategy model based on the global reward value.
[0014] Preferably, the method is executed in a cyclical iterative manner, repeating steps S1 to S5 until the core indicators of network energy consumption, latency and throughput reach a preset equilibrium state.
[0015] Beneficial effects: 1. This invention adopts a three-layer architecture consisting of a terminal layer, a micro base station layer, and a macro base station layer, which achieves clear division of functions and distributed processing. The terminal layer focuses on status perception and reporting, the micro base station layer is responsible for local scheduling and management, and the macro base station layer performs global coordination and decision-making. This effectively overcomes the processing bottlenecks and slow response problems of existing centralized scheduling schemes, and significantly improves the overall operating efficiency and real-time response capability of the system.
[0016] 2. This invention establishes an efficient feedback mechanism between the terminal layer and the micro base station layer, and implements cross-cell global optimization at the macro base station layer, achieving multi-level collaborative control. It can quickly adjust resource allocation in sudden service scenarios, while taking into account local load balancing and global energy consumption management. Compared with the single-point or static optimization methods of existing technologies, this invention achieves intelligent energy consumption control while ensuring user service quality and seamless switching experience, significantly reducing system operating costs.
[0017] 3. This invention combines intelligent methods such as hierarchical reinforcement learning and graph neural networks to construct a policy model with continuous learning and iterative update capabilities. Coupled with a hierarchical modular design that supports heterogeneous access and flexible expansion, the system can operate stably in large-scale scenarios with dynamic changes in user scale and business load. Compared with existing technologies that rely on fixed rules, this invention has the ability to self-evolve and self-optimize according to network environment and business characteristics, ensuring high performance and strong adaptability of the network in long-term operation. Attached Figure Description
[0018] Figure 1 This is a flowchart illustrating a preferred embodiment of the present invention; Figure 2 This is a schematic diagram of the flowchart algorithm of a preferred embodiment of the present invention; Figure 3 This is a schematic diagram of the network architecture of a preferred embodiment of the present invention. Detailed Implementation
[0019] The embodiments of the present invention will be described in detail below. The embodiments described below are implemented based on the technical solution of the present invention, and detailed implementation methods and specific operation processes are given. However, the protection scope of the present invention is not limited to the embodiments described below.
[0020] This invention designs an energy-saving and adaptive DRX control method based on a 6G multi-connection network. The technical solution includes the following steps, such as... Figure 1-2 As shown, it specifically includes: S1: Each terminal collects its own operating status and service association information and reports it to the micro base station it is connected to; the operating status includes DRX mode, active status, dormant status and data transmission requirements; the service association information includes the multi-connection association relationship between the terminal and the micro base station. S2: Each micro base station collects its own local operating status and aggregates the operating status and service-related information reported by all terminals connected to it to generate a micro base station local status vector. S3: Each micro base station generates and executes local DRX control actions based on its local state vector and global guidance parameters from the macro base station through the first strategy model; S4: The macro base station aggregates the information uploaded by all micro base stations, generates a global state vector, and generates a global DRX control policy based on the global state vector through the second strategy model. The global DRX control policy includes global guidance parameters sent to each micro base station. S5: Collect performance metrics after the network executes local DRX control actions and global DRX control strategies, calculate the global reward value, and update the first strategy model and the second strategy model based on the global reward value.
[0021] Specifically, for the first and second policy models, the first policy model employs a multi-agent deep deterministic policy gradient algorithm. Each micro-base station acts as an independent agent, making decisions within the centralized training and distributed execution framework of MADDPG. This algorithm allows each micro-base station to rely solely on local observation information (such as local state vectors and spatial feature vectors) during execution. However, during training, the policy model update utilizes global information from the macro-base station and other micro-base stations to evaluate the value of the Critic network, thereby learning cooperative policies that optimize local performance while promoting global objectives. The second policy model employs a deep Q-network algorithm. The macro-base station takes the global state vector as input and evaluates the long-term expected reward (Q-value) of different global DRX control policies (i.e., different "actions," such as setting different global guidance parameters) through the DQN's value network, selecting the action with the highest Q-value as the final policy output. DQN stabilizes the training process through experience replay and target network techniques, enabling the macro-base station to learn from historical experience and gradually approach the globally optimal decision.
[0022] Preferably, S1 further includes: Based on the operating status and service association information reported by each terminal, the DRX distribution characteristics of the terminal are calculated, and a multi-connection mapping relationship table between the terminal and the micro base station is constructed. DRX distribution characteristics include the proportion of active terminals, the proportion of DRX long-cycle mode, and the proportion of dormant terminals.
[0023] Preferably, S2 further includes: When generating the local state vector of the micro base station, the operating status and service-related information reported by the terminal are weighted and aggregated according to the preset contribution weight; Local operating status includes the DRX mode configured locally by the base station, the current service load, and the real-time energy consumption level.
[0024] Preferably, S3 further includes: Construct a micro base station topology graph, where nodes are micro base stations, and the weight of the edges between nodes is positively correlated with the number of terminals shared by the two base stations; A spatial modeling method is used to aggregate the states of neighboring nodes of each micro base station to generate a local spatial feature vector for each micro base station. When each micro base station generates a local DRX control action, its input information also includes a local spatial feature vector.
[0025] Preferably, S3 further includes: After generating local DRX control actions, the selection probability of DRX control actions is adjusted according to the consistency constraints of shared terminals. For terminals that are connected to multiple micro base stations at the same time, the most active priority rule is adopted, which includes the terminal selecting the mode with the most active DRX status among the connected micro base stations as its own operating mode.
[0026] Preferably, S3 further includes: When the network state changes abruptly, a fast adaptive mechanism is activated to update the parameters of the first strategy model in real time.
[0027] Specifically, sudden changes in network status include a surge in traffic or base station failures, while significant changes in the network environment include topology adjustments or changes in service types.
[0028] Preferably, S4 includes: The global DRX control strategy also includes the DRX control mode configuration of the macro base station itself; Global guidance parameters include the DRX long-cycle ratio and the hibernation ratio; When the network environment changes significantly, a rapid adaptation process is triggered to update the parameters of the second strategy model.
[0029] Preferably, S5 includes: The first and second strategy models are updated based on the global reward value. This includes decomposing the global reward value according to the micro base station dimension to obtain the local reward value of each micro base station. During the decomposition process, the energy consumption performance, latency control effect, throughput contribution and multi-connection consistency constraint satisfaction of each micro base station are taken into account. Each micro base station updates its first strategy model based on its local reward value; The macro base station updates the second strategy model based on the global reward value.
[0030] Specifically, the reward decomposition process employs a dynamic weight adjustment mechanism, including dynamic weight adaptation: real-time monitoring of the network's average energy consumption level. When this value exceeds a preset threshold, the weight coefficient of energy consumption indicators is significantly increased in the reward decomposition formula. Conversely, when network-wide energy consumption is low, the energy consumption weight is appropriately reduced, while the weight coefficients of service quality indicators such as latency and throughput are increased. Targeted penalties: for micro base stations that fail to meet the latency upper limit or throughput lower limit requirements, a fixed penalty term is applied to their decomposed local reward value, regardless of the overall energy consumption level. This mechanism guides micro base stations to prioritize energy conservation when global energy consumption is tight and strive to improve service quality when energy consumption is ample, achieving a dynamic balance.
[0031] In addition, for updating the first policy model and the second policy model, the process of "updating the first policy model" is a centralized training under the MADDPG algorithm framework. This includes each micro base station uploading its collected experience data (state, action, reward, next state) to a macro base station or a central trainer. The trainer uses all this experience data to calculate the update gradient for the Critic network of each micro base station. During the calculation, the input of the Critic network includes the action and state information of all micro base stations. This allows each micro base station to take into account the impact of the behavior of other base stations during training. The updated policy model parameters are then distributed to each micro base station. In this way, the local policies of each micro base station achieve system-level collaborative optimization while maintaining distributed execution. The process of "updating the second policy model" is the standard iterative learning process of DQN. This includes the macro base station storing each policy decision (state, action, reward, next state) as a sample in the experience replay buffer. During training, a batch of samples is randomly sampled from the buffer to calculate the loss function of the current Q network with respect to the target Q value. The parameters of the current Q network are updated by minimizing the loss using gradient descent. The parameters of the target network are also updated periodically. Since the input of DQN is the global state vector (including service traffic, terminal distribution, micro base station status, etc.) aggregated by the macro base stations, its output Q value evaluation has a global perspective, thus enabling it to learn DRX configuration preferences that tend to be globally optimal.
[0032] Preferably, the method is executed in a cyclical iterative manner, repeating steps S1 to S5 until the core indicators of network energy consumption, latency and throughput reach a preset equilibrium state.
[0033] Specifically, when the network topology changes or there are sudden fluctuations in services, a rapid adaptation process is automatically triggered to adjust the strategy model parameters.
[0034] Specifically, such as Figure 3 As shown, an energy-saving and adaptive DRX control method based on 6G multi-connection network is designed around a hierarchical multi-access network architecture, which includes three main structural parts: terminal layer structure, micro base station layer structure and macro base station layer structure.
[0035] The terminal layer structure mainly consists of user equipment (UE), a terminal status awareness module, an energy consumption statistics module, and a data collection and reporting unit. UEs are deployed at the bottom layer of the network, within the coverage area of micro base stations, and their locations are relatively dispersed. Each UE establishes a communication connection with adjacent micro base stations via a wireless access link, and simultaneously transmits its operational service data, energy consumption status information, and DRX (Discontinuous Reception) related parameters to its respective micro base station via the data collection and reporting unit.
[0036] The micro base station layer structure mainly consists of multiple small base station nodes, a state aggregation and modeling module, a collaborative control module, and an interface module for interacting with macro base stations. Micro base stations are distributed within the coverage area of macro base stations, forming a dense deployment. Each micro base station is directly connected to several user devices it covers, responsible for receiving and aggregating energy consumption and service status information uploaded from the terminal layer; simultaneously, it models and evaluates the traffic characteristics and DRX status of local users through the state aggregation and modeling module, and uploads the summarized information to the macro base station via the interface module. Lateral control information exchange can also be established between micro base stations when necessary to improve the consistency of local scheduling.
[0037] The macro base station layer structure mainly consists of macro base station nodes, a global state awareness module, a cross-layer coordination and decision-making module, and a control feedback downlink module. Macro base stations are typically located in the core coverage area of the network, maintaining stable connections with each micro base station via wide-area wireless links, forming a star topology. During system operation, the macro base station receives aggregated state information from the micro base station layer. The global state awareness module analyzes the overall network traffic demand, energy consumption distribution, and DRX operating efficiency. The cross-layer coordination and decision-making module then generates a globally optimized control strategy. Finally, the macro base station distributes the optimized strategy to each micro base station via downlink control signaling. The micro base stations then forward or execute the strategy to each terminal, thereby achieving multi-level DRX control and energy management.
[0038] The three structural components mentioned above form an organic collaborative relationship through layered information transmission and feedback links. The terminal layer is responsible for bottom-level data acquisition and status reporting, the micro base station layer undertakes intermediate processing and local modeling, and the macro base station layer performs global coordination and decision-making. The three cooperate with each other to form a closed-loop system that aggregates information from the bottom up and provides feedback control from the top down, enabling the entire system to achieve efficient DRX control and energy management in multi-access network environments.
[0039] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.
Claims
1. An energy-saving and adaptive DRX control method based on a 6G multi-connection network, characterized in that, include: S1: Each terminal collects its own operating status and service association information and reports it to the micro base station it is connected to; the operating status includes DRX mode, active status, dormant status and data transmission requirements; the service association information includes the multi-connection association relationship between the terminal and the micro base station; S2: Each micro base station collects its own local operating status and aggregates the operating status and service-related information reported by all terminals connected to it to generate a micro base station local status vector. S3: Each micro base station generates and executes local DRX control actions based on its local state vector and global guidance parameters from the macro base station through the first strategy model; S4: The macro base station aggregates the information uploaded by all micro base stations, generates a global state vector, and generates a global DRX control strategy based on the global state vector through the second strategy model. The global DRX control strategy includes global guidance parameters sent to each micro base station. S5: Collect performance metrics after the network executes local DRX control actions and global DRX control strategies, calculate the global reward value, and update the first strategy model and the second strategy model based on the global reward value.
2. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S1 further includes: Based on the operating status and service association information reported by each terminal, the DRX distribution characteristics of the terminal are calculated, and a multi-connection mapping relationship table between the terminal and the micro base station is constructed. The DRX distribution characteristics include the proportion of active terminals, the proportion of DRX long-cycle mode, and the proportion of dormant terminals.
3. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S2 further includes: When generating the local state vector of the micro base station, the operating status and service-related information reported by the terminal are weighted and aggregated according to the preset contribution weight; The local operating status includes the DRX mode configured locally by the base station, the current service load, and the real-time energy consumption level.
4. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S3 further includes: Construct a micro base station topology graph, where nodes are micro base stations, and the weight of the edges between nodes is positively correlated with the number of terminals shared by the two base stations; A spatial modeling method is used to aggregate the states of neighboring nodes of each micro base station to generate a local spatial feature vector for each micro base station. When each micro base station generates a local DRX control action, its input information also includes the local spatial feature vector.
5. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S3 further includes: After generating local DRX control actions, the selection probability of DRX control actions is adjusted according to the consistency constraints of shared terminals. For terminals that are connected to multiple micro base stations at the same time, the most active priority rule is adopted, which includes the terminal selecting the mode with the most active DRX status among the connected micro base stations as its own operating mode.
6. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S3 further includes: When the network state changes abruptly, a fast adaptive mechanism is activated to update the parameters of the first strategy model in real time.
7. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S4 includes: The global DRX control strategy also includes the DRX control mode configuration of the macro base station itself; The global guidance parameters include the DRX long-cycle ratio and the sleep ratio; When the network environment changes significantly, a rapid adaptation process is triggered to update the parameters of the second strategy model.
8. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, S5 includes: The step of updating the first strategy model and the second strategy model based on the global reward value includes decomposing the global reward value according to the micro base station dimension to obtain the local reward value of each micro base station. During the decomposition process, the energy consumption performance, latency control effect, throughput contribution and multi-connection consistency constraint satisfaction of each micro base station are taken into account. Each micro base station updates its first strategy model based on its local reward value; The macro base station updates the second strategy model based on the global reward value.
9. The energy-saving and adaptive DRX control method based on a 6G multi-connection network according to claim 1, characterized in that, The method is executed in a cyclical iterative manner, repeating steps S1 to S5 until the core indicators of network energy consumption, latency and throughput reach a preset equilibrium state.