Communication network optimization method and device and medium

Abstract

The invention provides a communication network optimization method and device and a medium, relates to the technical field of communication, and is used for solving the problem of how to optimize coding configuration parameters of communication data in real time so as to avoid network communication abnormity. The method comprises the following steps: selecting a transmission path of communication data, and obtaining real-time network state information of the transmission path; determining a receiving terminal of the communication data, and obtaining real-time terminal capability information of the receiving terminal; and optimizing coding configuration parameters of the communication data according to the real-time network state information and the real-time terminal capability information. By selecting the transmission path of the communication data, determining the receiving terminal of the communication data, comprehensively considering the network state of the transmission path and the terminal capability of the receiving terminal, and selecting the optimal coding configuration parameter for the communication data, the communication network operates in the optimal coding state in real time, so that the transmission quality of the communication data is improved, and the user experience is improved. The privacy number call quality is improved, and the user experience is improved.

Description

Technical Field

[0001] The present disclosure at least relates to the field of communication technology, and in particular to a communication network optimization method, device and medium. Background Art

[0002] Although modern mobile communication systems have been widely used, they still face problems such as unstable call quality, dropped calls due to network congestion, and cross-border communication delays. Traditional optimization methods often focus on hardware upgrades or simple protocol adjustments, and it is difficult to fully tap the optimization potential of the software and algorithm layers. As a result, in practice, when faced with sudden network communication anomalies, manual parameter adjustment based on human experience is still relied upon to solve communication failures, which results in large parameter adjustment errors and high costs. Summary of the invention

[0003] The technical problem to be solved by the present disclosure is to provide a communication network optimization method, device and medium in view of the above-mentioned shortcomings, so as to solve the problem of how to optimize the coding configuration parameters of communication data in real time to avoid network communication abnormalities.

[0004] In a first aspect, the present disclosure provides a communication network optimization method, the method comprising:

[0005] Select a transmission path for communication data and obtain real-time network status information of the transmission path;

[0006] Determine a receiving terminal of the communication data and obtain real-time terminal capability information of the receiving terminal;

[0007] Optimize the encoding configuration parameters of communication data based on real-time network status information and real-time terminal capability information.

[0008] Further, selecting a transmission path for communication data and obtaining real-time network status information of the transmission path specifically includes:

[0009] Real-time monitoring of the real-time network status information of each communication line in the communication network;

[0010] Obtaining terminal location information and service quality requirement information of communication data to be transmitted;

[0011] According to the terminal location information and service quality demand information, combined with the real-time network status information of each communication line, the optimal transmission path for communication data is selected based on the intelligent routing selection strategy;

[0012] Get real-time network status information for optimal transmission paths.

[0013] Furthermore, real-time monitoring of the real-time network status information of each communication line in the communication network includes:

[0014] Pre-deploy signal quality monitoring modules in key network nodes of the communication network;

[0015] The signal quality monitoring module collects the real-time network status information of each key network node in real time;

[0016] Determine the corresponding key network nodes on each communication line in the communication network, and obtain the real-time network status information of each corresponding key network node.

[0017] Further, wherein:

[0018] The real-time network status information includes at least one of signal strength, signal-to-noise ratio, packet loss rate, delay, jitter, bit error rate, frequency offset, noise power ratio, service cell quality, and neighboring cell interference level;

[0019] The intelligent routing selection strategy includes at least one of a bandwidth maximization strategy, a minimum delay strategy, a packet loss rate minimum strategy, a hybrid cost optimization strategy, a predictive path selection strategy, a dynamic rerouting mechanism, a geographic fence and a load balancing strategy;

[0020] The real-time terminal capability information includes at least one of the processing capability, display limitation, current power status, and current scene mode of the receiving terminal.

[0021] Further, according to the real-time network status information and the real-time terminal capability information, the encoding configuration parameters of the communication data are optimized, specifically including:

[0022] The real-time network status information and the real-time terminal capability information are input into a pre-trained coding parameter dynamic adjustment model based on a deep Q network DQN algorithm, so that the coding parameter dynamic adjustment model outputs the optimal coding configuration parameters of the communication data, and the coding configuration parameters include at least one of bit rate, sampling rate, resolution and frame rate, coding algorithm, error recovery mechanism, multiplexed transmission, and transmission protocol.

[0023] Furthermore, before optimizing the coding configuration parameters of the communication data, the method further includes training a coding parameter dynamic adjustment model based on the DQN algorithm, specifically including:

[0024] Define the deep neural network approximate Q function Q(s,a;θ) and experience replay mechanism E(s,a,r,s′) of the DQN algorithm, where s represents the state space, a represents the action space, θ represents the parameters of the deep neural network, r represents the reward function, and s′ represents the new state in the s training iteration;

[0025] The deep neural network is copied into a target network and a prediction network, and the loss function L(θ)=E[(r+γ·max(a′,Q(s′,a′;θ -1 ))-Q(s,a;θ)) 2], where E represents the expectation of the state-action-reward-new state quadruple obtained from experience playback, γ represents the discount factor, max(a′, Q(s′, a′; θ -1 )) is the maximum expected return given by the target network at the next state s′, θ -1 represents the parameter of the target network's delayed copy of the parameter θ of the prediction network, and Q(s,a;θ) represents the prediction result of the prediction network;

[0026] Collect training samples, input the real-time network status information and real-time terminal capability information in the training samples into s, the coding configuration parameters into a, and the communication quality evaluation index into r, and use the Adam optimizer to optimize L(θ) to obtain a dynamic adjustment model of coding parameters that meets the preset training end conditions.

[0027] Furthermore, after optimizing the coding configuration parameters of the communication data, the method further includes collecting new training samples to retrain the coding parameter dynamic adjustment model based on the DQN algorithm, specifically including:

[0028] Collecting communication quality evaluation indicators after optimizing the coding configuration parameters of communication data, and judging whether the coding parameter dynamic adjustment model based on the DQN algorithm needs to be trained again according to the newly collected communication quality evaluation indicators;

[0029] If so, the real-time network status information, real-time terminal capability information, coding configuration parameters and newly collected communication quality evaluation indicators are correspondingly combined into new training samples, and the coding parameter dynamic adjustment model based on the DQN algorithm is trained again.

[0030] Furthermore, the method further includes obtaining terminal information, specifically including:

[0031] Receiving a measurement report from the terminal, and obtaining the terminal location information according to the measurement report;

[0032] Receiving hardware and software capability information from a receiving terminal, and obtaining processing capability, display limitation, and current power status of the receiving terminal as part of the real-time terminal capability information based on the hardware and software capability information;

[0033] Detecting the current use state of the receiving terminal, obtaining the current scene mode of the receiving terminal as another part of the real-time terminal capability information according to the current use state, the current scene mode including at least one of a conference mode, a streaming mode, and a power saving mode, and obtaining the service quality requirement information of the communication data according to the current scene mode;

[0034] A communication quality evaluation index is received from a receiving terminal, where the communication quality evaluation index includes at least one of a mean opinion score (MOS) index and a quality of service (QoS) index.

[0035] In a second aspect, the present disclosure provides a communication network optimization device, the device comprising:

[0036] A network status unit, used to select a transmission path for communication data and obtain real-time network status information of the transmission path;

[0037] A terminal capability unit, used to determine a receiving terminal of communication data and obtain real-time terminal capability information of the receiving terminal;

[0038] The configuration parameter unit is connected to the network status unit and the terminal capability unit, and is used to optimize the encoding configuration parameters of the communication data according to the real-time network status information and the real-time terminal capability information.

[0039] In a third aspect, the present disclosure provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the communication network optimization method as described above is implemented.

[0040] The present disclosure provides a communication network optimization method, device and medium. By selecting a transmission path for communication data and determining a receiving terminal for the communication data, the network status of the transmission path and the terminal capability of the receiving terminal are comprehensively considered to select the optimal coding configuration parameters for the communication data, so that the communication network can operate in the optimal coding state in real time, thereby improving the communication data transmission quality, such as the call quality of a private number, and improving the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flow chart of a communication network optimization method according to an embodiment of the present disclosure;

[0042] Figure 2 It is a structural diagram of a communication network optimization device according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the technical solution of the present disclosure, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0044] It should be understood that the specific embodiments and drawings described herein are only used to explain the present disclosure rather than to limit the present disclosure.

[0045] It can be understood that, in the absence of conflict, the various embodiments of the present disclosure and the various features in the embodiments can be combined with each other.

[0046] It can be understood that, for the convenience of description, the drawings of the present disclosure only show parts related to the present disclosure, while parts irrelevant to the present disclosure are not shown in the drawings.

[0047] It can be understood that each module or unit involved in the embodiments of the present disclosure may correspond to only one entity structure, or may be composed of multiple entity structures, or multiple modules or units may be integrated into one entity structure.

[0048] It will be understood that, in the absence of conflict, the functions and steps marked in the flowcharts and block diagrams of the present disclosure may occur in an order different from that marked in the drawings.

[0049] It is understood that the flowcharts and block diagrams of the present disclosure illustrate the possible architectures, functions, and operations of the systems, devices, equipment, and methods according to the various embodiments of the present disclosure. Each box in the flowchart or block diagram may represent a module, unit, program segment, or code, which contains executable instructions for implementing the specified functions. Moreover, each box or combination of boxes in the block diagram and flowchart may be implemented by a hardware-based device that implements the specified functions, or may be implemented by a combination of hardware and computer instructions.

[0050] It can be understood that the modules and units involved in the embodiments of the present disclosure can be implemented by software or hardware, for example, the modules and units can be located in a processor.

[0051] In order to facilitate the understanding of the present disclosure, some technical terms involved in the present disclosure are first explained:

[0052] Telephone communication network: The telephone communication network is a telecommunications network for interactive voice communication and open telephone services, referred to as the telephone network. It is a professional network with the largest telecommunications business volume and the widest service area. It is compatible with many other non-telephone service networks and is the basic form and foundation of the telecommunications network, including local telephone networks, long-distance telephone networks and international telephone networks. The telephone network adopts a telephone exchange method and is mainly composed of four parts: user terminal equipment that sends and receives telephone signals, switching equipment that performs circuit switching, lines connecting user terminals and switching equipment, and links between switching equipment.

[0053] VoIP: Voice over Internet Protocol, is a voice communication technology that uses the Internet Protocol (IP) to achieve voice calls and multimedia conferences, that is, to communicate over the Internet. Other informal names include IP telephony, Internet telephony, broadband telephony and broadband phone service.

[0054] Communication quality: The degree to which telecommunications services are effective in meeting users’ communication needs. The basic content includes four aspects: speed, accuracy and security, convenience, and reliability.

[0055] Signal Strength: A measure of the power level of the received signal, usually in dBm or asu (Arbitrary Strength Unit). A strong signal usually means better communication quality.

[0056] Signal-to-Noise Ratio (SNR): It indicates the ratio of signal power to noise power, expressed in dB. A high SNR indicates a purer signal, which is beneficial to improving call quality.

[0057] Packet Loss Rate: The proportion of data packets that fail to reach the receiving end during data transmission, expressed as a percentage. A high packet loss rate will directly lead to voice interruption or call interruption.

[0058] Latency: The time it takes for a data packet to be sent and received, measured in milliseconds (ms). Low latency is critical for real-time communication and can reduce the perceived delay on both sides of the call.

[0059] Jitter: The variation in the time interval between data packets arriving reflects the instability of network transmission. Too high jitter will affect voice synchronization and cause unclear calls.

[0060] Bit Error Rate (BER): The ratio of the number of bits with data transmission errors to the total number of bits transmitted, indicating the accuracy of data transmission. Low BER is the basis of high-quality communication.

[0061] Frequency Offset: It refers to the difference between the received signal frequency and the transmitted signal frequency. It is particularly important in mobile communications and affects the demodulation quality.

[0062] Noise Power Ratio: The ratio of the network background noise level to the signal strength, which affects the clarity of the call.

[0063] Serving Cell Quality: In a multi-cell environment, measuring the signal quality of the currently connected serving cell is crucial for handover decisions.

[0064] Neighbor Cell Interference Level: The degree of signal interference from neighboring cells. High interference may lead to deterioration of call quality.

[0065] Embodiment 1:

[0066] like Figure 1 As shown, the present disclosure provides a communication network optimization method, the method comprising:

[0067] S1. Select a transmission path for communication data and obtain real-time network status information of the transmission path;

[0068] S2. Determine a receiving terminal of the communication data and obtain real-time terminal capability information of the receiving terminal;

[0069] S3. Optimize the encoding configuration parameters of the communication data according to the real-time network status information and the real-time terminal capability information.

[0070] In this embodiment, the method selects the transmission path of the communication data and determines the receiving terminal of the communication data, comprehensively considers the network status of the transmission path and the terminal capability of the receiving terminal, selects the optimal coding configuration parameters for the communication data, and enables the communication network to operate in the optimal coding state in real time, thereby improving the communication data transmission quality and user experience. Figure 1 The method shown is applicable to Figure 2 The device shown.

[0071] Specifically, the method includes three main steps: the first step is to monitor the network status, which mainly includes dynamically selecting the optimal path to transmit data and analyzing the network status in real time based on the current optimal path; the second step is to evaluate the terminal capability, which mainly involves identifying and evaluating the receiving end; the third step is to dynamically adjust the coding parameters, which mainly involves obtaining the optimal coding configuration parameters based on the current network status and adjusting the coding parameters in real time, so as to always operate in the state of the optimal coding configuration to achieve the best communication experience. It is easy to understand that there is no necessary order between the first step and the second step.

[0072] More specifically, a specific demand background proposed in this embodiment is that a privacy number mobile communication network failure occurred in a certain operator, resulting in abnormal interruption of some privacy number outbound call channels or call delays. At that time, it mainly relied on relevant technical personnel and operation and maintenance personnel to support the platform overnight and manually adjust parameters based on manual experience to solve the platform communication failure. The privacy number is a telephone communication between business personnel (such as drivers, couriers, etc.) and users in a certain business scenario (such as taxis, food delivery, etc.), which meets the need to hide the real number. The privacy number and regular calls are registered to different business platforms (servers). If there is an abnormality in the privacy number channel call, the regular call is not affected.

[0073] In view of this, an intelligent solution is proposed to solve communication failures and improve the quality of communication networks. Through continuous learning and training of a large amount of historical call data, the optimal call configuration strategy is output to replace the problem of unreliable configuration of call parameters based on manual experience. Specifically, a communication network optimization method integrating DQN (Deep Q-Network) algorithm and dynamic coding is provided, and the communication quality optimization of mobile phone communication networks is specifically targeted. That is, an innovative mobile phone communication network optimization system and method is proposed, which integrates the DQN algorithm in the field of machine learning and the dynamic coding of the communication network. Through comprehensive and intelligent data collection of the call network and optimization of network resources, it can effectively improve the call clarity, reduce problems such as traffic jams and noise, and cope with high-concurrency call scenarios.

[0074] The system consists of several modules including signal quality monitoring module, intelligent routing selection module, dynamic coding adaptation module, network load balancing module, etc., forming a complete network optimization platform system.

[0075] The method comprises: performing data collection and preprocessing through a signal quality monitoring module, deploying sensors of the signal quality collection and monitoring module at key nodes of a communication network, and collecting network parameters during a call period, including data such as packet loss rate, delay, jitter, and bandwidth utilization rate; dynamically selecting an optimal path to transmit data by an intelligent routing selection module according to factors such as current network conditions, user locations, and service quality requirements, and the optimal path helps the encoder to dynamically adjust the encoding strategy according to the characteristics of the input data, thereby making the best decision within a limited time and ensuring the efficiency and real-time performance of the encoding process; based on the current optimal path, real-time analysis of network conditions and terminal capabilities is performed through a dynamic encoding adaptation module, which performs deep reinforcement learning on the encoding parameters of the communication network based on the reward function of a DQN algorithm to obtain optimal encoding configuration parameters, and during system operation, adjusts audio encoding parameters in real time according to the current network conditions, and operates in an optimal encoding state to achieve the best call experience; a network load balancing module ensures high availability and scalability of network services, and intelligently distributes incoming network traffic to multiple back-end servers to avoid single-point overload and optimize resource utilization efficiency.

[0076] Among them, the main innovation lies in the dynamic coding adaptation module. The dynamic coding adaptation module mainly analyzes the network status and terminal capabilities in real time, obtains the optimal coding configuration parameters based on the learning of the DQN algorithm, obtains the real-time coding status, and applies the DQN algorithm model to the privacy number communication network optimization scenario to achieve the purpose and effect of dynamic coding configuration of communication network parameters, solving the traditional error and cost problems of adjusting parameters based on manual experience. Based on the innovation of the dynamic coding adaptation module, it cooperates with the signal quality acquisition and monitoring module, the intelligent routing selection module and the network load balancing module to form a more complete communication network optimization method system.

[0077] In one implementation, S1, selecting a transmission path for communication data and acquiring real-time network status information of the transmission path, specifically includes:

[0078] Real-time monitoring of the real-time network status information of each communication line in the communication network;

[0079] Obtaining terminal location information and service quality requirement information of communication data to be transmitted;

[0080] According to the terminal location information and service quality demand information, combined with the real-time network status information of each communication line, the optimal transmission path for communication data is selected based on the intelligent routing selection strategy;

[0081] Get real-time network status information for optimal transmission paths.

[0082] In this embodiment, the communication network optimization system is a comprehensive system including a signal quality acquisition and monitoring module, an intelligent routing selection module, a dynamic coding adaptation module and a network load balancing module, wherein the dynamic coding adaptation module implements a dynamic coding adaptation mechanism through a DQN algorithm, which can adjust the communication network coding parameters in real time according to the current network conditions to ensure that the best call experience can be provided under any circumstances. The dynamic coding adaptation module is a key technology for optimizing communication quality and bandwidth utilization. Especially when the network conditions fluctuate greatly, it automatically adjusts the coding parameters of the communication network by analyzing the network conditions and terminal capabilities in real time to achieve the best media transmission effect. The detailed steps of the dynamic coding adaptation module include: the first step is to monitor the network status. The module continuously monitors the real-time status of the network, including bandwidth, delay, packet loss rate, etc. These parameters are obtained through the signal quality acquisition and monitoring module. When the network conditions change, the module can respond quickly. In order to obtain relevant information in this step, it is necessary to cooperate with the signal quality acquisition and monitoring module and the intelligent routing selection module, and it may also be necessary to cooperate with the network load balancing module.

[0083] In one embodiment, real-time monitoring of real-time network status information of each communication line in the communication network specifically includes:

[0084] Pre-deploy signal quality monitoring modules in key network nodes of the communication network;

[0085] The signal quality monitoring module collects the real-time network status information of each key network node in real time;

[0086] Determine the corresponding key network nodes on each communication line in the communication network, and obtain the real-time network status information of each corresponding key network node.

[0087] In this embodiment, the signal quality acquisition and monitoring module is a key component of the communication system and network equipment. Its main responsibility is to continuously evaluate and ensure the signal strength, stability and reliability during data transmission. The specific workflow of the module can be roughly divided into several steps: signal acquisition, signal preprocessing, signal analysis (including demodulation and decoding, quality parameter evaluation), threshold comparison and alarm, data recording and reporting. This module comprehensively monitors the key indicators that affect call quality, ensures the accuracy and real-time nature of the data, and monitors the signal strength, noise level and other parameters of each call line in real time. Detailed parameters include but are not limited to: signal strength, signal-to-noise ratio, packet loss rate, delay, jitter, bit error rate, frequency offset, noise power ratio, service cell quality, and neighboring area interference level. Through continuous monitoring and analysis of these parameters, basic data support is provided for the intelligent optimization algorithm, so that the system can respond and adjust quickly, thereby ensuring a high-quality experience in telephone communication.

[0088] The intelligent routing selection module is one of the key components to ensure high-quality communication. It dynamically selects the optimal path to transmit data based on factors such as the current network status, user location, and service quality requirements. Several commonly used strategies for intelligent routing selection include: bandwidth maximization strategy, which gives priority to the path with the largest available bandwidth. It is suitable for scenarios that require high-speed data transmission, such as high-definition video calls or instant transmission of large data files, to ensure a smooth and unimpeded communication experience; minimum delay strategy, which selects the path with the minimum delay time. For real-time interactive applications such as VoIP calls, low latency is the key to ensuring call continuity and real-time performance; minimum packet loss rate strategy, which gives priority to paths with low packet loss rates. In an unstable network environment, reducing packet loss can effectively improve call quality and reduce voice interruptions; mixed cost optimization strategy, which combines multiple factors such as path cost (such as transmission costs), bandwidth, delay, and packet loss rate, and uses preset weights to optimize the routing. Recalculate the comprehensive score and select the route with the highest comprehensive cost-effectiveness, which is suitable for scenarios that need to balance cost and service quality; predictive path selection, using algorithms to predict future network conditions, avoid expected congested areas or faulty nodes in advance, and select the path that is expected to perform best in the future; dynamic rerouting mechanism, continuously monitor network conditions during data transmission, and immediately trigger re-evaluation and switch to the optimal alternative path once the current path performance degrades, to ensure that call continuity is not affected; geo-fencing and load balancing, for high-density user areas, divide management areas through geo-fencing technology, and intelligently allocate user connections according to the load of base stations in each area, to avoid overloading of a single base station, and to achieve effective use of network resources. These strategies can be applied individually or in combination according to actual conditions to flexibly respond to complex and changing network environments and ensure call quality.

[0089] The network load balancing module is an important component to ensure high availability and scalability of network services. It intelligently distributes incoming network traffic to multiple backend servers to avoid single-point overload and optimize resource utilization efficiency. Network load balancing is a strategy for the operating status of the entire communication system, which may affect the results of intelligent routing selection and thus affect the results of dynamic encoding. The working principle of the network load balancing module includes: 1) Traffic distribution strategy, which consists of the following steps: Round Robin, the simplest distribution method, which distributes requests to each backend server in turn to ensure that the load of each server is roughly equal; Least Connections, which distributes new requests to the server with the least number of active connections, which is suitable for long-connection services and can more efficiently utilize server resources; Hash, which performs hash calculations based on the source IP address, URL or other request attributes, and always directs requests with the same characteristics to the same server to ensure session consistency; Weighted distribution, which assigns a weight value to the server according to its actual processing capacity. Traffic distribution is based on server weights to adapt to servers with different performance. 2) Health check: Send detection requests to the backend server regularly, and judge whether the server is healthy based on the response. If a server is found to be abnormal, it will be automatically removed from the service list until it is restored to a healthy state and then re-added to ensure the stability of the overall service. 3) Session persistence: In applications that need to maintain user session status, the load balancer can ensure that requests from the same user are forwarded to the same server for processing, ensuring the consistency of application logic. 4) Dynamic expansion and contraction: With the automated deployment tools of the cloud platform, the number of backend servers can be dynamically increased or decreased according to actual traffic, achieving elastic scaling to cope with sudden traffic peaks. 5) SSL (Secure Sockets Layer) offloading: SSL encryption and decryption are processed at the load balancing layer to reduce the computing burden of the backend server, improve security and processing speed. 6) Intelligent routing: Based on information such as geographic location, user type, content type, etc., requests are routed to the most appropriate server cluster to optimize user experience. 7) Monitoring and alarm: Real-time monitoring of traffic distribution, server health status and system performance indicators, and alarms are triggered immediately in case of abnormalities, facilitating timely intervention by the operation and maintenance team. Through the above mechanism, the network load balancing module realizes efficient and stable management of network traffic and is the key infrastructure for large-scale distributed systems and services.

[0090] In one embodiment, wherein:

[0091] The real-time network status information includes at least one of signal strength, signal-to-noise ratio, packet loss rate, delay, jitter, bit error rate, frequency offset, noise power ratio, service cell quality, and neighboring cell interference level;

[0092] The intelligent routing selection strategy includes at least one of a bandwidth maximization strategy, a minimum delay strategy, a packet loss rate minimum strategy, a hybrid cost optimization strategy, a predictive path selection strategy, a dynamic rerouting mechanism, a geographic fence and a load balancing strategy;

[0093] The real-time terminal capability information includes at least one of the processing capability, display limitation, current power status, and current scene mode of the receiving terminal.

[0094] In this embodiment, the detailed steps of the dynamic coding adaptation module also include: the second step is to evaluate the terminal capability, specifically including identifying and evaluating the processing capability (such as CPU speed, memory size) and display limitations (resolution, frame rate) of the receiving end, as well as the current power status, to ensure that the coding parameters match the terminal, to avoid resource waste or user experience degradation. The terminal refers to the user terminal / communication terminal. There are many ways to obtain the resources and data of the terminal, such as: the terminal device can regularly measure the received signal strength and report this information to the network; the terminal device can estimate the channel status and feed this information back to the base station; the terminal device will report its hardware and software capabilities when connected to the network, such as the supported frequency band, modulation mode, maximum data rate, etc., which helps the network select the appropriate configuration and service for the terminal; the network can infer the resource usage of the terminal by analyzing the traffic data of the user plane, such as data throughput, active time period, etc.; the network can test the response time and packet loss rate of the terminal by sending a detection packet, thereby inferring the network performance and resource status of the terminal.

[0095] In one implementation, S3, optimizing the encoding configuration parameters of the communication data according to the real-time network status information and the real-time terminal capability information, specifically includes:

[0096] The real-time network status information and the real-time terminal capability information are input into a pre-trained coding parameter dynamic adjustment model based on a deep Q network DQN algorithm, so that the coding parameter dynamic adjustment model outputs the optimal coding configuration parameters of the communication data, and the coding configuration parameters include at least one of bit rate, sampling rate, resolution and frame rate, coding algorithm, error recovery mechanism, multiplexed transmission, and transmission protocol.

[0097] In this embodiment, the detailed steps of the dynamic coding adaptation module also include: the third step is to dynamically adjust the coding parameters based on the DQN algorithm, including deep reinforcement learning of the coding parameters of the communication network according to the reward function, and adjusting the coding parameters in real time according to the current network conditions. The parameters output by the algorithm are the latest coding parameters of the communication network after learning and optimization, and can be used as the data basis for subsequent continuous learning, so that the communication network always operates in the state of the optimal coding configuration to achieve the best call experience. The parameters and control schemes involved include: 1) Bit rate, which is the amount of data transmitted per second and determines the speed of data transmission. A higher bit rate can transmit more data, but it also requires greater bandwidth and support. For example, audio bit rate is the amount of audio data transmitted per second and determines the quality of the audio. A higher bit rate can provide higher quality audio. According to the real-time changes in network bandwidth, the bit rate is dynamically increased or decreased to achieve bit rate adaptation. When the network conditions are good, the bit rate is increased to obtain higher quality audio and video. When the network is congested, the bit rate is reduced to ensure smooth communication. 2) Audio sampling rate, such as the number of audio samples collected per second, determines the frequency response range of the audio. A higher sampling rate can capture higher frequency sounds, but it also increases the amount of data. 3) Resolution and frame rate. For video communication, the resolution and frame rate of the video are automatically adjusted according to the network conditions and the terminal display capabilities. For example, in Reduce resolution and frame rate in low-bandwidth environments to reduce data transmission volume; 4) Coding algorithm / encoder, the algorithm used for data compression and transmission, select the most suitable encoder (such as H.264, H.265, Opus), as well as their profiles and levels according to different scenarios and requirements to balance compression efficiency and encoding quality; 5) Error recovery mechanism, in the case of high packet loss, adopt forward error correction (FEC), redundant data packets or fast retransmission mechanisms to enhance the data's ability to resist packet loss, improve the reliability and stability of data transmission, and ensure communication continuity; 6) Multiplexing, merge multiple data streams into a single composite signal for transmission in the same channel, improve channel utilization, and optimize bandwidth management; 7) Transmission protocol: The network protocol used for data transmission, such as TCP, UDP, etc., determines the mode and reliability of data transmission. TCP provides reliable transmission, but with higher latency. UDP provides low-latency transmission, but with lower reliability. These coding parameters play an important role in different application scenarios. By configuring these parameters reasonably, the performance of the communication network can be optimized and the quality and efficiency of data transmission can be improved.

[0098] In one embodiment, before optimizing the coding configuration parameters of the communication data, the method further includes training a coding parameter dynamic adjustment model based on a DQN algorithm, specifically including:

[0099] Define the deep neural network approximate Q function Q(s,a;θ) and experience replay mechanism E(s,a,r,s′) of the DQN algorithm, where s represents the state space, a represents the action space, θ represents the parameters of the deep neural network, r represents the reward function, and s′ represents the new state in the s training iteration;

[0100] The deep neural network is copied into a target network and a prediction network, and the loss function L(θ)=E[(r+γ·max(a′,Q(s′,a′;θ -1 ))-Q(s,a;θ)) 2 ], where E represents the expectation of the state-action-reward-new state quadruple obtained from experience playback, γ represents the discount factor, max(a′, Q(s′, a′; θ -1 )) is the maximum expected return given by the target network at the next state s′, θ -1 represents the parameter of the target network's delayed copy of the parameter θ of the prediction network, and Q(s,a;θ) represents the prediction result of the prediction network;

[0101] Collect training samples, input the real-time network status information and real-time terminal capability information in the training samples into s, the coding configuration parameters into a, and the communication quality evaluation index into r, and use the Adam optimizer to optimize L(θ) to obtain a dynamic adjustment model of coding parameters that meets the preset training end conditions.

[0102] In this embodiment, the communication network optimization method integrating the DQN algorithm and dynamic coding includes: a reward function based on the DQN algorithm, deep reinforcement learning of the coding parameters of the communication network according to the reward function, real-time adjustment of the coding parameters according to the current network conditions, obtaining the optimal coding configuration parameters, and always running in the state of the optimal coding configuration to achieve the best call experience.

[0103] Reinforcement Learning (RL) is a very suitable technical direction for machine learning model construction. Reinforcement learning allows the model to learn the optimal strategy by interacting with the environment without explicitly providing correct or incorrect labels, which is highly consistent with telephone communication application scenarios. Reinforcement learning algorithms (such as Deep Q-Network, DQN) are used as the core because they can autonomously learn and make decisions in uncertain environments, and are suitable for solving dynamic coding adaptation problems. DQN combines Q-learning with deep neural networks, can process high-dimensional input spaces (such as complex network state features), and learn the value mapping from state to action, showing superior performance in many sequential decision-making problems.

[0104] Here are some key steps and details of DQN implementation:

[0105] 1) Environment modeling: define the state space and action space of the environment. In the scenario of dynamic coding adaptation, the state space may include indicators such as the bandwidth, delay, and packet loss rate of the current network, while the action space corresponds to different coding rates or quality level selections. Including: state space definition, quantifying the network state (such as bandwidth, delay, packet loss rate, etc.) as the input state of the model; action space design, the action space is defined as a set of optional coding rates, and the model needs to learn to select the most appropriate coding rate under a given state; applicability, suitable for dynamic coding adaptation scenarios, can directly output the optimal coding rate selection based on the network state; reward mechanism, set the reward function based on call quality evaluation indicators (such as MOS score (Mean Opinion Score, used to estimate the quality of human communication in voice communication), packet loss rate, QoS (Quality of Service), etc.), and guide the model to optimize long-term call quality.

[0106] 2) Q-Network structure design: DQN uses a deep neural network (NN) to approximate the Q function, expressed as: Q(state s, action a; network weight parameter θ), where: s represents the state, a represents the action, θ represents the network weight parameter, the network input is the current environment state, and the output is the expected reward value for each possible action.

[0107] 3) Experience Replay: In order to avoid learning bias caused by continuous similar states, DQN introduces an experience replay mechanism, which is expressed as: E(state s, action a, reward r, new state s'). After each decision, the tuple of E(s, a, r, s') is stored in the experience pool. During learning, samples are randomly extracted from the experience pool for batch update, which can break the correlation between data and improve learning efficiency and stability.

[0108] 4) Fixed Q-targets: In order to stabilize the learning process, DQN uses two networks with the same network structure but independent parameters. One is the target network, which is used to calculate the target value Q (state s, action a; network weight parameter θ), which is obtained by combining the Bellman equation; the other is the evaluation network, which is used to generate the current strategy. The prediction network periodically (for example, every certain number of steps) copies the parameters of the target network Q to reduce target fluctuations and promote more stable learning.

[0109] 5) Loss function and optimization: The loss function is defined by comparing the difference between the predicted Q value (the current strategy given by the prediction network) and the target Q value (given by the target network combined with the Bellman equation), and the Adam optimizer is used to minimize this loss, thereby updating the weight parameters of the prediction network. The mathematical expression is:

[0110] L(θ)=E[(r+γ·max(a', Q(s',a';θ -1 ))-Q(s,a;θ)) 2 ]

[0111] Where: L(θ) is the loss function with respect to the network parameters (θ); E represents the expectation of a state-action-reward-new-state quadruple uniformly sampled from the experience replay pool; r is the reward immediately after taking action a; γ is the discount factor used to measure the current value of future rewards; max(a', Q(s', a'; θ -1 )) is the maximum expected return given by the target network at the next state (s'), θ -1 Represents the parameters of the target network, which is a delayed copy of the prediction network parameters θ, used to calculate the target Q value and increase the stability of learning; Q(s,a;θ) is the prediction network's Q value prediction for the current state s and the selected action a.

[0112] In DQN, the combination of the target network and the Bellman equation is a key technology to ensure learning stability and improve performance. The detailed process includes: 1) The Bellman equation is a recursive relationship used to describe the optimal value function in reinforcement learning. For the Q-value function, the Bellman equation can be expressed as Q(s,a)=r+γ·max(a',Q(s',a')); 2) The target network is a fixed network used to provide the target Q value, thereby avoiding the instability of Q-value estimation. In DQN, the parameters of the target network are usually copied from the evaluation network (i.e., the network currently being trained) every certain number of steps (such as every 10,000 steps); 3) Combined with the target Network and Bellman equation: collect experience, perform actions in the environment, collect states (s), actions (a), rewards (r), and new states (s'); calculate immediate rewards: get immediate rewards (r); use the target network to calculate the maximum Q value of the new state, and use the target network to calculate the maximum Q value under the new state (s'): max(a', Q(s',a')); calculate the target Q value, according to the Bellman equation, calculate the target Q value y = r + γ max(a', Q(s',a')); update the evaluation network, use the difference between the target Q value (y) and the Q value calculated by the evaluation target network, and update the parameters of the evaluation network through back propagation.

[0113] The pseudo code expression of the loss function is as follows:

[0114]

[0115]

[0116] #D represents the experience replay pool, and U(D) represents uniform sampling from D

[0117] #gamma is the discount factor, theta and theta^- represent the parameters of the online network and the target network respectively

[0118] The loss function optimization process usually involves using an optimization algorithm such as Adam to minimize this loss function, thereby updating the weight parameter (θ) of the prediction network so that the predicted Q value gradually approaches the ideal value, that is, performing the following update steps through the gradient descent method. The mathematical expression of the optimization process is: θ t+1 =θ t -α▽ θ L(θ), where α is the learning rate, which controls the step size of parameter updates.

[0119] The Adam optimizer (Adaptive Moment Estimation) is an efficient gradient descent algorithm that combines the ideas of RMSProp (Root Mean Square Propagation) and momentum to provide adaptive learning rates for different parameters. The following are the key steps performed by the Adam optimizer:

[0120] 1) Initialization: First, for each parameter θ i Initialize a first-order moment estimate m i (momentum term, initially 0) and a second-order moment estimate v i (variance term, initial value is 0), and set the hyperparameter β 1 (usually 0.9) and β 2 (usually 0.99) and a small constant ε (to prevent division by zero, usually 10 -8 ).

[0121] 2) Gradient calculation: At each iteration, the loss function is calculated with respect to the parameter θ i The gradient g i .

[0122] 3) Momentum update: Update the first-order moment estimate m i and the second-order moment estimate v i , using exponentially weighted average to consider historical gradients:

[0123] m i =β 1 m i-1 +(1-β 1)(g i )

[0124] v i =β 2 v i-1 +(1-β 2 )(g i ) 2

[0125] Here, m i Equivalent to the momentum term, accumulating the past gradient direction; v i The square of the gradient is accumulated and used to adjust the scale of the learning rate.

[0126] 4) Bias correction: Since mi and vi are both set to 0 during initialization, they will be biased in the early iterations, so bias correction is performed on them:

[0127]

[0128] Where t is the current iteration number.

[0129] 5) Parameter update: Use the corrected momentum term and variance term to update the parameters:

[0130]

[0131] Among them, α is the global learning rate, which controls the step size of the update.

[0132] The Adam optimizer automatically adjusts the learning rate of each parameter through the above steps. It is particularly suitable for optimization problems with sparse gradients or non-stationary objective functions. It can accelerate convergence and provide more stable performance. When using the Adam optimizer, in order to facilitate more accurate calculation results, there are several important points to note: learning rate selection. Although Adam can adaptively adjust the learning rate of each parameter, the selection of the global learning rate α is still critical. Usually a smaller default value, such as 10 -3 is a good starting point, but further fine-tuning may be required depending on the specific task; β1 and β2 settings. Adam's performance is sensitive to the hyperparameters β1 (momentum term) and β2 (variance term). By default, β1 = 0.9 and β2 = 0.99 work well, but some complex models or specific tasks may require adjustment of these values; bias correction, make sure the implementation includes |m i | and |v i|Bias correction, especially in the early iterations, which can avoid the deviation caused by initialization to 0; warm-up phase, for some models, especially large or complex models, starting with a smaller learning rate or gradually increasing the learning rate (warm-up strategy) can improve training stability; decay, during the training process, sometimes the global learning rate α is gradually reduced (learning rate decay), which helps to find a more refined optimal solution in the later stages of training; memory consumption, Adam requires more memory than other optimizers to store the first-order moment and second-order moment estimates of each parameter, which may become a consideration in large-scale models or resource-constrained environments; NaN problem, if the loss is found to become NaN, check whether there is a gradient explosion problem, you may need to adjust the learning rate, clip the gradient, or check whether the activation function is appropriate.

[0133] 6) Exploration and exploitation based on the ε-greedy strategy: At each decision step, the ε-greedy strategy is used to balance exploration and exploitation. In the early stage, actions are randomly selected with a certain probability ε to explore the environment. The rest of the time, the action with the highest expected return under the current strategy is selected. As learning progresses, the value of ε usually decays from a higher initial value to a smaller value to reduce the proportion of exploration and increase the proportion of decisions made according to the current strategy, that is, to utilize the learned knowledge. First, for the current state (s), among all optional actions (a), the expected return value (Q(s,a)) of each action is calculated according to the prediction network (or current strategy). Secondly, select an action and randomly select any action with probability ε to ensure exploration of unknown state-action combinations. Select the action with the largest Q(s,a) value with probability 1-ε, that is, greedy selection, to ensure the use of the best known strategy. In mathematical expression, assuming that there are N optional actions, select action a i The probability of is expressed mathematically as:

[0134]

[0135] The Adam optimization algorithm is used to continuously optimize the loss function to minimize the function. The pseudo code expression of the ε-greedy strategy rule is as follows:

[0136] def calculate_probability(action_i,Q_values,epsilon,N):

[0137] # Initialize action probability to 0

[0138] probability=0

[0139] #Check whether the current action is the action with the largest Q value

[0140] if action_i==argmax(Q_values):

[0141] #If yes, then apply the corresponding probability calculation formula

[0142] probability=epsilon / N+(1-epsilon)

[0143] else:

[0144] #Otherwise, apply the default exploration probability

[0145] probability = epsilon / N

[0146] return probability

[0147] Therefore, the DQN algorithm can continuously learn relevant parameter configuration strategies in a dynamic communication network environment and continuously optimize its decision-making strategy to achieve efficient and adaptive coding rate adjustment.

[0148] The process of performing real-time coding adaptation includes: 1) State observation: real-time monitoring and input of the network status of the current call into the machine learning model. 2) Decision making: the model predicts and selects the optimal coding rate based on the current status. 3) Coding adjustment: the dynamic encoder adjusts the coding parameters in real time according to the model output to achieve the best balance between bandwidth and quality. 4) Feedback learning: collect call quality feedback after each adaptation to continuously optimize the decision-making ability of the model.

[0149] Through the above steps, the benefits of the DQN algorithm in dynamic coding include: using a large amount of historical call data to train the model offline; ensuring that the model has a certain degree of interpretability to facilitate understanding of its decision logic; while ensuring the accuracy of parameter adaptation, optimizing the model structure and algorithm, reducing computing resource consumption, and improving the overall efficiency of the system; implementing this dynamic coding adaptation solution is expected to bring significant improvements in call quality, especially when network conditions are poor, it can effectively reduce abnormal call interruptions and voice distortion or voice delay, and improve user call satisfaction. At the same time, through intelligent management of bandwidth resources, it can also reduce network congestion and improve the stability and efficiency of the entire communication system.

[0150] In one embodiment, after optimizing the coding configuration parameters of the communication data, the method further includes collecting new training samples to retrain the coding parameter dynamic adjustment model based on the DQN algorithm, specifically including:

[0151] Collecting communication quality evaluation indicators after optimizing the coding configuration parameters of communication data, and judging whether the coding parameter dynamic adjustment model based on the DQN algorithm needs to be trained again according to the newly collected communication quality evaluation indicators;

[0152] If so, the real-time network status information, real-time terminal capability information, coding configuration parameters and newly collected communication quality evaluation indicators are correspondingly combined into new training samples, and the coding parameter dynamic adjustment model based on the DQN algorithm is trained again.

[0153] In this embodiment, after the parameters are adjusted, a quality feedback process loop is entered: the effect of the current encoding parameter settings is evaluated through quality feedback from the receiving end (such as MOS score, QoS index, etc.), and further fine-tuning is performed based on the feedback results to form a closed-loop control and continuously optimize the communication quality. The model parameter data is fine-tuned, and the best configuration solution is found through continuous fine-tuning.

[0154] In one embodiment, the method further includes obtaining terminal information, specifically including:

[0155] Receiving a measurement report from the terminal, and obtaining the terminal location information according to the measurement report;

[0156] Receiving hardware and software capability information from a receiving terminal, and obtaining processing capability, display limitation, and current power status of the receiving terminal as part of the real-time terminal capability information based on the hardware and software capability information;

[0157] Detecting the current use state of the receiving terminal, obtaining the current scene mode of the receiving terminal as another part of the real-time terminal capability information according to the current use state, the current scene mode including at least one of a conference mode, a streaming mode, and a power saving mode, and obtaining the service quality requirement information of the communication data according to the current scene mode;

[0158] A communication quality evaluation index is received from a receiving terminal, where the communication quality evaluation index includes at least one of a mean opinion score (MOS) index and a quality of service (QoS) index.

[0159] In this embodiment, the dynamic adjustment of coding parameters and the quality feedback process are continuously executed, and finally an intelligent preset strategy of coding parameters is formed and output, providing a variety of preset coding strategies, such as conference mode, streaming mode, power saving mode and other different scenarios to meet different user communication needs. At the same time, during the operation of the system, the administrator or user is allowed to manually select or customize the strategy adjustment according to the actual situation and combined with the manual experience value. The main methods are:

[0160] 1) Choose according to needs: Identify the user's main needs. For example, if you need to conduct an efficient remote meeting, choose the conference mode. If you plan to watch online videos, the streaming mode is a better choice. If the device is low on power, turn on the power saving mode to help you continue to use the device.

[0161] 2) Device settings: Most devices allow users to manually switch between different modes. For example, in the settings of a smartphone or tablet, you can usually find options such as "power saving mode" and "performance mode". For conference mode and streaming mode, this can be achieved through the settings within the application, such as the "HD video" option in video conferencing software, or the "smooth mode" in a video player.

[0162] 3) Intelligent judgment: Some devices and software can automatically adjust the working mode according to the current application scenario. For example, when a video call is detected, it automatically switches to conference mode to optimize the audio and video experience.

[0163] Through the above mechanism, the dynamic coding adaptation module can ensure the best possible communication experience under various network conditions, while making rational use of network resources. It is an indispensable part of modern communication systems.

[0164] In summary, this embodiment uses a dynamic coding adaptation module to perform real-time analysis of network conditions and terminal capabilities. Based on the reward function of the DQN algorithm, the module performs deep reinforcement learning on the coding parameters of the mobile communication network to obtain the optimal coding configuration parameters. During system operation, the audio coding parameters are adjusted in real time according to the current network conditions, and the system operates in the optimal coding state to achieve the best call experience. The DQN algorithm is applied in the communication network optimization scenario to realize intelligent dynamic coding of the system, replacing the manual coding method that relies on human experience, so as to achieve dynamic optimization of the communication network and improve the user communication experience.

[0165] Embodiment 2:

[0166] like Figure 2 As shown, the present disclosure provides a communication network optimization device, the device comprising:

[0167] The network status unit 1 is used to select a transmission path for communication data and obtain real-time network status information of the transmission path;

[0168] Terminal capability unit 2, used to determine a receiving terminal of communication data and obtain real-time terminal capability information of the receiving terminal;

[0169] The configuration parameter unit 3 is connected to the network status unit 1 and the terminal capability unit 2, and is used to optimize the encoding configuration parameters of the communication data according to the real-time network status information and the real-time terminal capability information.

[0170] In one implementation, the network status unit 1 specifically includes:

[0171] A network monitoring subunit, used to monitor the real-time network status information of each communication line in the communication network in real time;

[0172] The terminal information subunit is used to obtain the terminal location information and service quality requirement information of the communication data to be transmitted;

[0173] An intelligent routing subunit, connected to the network monitoring subunit and the terminal information subunit, is used to select an optimal transmission path for communication data based on an intelligent routing selection strategy according to the terminal location information and service quality requirement information, combined with the real-time network status information of each communication line;

[0174] The path information subunit is connected to the intelligent routing subunit and is used to obtain real-time network status information of the optimal transmission path.

[0175] In one implementation, the network monitoring subunit is specifically configured to:

[0176] Pre-deploy signal quality monitoring modules in key network nodes of the communication network;

[0177] The signal quality monitoring module collects the real-time network status information of each key network node in real time;

[0178] Determine the corresponding key network nodes on each communication line in the communication network, and obtain the real-time network status information of each corresponding key network node.

[0179] In one embodiment, wherein:

[0180] The real-time network status information includes at least one of signal strength, signal-to-noise ratio, packet loss rate, delay, jitter, bit error rate, frequency offset, noise power ratio, service cell quality, and neighboring cell interference level;

[0181] The intelligent routing selection strategy includes at least one of a bandwidth maximization strategy, a minimum delay strategy, a packet loss rate minimum strategy, a hybrid cost optimization strategy, a predictive path selection strategy, a dynamic rerouting mechanism, a geographic fence and a load balancing strategy;

[0182] The real-time terminal capability information includes at least one of the processing capability, display limitation, current power status, and current scene mode of the receiving terminal.

[0183] In one implementation, the configuration parameter unit 3 is specifically configured to:

[0184] The real-time network status information and the real-time terminal capability information are input into a pre-trained coding parameter dynamic adjustment model based on a deep Q network DQN algorithm, so that the coding parameter dynamic adjustment model outputs the optimal coding configuration parameters of the communication data, and the coding configuration parameters include at least one of bit rate, sampling rate, resolution and frame rate, coding algorithm, error recovery mechanism, multiplexed transmission, and transmission protocol.

[0185] In one embodiment, the device further includes a model training unit connected to the configuration parameter unit 3, and configured to train a dynamic adjustment model of encoding parameters based on the DQN algorithm, specifically including:

[0186] Define subunits for defining the deep neural network approximate Q function Q(s,a;θ) and experience replay mechanism E(s,a,r,s′) of the DQN algorithm, where s represents the state space, a represents the action space, θ represents the parameters of the deep neural network, r represents the reward function, and s′ represents the new state in s training iteration;

[0187] The copy subunit is connected to the definition subunit and is used to copy the deep neural network into a target network and a prediction network, and define the loss function L(θ)=E[(r+γ·max(a′,Q(s′,a′;θ -1 ))-Q(s,a;θ)) 2 ], where E represents the expectation of the state-action-reward-new state quadruple obtained from experience playback, γ represents the discount factor, max(a′, Q(s′, a′; θ -1 )) is the maximum expected return given by the target network at the next state s′, θ -1 represents the parameter of the target network's delayed copy of the parameter θ of the prediction network, and Q(s,a;θ) represents the prediction result of the prediction network;

[0188] The training optimization subunit is connected to the copy subunit and is used to collect training samples, input the real-time network status information and real-time terminal capability information in the training samples into s, the coding configuration parameters into a, and the communication quality evaluation index into r, and use the Adam optimizer to optimize L(θ) to obtain a dynamic adjustment model of coding parameters that meets the preset training end conditions.

[0189] In one embodiment, the model training unit further includes a collection subunit, which is connected to the training optimization subunit and is used to collect new training samples to retrain the encoding parameter dynamic adjustment model based on the DQN algorithm, specifically for:

[0190] Collecting communication quality evaluation indicators after optimizing the coding configuration parameters of communication data, and judging whether the coding parameter dynamic adjustment model based on the DQN algorithm needs to be trained again according to the newly collected communication quality evaluation indicators;

[0191] If so, the real-time network status information, real-time terminal capability information, coding configuration parameters and newly collected communication quality evaluation indicators are correspondingly combined into new training samples, and the coding parameter dynamic adjustment model based on the DQN algorithm is trained again.

[0192] In one embodiment, the device further includes a terminal information subunit, which is used to obtain terminal information, specifically for:

[0193] Receiving a measurement report from the terminal, and obtaining the terminal location information according to the measurement report;

[0194] Receiving hardware and software capability information from a receiving terminal, and obtaining processing capability, display limitation, and current power status of the receiving terminal as part of the real-time terminal capability information based on the hardware and software capability information;

[0195] Detecting the current use state of the receiving terminal, obtaining the current scene mode of the receiving terminal as another part of the real-time terminal capability information according to the current use state, the current scene mode including at least one of a conference mode, a streaming mode, and a power saving mode, and obtaining the service quality requirement information of the communication data according to the current scene mode;

[0196] A communication quality evaluation index is received from a receiving terminal, where the communication quality evaluation index includes at least one of a mean opinion score (MOS) index and a quality of service (QoS) index.

[0197] Embodiment 3:

[0198] Embodiment 3 of the present disclosure provides a computer-readable storage medium, in which a computer program is stored. When the computer program is executed by a processor, the communication network optimization method as described in Embodiment 1 or the communication network optimization device as described in Embodiment 2 is implemented.

[0199] The computer-readable storage medium includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program elements or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable read only memory), flash memory or other memory technology, CD-ROM (Compact Disc Read-Only Memory), digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0200] In addition, the present disclosure may also provide a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor runs the computer program stored in the memory, the processor executes the communication network optimization method as described in Example 1. The computer device may be the communication network optimization device as described in Example 2.

[0201] The memory is connected to the processor, the memory may be a flash memory or a read-only memory or other memory, and the processor may be a central processing unit or a single-chip microcomputer.

[0202] Embodiments 1-3 of the present disclosure provide a communication network optimization method, device and medium, which select a transmission path for communication data and determine a receiving terminal for communication data, comprehensively consider the network status of the transmission path and the terminal capability of the receiving terminal, select the optimal coding configuration parameters for the communication data, and enable the communication network to operate in the optimal coding state in real time, thereby improving the communication data transmission quality, such as the call quality of a private number, and improving the user experience.

[0203] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present disclosure, but the present disclosure is not limited thereto. For those of ordinary skill in the art, various modifications and improvements can be made without departing from the spirit and substance of the present disclosure, and these modifications and improvements are also considered to be within the scope of protection of the present disclosure.

Claims

1. A communication network optimization method, characterized in that: The method comprises: Select a transmission path for communication data and obtain real-time network status information of the transmission path; Determine a receiving terminal of the communication data and obtain real-time terminal capability information of the receiving terminal; Optimize the encoding configuration parameters of communication data based on real-time network status information and real-time terminal capability information.

2. The method according to claim 1, characterized in that Select the transmission path for communication data and obtain real-time network status information of the transmission path, including: Real-time monitoring of the real-time network status information of each communication line in the communication network; Obtaining terminal location information and service quality requirement information of communication data to be transmitted; According to the terminal location information and service quality demand information, combined with the real-time network status information of each communication line, the optimal transmission path for communication data is selected based on the intelligent routing selection strategy; Get real-time network status information for optimal transmission paths.

3. The method according to claim 2, characterized in that Real-time monitoring of the real-time network status information of each communication line in the communication network, including: Pre-deploy signal quality monitoring modules in key network nodes of the communication network; The signal quality monitoring module collects the real-time network status information of each key network node in real time; Determine the corresponding key network nodes on each communication line in the communication network, and obtain the real-time network status information of each corresponding key network node.

4. The method according to claim 2, characterized in that: in: The real-time network status information includes at least one of signal strength, signal-to-noise ratio, packet loss rate, delay, jitter, bit error rate, frequency offset, noise power ratio, service cell quality, and neighboring cell interference level; The intelligent routing selection strategy includes at least one of a bandwidth maximization strategy, a minimum delay strategy, a packet loss rate minimum strategy, a hybrid cost optimization strategy, a predictive path selection strategy, a dynamic rerouting mechanism, a geographic fence and a load balancing strategy; The real-time terminal capability information includes at least one of the processing capability, display limitation, current power status, and current scene mode of the receiving terminal.

5. The method according to any one of claims 1 to 4, characterized in that: According to the real-time network status information and real-time terminal capability information, the encoding configuration parameters of the communication data are optimized, including: The real-time network status information and the real-time terminal capability information are input into a pre-trained coding parameter dynamic adjustment model based on a deep Q network DQN algorithm, so that the coding parameter dynamic adjustment model outputs the optimal coding configuration parameters of the communication data, and the coding configuration parameters include at least one of bit rate, sampling rate, resolution and frame rate, coding algorithm, error recovery mechanism, multiplexed transmission, and transmission protocol.

6. The method according to claim 5, characterized in that Before optimizing the coding configuration parameters of the communication data, the method further includes training a coding parameter dynamic adjustment model based on the DQN algorithm, specifically including: Define the deep neural network approximate Q function Q(s,a;θ) and experience replay mechanism E(s,a,r,s′) of the DQN algorithm, where s represents the state space, a represents the action space, θ represents the parameters of the deep neural network, r represents the reward function, and s′ represents the new state in the s training iteration; The deep neural network is copied into a target network and a prediction network, and the loss function L(θ)=E[(r+γ·max(a′,Q(s′,a′;θ -1 ))-Q(s,a;θ)) 2 ], where E represents the expectation of the state-action-reward-new state quadruple obtained from experience playback, γ represents the discount factor, max(a′, Q(s′, a′; θ -1 )) is the maximum expected return given by the target network at the next state s′, θ -1 represents the parameter of the target network's delayed copy of the parameter θ of the prediction network, and Q(s,a;θ) represents the prediction result of the prediction network; Collect training samples, input the real-time network status information and real-time terminal capability information in the training samples into s, the coding configuration parameters into a, and the communication quality evaluation index into r, and use the Adam optimizer to optimize L(θ) to obtain a dynamic adjustment model of coding parameters that meets the preset training end conditions.

7. The method according to claim 6, characterized in that After optimizing the coding configuration parameters of the communication data, the method further includes collecting new training samples to retrain the coding parameter dynamic adjustment model based on the DQN algorithm, specifically including: Collecting communication quality evaluation indicators after optimizing the coding configuration parameters of communication data, and judging whether the coding parameter dynamic adjustment model based on the DQN algorithm needs to be trained again according to the newly collected communication quality evaluation indicators; If so, the real-time network status information, real-time terminal capability information, coding configuration parameters and newly collected communication quality evaluation indicators are correspondingly combined into new training samples, and the coding parameter dynamic adjustment model based on the DQN algorithm is trained again.

8. The method according to claim 7, characterized in that The method further includes obtaining terminal information, specifically including: Receiving a measurement report from the terminal, and obtaining the terminal location information according to the measurement report; Receiving hardware and software capability information from a receiving terminal, and obtaining processing capability, display limitation, and current power status of the receiving terminal as part of the real-time terminal capability information based on the hardware and software capability information; Detecting the current use state of the receiving terminal, obtaining the current scene mode of the receiving terminal as another part of the real-time terminal capability information according to the current use state, the current scene mode including at least one of a conference mode, a streaming mode, and a power saving mode, and obtaining the service quality requirement information of the communication data according to the current scene mode; A communication quality evaluation index is received from a receiving terminal, where the communication quality evaluation index includes at least one of a mean opinion score (MOS) index and a quality of service (QoS) index.

9. A communication network optimization device, characterized in that: The device comprises: A network status unit, used to select a transmission path for communication data and obtain real-time network status information of the transmission path; A terminal capability unit, used to determine a receiving terminal of communication data and obtain real-time terminal capability information of the receiving terminal; The configuration parameter unit is connected to the network status unit and the terminal capability unit, and is used to optimize the encoding configuration parameters of the communication data according to the real-time network status information and the real-time terminal capability information.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the communication network optimization method according to any one of claims 1 to 8 is implemented.

Images