A gpu-based multi-client real-time spectrum monitoring system and method thereof

Abstract

The application discloses a GPU-based multi-client real-time spectrum monitoring system and a method thereof, and belongs to the technical field of spectrum analysis.The method comprises the following steps: acquiring the spectrum state, behavior information, host state and service quality requirement of a client; calculating the interference intensity coefficient, behavior state coefficient, host state coefficient and service quality coefficient of the client through a corresponding model; dynamically allocating the computing power of each client by using a power distribution model based on the above coefficients; and finally outputting the target refresh rate of each client by a refresh rate optimization model in combination with the allocated power and service quality coefficient.The application solves the problem that the resource allocation in the prior art is rigid and cannot adapt to the differentiated requirements of multi-clients by multi-dimensional state sensing and modeling, realizes intelligent, dynamic and efficient allocation of system computing resources, guarantees the service quality of multi-services, and improves the overall energy efficiency and stability of the system.

Description

Technical Field

[0001] This invention belongs to the field of spectrum analysis technology, and in particular relates to a GPU-based multi-client real-time spectrum monitoring system and method. Background Technology

[0002] With the rapid development of wireless communication technology, real-time monitoring and analysis of the electromagnetic environment has become increasingly important. Real-time spectrum monitoring systems based on graphics processing units (GPUs), leveraging their powerful parallel computing capabilities, can efficiently process high-speed data streams from devices such as software-defined radios (SDRs) and perform large-scale fast Fourier transforms (FFTs), thereby enabling real-time analysis of the broadband spectrum. This has become a crucial technological direction in the modern spectrum monitoring field. Such systems typically need to serve multiple clients; therefore, how to efficiently and fairly allocate system resources to meet the real-time requirements of different clients is a key issue.

[0003] Existing multi-client spectrum monitoring solutions primarily focus on using GPUs to replace CPUs for FFT calculations to improve processing speed, but their resource scheduling strategies are relatively simple and static. Common practices include using round-robin or fixed-priority scheduling to provide the same quality of service to all clients, or simply allocating resources coarsely based on the client's identity level. These methods fail to fully consider the client's real-time spectrum environment, the reliability of its historical behavior, the real-time load status of the host system, and the differentiated quality of service requirements of different services.

[0004] Therefore, the existing technology has the following main drawbacks: First, the system resource allocation is rigid and cannot be dynamically adjusted according to the real-time changes in the client's spectrum interference level, which may result in important monitoring tasks not receiving sufficient resources when interference occurs; second, there is a lack of comprehensive assessment of the credibility of client behavior, making it difficult to prevent the abuse of system resources by malicious requests or scanning behavior; finally, the host's own power consumption, temperature and other hard constraints are not integrated into the modeling of the utilization of resources such as GPU, video memory and network, making it impossible to make intelligent global load reduction decisions when the system faces overheating or overload pressure, thus affecting the overall stability and energy efficiency of the system. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a GPU-based multi-client real-time spectrum monitoring system and method, which solves the aforementioned problems.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a GPU-based multi-client real-time spectrum monitoring method, comprising the following steps:

[0007] Obtain the client's spectrum status information, client behavior information, host status information, and service quality requirements information;

[0008] Based on the aforementioned spectrum state information, the client interference intensity coefficient is calculated using the client interference intensity model.

[0009] Based on the client's behavioral information, the client's behavioral state coefficients are calculated using the client's behavioral state model.

[0010] Based on the host's state information, host state coefficients are calculated using a host state model.

[0011] Based on the service quality requirements information of the aforementioned business, the business quality coefficient is calculated through the business quality model.

[0012] Based on the client interference intensity coefficient, the client behavior state coefficient, and the host state coefficient, the computing power allocated to each client is dynamically calculated and output through a computing power allocation model.

[0013] Based on the computing power allocated to the client, the client's baseline refresh rate, and the service quality coefficient, the target refresh rate of the client is calculated and output through a refresh rate optimization model.

[0014] Based on the above technical solutions, the present invention also provides the following optional technical solutions:

[0015] A further technical solution: The refresh rate optimization model is expressed as follows:

[0016]

[0017] in, Indicates the first The target refresh rate for each client. Indicates the first Computing power per client Indicates the first The baseline refresh rate for each client. Indicates the first The service quality coefficient of each client This represents the threshold for the business quality coefficient. This represents the business quality adjustment coefficient. This represents the reference value for calculating power.

[0018] A further technical solution: The power allocation model is expressed as follows:

[0019]

[0020] in, Indicates the first Computing power per client This represents the total computing power that the system can allocate. Indicates the first Interference intensity coefficient of each client, Indicates the first Each client behavior state coefficient Represents the host status coefficient. This indicates the total number of current clients.

[0021] Further technical solution: Based on the service quality requirement information of the aforementioned business, the steps for calculating the service quality coefficient through a service quality model are as follows:

[0022] The allowed delay time and average task duration are subjected to max-min normalization to obtain the delay time index and the task duration index.

[0023] A service quality model is constructed based on the latency index, the task duration index, and the allowable packet loss rate. The service quality model is expressed as follows:

[0024]

[0025] in, Indicates the business quality coefficient. Indicates the delay duration index, This represents the packet loss rate index. This indicates the task duration index. Represents the weight coefficient and The The Furthermore, the higher the value, the greater the client's business demand, and the more priority should be given to ensuring service.

[0026] Import the latency index, task duration index, and allowable packet loss rate of each client into the business quality model to obtain the business quality coefficient of each client.

[0027] A further technical solution: Based on the host's state information, the steps for calculating the host state coefficients using a host state model are as follows:

[0028] Based on the current total power consumption and temperature, a power consumption-temperature-pressure model is constructed to output the power consumption-temperature-pressure coefficient. The power consumption-temperature-pressure model is expressed as follows:

[0029]

[0030] in, This represents the power consumption-temperature-pressure coefficient. This indicates the current total power consumption. Indicates the host temperature. Indicates the maximum allowable power consumption. Indicates the maximum permissible temperature. Represents the weight coefficient and The ;

[0031] A host state model is constructed based on GPU utilization, memory usage, and network bandwidth usage under the power consumption-temperature-pressure coefficient. The host state model is represented as follows:

[0032]

[0033] in, Represents the host status coefficient. This represents the power consumption-temperature-pressure coefficient. Indicates GPU utilization. This indicates the video memory usage rate. Indicates network bandwidth utilization, the The larger the value, the more abundant the host resources.

[0034] A further technical solution: Based on the client's behavioral information, the steps for calculating the client's behavioral state coefficients using a client behavioral state model are as follows:

[0035] The request frequency index and the historical interaction count index are obtained by performing max-min normalization on the request frequency and the historical interaction count index.

[0036] A client behavior state model is constructed based on the request frequency index, the historical interaction count index, and the percentage of permission levels. This client behavior state model is represented as follows:

[0037]

[0038] in, This represents the customer behavior status coefficient. This indicates the request frequency index. This represents the percentage of historical interactions. Indicates the percentage of different permission levels. Represents the weight coefficient and The The Furthermore, the higher the value, the higher the client's credibility, requiring additional security services.

[0039] Import the request frequency index, historical interaction count index, and permission level ratio of each client into the client behavior state model to output the behavior state coefficients of each client.

[0040] A further technical solution: Based on the aforementioned spectrum state information, the steps for calculating the client interference intensity coefficient using the client interference intensity model are as follows:

[0041] The noise floor rise (the absolute difference between the current noise floor and the historical average noise floor), the peak difference (the difference between the highest peak value and the overall average peak value), and the excess time length (the length of time the signal power exceeds the standard value of the signal power) are processed by maximum-minimum normalization to obtain the noise floor rise index, the peak difference index, and the excess time length index.

[0042] A client interference intensity model is constructed based on the noise floor rise index, the peak spectral difference index, and the excess time length index. The client interference intensity model is expressed as follows:

[0043]

[0044] in, Indicates the client interference strength coefficient. Indicates the noise floor rise index. Indicates the peak difference index of the spectrum. This represents the excess time length index. Represents the weight coefficient and The The Furthermore, the larger the value, the more significant the client-side interference;

[0045] The noise floor rise index, peak spectral difference index, and excess time length index of each client are imported into the client interference intensity model to output the interference intensity coefficient of each client.

[0046] A GPU-based multi-client real-time spectrum monitoring system employs the aforementioned GPU-based multi-client real-time spectrum monitoring method.

[0047] Beneficial effects

[0048] This invention provides a GPU-based multi-client real-time spectrum monitoring system and method, which has the following advantages compared with the prior art:

[0049] 1. By constructing four evaluation models—client interference intensity, behavioral status, host status, and service quality—the system can comprehensively perceive internal and external status and dynamically allocate computing power to the most needed clients on demand and according to priority, which greatly improves the system's resource utilization efficiency and service fairness.

[0050] 2. This invention introduces a host status monitoring model based on power consumption and temperature stress, enabling the system to keenly perceive its own operating status. When faced with risks of overheating, excessive power consumption, or resource saturation, the system can automatically and smoothly reduce the global load, effectively preventing hardware overload and improving the system's robustness and reliability under harsh conditions such as high temperature and high load.

[0051] 3. This invention can identify the priority of different services through a service quality model, and combined with a refresh rate optimization mechanism, the system can provide high refresh rate and low latency high-quality services for high-requirement services (such as real-time control and emergency monitoring), while appropriately restricting the resource usage of low-priority or non-critical services, thereby achieving refined service quality management and meeting the needs of diverse application scenarios.

[0052] 4. The client behavior state model can effectively identify abnormal behaviors such as high-frequency requests and malicious scanning. By reducing their resource allocation priority, it can suppress resource abuse, protect the system from attacks or misoperations, and provide a safe and reliable operating environment for legitimate and important business connections. Attached Figure Description

[0053] Figure 1 This is a schematic diagram of the process of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0056] Please see Figure 1 The present invention provides a GPU-based multi-client real-time spectrum monitoring method, comprising the following steps:

[0057] Obtain the client's spectrum status information, client behavior information, host status information, and service quality requirements information;

[0058] Based on the aforementioned spectrum state information, the client interference intensity coefficient is calculated using the client interference intensity model.

[0059] Based on the client's behavioral information, the client's behavioral state coefficients are calculated using the client's behavioral state model.

[0060] Based on the host's state information, host state coefficients are calculated using a host state model.

[0061] Based on the service quality requirements information of the aforementioned business, the business quality coefficient is calculated through the business quality model.

[0062] Based on the client interference intensity coefficient, the client behavior state coefficient, and the host state coefficient, the computing power allocated to each client is dynamically calculated and output through a computing power allocation model.

[0063] Based on the computing power allocated to the client, the client's baseline refresh rate, and the service quality coefficient, the target refresh rate of the client is calculated and output through a refresh rate optimization model.

[0064] Specifically, the system constructs a dynamic resource allocation decision chain through real-time acquisition and modeling of four types of information sources. The interference intensity coefficient identifies incremental client resource demands in high-interference scenarios, the behavioral state coefficient suppresses resource consumption by low-trust clients, and the host state coefficient prevents system overload. The power allocation model integrates these three types of coefficients to allocate total computing power to each client on demand. Based on power allocation, the baseline refresh rate is dynamically adjusted in conjunction with the service quality coefficient, ensuring that high-priority services receive sufficient computing resources while increasing the frequency of spectrum data updates. This mechanism forms a four-tiered control system encompassing environmental interference response, abnormal behavior filtering, system load balancing, and service demand assurance.

[0065] Compared to existing technologies, traditional solutions allocate resources based solely on fixed rules, failing to establish a correlation model between environmental interference and resource demand, and thus unable to dynamically adjust power allocation. This solution maps spectral state changes to resource demand changes using interference intensity coefficients, enabling resource reallocation under interference scenarios. Existing technologies lack client behavior evaluation mechanisms; this solution constructs a client trustworthiness evaluation system using behavior state coefficients, effectively suppressing resource abuse. Traditional methods monitor GPU or network resources individually; this solution achieves multi-dimensional joint modeling of resource occupancy and physical state through host state coefficients, improving system stability.

[0066] Through the above technical solutions, this application realizes intelligent resource scheduling in multi-client scenarios. In sudden interference scenarios, high-interference clients can automatically obtain additional computing power to ensure the continuity of critical monitoring tasks; for clients with abnormally high frequency requests, the system reduces their computing power allocation to avoid resource crowding; when the host temperature or power consumption approaches the threshold, the global power allocation coefficient is automatically reduced to prevent hardware overload damage; the refresh rate of high-priority services is dynamically increased according to service quality requirements to ensure the effectiveness of real-time monitoring data.

[0067] Preferably, the spectrum status information includes noise floor rise (the absolute difference between the current noise floor and the historical average noise floor), spectrum peak difference (the difference between the highest peak value and the overall average peak value), and excess time length (the length of time the signal power exceeds the standard signal power value). The client behavior information includes request frequency (the number of requests in 24 hours), historical interaction count, and permission level percentage. The host status information includes current total power consumption, host temperature, GPU utilization, video memory usage, and network bandwidth usage. The service quality requirement information includes allowed latency, allowed data packet loss rate, and average task duration.

[0068] The noise floor rise value refers to the absolute difference between the current noise floor and the historical average noise floor. This can be calculated by collecting environmental noise data in real time and comparing it with a historical database, and is used to detect sudden interference events. The peak frequency difference value refers to the difference between the highest peak value and the overall average peak value. This can be calculated by extracting peak features using a spectrum analysis algorithm and then calculating the variance or standard deviation, and is used to identify abnormal signal occupancy behavior. The excess time length refers to the length of time the signal power exceeds the standard value. This can be calculated by setting a threshold and statistically analyzing the percentage of excess periods, and is used to assess the intensity of persistent interference. The request frequency refers to the number of requests within 24 hours. This can be calculated by logging and time window statistics, and is used to identify high-frequency abnormal access behavior. The historical interaction count refers to the number of historical communications between the client and the system. This can be calculated by cumulative counting or sliding window counting, and is used to assess client trustworthiness. The permission level percentage refers to the proportion of client permission levels allocated in the system. This can be calculated by querying the role and permission database, and is used for differentiated service priority allocation. The current total power consumption refers to the real-time power consumption of the host, which can be collected by the power management module or sensors, and is used to prevent system overload risks. Host temperature refers to the real-time temperature of the hardware device, which can be monitored using temperature sensors to determine heat dissipation pressure. GPU utilization refers to the proportion of graphics processor computing resources used, which can be obtained in real-time through driver interfaces to quantify the computing load. Video memory utilization refers to the proportion of video memory resources used, which can be statistically analyzed by the memory management module to assess storage resource consumption. Network bandwidth utilization refers to the proportion of network bandwidth used, which can be measured using network traffic monitoring tools to identify communication resource bottlenecks. Allowable latency refers to the maximum tolerable response time for the business, which can be set with thresholds through the Quality of Service (QoS) protocol to ensure real-time requirements. Allowable packet loss rate refers to the acceptable proportion of data loss for the business, which can be configured through communication protocol parameters to ensure data integrity. Average task duration refers to the average time required for business processing, which can be statistically analyzed through task scheduling logs to optimize resource allocation duration.

[0069] Specifically, the noise floor rise value in the spectrum status information quantifies environmental noise fluctuations, dynamically reflecting the impact of sudden interference on the system; the spectrum peak difference value identifies abnormal spectrum occupancy behavior by analyzing signal distribution characteristics; and the excess time length assesses the persistence of interference by statistically analyzing the duration of signals exceeding limits. The client behavior information, combining request frequency with historical interaction counts, effectively distinguishes between normal requests and malicious scanning behavior; the permission level ratio provides a basis for differentiated services for clients with different levels of trust. The host status information, with joint monitoring of total power consumption and temperature, prevents system throttling or crashes caused by hardware overheating; the ternary parameter system of GPU utilization, memory usage, and network bandwidth usage accurately quantifies the real-time consumption status of computing, storage, and communication resources. The latency and task duration parameters in the business service quality requirement information ensure that real-time tasks receive priority resource allocation; and the allowable data packet loss rate parameter provides data integrity assurance for critical businesses. The synergistic effect of these parameters overcomes the limitations of traditional single-dimensional monitoring, providing refined input for dynamic resource scheduling through multi-source data fusion.

[0070] Preferably, the step of calculating the client interference intensity coefficient based on the spectrum state information using the client interference intensity model is as follows:

[0071] The noise floor rise (the absolute difference between the current noise floor and the historical average noise floor), the peak difference (the difference between the highest peak value and the overall average peak value), and the excess time length (the length of time the signal power exceeds the standard value of the signal power) are processed by maximum-minimum normalization to obtain the noise floor rise index, the peak difference index, and the excess time length index.

[0072] A client interference intensity model is constructed based on the noise floor rise index, the peak spectral difference index, and the excess time length index. The client interference intensity model is expressed as follows:

[0073]

[0074] in, Indicates the client interference strength coefficient. Indicates the noise floor rise index. Indicates the peak difference index of the spectrum. This represents the excess time length index. Represents the weight coefficient and The The Furthermore, the larger the value, the more significant the client-side interference;

[0075] The noise floor rise index, peak spectral difference index, and excess time length index of each client are imported into the client interference intensity model to output the interference intensity coefficient of each client.

[0076] The noise floor rise refers to the absolute difference between the current noise floor and the historical average noise floor. Specifically, a sliding window algorithm can be used to calculate the historical average noise floor, and the absolute difference is taken after obtaining the current noise floor value through real-time sampling. This reflects the degree of abnormal fluctuations in environmental noise. The spectral peak difference refers to the difference between the highest peak value and the overall peak average. Specifically, peak statistics can be calculated after obtaining spectral data through Fast Fourier Transform, used to capture abrupt changes in spectral energy distribution. The excess time duration refers to the duration for which the signal power exceeds a preset standard value. This can be implemented using a threshold comparator combined with a timer, used to quantify the sustained impact of abnormal signals. Weighting coefficients. The specific settings can be determined by using preset empirical values ​​or by using the analytic hierarchy process. By constructing a judgment matrix, the relative importance weights of each indicator can be calculated, enabling the model to adapt to the interference assessment needs of different scenarios.

[0077] Specifically, this technical solution constructs a multi-dimensional interference assessment system. First, it normalizes the noise floor rise, spectral peak difference, and excess time length, transforming interference characteristics of different dimensions into comparable exponential forms. The noise floor rise index, by calculating the deviation between the current noise floor and the historical average, can effectively detect abnormal increases in environmental noise; for example, this index increases significantly during sudden electromagnetic interference events. The spectral peak difference index, by analyzing the distribution characteristics of spectral peaks, can identify abnormal energy concentration phenomena, such as spectral spikes caused by illegal signal transmissions. The excess time length index, by statistically analyzing the duration of the exceeding signal, can reflect the persistence of the interference signal, such as long-term co-channel interference sources. These three indices are weighted and combined to form an interference intensity coefficient, where the weighting coefficients can be dynamically adjusted according to the actual monitoring scenario. For example, in electromagnetically sensitive areas, the weighting coefficient for noise floor rise can be increased to 0.5, and in spectral congestion scenarios, the weighting coefficient for spectral peak difference can be set to 0.6. The final generated interference intensity coefficient serves as a quantitative indicator, providing data support for subsequent calculations of dynamic power allocation.

[0078] Compared to existing technologies, traditional methods judge interference levels based on a single-dimensional indicator (such as signal strength threshold), failing to distinguish between transient and persistent interference and neglecting changes in spectral characteristics. This proposed solution integrates three key features—noise floor, spectral peak distribution, and time dimension—to construct a multi-dimensional interference assessment model capable of simultaneously capturing the impact of both sudden interference events and persistent interference sources. For example, when short-term strong interference occurs, the spectral peak difference index responds rapidly; when long-term background noise pollution exists, the noise floor rise index and excess time length index continuously increase. This multi-dimensional quantitative assessment method significantly improves the sensitivity and accuracy of interference detection compared to existing technologies.

[0079] Through the above technical solution, this application can quantitatively assess the comprehensive interference level of the spectrum environment in which the client is located in real time, providing an accurate basis for dynamically adjusting the allocation of computing resources. When the interference intensity coefficient of a client exceeds a set threshold, the system can automatically increase its allocated computing power to ensure stable spectrum monitoring performance even when electromagnetic interference intensifies. For example, in an airport radar monitoring scenario, when sudden radio interference in the surrounding area causes the noise floor rise index and excess time length index to rise simultaneously, the system can prioritize ensuring the computing resources of the monitoring client to avoid monitoring interruption of critical air traffic control communication frequency bands.

[0080] Preferably, the step of calculating the client behavior state coefficients based on the client's behavior information using the client behavior state model is as follows:

[0081] The request frequency index and the historical interaction count index are obtained by performing max-min normalization on the request frequency and the historical interaction count index.

[0082] A client behavior state model is constructed based on the request frequency index, the historical interaction count index, and the percentage of permission levels. This client behavior state model is represented as follows:

[0083]

[0084] in, This represents the customer behavior status coefficient. This indicates the request frequency index. This represents the percentage of historical interactions. Indicates the percentage of different permission levels. Represents the weight coefficient and The The Furthermore, the higher the value, the higher the client's credibility, requiring additional security services.

[0085] Import the request frequency index, historical interaction count index, and permission level ratio of each client into the client behavior state model to output the behavior state coefficients of each client.

[0086] Request frequency refers to the number of service requests made by the client within a preset time period, which can be implemented using a 24-hour sliding window to quantify the activity level of client behavior. Historical interaction count refers to the cumulative number of interactions between the client and the system since establishing a connection, which can be achieved through log database queries to assess the client's historical trustworthiness. Permission level ratio refers to the ratio of the system access permission levels granted to the client to the highest permission level, which can be implemented using a role-based permission matrix mapping to reflect the system's preset level of trust in the client. Weighting coefficients. This refers to the relative importance parameters of each evaluation dimension, which can be achieved through the analytic hierarchy process or by assigning values ​​based on expert experience. These parameters are used to adjust the evaluation focus according to the application scenario.

[0087] Specifically, the client behavior state model generates a credibility coefficient by quantitatively analyzing and comprehensively evaluating behavioral characteristics across three dimensions. Request frequency is converted after normalization. The index, high-frequency abnormal requests will lead to Approaching 1, at this point Items approaching 0 result in a lower credibility score. Historical interaction counts are generated after normalization. The index shows that clients with long-term stable interaction will receive higher scores. Values ​​that positively improve the credibility assessment results. Percentage of permission levels. As static parameters, they directly participate in the calculation, ensuring that high-privilege clients receive basic trust guarantees. By adjusting... Weight allocation, for example, setting =0.5、 =0.3、 =0.2, which enhances the ability to suppress abnormal requests. The final output... The coefficients dynamically reflect the characteristics of client behavior, providing a quantitative basis for subsequent resource allocation.

[0088] Compared to existing technologies, traditional methods rely solely on static permission levels for resource allocation, failing to identify high-frequency malicious requests or inactive trusted clients. This solution integrates dynamic behavioral indicators with static permission parameters to construct a multi-dimensional evaluation model. While retaining the fundamental role of permission levels, it introduces a reverse suppression mechanism based on request frequency and a positive incentive mechanism based on historical interactions, effectively distinguishing between normal clients and potentially malicious clients.

[0089] Through the above technical solution, this application solves the resource allocation imbalance problem caused by the simplification of client trust assessment and realizes dynamic trust assessment based on behavioral characteristics. For example, when a low-privilege client is detected to suddenly generate high-frequency requests, its... The coefficient will be significantly reduced, thus limiting power allocation calculations and preventing abnormal resource consumption. Meanwhile, high-privilege clients with long-term interaction can obtain stable high... The scoring system ensures the quality of service for critical business operations. This evaluation mechanism optimizes resource utilization while ensuring system security.

[0090] Preferably, the step of calculating the host state coefficients based on the host state information using the host state model is as follows:

[0091] Based on the current total power consumption and temperature, a power consumption-temperature-pressure model is constructed to output the power consumption-temperature-pressure coefficient. The power consumption-temperature-pressure model is expressed as follows:

[0092]

[0093] in, This represents the power consumption-temperature-pressure coefficient. This indicates the current total power consumption. Indicates the host temperature. Indicates the maximum allowable power consumption. Indicates the maximum permissible temperature. Represents the weight coefficient and The ;

[0094] A host state model is constructed based on GPU utilization, memory usage, and network bandwidth usage under the power consumption-temperature-pressure coefficient. The host state model is represented as follows:

[0095]

[0096] in, Represents the host status coefficient. This represents the power consumption-temperature-pressure coefficient. Indicates GPU utilization. This indicates the video memory usage rate. Indicates network bandwidth utilization, the The larger the value, the more abundant the host resources.

[0097] The power consumption-temperature stress coefficient refers to the comprehensive impact of host power consumption and temperature on system stability, quantified through normalization. Specifically, it can be calculated by real-time acquisition of total host power consumption and temperature sensor data, combined with a preset maximum allowable value. This coefficient reflects the degree to which the hardware safety boundary is approached; when power consumption or temperature approaches the threshold, a load reduction protection mechanism is automatically triggered. The host status coefficient is a quantitative indicator that comprehensively evaluates the availability of host resources. Specifically, it can be calculated by multiplying the power consumption-temperature stress coefficient with GPU utilization, memory usage, and network bandwidth usage. This coefficient, through the coupled calculation of multi-dimensional resource usage status, reflects the overall system load level in real time, providing a basis for dynamically adjusting resource allocation.

[0098] Specifically, the power consumption-temperature stress model calculates the current total power consumption and host temperature proportionally to their maximum allowable values, and uses a minimum value function to limit the stress coefficient from exceeding a threshold, effectively avoiding model distortion under extreme conditions. The introduction of weighting coefficients allows adjustment of the contribution ratios of power consumption and temperature to the stress coefficient based on system characteristics. The host state model combines hardware constraint stress with real-time resource utilization through a product approach. When any resource approaches full load or a hard constraint approaches a critical value, the state coefficient exhibits a non-linear decreasing trend. This modeling approach enables the system to proactively trigger a global resource reallocation strategy when single-point bottlenecks such as excessive GPU utilization, near-saturation of video memory, or limited network bandwidth occur.

[0099] Compared to existing technologies, traditional methods typically monitor temperature or power consumption independently, without establishing a correlation model between hardware constraints and resource utilization. Existing solutions only consider single parameters such as CPU or GPU utilization during resource scheduling, lacking collaborative analysis of critical resources such as video memory and network bandwidth. This solution achieves joint evaluation of hardware security boundaries and multi-dimensional resource status through a dual-coupling model, enabling the system to proactively adjust the load when temperature rises in the early stages or power consumption approaches a threshold, rather than passively responding to fault alarms.

[0100] Through the above technical solution, this application achieves coordinated monitoring of the host system's hard constraints and soft resource occupancy status. It can automatically reduce computing power allocation when the GPU's temperature rises due to high load, prioritize releasing non-critical task resources when memory usage is too high, and dynamically adjust data transmission frequency when network bandwidth is limited. This intelligent load reduction mechanism effectively avoids system crashes caused by local resource overload and extends the lifespan of hardware devices through early intervention.

[0101] Preferably, the step of calculating the service quality coefficient based on the service quality requirement information of the service is as follows:

[0102] The allowed delay time and average task duration are subjected to max-min normalization to obtain the delay time index and the task duration index.

[0103] A service quality model is constructed based on the latency index, the task duration index, and the allowable packet loss rate. The service quality model is expressed as follows:

[0104]

[0105] in, Indicates the business quality coefficient. Indicates the delay duration index, This represents the packet loss rate index. This indicates the task duration index. Represents the weight coefficient and The The Furthermore, the higher the value, the greater the client's business demand, and the more priority should be given to ensuring service.

[0106] Import the latency index, task duration index, and allowable packet loss rate of each client into the business quality model to obtain the business quality coefficient of each client.

[0107] The allowable latency time refers to the maximum acceptable response time threshold for the service, which can be calculated using timestamp differences, and is used to measure the service's sensitivity to real-time performance. The average task duration refers to the average runtime of the service over a historical period, which can be obtained through a sliding window statistical method, and is used to assess the service's stability requirements for computing resource usage. The service quality model is a linearly weighted calculation model that integrates latency sensitivity, task stability, and packet loss tolerance. The priority of each dimension can be adjusted by configuring weight coefficients, and it is used to quantify the comprehensive service quality requirements of different services.

[0108] Specifically, the latency index is calculated as (current latency - minimum latency) / (maximum latency - minimum latency), reflecting the business's relative requirements for real-time performance. The task duration index characterizes the intensity of the business's demand for long-term resource consumption. A ternary linear model is then constructed, comprising the latency index, task duration index, and allowable packet loss rate, where the latency-related terms adopt... The format allows businesses with stricter latency requirements to receive higher scores; packet loss rate-related items are assessed through... This processing enables clients with stringent packet loss rate requirements to prioritize resource protection. By dynamically configuring weighting coefficients, the system can adjust the evaluation focus for different scenarios. For example, in real-time monitoring scenarios, the latency weighting coefficient can be increased to 0.5, while in data transmission scenarios, the packet loss rate weighting coefficient can be increased to 0.6. The final output service quality coefficient is calculated using linear weighting and constrained within the [0,1] interval, forming a quantitative indicator that can be compared horizontally.

[0109] Compared to existing technologies, traditional methods typically allocate resources based on fixed priorities or a single dimension, such as client identity level or historical request count. This solution, however, integrates three key dimensions—latency, duration, and packet loss rate—and introduces a configurable weighting mechanism to dynamically adapt to varying quality requirements across different business scenarios. Existing technologies lack consideration for task duration, making it difficult to differentiate the impact of short-term, bursty tasks versus long-term, stable tasks on system resource consumption. This solution, however, effectively identifies business types requiring continuous resource guarantees through a task duration index.

[0110] Through the above technical solution, this application achieves dynamic fusion evaluation of multi-dimensional service quality parameters, solving the problem of rigid resource allocation caused by the single evaluation dimension in traditional methods. The system can accurately identify high-priority services based on the combined characteristics of real-time, stability, and reliability requirements, prioritizing the refresh rate requirements of critical tasks during GPU resource allocation. For example, for urgent spectrum monitoring tasks, their low latency requirements and high packet loss sensitivity can automatically receive higher weights in the model, thus maintaining the target refresh rate even when host resources are strained.

[0111] Preferably, the power allocation model is expressed as:

[0112]

[0113] in, Indicates the first Computing power per client This represents the total computing power that the system can allocate. Indicates the first Interference intensity coefficient of each client, Indicates the first Each client behavior state coefficient Represents the host status coefficient. This indicates the total number of current clients.

[0114] The client interference intensity coefficient refers to the degree of interference generated by the client on the system, quantified by spectrum state information, and is used to prioritize the allocation of computing resources to clients in high-interference scenarios. The client behavior state coefficient is a reliability index evaluated through client behavior information, used to suppress resource consumption by abnormal clients. The host state coefficient is a global parameter reflecting the real-time load state of the host, used to dynamically adjust the total power allocation when the host is overloaded. The total allocatable computing power of the system refers to the total amount of computing resources currently available to all clients, which can be determined based on the host's hardware performance threshold or a preset strategy.

[0115] Specifically, this model uses the product of client interference intensity and behavioral state as the resource allocation weight through normalization, and combines it with the host state coefficient for global adjustment. For example, when the host temperature rises or the GPU utilization exceeds a threshold, the host state coefficient decreases, causing the computing power of all clients to be reduced proportionally, thereby proactively mitigating the risk of system overload. Simultaneously, the product of interference intensity coefficient and behavioral state coefficient ensures that clients with high interference and high credibility receive more resources, while clients with low interference or low credibility are allocated fewer resources. The summation operation in the denominator further normalizes the weights, preventing a single client parameter anomaly from affecting the overall fairness of the allocation.

[0116] Compared to existing technologies, current solutions typically allocate resources using fixed priorities or round-robin methods, failing to consider the impact of real-time host load status on global resource allocation. For example, traditional methods may maintain the original power allocation even when the host is overheating, potentially leading to system crashes. This solution, however, introduces a host state coefficient as a dynamic adjustment factor, automatically reducing the computing power of all clients when the host load is too high, thus ensuring system stability. Furthermore, existing technologies do not jointly model client interference intensity and behavioral credibility, while this solution achieves fine-grained control of resource allocation through the product of these two factors.

[0117] Through the above technical solution, this application solves the problem that static resource allocation cannot adapt to real-time changes in host load, and realizes dynamic adjustment of global computing power when the host is overloaded or at high temperature, avoiding the risk of system crash. Simultaneously, by coupling client interference intensity and behavioral reliability parameters, the resource guarantee capability of high-priority clients is optimized, and resource abuse by low-reliability clients is suppressed. For example, when the host temperature reaches a critical value, the host state coefficient decreases to 0.5, and the computing power of all clients is automatically reduced to 50% of its original value, thereby effectively reducing system load while maintaining basic services.

[0118] The refresh rate optimization model is expressed as follows:

[0119]

[0120] in, Indicates the first The target refresh rate for each client. Indicates the first Computing power per client Indicates the first The baseline refresh rate for each client. Indicates the first The service quality coefficient of each client This represents the threshold for the business quality coefficient. This represents the business quality adjustment coefficient. This represents the reference value for calculating power.

[0121] Among them, the base refresh rate It refers to the first The initial refresh rate for each client under standard resource allocation can be implemented using a preset value or a historical average refresh rate, reflecting the client's basic refresh requirements. (Calculation power) This refers to dynamically allocating to the first... The computing resources of each client are used to characterize the actual proportion of current resource allocation. Service Quality Coefficient It refers to the first Service quality priority for each client service is used to quantify the criticality of the service. Service quality coefficient threshold. This refers to the benchmark value used to distinguish between high and low service priorities. Specifically, it can be implemented using a system-preset fixed value or a dynamically adjusted median, and is used to trigger the effect of service quality on refresh rate gain or suppression. The service quality adjustment coefficient λ is a parameter that controls the intensity of the impact of service quality deviation on refresh rate. It can be set using empirical values ​​or adaptive algorithms, and is used to balance the relationship between resource allocation and service priority.

[0122] Specifically, the refresh rate optimization model is based on a baseline refresh rate and calculates the power ratio. The impact of dynamically adjusted resource allocation on refresh rate is considered. When a client receives computing power higher than the baseline value, its refresh rate increases proportionally; conversely, it decreases. Simultaneously, the model incorporates a service quality deviation term. When the service quality coefficient is higher than the threshold, the deviation term is positive, amplifying the refresh rate increase through λ to prioritize high-priority services. When the service quality coefficient is lower than the threshold, the deviation term is negative, suppressing refresh rate growth to avoid resource waste. For example, when host resources are scarce, the computing power of high-interference clients may be reduced, but if their service quality coefficient is significantly higher than the threshold, a higher refresh rate can still be maintained through the deviation term, thus balancing resource allocation efficiency and service assurance for critical services.

[0123] Compared to existing technologies, traditional solutions typically employ fixed priority or round-robin scheduling, failing to dynamically adjust the refresh rate based on real-time computing power and service demands. For example, existing technologies may allocate resources solely based on client identity levels, resulting in low-interference but high-priority services not receiving sufficient refresh rates in a timely manner. This solution, by coupling the computing power allocation ratio with service quality deviations, can automatically adjust the refresh rate when resources are limited, preventing high-interference clients from excessively consuming resources while ensuring the service quality of critical services.

[0124] Through the above technical solution, this application achieves dynamic optimization of the client refresh rate, solving the priority misalignment problem caused by rigid resource allocation. Specifically, the combination of computing power allocation and service quality coefficient enables the system to automatically balance resource allocation when the host load fluctuates, prioritizing the real-time needs of high-priority services while suppressing resource consumption by low-quality or high-interference clients, thereby improving overall resource utilization efficiency and system responsiveness.

[0125] A GPU-based multi-client real-time spectrum monitoring system employs the aforementioned GPU-based multi-client real-time spectrum monitoring method.

[0126] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0127] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A GPU-based multi-client real-time spectrum monitoring method, characterized in that, Includes the following steps: Obtain the client's spectrum status information, client behavior information, host status information, and service quality requirements information; Based on the aforementioned spectrum state information, the client interference intensity coefficient is calculated using the client interference intensity model. Based on the client's behavioral information, the client's behavioral state coefficients are calculated using the client's behavioral state model. Based on the host's state information, host state coefficients are calculated using a host state model. Based on the service quality requirements information of the aforementioned business, the business quality coefficient is calculated through the business quality model. Based on the client interference intensity coefficient, the client behavior state coefficient, and the host state coefficient, the computing power allocated to each client is dynamically calculated and output through a computing power allocation model. Based on the computing power allocated to the client, the client's baseline refresh rate, and the service quality coefficient, the target refresh rate of the client is calculated and output through a refresh rate optimization model.

2. The GPU-based multi-client real-time spectrum monitoring method according to claim 1, characterized in that, The spectrum status information includes noise floor rise, spectrum peak difference, and excess time length. The client behavior information includes request frequency, historical interaction count, and permission level percentage. The host status information includes current total power consumption, host temperature, GPU utilization, video memory usage, and network bandwidth usage. The service quality requirement information includes allowed latency, allowed data packet loss rate, and average task duration.

3. The GPU-based multi-client real-time spectrum monitoring method according to claim 2, characterized in that, The refresh rate optimization model is expressed as follows: ； in, Indicates the first The target refresh rate for each client. Indicates the first Computing power per client Indicates the first The baseline refresh rate for each client. Indicates the first The service quality coefficient of each client This represents the threshold for the business quality coefficient. This represents the business quality adjustment coefficient. This represents the reference value for calculating power.

4. The GPU-based multi-client real-time spectrum monitoring method according to claim 3, characterized in that, The power allocation model is expressed as follows: ； in, Indicates the first Computing power per client This represents the total computing power that the system can allocate. Indicates the first Interference intensity coefficient of each client, Indicates the first Each client behavior state coefficient Represents the host status coefficient. This indicates the total number of current clients.

5. The GPU-based multi-client real-time spectrum monitoring method according to claim 4, characterized in that, The steps for calculating the service quality coefficient based on the service quality requirement information of the aforementioned business using the service quality model are as follows: The allowed delay time and average task duration are subjected to max-min normalization to obtain the delay time index and the task duration index. A service quality model is constructed based on the latency index, the task duration index, and the allowable packet loss rate. The service quality model is expressed as follows: ； in, Indicates the business quality coefficient. Indicates the delay duration index, This represents the packet loss rate index. This indicates the task duration index. Represents the weight coefficient and The The Furthermore, the higher the value, the greater the client's business demand, and the more priority should be given to ensuring service. Import the latency index, task duration index, and allowable packet loss rate of each client into the business quality model to obtain the business quality coefficient of each client.

6. The GPU-based multi-client real-time spectrum monitoring method according to claim 4, characterized in that, The steps for calculating the host state coefficients based on the host state information and the host state model are as follows: Based on the current total power consumption and temperature, a power consumption-temperature-pressure model is constructed to output the power consumption-temperature-pressure coefficient. The power consumption-temperature-pressure model is expressed as follows: ； in, This represents the power consumption-temperature-pressure coefficient. This indicates the current total power consumption. Indicates the host temperature. Indicates the maximum allowable power consumption. Indicates the maximum permissible temperature. Represents the weight coefficient and The ; A host state model is constructed based on GPU utilization, memory usage, and network bandwidth usage under the power consumption-temperature-pressure coefficient. The host state model is represented as follows: ； in, Represents the host status coefficient. This represents the power consumption-temperature-pressure coefficient. Indicates GPU utilization. This indicates the video memory usage rate. Indicates network bandwidth utilization, the The larger the value, the more abundant the host resources.

7. The GPU-based multi-client real-time spectrum monitoring method according to claim 4, characterized in that, The steps for calculating the client behavior state coefficients based on the client's behavior information using the client behavior state model are as follows: The request frequency index and the historical interaction count index are obtained by performing max-min normalization on the request frequency and the historical interaction count index. A client behavior state model is constructed based on the request frequency index, the historical interaction count index, and the percentage of permission levels. This client behavior state model is represented as follows: ； in, This represents the customer behavior status coefficient. This indicates the request frequency index. This represents the percentage of historical interactions. Indicates the percentage of different permission levels. Represents the weight coefficient and The The Furthermore, the higher the value, the higher the client's credibility, and the more reliable the service needs to be. Import the request frequency index, historical interaction count index, and permission level ratio of each client into the client behavior state model to output the behavior state coefficients of each client.

8. The GPU-based multi-client real-time spectrum monitoring method according to claim 4, characterized in that, Based on the aforementioned spectrum state information, the steps for calculating the client interference intensity coefficient using the client interference intensity model are as follows: The noise floor rise value, the peak difference value of the spectrum, and the excess time length are subjected to maximum-min normalization to obtain the noise floor rise index, the peak difference index of the spectrum, and the excess time length index. A client interference intensity model is constructed based on the noise floor rise index, the peak spectral difference index, and the excess time length index. The client interference intensity model is expressed as follows: ； in, Indicates the client interference strength coefficient. This represents the noise floor rise index. Indicates the peak difference index of the spectrum. This represents the excess time length index. Represents the weight coefficient and The The Furthermore, the larger the value, the more significant the client-side interference; The noise floor rise index, peak spectral difference index, and excess time length index of each client are imported into the client interference intensity model to output the interference intensity coefficient of each client.

9. A GPU-based multi-client real-time spectrum monitoring system, characterized in that, The GPU-based multi-client real-time spectrum monitoring method described in any one of claims 1-8 is adopted.

Images