Anti-jamming communication method and system based on delay perception

Abstract

The application discloses a kind of anti-interference communication method and system based on delay perception, it is related to communication security technical field.The method comprises the following steps: determining the service remaining delay budget and fast state urgency of communication session in current transmission period;According to service remaining delay budget and fast state urgency, determine and execute the anti-interference response strategy of communication session, to realize anti-interference communication;Wherein, anti-interference response strategy includes only executing fast path decision, or parallelly executing fast path decision and adaptive decision, fast path decision is used to select anti-interference action from current security action set and execute in current transmission period, adaptive decision is used to update candidate security action set of next transmission period, after candidate security action set is updated, as the current security action set of next transmission period.The application realizes the dynamic balance of timeliness of anti-interference response and anti-interference decision quality.

Description

Technical Field

[0001] This invention relates to the field of communication security technology, and in particular to an anti-interference communication method and system based on time delay awareness. Background Technology

[0002] Wireless services such as industrial control, power line inspection, emergency communication, unmanned system control, port and mine dispatching, vehicle coordination, and industrial park production scheduling have stringent requirements for low latency, high reliability, and continuous online capability. These services are highly susceptible to suppression interference, frequency sweeping interference, pulse interference, or reactive interference, which can easily lead to packet timeouts, control command interruptions, or complete communication session disconnections.

[0003] To combat the aforementioned interference, existing technologies typically employ anti-interference measures such as frequency hopping, spread spectrum, power control, modulation and coding adjustment, and redundant transmission. While these traditional methods can mitigate the impact of interference to some extent, their decision-making logic is mostly based on preset thresholds or fixed rules, lacking the ability to proactively adapt to dynamic environmental changes. Some solutions achieve dynamic strategy optimization through spectrum sensing and environmental learning. For example, patent application CN202411950760.9 discloses a dynamic spreading factor anti-interference method based on annealed Q-learning. This method obtains the current interference state through broadband spectrum sensing, constructs a state space, action space, and reward function with system throughput as the optimization objective, and uses a Q-value table to determine the spreading factor for the next time slot. This allows for anti-interference decisions to be made in dynamic interference environments, avoiding the delay and overhead caused by channel switching.

[0004] In the process of implementing the embodiments of the present invention, at least the following problems were found in the related art: Dynamic anti-interference methods based on online learning inherently consume computation and processing time in their decision-making process. When interference frequencies change rapidly, interference durations are shorter than the decision cycle, or service latency budgets are extremely tight, a significant contradiction arises between the time-consuming anti-interference decision-making process and the service's tolerable latency. Each interference instance necessitates re-evaluating numerous candidate frequencies, spreading parameters, or redundancy strategies, which not only increases processing latency but may also compromise link stability due to frequent parameter switching, failing to achieve a dynamic balance between the timeliness of anti-interference response and the quality of anti-interference decisions. Summary of the Invention

[0005] To address the shortcomings of the existing technology, the present invention provides a time-delay-aware anti-interference communication method and system to achieve a dynamic balance between the timeliness of anti-interference response and the quality of anti-interference decision-making.

[0006] According to one aspect of the present invention, the present invention provides a delay-aware anti-interference communication method, comprising: determining the remaining service delay budget and fast state urgency of a communication session in the current transmission cycle; determining and executing an anti-interference response strategy for the communication session based on the remaining service delay budget and fast state urgency to achieve anti-interference communication; wherein the anti-interference response strategy includes fast path decision and adaptive decision, the fast path decision being used to select an anti-interference action from the current set of security actions and execute it in the current transmission cycle, and the adaptive decision being used to update the set of candidate security actions for the next transmission cycle, the updated set of candidate security actions serving as the current set of security actions for the next transmission cycle.

[0007] Optionally, the remaining latency budget for the service is determined, including: obtaining the maximum tolerable latency, consumed transmission latency, queue waiting latency, and policy execution expected latency of the communication session in the current transmission cycle; determining the security protection margin based on the statistical deviation between the actual end-to-end latency and the expected end-to-end latency of multiple historical transmission cycles; and obtaining the remaining latency budget for the service by subtracting the consumed transmission latency, queue waiting latency, policy execution expected latency, and security protection margin from the maximum tolerable latency for the service.

[0008] Optionally, determining the urgency of the fast state includes: constructing a fast state vector for the communication session in the current transmission cycle, wherein the fast state vector consists of at least two of the following: the normalized value of the queue growth rate, the link quality degradation rate, the proportion of abnormal transmission of service packets, and the proportion of ready actions in the current set of security actions; assigning corresponding weights to each component of the fast state vector, and summing the weighted components to obtain the urgency of the fast state.

[0009] Optionally, the adaptive decision includes a first adaptive decision and a second adaptive decision; based on the remaining latency budget and the urgency of the fast state, the anti-interference response strategy for the communication session is determined, including: combining a latency threshold and an urgency threshold to determine the latency-urgency state interval of the communication session in the current transmission cycle; based on the latency-urgency state interval, the decision mode of the anti-interference response strategy is determined, wherein the decision mode includes a single decision mode and a combined decision mode, the single decision mode is to execute only the fast path decision, and the combined decision mode is to execute the fast path decision and the adaptive decision in parallel; when the decision mode is the combined decision mode, the first adaptive decision or the second adaptive decision is determined according to the adaptive decision level corresponding to the latency-urgency state interval, wherein the first adaptive decision updates the candidate security action set for the next transmission cycle within the first candidate action range, and the second adaptive decision updates the candidate security action set for the next transmission cycle within the second candidate action range, the second candidate action range being larger than the first candidate action range.

[0010] Optionally, the fast path decision is used to select anti-interference actions from the current set of security actions, including: calculating a migration judgment value based on the short-window spectrum situation map of the communication session to characterize whether strong interference is migrating to the current operating frequency range; if the migration judgment value is greater than or equal to the judgment threshold, selecting anti-interference actions that meet the time delay constraints and reliability constraints from the current set of security actions to generate an executable action set; calculating the fast selection score corresponding to each anti-interference action in the executable action set, and selecting the target anti-interference action from the executable action set according to the fast selection score as the anti-interference action to be executed output by the fast path decision.

[0011] Optionally, based on the short-window spectrum situation map of the communication session, a migration determination value is calculated to characterize whether strong interference is migrating to the current operating frequency range. This includes: predicting the predicted centroid and predicted bandwidth of the interference block in future scheduling cycles using a trajectory fitting algorithm based on the spectral centroid movement trajectory of the interference block in the short-window spectrum situation map; determining the current operating frequency range of the communication session and the predicted interference range composed of the predicted centroid and predicted bandwidth, and calculating the predicted overlap between the current operating frequency range and the predicted interference range; and weighting and summing the predicted overlap, the normalized super-noise energy of the current operating frequency range, the window occupancy rate of the frequency unit in the short-window spectrum situation map, and the link quality degradation rate to obtain the migration determination value.

[0012] Optionally, the fast selection score corresponding to each anti-interference action in the set of executable actions is calculated, including: obtaining the actual success benefit of each anti-interference action to be evaluated in the set of executable actions during a preset time period; determining the rule matching degree between the anti-interference action to be evaluated and the current interference type and the current link state, and obtaining the first action switching cost and the first expected execution time corresponding to the anti-interference action to be evaluated; and weighting and summing the success benefit and rule matching degree as positive factors, and the first action switching cost and the first expected execution time as negative factors to obtain the fast selection score of the anti-interference action to be evaluated.

[0013] Optionally, the adaptive decision-making process is used to update the candidate security action set for the next transmission cycle, including: encoding the short-window spectrum situation map of the communication session, service link quality indicators, and historical action benefits into a context vector; based on the context vector, determining the estimated benefit and upper bound of the estimated benefit confidence interval for each candidate anti-interference action to be evaluated, according to the parameter matrix and benefit vector corresponding to each candidate anti-interference action to be evaluated in the candidate security action set; obtaining the second action switching cost and the second estimated execution time corresponding to the candidate anti-interference action to be evaluated; calculating the priority of the candidate anti-interference action to be evaluated based on the estimated benefit, the upper bound of the estimated benefit confidence interval, the second action switching cost, and the second estimated execution time; and sorting and updating the candidate anti-interference actions in the candidate security action set in descending order of priority, wherein the candidate anti-interference actions with higher priority are preferentially selected in the next transmission cycle.

[0014] Optionally, the priority of the candidate anti-interference actions to be evaluated is calculated based on the estimated benefit, the upper bound of the estimated benefit confidence interval, the switching cost of the second action, and the second estimated execution time, including calculating the priority of the candidate anti-interference actions to be evaluated according to the following formula: , where P t(a2) U represents the priority score of candidate anti-interference action a2 to be evaluated in the t-th transmission cycle. t(a2） To estimate the revenue, To estimate the upper bound of the confidence interval for returns, x t Let x be the context vector of the communication session in the t-th transmission cycle. t T For x t The transpose of A a2 Let C be the parameter matrix of a2, δ be the exploration coefficient, and C be the parameter matrix of a2. sw(a2) T is the switching cost of the second action corresponding to a2. exec(a2) Let a2 be the second estimated execution time, and μ1 and μ2 be C, respectively. sw(a2) and T exec(a2) The corresponding penalty coefficient.

[0015] According to another aspect of the present invention, the present invention also provides a latency-aware anti-interference communication system, comprising: a service state determination module configured to determine the remaining latency budget and fast state urgency of a communication session in the current transmission cycle; and an anti-interference response strategy generation module configured to determine and execute an anti-interference response strategy for the communication session based on the remaining latency budget and fast state urgency to achieve anti-interference communication; wherein the anti-interference response strategy includes executing only fast path decision, or executing fast path decision and adaptive decision in parallel, the fast path decision being used to select an anti-interference action from the current security action set and execute it in the current transmission cycle, and the adaptive decision being used to update the candidate security action set for the next transmission cycle, the updated candidate security action set being used as the current security action set for the next transmission cycle.

[0016] According to another aspect of the embodiments of the present invention, the present invention also provides a time-delay-aware anti-interference communication device, including a processor and a memory storing program instructions, wherein the processor is configured to execute the aforementioned time-delay-aware anti-interference communication method when executing the program instructions.

[0017] The time-delay-aware anti-interference communication method and system provided by this invention have at least the following beneficial technical effects: In the current transmission cycle of a communication session, the remaining latency budget and the urgency of the fast state are calculated in real time, and the anti-interference response strategy is dynamically selected based on these criteria: only fast path decision is executed, which directly selects anti-interference actions from the current set of security actions and executes them immediately, minimizing decision-making time and ensuring the timeliness of anti-interference response in the current transmission cycle; fast path decision and adaptive decision are executed in parallel, with fast path decision ensuring the timeliness of anti-interference response in the current transmission cycle, and adaptive decision updating the candidate set of security actions for the next transmission cycle in the background; the candidate set of security actions optimized by adaptive decision serves as the current set of security actions for the next transmission cycle, so that fast path decision is always based on a continuously evolving high-quality anti-interference action library. Therefore, by using the remaining latency budget and the urgency of the rapid state as the two-dimensional switching criteria for decision-making paths, the overall state of the current communication session can be accurately assessed and corresponding anti-interference response strategies can be determined. When latency is tight or under urgent circumstances, priority is given to ensuring the rapid response of the current anti-interference decision. When latency is ample or the urgency is low, the quality optimization of future anti-interference decisions can be promoted simultaneously. The timeliness of the anti-interference response is not lost in the pursuit of optimal decision-making, and the quality of the anti-interference decision is not stagnated in the pursuit of latency, thus achieving a dynamic balance between the timeliness of the anti-interference response and the quality of the anti-interference decision. Attached Figure Description

[0018] Figure 1 A flowchart illustrating an anti-interference communication method based on delay awareness provided in an embodiment of the present invention; Figure 2 A flowchart illustrating another delay-aware anti-interference communication method provided in an embodiment of the present invention; Figure 3 A flowchart illustrating another delay-aware anti-interference communication method provided in an embodiment of the present invention; Figure 4 A flowchart illustrating another delay-aware anti-interference communication method provided in an embodiment of the present invention; Figure 5 A flowchart illustrating another delay-aware anti-interference communication method provided in an embodiment of the present invention; Figure 6 A schematic diagram illustrating the principle of symbolic scattering mapping and interleaving actions provided in an embodiment of the present invention; Figure 7 A schematic diagram of the structure of an anti-interference communication system based on time delay awareness provided in an embodiment of the present invention; Figure 8 This is a schematic diagram of the structure of an anti-interference communication device based on time delay awareness, provided in an embodiment of the present invention. Detailed Implementation

[0019] To better understand the above technical solutions, a detailed description of the solutions will be provided below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are merely some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0020] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms, and “multiple” generally includes at least two unless the context clearly indicates otherwise.

[0021] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that an article or device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such an article or device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the article or device that includes said element.

[0022] Combination Figure 1 As shown, this embodiment of the invention provides an anti-interference communication method based on time delay awareness, comprising the following steps: S10, determine the remaining latency budget and fast state urgency of the communication session in the current transmission cycle.

[0023] The remaining service delay budget characterizes the total amount of time available for the communication session's service data packets to complete remaining transmission, queue waiting, and anti-interference strategy execution within the current transmission cycle, from the current moment to the deadline of their maximum tolerable delay. The rapid state urgency characterizes the overall real-time risk level faced by the communication link within the current transmission cycle due to factors such as continuous queue growth, link quality deterioration, abnormal service packet transmission, and insufficient reliable actions that can be immediately executed from the current set of security actions. A smaller remaining service delay budget indicates less time available for decision optimization; a higher rapid state urgency indicates greater real-time risk to the current link. The remaining service delay budget and rapid state urgency together constitute the switching criteria for the decision path from two dimensions: time margin and risk level. When the remaining service delay budget is insufficient or the rapid state urgency is high, priority is given to ensuring the timeliness of anti-interference response, rather than investing time in decision optimization within the current transmission cycle.

[0024] In some possible implementations, determining the remaining latency budget for the service includes: obtaining the maximum tolerable latency, consumed transmission latency, queue waiting latency, and policy execution expected latency of the communication session in the current transmission cycle; determining the security protection margin based on the statistical deviation between the actual end-to-end latency and the expected end-to-end latency of multiple historical transmission cycles; and obtaining the remaining latency budget for the service by subtracting the consumed transmission latency, queue waiting latency, policy execution expected latency, and security protection margin from the maximum tolerable latency for the service.

[0025] Furthermore, the remaining latency budget for the service is calculated using the following formula:

[0026] Where B is the remaining latency budget for the service. To the maximum tolerable latency for the business, The consumed transmission delay, For queue waiting delay, The expected delay for strategy execution. For safety protection margin.

[0027] Maximum tolerable latency for the business This refers to the maximum end-to-end latency that a service data packet can withstand from the sender to the receiver, determined during the communication session establishment phase based on the service type and QoS requirements; consumed transmission latency. This refers to the time elapsed from when a data packet enters the transmission queue to the current moment, obtained from the difference in timestamps recorded by the MAC layer or baseband processing module; queue waiting delay. This refers to the estimated time a data packet will still need to wait in the current sending queue, estimated based on the current queue length and average service rate; policy execution estimated latency. This refers to the estimated additional time overhead required to execute the anti-interference response strategy, determined based on the type of response action to be performed; security protection margin. It refers to the extra time protection reserved to cope with uncertainties such as delay estimation errors, sudden fluctuations in channel status, and system processing jitter.

[0028] Furthermore, the statistical deviation between the actual end-to-end delay and the expected end-to-end delay for multiple historical transmission cycles includes the average or maximum deviation between the actual end-to-end delay and the expected end-to-end delay for each historical transmission cycle within the sliding window. The security protection margin is determined based on the statistical deviation between the actual end-to-end delay and the expected end-to-end delay for multiple historical transmission cycles, and a preset margin amplification factor.

[0029] For example, when the statistical deviation is the maximum deviation, the safety protection margin Calculated using the following formula:

[0030] Where W is the sliding window for the most recent transmission cycles, Let be the actual end-to-end delay of the service data packet in the i-th transmission period. Let be the estimated end-to-end delay of the service data packets in the i-th transmission period. This represents the maximum deviation between the actual end-to-end delay and the expected end-to-end delay for each historical transmission cycle within the sliding window. It is a margin amplification factor greater than or equal to 1.

[0031] For example, when the statistical deviation is the average deviation, the safety protection margin Calculated using the following formula:

[0032] Among them, | W | represents the number of transmission cycles within the sliding window.

[0033] Safety protection margin This calculation method enables the security protection margin to be adaptively adjusted according to the latency jitter statistics of the actual communication environment.

[0034] The calculation method for the remaining latency budget of the aforementioned services precisely quantifies the total amount of remaining time available for anti-interference decisions within the current transmission cycle by deducting four time overheads—consumed transmission latency, queue waiting latency, expected policy execution latency, and security protection margin—from the maximum tolerable latency of the services. (Security protection margin) The introduction of this feature provides a buffer for the remaining latency budget of services to cope with estimation deviations and sudden jitter, further ensuring the latency determinism of service transmission and serving as an important foundation for achieving latency-aware adaptive anti-interference decision-making.

[0035] S20: Based on the remaining latency budget and the urgency of the rapid state, determine and execute the anti-interference response strategy for the communication session to achieve anti-interference communication.

[0036] The anti-interference response strategy includes performing only fast path decision, or performing fast path decision and adaptive decision in parallel. Fast path decision is used to select anti-interference action from the current set of security actions and execute it in the current transmission cycle. Adaptive decision is used to update the set of candidate security actions for the next transmission cycle. After the candidate security action set is updated, it becomes the current set of security actions for the next transmission cycle.

[0037] The anti-interference communication method based on latency awareness, as described in this invention, uses the remaining latency budget and the urgency of the rapid state as a two-dimensional switching criterion for decision-making paths. It accurately assesses the overall state of the current communication session and determines the corresponding anti-interference response strategy. When latency is tight or under urgent circumstances, it prioritizes ensuring the rapid response of the current anti-interference decision. When latency is ample or the urgency is low, it simultaneously promotes the quality optimization of future anti-interference decisions. The timeliness of the anti-interference response is not lost in the pursuit of optimal decision-making, and the quality of the anti-interference decision is not stagnated in the pursuit of latency, thus achieving a dynamic balance between the timeliness of the anti-interference response and the quality of the anti-interference decision.

[0038] Optionally, combined Figure 2 As shown, in step 10, determining the urgency of the rapid state includes the following steps: S101, Construct the fast state vector of the communication session in the current transmission cycle.

[0039] The fast state vector consists of at least two of the following: the queue growth rate normalization value, the link quality degradation rate, the proportion of abnormal service packet transmission, and the proportion of ready actions in the current set of security actions. The queue growth rate normalization value is a normalized index obtained by dividing the increase in queue length of the communication session in the current transmission cycle by the total queue capacity. It is used to characterize the backlog trend of service data in the transmission queue. The link quality degradation rate is an index obtained by weighting the decrease slope of the signal-to-interference-plus-noise ratio and the increase slope of the block error rate in the most recent observation windows. It is used to characterize the degradation trend and speed of communication link quality. The abnormal transmission ratio of service packets is the proportion of service packets that have experienced abnormal transmission events such as retransmission, timeout, or failure to return acknowledgment characters in the most recent few transmitted service packets. It is used to characterize the degree of abnormality of service transmission under the current anti-interference action. The ready action ratio in the current security action set is the proportion of the number of immediately executable actions in the current security action set that simultaneously meet the constraints (such as reliability constraints, expected execution time constraints, and protocol compatibility constraints) to the total number of actions in the current security action set. It is used to characterize the sufficiency of effective backup actions that can be directly selected for fast path decision-making in the current security action set.

[0040] For example, the fast state vector is represented as:

[0041] in, For fast state vectors, This is the normalized value of the queue growth rate. For the rate of link quality degradation, This refers to the proportion of abnormal transmissions of service packets. This represents the proportion of ready actions in the current set of safety actions.

[0042] S102, assign corresponding weights to each component of the fast state vector, and sum the weighted components to obtain the fast state urgency.

[0043] In some possible implementations, each component of the fast state vector is assigned a corresponding weight. This includes: calculating the fluctuation metric of the target component in the fast state vector over the most recent K transmission cycles, where the fluctuation metric includes the sliding standard deviation or the absolute value of the slope of change; normalizing the fluctuation metric of the target component to obtain its corresponding weight, with components exhibiting more dramatic fluctuations receiving higher weights. By adaptively increasing the weight of this component, the fast state urgency can more sensitively capture the most prominent risk factors, avoiding the weakening of key risk signals caused by the averaging of multiple components under fixed weights, thereby improving the timeliness and accuracy of anti-interference decision path selection.

[0044] In some possible implementations, the urgency of a fast state is calculated using the following formula:

[0045] Among them, H t Let λ be the urgency of the initial fast state. q , λ g , λ e and λ r q t g t e t and r t The corresponding weights.

[0046] In some possible implementations, each component of the fast state vector is assigned a corresponding weight, and the weighted components are summed to obtain the fast state urgency. This includes: weighting and summing the components of the fast state vector and their corresponding weights to obtain the initial fast state urgency; determining a nonlinear correction factor based on the difference between the ready action ratio and a preset action sufficiency threshold; and multiplying the initial urgency by the nonlinear correction factor to obtain the fast state urgency. Specifically, when the ready action ratio is less than the action sufficiency threshold, the nonlinear correction factor is greater than 1, and the nonlinear correction factor is negatively correlated with the ready action ratio; when the ready action ratio is greater than or equal to the action sufficiency threshold, the nonlinear correction factor is equal to 1.

[0047] For example, the initial fast state urgency is calculated using the following formula:

[0048] Among them, H t0 Let λ be the urgency of the initial fast state. q , λ g , λ e and λ r q t g t e t and r t The corresponding weights.

[0049] For example, the urgency of a rapid state is calculated using the following formula:

[0050] Among them, H t To assess the urgency of the rapid state, H t0 Let θ represent the urgency of the initial fast state, and θ be the nonlinear correction factor.

[0051] The ready action ratio represents the sufficiency of effective backup actions in the current set of safe actions that can be directly selected for the fast path. By introducing the ready action ratio as a non-linear correction factor, the normal fast state urgency assessment is maintained when there are sufficient backup actions, and the fast state urgency is non-linearly amplified when there are insufficient backup actions, forcing the system to prioritize fast path decision-making to avoid the risk of no actions available, and further enhancing the system's early warning capability for the depletion of critical resources.

[0052] Optionally, in step S20, the adaptive decision includes a first adaptive decision and a second adaptive decision; determining the anti-interference response strategy for the communication session based on the remaining service delay budget and the urgency of the fast state includes the following steps: combining a delay threshold and an urgency threshold to determine the threshold of the remaining service delay budget and the urgency of the fast state, and determining the delay-urgency state interval of the communication session in the current transmission cycle; determining the decision mode of the anti-interference response strategy based on the delay-urgency state interval, wherein the decision mode includes a single decision mode and a combined decision mode, the single decision mode is to execute only the fast path decision, and the combined decision mode is to execute the fast path decision and the adaptive decision in parallel; when the decision mode is the combined decision mode, determining whether to execute the first adaptive decision or the second adaptive decision based on the adaptive decision level corresponding to the delay-urgency state interval, wherein the first adaptive decision updates the candidate security action set for the next transmission cycle within the first candidate action range, and the second adaptive decision updates the candidate security action set for the next transmission cycle within the second candidate action range, the second candidate action range being larger than the first candidate action range.

[0053] In some possible implementations, the latency-urgency state interval includes a first state interval, a second state interval, and a third state interval. When the remaining latency budget is less than the first latency threshold, or the urgency of the fast state is greater than or equal to the first urgency threshold, the communication session is determined to be in the first state interval in the current transmission cycle, and the corresponding decision mode is a single decision mode. When the remaining latency budget is greater than or equal to the first latency threshold and less than the second latency threshold, and the urgency of the fast state is less than the first urgency threshold, or when the remaining latency budget is less than the second latency threshold, and the urgency of the fast state is greater than or equal to the second urgency threshold and less than the first urgency threshold, the communication session is determined to be in the second state interval in the current transmission cycle, and the corresponding decision mode is a combined decision mode, with the corresponding adaptive decision level being the first adaptive decision. When the remaining latency budget is greater than or equal to the second latency threshold, and the urgency of the fast state is less than the second urgency threshold, the communication session is determined to be in the third state interval in the current transmission cycle, with the corresponding decision mode being a combined decision mode, and the corresponding adaptive decision level being the second adaptive decision.

[0054] The adaptive decision level is an optimization depth level assigned to adaptive decisions based on the latency-urgency state interval of the communication session in the current transmission cycle. When the communication session is in the first state interval, the adaptive decision level is no adaptive decision; when the communication session is in the second state interval, the adaptive decision level is first adaptive decision; and when the communication session is in the third state interval, the adaptive decision level is second adaptive decision.

[0055] The first candidate action range refers to the finite set of candidate anti-interference actions that the first adaptive decision is allowed to update in priority when executing the fast path decision and the first adaptive decision in parallel.

[0056] In some possible implementations, the first candidate action range is obtained as follows: anti-interference actions that meet preset conditions are selected from the current set of security actions to form the first candidate action range; the preset conditions include: the parameter distance between the candidate anti-interference action and the currently executed anti-interference action is less than a parameter distance threshold; the parameter distance threshold indicates that only one level change occurs within a single action domain in the frequency domain, waveform domain, or redundancy domain. For example, a single level change in the frequency domain action is switching to an adjacent frequency point or adjacent sub-band; a single level change in the waveform action is decreasing the modulation order by one level or increasing the coding redundancy by one level; and a single level change in the redundancy action is adding one repeated transmission or enabling short packet redundancy. The limitation of the first candidate action range ensures that the computational cost of the first adaptive decision matches the medium latency budget interval, completing the optimization and ranking of candidate schemes within the neighborhood of the current anti-interference action within a finite time.

[0057] The second candidate action range refers to the larger set of candidate anti-interference actions that the second adaptive decision is allowed to update in priority when executing the fast path decision and the second adaptive decision in parallel.

[0058] In some possible implementations, the second candidate action range is obtained in the following manner: respectively perform independent security screening on available frequency domain actions, available waveform actions and available redundancy actions from the current security action set and the preconfigured available action set to obtain atomic actions of each domain; perform cross-domain combination on the atomic actions that pass the compatibility screening to obtain joint candidate anti-interference actions, which form the second candidate action range. The compatibility screening is used to eliminate atomic action combinations that have radio frequency resource conflicts, protocol parameter conflicts, or a joint execution time exceeding the delay budget of the next transmission cycle. Exemplarily, the frequency domain action screening includes available frequency points, subbands or subcarrier groups, etc.; the waveform action screening includes available spread spectrum parameters, modulation and coding schemes, interleaving depths, etc.; the redundancy action screening includes available short packet redundancy, repeated transmission, and dual-link duplication configuration, etc. The second candidate action range is not limited to the neighborhood of the currently executing anti-interference action, so that the computational overhead of the second adaptive decision matches the sufficient delay budget interval, and potentially better anti-interference strategies are explored in a broader action space.

[0059] As an optional example, in conjunction with Figure 3 shown in the figure, in step S20, determining the anti-interference response strategy for the communication session according to the service remaining delay budget and the fast state urgency includes the following steps: S201, in the case that B<B1, or H t ≥H1, the anti-interference response strategy is determined to be executing only the fast path decision.

[0060] S202, in the case that B1≤B<B2 and H t <H1, or B<B2 and H2≤H t <H1, the anti-interference response strategy is determined to be executing the fast path decision and the first adaptive decision in parallel.

[0061] S203, in the case that B≥B2 and H t <H2, the anti-interference response strategy is determined to be executing the fast path decision and the second adaptive decision in parallel.

[0062] Wherein, B is the service remaining delay budget, B1 and B2 are the first delay threshold and the second delay threshold respectively, H t is the fast state urgency, H1 and H2 are the first urgency threshold and the second urgency threshold respectively, B1<B2, H1>H2.

[0063] By dividing the adaptive decision into two levels, the first adaptive decision and the second adaptive decision, and combining with the service remaining delay budget B and the fast state urgency H tMultiple threshold ranges are precisely matched, enabling step-by-step adjustment of the adaptive decision-making depth. The search range of adaptive decision-making expands step-by-step with the increase of available latency budget and shrinks step-by-step with the increase of urgency, so that the optimization depth and latency constraints are always dynamically matched, maximizing the use of available latency resources to improve decision quality while ensuring timely anti-interference response.

[0064] Optionally, combined Figure 4 As shown, in step S20, the fast path decision is used to select an anti-interference action from the current set of safety actions, including the following steps: S211, based on the short-window spectrum situation diagram of the communication session, calculate the migration judgment value used to characterize whether strong interference is migrating to the current operating frequency range.

[0065] Here, the short-window spectrum situation map refers to a two-dimensional representation structure reflecting the real-time state and short-term trends of the electromagnetic environment, formed by local spectrum observation data collected within a preset short time window (e.g., a sliding window of 1 to 10 scheduling cycles or 3 to 10 frames). The short-window spectrum situation map consists of an energy matrix of L consecutive observation times and N frequency units. This indicates that it is used to capture rapidly changing dynamic characteristics of interference, meeting the real-time requirements of spectrum sensing in ultra-low latency scenarios.

[0066] In some possible implementations, based on the short-window spectrum situation map of the communication session, a migration determination value is calculated to characterize whether strong interference is migrating to the current operating frequency range. This includes: predicting the predicted centroid and predicted bandwidth of the interference block in future scheduling cycles using a trajectory fitting algorithm based on the spectral centroid movement trajectory of the interference block in the short-window spectrum situation map; determining the current operating frequency range of the communication session and the predicted interference range composed of the predicted centroid and predicted bandwidth, and calculating the predicted overlap between the current operating frequency range and the predicted interference range; and weighting and summing the predicted overlap, the normalized super-noise energy of the current operating frequency range, the window occupancy rate of the frequency unit in the short-window spectrum situation map, and the link quality degradation rate to obtain the migration determination value.

[0067] Among them, the predicted overlap refers to the proportion of the overlap between the current operating frequency range and the predicted interference range in the frequency domain to the current operating frequency range, which is used to quantify the degree of spatial coverage threat posed by the predicted interference to the current communication link; the normalized super-noise energy of the current operating frequency range refers to the normalized metric of the portion of the received signal energy exceeding the noise floor estimate within the frequency range corresponding to the operating frequency point, subband, or subcarrier group used by the current communication session, which is used to characterize the intensity of the interference signal affecting the current operating frequency range; the window occupancy rate of the frequency unit in the short-window spectrum situation diagram refers to the proportion of the number of observation times in which the received energy of a certain frequency unit exceeds the preset super-noise judgment threshold in the L consecutive observation times constituting the short-window spectrum situation diagram, which is used to characterize the proportion of time that the frequency unit is continuously occupied by the interference signal; the link quality degradation rate is an index obtained by weighting the decrease slope of the signal-to-interference-plus-noise ratio and the increase slope of the block error rate in the most recent multiple observation windows, which is used to characterize the degradation trend and degradation speed of the communication link quality.

[0068] In practical applications, interference blocks are determined as follows: based on the energy matrix of the short-window spectral situation diagram, the noise floor estimate of each frequency unit is determined. ; to energy Exceed Frequency units with an over-noise threshold (ηE, e.g., 6 dB) are marked as interfered units; interfered units that continuously exceed ηE on the frequency axis are aggregated into an interference block. The spectral centroid of each interference block is calculated using the over-noise energy of the interfered units as weights. and bandwidth In the presence of multiple interfering blocks, the interfering block that is closest to the current operating frequency range after prediction or has the greatest prediction overlap is selected as the final interfering block.

[0069] Furthermore, based on the spectral centroid movement trajectory of the interfering block in the short-window spectral situation map, a trajectory fitting algorithm is used to predict the predicted centroid of the interfering block in future scheduling cycles. This includes: extracting the spectral centroid sequence and bandwidth sequence of the target interfering block at L consecutive observation times from the short-window spectral situation map; calculating the weighted average of each spectral centroid in the spectral centroid sequence using the super-noise energy of the frequency unit covered by the target interfering block as the weight; performing trajectory fitting using a weighted recursive least squares algorithm with a forgetting factor based on the spectral centroid sequence to obtain an estimate of the centroid movement speed of the target interfering block; extrapolating the predicted centroid of the target interfering block in the next h scheduling cycles along the trajectory fitting direction based on the centroid movement speed estimate and the spectral centroid at the current observation time; calculating the fitting residual sequence of the spectral centroid sequence, and constructing a confidence interval for the predicted centroid based on the statistical distribution characteristics of the fitting residual sequence to obtain the predicted centroid interval. The forgetting factor assigns higher weights to spectral centroids closer to the current time during the fitting process.

[0070] Specifically, based on the spectral centroid sequence, a weighted recursive least squares algorithm with a forgetting factor is used for trajectory fitting to obtain an estimate of the centroid movement speed of the target interference block. This includes: assigning weights to the centroid values ​​at each observation time in the spectral centroid sequence according to an exponential decay rule, with higher weights for observations closer to the current time; constructing a weighted linear regression model with time as the independent variable and the spectral centroid as the dependent variable, and solving for the centroid movement speed estimate by minimizing the sum of squared residuals weighted by the weights; wherein, the forgetting factor is also used to control the decay rate of the weights, and the value of the forgetting factor is negatively correlated with the severity of the dynamic changes of interference in the electromagnetic environment in which the communication session is located, so that the response speed of the trajectory fitting to the sudden changes in the motion state of the interference adaptively improves as the dynamic changes of the interference intensify.

[0071] Specifically, based on the statistical distribution characteristics of the fitted residual sequence, a confidence interval for the predicted centroid is constructed, resulting in the predicted centroid interval. This includes: calculating the fitted residuals at each observation time in the spectral centroid sequence and calculating the standard deviation of the fitted residuals; calculating the half-width of the predicted centroid interval based on the preset confidence level and the standard deviation of the fitted residuals, combined with the time distance between the predicted time and the observation time; subtracting the half-width from the predicted centroid as the lower bound of the predicted centroid interval, and adding the half-width to the predicted centroid as the upper bound of the predicted centroid interval, thus forming the predicted centroid interval.

[0072] For example, the half-width of the predicted centroid interval is calculated using the following formula:

[0073] in, To predict the half-width of the centroid interval, The upper α / 2 quantile of the t-distribution with L-2 degrees of freedom (when L is large, the quantile of the standard normal distribution can be used). Approximate, such as at a 95% confidence level ), The standard deviation of the fitted residual is estimated, where L is the number of observation samples used for fitting (length of the spectral centroid sequence). To predict the target time (the time label of the h-th future scheduling cycle). Let L be the average of the time labels for the L observation times. This is the sum of squares of the deviations at the observation time.

[0074] In the formula for calculating the half-width of the predicted centroid interval, the first term "1" within the square root corresponds to the random fluctuation component of the new observation itself, the second term "1 / L" corresponds to the statistical uncertainty component of the estimated fitted mean based on finite historical samples, and the third term corresponds to the factor due to the prediction time. Far from the observation center The resulting trend extrapolation risk amplification component, along with the other two factors, constitutes the total metric of prediction uncertainty. As the prediction time gradually moves away from the observation window, the proportion of the extrapolation risk term gradually increases, and the prediction interval automatically widens to compensate for the extrapolation uncertainty. This ensures that even if the disturbance trend deviates to a certain degree of nonlinearity, the prediction interval can still cover the frequency range where the disturbance may occur in the future with a preset confidence level, reducing the risk of missed predictions due to single-point prediction bias.

[0075] Furthermore, determining the predicted interference interval, which consists of the predicted centroid and the predicted bandwidth, includes: extracting the bandwidth sequence of the target interference block at L consecutive observation times from the short-window spectral situation map, and determining the predicted bandwidth based on the statistical characteristics of the bandwidth sequence; and constructing the predicted interference interval based on the predicted bandwidth and the upper and lower bounds of the predicted centroid interval, so that the predicted interference interval covers the frequency range in which the target interference block may appear in future scheduling cycles in the form of intervals.

[0076] For example, the predicted interference interval is represented as:

[0077] in, To predict the lower bound of the centroid interval, To predict the upper bound of the centroid interval, For predicting bandwidth.

[0078] It is understandable that the predicted interference range can also be obtained directly using the predicted centroid and the predicted bandwidth. For example, the predicted interference range is represented as:

[0079] in, To predict the center of mass, For predicting bandwidth.

[0080] The following is a specific example illustrating how to predict interference intervals.

[0081] Assume the current operating frequency range of the communication session The short-window spectral situation map covers L=10 consecutive observation times (time labels t1 to t10). The spectral centroid sequence of the target interference block is extracted from the short-window spectral situation map. ={2425.0,2424.0,2423.2,2422.3,2421.5,2420.6,2419.8,2419.0,2418.2,2417.3} (unit: MHz), showing a continuous trend of drifting towards lower frequencies; bandwidth sequence (Unit: MHz), bandwidth is relatively stable.

[0082] A weighted recursive least squares algorithm with a forgetting factor was used for trajectory fitting. The forgetting factor γ was set to 0.85, and the weights of each observation time were distributed exponentially, with the most recent time t10 having the largest weight. The centroid movement velocity was obtained through weighted linear regression. That is, each observation cycle shifts approximately 0.85MHz towards lower frequencies. Based on the fitted centroid value and shift velocity at the current time t10, the predicted centroid for the next h=3 scheduling cycles is extrapolated: Calculate the fitting residuals for 10 observation times and statistically analyze their standard deviations. Set the confidence level to 95% (corresponding to...) Considering the distance between the predicted time t13 and the average of the observed times, calculate the half-width of the prediction interval. .

[0083] Therefore, the predicted centroid interval is [2414.75 - 0.35, 2414.75 + 0.35] = [2414.40MHz, 2415.10MHz]. The predicted bandwidth is taken as the average bandwidth of the most recent 10 observation times, i.e. Then predict the interference range =[2414.40 - 8.01 / 2, 2415.10 + 8.01 / 2] = [2410.40MHz, 2419.11MHz].

[0084] The weighting mechanism incorporating a forgetting factor makes trajectory fitting more sensitive to recent changes in the state of disturbance motion. When the disturbance source changes speed or turns, it converges to the new motion trend more quickly, avoiding the prediction lag caused by historical data delays in equal-weighted fitting. Simultaneously, a prediction confidence interval is constructed based on the statistical characteristics of the fitting residuals, explicitly quantifying the uncertainty of the model estimation as the width of the prediction interval. This allows the predicted disturbance interval to cover the range of possible future disturbance frequencies in a probabilistic sense, rather than simply providing a deterministic prediction value. When the disturbance motion exhibits nonlinear or random fluctuation characteristics, this interval prediction method effectively reduces the risk of missed detection due to single-point prediction bias, providing a more conservative and reliable threat space input for subsequent prediction overlap calculations and migration determination. In electromagnetic environments with complex and variable disturbance motion patterns, it significantly improves the robustness and prediction accuracy of fast path decision-making.

[0085] For example, calculating the prediction overlap between the current operating frequency range and the predicted interference range includes obtaining the prediction overlap by calculating the following formula:

[0086] in, This is the current operating frequency range. This is used to predict the interference range. The larger the predicted overlap value, the higher the predicted interference coverage of the current operating frequency range, and the more serious the threat.

[0087] For example, the window occupancy rate of the frequency cells in the short-window spectrum situation diagram is calculated according to the following formula:

[0088] in, In the short-window spectrum situation diagram, L represents the window occupancy rate of frequency cell k, and L is the observation window length. Let k be the received energy value at the i-th observation time and frequency unit k. This is the noise floor estimate for frequency unit k. I is the preset noise threshold, and I(·) is the indicator function. It takes the value 1 when the condition in parentheses is met, and takes the value 0 otherwise.

[0089] Furthermore, the predicted overlap, the normalized super-noise energy of the current operating frequency range, the window occupancy rate of the frequency unit in the short-window spectrum situation diagram, and the link quality degradation rate are weighted and summed to obtain the migration decision value, including the migration decision value calculated according to the following formula:

[0090] in, This is the migration determination value. This represents the normalized super-noise energy within the current operating frequency range. For window occupancy, To predict overlap, For the rate of link quality degradation, They are respectively The corresponding weights .

[0091] The calculation of the migration judgment value integrates four dimensions: predicted overlap, current frequency band energy, window occupancy rate, and link quality deterioration rate. It takes into account both the spatial movement trend of interference and the real-time changes in the current frequency band interference level and link quality, effectively suppressing false triggering caused by fluctuations in a single indicator.

[0092] S212, if the migration decision value is greater than or equal to the decision threshold, select anti-interference actions that meet the time delay constraint and reliability constraint from the current set of security actions, and generate a set of executable actions.

[0093] Here, the latency constraint means that the total estimated time overhead for the candidate anti-interference action to complete the switching preparation, parameter configuration, and execution from the current moment should be less than the remaining latency budget B of the current transmission cycle, eliminating the possibility of service timeout due to the execution of time-consuming actions; the reliability constraint means that the pre-verified reliability index of the candidate anti-interference action in the historical execution record or theoretical performance model is greater than the preset minimum reliability threshold. The pre-verified reliability index can be determined based on statistics such as the transmission success rate, signal-to-interference-plus-noise ratio recovery degree, or block error rate improvement of the candidate anti-interference action in the most recent actual executions, ensuring the quality of the action output by the fast path decision.

[0094] In some possible implementations, anti-interference actions that meet the time delay and reliability constraints are selected from the current set of security actions to generate an executable action set. This includes: identifying interference features on the short-window spectrum situation map to determine the interference type of the current interference; selecting anti-interference actions that match the interference type from the current set of security actions based on the interference type of the current interference to generate a subset of candidate actions; and performing time delay constraint checks and reliability constraint checks on each anti-interference action in the subset of candidate actions, and forming an executable action set by combining the anti-interference actions that pass the time delay constraint checks and reliability constraint checks.

[0095] For example, the interference types include one or more of the following: narrowband interference, partial bandwidth interference, interference with a clear migration direction, wideband blocking interference, interference with an uncertain migration direction, impulse interference, short-term burst interference, and multi-band simultaneous blocking interference. When the current interference is narrowband interference, partial bandwidth interference, or interference with a clear migration direction, frequency avoidance actions (e.g., switching to a clean neighboring frequency, switching to a spare subband, or switching to a non-contiguous subcarrier group) are prioritized. When the current interference is wideband blocking interference or interference with an uncertain migration direction, waveform enhancement actions (e.g., increasing the spreading factor, decreasing the modulation order, increasing coding redundancy, adjusting the receiver filter, or beam switching) are prioritized. When the current interference is impulse interference or short-term burst interference, time-domain diversity actions (e.g., interleaving depth adjustment, symbol scatter mapping, or short packet redundant transmission) are prioritized. When the current interference is multi-band simultaneous blocking interference, pre-configured spatial diversity actions (e.g., dual-link replication or multi-connection backup actions) are invoked.

[0096] The type matching priority rule based on interference features avoids invalid actions from occupying filtering resources. At the same time, the dual constraint verification further filters from the two dimensions of execution time and expected effect, ensuring that the output action is not only feasible within the current latency budget, but also has a verified reliability basis. Thus, under the ultra-low latency constraint, the relevance and security of the decision output are taken into account.

[0097] This can be understood as follows: if the migration judgment value is less than the judgment threshold, the currently executing anti-interference action remains unchanged, and a new anti-interference action selection process is not triggered.

[0098] S213, calculate the fast selection score corresponding to each anti-interference action in the set of executable actions, and select the target anti-interference action from the set of executable actions based on the fast selection score, as the anti-interference action to be executed in the fast path decision output.

[0099] In some possible implementations, the fast selection score corresponding to each anti-interference action in the set of executable actions is calculated, including: obtaining the actual success benefit of each anti-interference action to be evaluated in the set of executable actions during a preset time period; determining the rule matching degree between the anti-interference action to be evaluated and the current interference type and the current link state, and obtaining the first action switching cost and the first estimated execution time corresponding to the anti-interference action to be evaluated; and weighting and summing the success benefit and rule matching degree as positive factors, and the first action switching cost and the first estimated execution time as negative factors to obtain the fast selection score of the anti-interference action to be evaluated.

[0100] Among them, the success benefit of the anti-interference action to be evaluated in the preset time period refers to the statistical quantification value of the transmission effect feedback obtained when the anti-interference action to be evaluated is actually executed in the recent period. It can be obtained by normalizing and comprehensively calculating based on indicators such as transmission success rate, throughput, end-to-end latency improvement, or signal-to-interference-plus-noise ratio recovery degree after each execution. The rule matching degree between the anti-interference action to be evaluated and the current interference type and the current link state refers to the quantification index of the applicability of the anti-interference action to be evaluated to the current interference scenario based on the preset interference type identification results and link state characteristics. It can be determined based on the pre-built interference-action matching rule base. The rule base defines the matching relationship and corresponding matching score between different interference characteristics (such as narrowband interference, wideband blocking, pulse burst, frequency sweeping interference, etc.) and different anti-interference action types (such as frequency avoidance, spread spectrum enhancement, modulation order reduction, redundant transmission, etc.); the first action switching cost refers to the quantitative measure of the additional overhead required to switch from the currently executing anti-interference action to the anti-interference action a1 to be evaluated (such as the signaling interaction delay during the switching process); the first estimated execution time refers to the estimated time overhead required for the anti-interference action a1 to be evaluated to be selected from the time it is completed and produces an anti-interference effect (such as the parameter configuration time of the action).

[0101] For example, the success benefit and rule matching degree are used as positive factors, and the first action switching cost and the first estimated execution time are used as negative factors. A weighted sum is then performed to obtain the quick selection score for the anti-interference action to be evaluated, including the quick selection score calculated according to the following formula:

[0102] in, The score for the fast selection of the anti-interference action a1 to be evaluated is given. To evaluate the actual success benefits of the anti-interference action a1 during the preset time period, To evaluate the rule matching degree between the anti-interference action a1 to be evaluated and the current interference type and current link state, The switching cost of the first action corresponding to the anti-interference action a1 to be evaluated is... The first estimated execution time corresponding to the anti-interference action a1 to be evaluated is... They are respectively The corresponding weights.

[0103] Positive factors ensure that the system prioritizes actions with good historical performance and high adaptability to the current interference scenario, making full use of existing experience and expert knowledge. Negative factors, on the other hand, constrain candidate actions from two dimensions: latency overhead and switching cost. This makes actions with high computational overhead or high switching cost disadvantaged in the score comparison, thus suppressing the selection of unnecessary frequent switching and time-consuming actions. With extremely low computational complexity, the system achieves multi-dimensional rational selection of candidate actions, ensuring the timeliness and rationality of fast path decision output under ultra-low latency constraints.

[0104] In some possible implementations, the fast selection score corresponding to each anti-interference action in the set of executable actions is calculated, including: obtaining the actual success benefit of each anti-interference action to be evaluated in the set of executable actions during a preset time period; determining the rule matching degree between the anti-interference action to be evaluated and the current interference type and the current link state, and obtaining the first action switching cost, the first expected execution time and the first interference exposure risk corresponding to the anti-interference action to be evaluated; and weighting and summing the success benefit and rule matching degree as positive factors, and the first action switching cost, the first expected execution time and the first interference exposure risk as negative factors, to obtain the fast selection score of the anti-interference action to be evaluated.

[0105] Among them, the first interference exposure risk refers to the quantitative measure of the degree of risk that the communication signal will be intercepted, tracked or targeted by the interference source when the anti-interference action a1 to be evaluated is performed (e.g., the low interception characteristic index of the transmission waveform corresponding to the anti-interference action a1 to be evaluated).

[0106] For example, the success benefits and rule matching degree are used as positive factors, and the switching cost of the first action, the first expected execution time, and the risk of the first interference exposure are used as negative factors. These factors are then weighted and summed to obtain the quick selection score for the anti-interference action to be evaluated. This includes calculating the quick selection score according to the following formula:

[0107] in, The score for the fast selection of the anti-interference action a1 to be evaluated is given. To evaluate the actual success benefits of the anti-interference action a1 during the preset time period, To evaluate the rule matching degree between the anti-interference action a1 to be evaluated and the current interference type and current link state, The switching cost of the first action corresponding to the anti-interference action a1 to be evaluated is... The first estimated execution time corresponding to the anti-interference action a1 to be evaluated is... To assess the first interference exposure risk corresponding to the anti-interference action a1, They are respectively The corresponding weights.

[0108] The newly added interference exposure risk factor enables fast path decision-making to not only focus on immediate anti-interference effectiveness but also to proactively consider feedback from interference source behavior. For example, blindly increasing transmission power in an interference environment may temporarily restore the link, but it could also expose the location of communication nodes or trigger power tracking by the interference source, leading to stronger subsequent interference. By explicitly incorporating interference exposure risk into the score calculation, short-sighted and aggressive behavior is effectively suppressed while ensuring current transmission reliability, reducing the overall vulnerability of communication sessions, and achieving a balance between immediate anti-interference and long-term link concealment in dynamic adversarial environments.

[0109] In the fast path decision-making scheme, firstly, interference trajectory prediction and migration determination based on short-window spectrum situation maps can identify threats in advance before interference actually covers the current operating frequency range, shifting the initiation time of avoidance actions from "passively discovering interference" to "actively anticipating threats." Secondly, when generating the set of executable actions, dual screening conditions of latency constraints and reliability constraints are applied simultaneously to ensure that anti-interference actions are not only feasible within the expected execution time, but also that their pre-verified reliability is not lower than the minimum threshold, guaranteeing the determinism and effectiveness of the fast path decision output. Thirdly, a quantitative comparison of anti-interference actions is performed by quickly selecting scores, integrating historical execution benefits, rule matching degree, and multi-dimensional cost factors to complete the action optimization. In this way, fast path decision-making achieves efficient anti-interference response under ultra-low latency constraints.

[0110] Optionally, in step S20, the fast path decision is used to select anti-interference actions from the current set of security actions, and further includes the following steps: determining the current interference type of the communication session based on the short window spectrum situation map; when the current interference type is Orthogonal Frequency Division Multiplexing (OFDM) partial bandwidth interference, frequency sweeping interference, or random subcarrier interference, the symbol scatter mapping and interleaving action is taken as the anti-interference action to be executed by the fast path decision output, wherein the symbol scatter mapping and interleaving action includes: mapping multiple modulation components of the same service symbol or the same coding block to multiple non-contiguous subcarriers or subcarrier groups according to a preset frequency interval; and interleaving the mapped modulation component sequence to scramble the time domain or frequency domain transmission order.

[0111] Figure 6 This diagram illustrates the principle of symbolic scattering and interleaving. Each small square represents a subcarrier carrying a modulation component. A blank square indicates that the modulation component carried on that subcarrier was correctly received after transmission; a square with an "X" indicates that the modulation component carried on that subcarrier was transmitted incorrectly due to interference. (Combined with...) Figure 6 As shown, discontinuous subcarriers or subcarrier groups carry multiple modulation components of the same service symbol. After interleaving and dispersion, the modulation components carried on the disturbed subcarrier are discretized into randomly distributed isolated errors during deinterleaving and demapping at the receiver, rather than continuous errors concentrated in the same coding block, thus falling within the error-correctable range of error-correcting coding. The triggering conditions for symbol dispersion mapping and interleaving actions precisely correspond to the spectral characteristics of OFDM partial bandwidth interference, frequency sweep interference, and random subcarrier interference. This allows the interference type identification results to directly drive the action selection for fast path decision-making, making the decision-making link simple and efficient, and further compressing the overall response delay from interference perception to policy execution.

[0112] Optionally, combined Figure 5 As shown, in step S20, the adaptive decision is used to update the set of candidate security actions for the next transmission cycle, including the following steps: S221 encodes the short-window spectrum situation map, service link quality indicators, and historical action benefits of the communication session into a context vector.

[0113] Among them, context vector This refers to a multi-dimensional real-valued vector formed by extracting and normalizing features from the multi-source heterogeneous state information of a communication session during the t-th transmission cycle and historical time period, and then concatenating them in a preset order. This vector represents a comprehensive contextual feature of the current electromagnetic environment, link status, and historical decision-making effectiveness. The context vector includes at least several of the following components: normalized super-noise energy, window occupancy rate, predicted overlap, interference centroid movement speed, signal-to-interference-plus-noise ratio margin, block error rate, end-to-end delay, retransmission count, queue growth rate, and recent success rate, average hold time, and most recent usage interval of candidate anti-interference actions. Each component is normalized to the [0,1] interval, ensuring comparability of indicators with different dimensions in subsequent revenue estimation and priority calculation. The introduction of the context vector enables the system to uniformly map complex and varied communication scenarios to the same vector space, providing a structured input basis for revenue generalization based on historical experience.

[0114] S222, based on the context vector, determines the estimated benefit and the upper bound of the confidence interval of the estimated benefit of each candidate anti-interference action to be evaluated based on the parameter matrix and benefit vector corresponding to each candidate anti-interference action in the candidate security action set.

[0115] Wherein, parameter matrix This refers to a d×d-dimensional symmetric positive definite matrix, independently maintained for each candidate anti-interference action a2 to be evaluated in the candidate safety action set, used for solving the weight vector of the linear payoff model using online ridge regression; payoff vector It refers to a d×1 dimensional vector that is independently maintained for each candidate anti-interference action in the candidate safety action set and is used to solve the weight vector of the linear revenue model in online ridge regression.

[0116] In some possible implementations, the candidate anti-interference action a2 to be evaluated is actually executed and the normalized actual benefit is obtained. Then, the parameter matrix and payoff vector Perform online incremental updates as follows:

[0117] in, Forgetting factor, , Initialize to λI, where λ is the regularization coefficient and I is the identity matrix.

[0118] Parameter matrix and payoff vector As a sufficient statistical measure of historical observation samples for this action, the computational complexity of its model update is O(d²) (where d is the dimension of the context vector), which is independent of the total number of historical samples, and the storage overhead is also constant at O(d²), avoiding the computational and storage burden of traditional batch retraining schemes that increases linearly with the accumulation of samples. Simultaneously, the introduction of the forgetting factor ρ causes the contribution of historical samples to decay exponentially over time, allowing the model to automatically downplay the impact of outdated experience on current gain estimation and maintain sensitivity to recent link feedback in adversarial environments where interference strategies change dynamically. The online incremental update scheme achieves lightweight continuous learning of the candidate anti-interference action gain model with constant low computational and storage overhead, enabling the adaptive decision-making gain model to continuously track non-stationary electromagnetic environments, providing efficient and stable online learning support for the accuracy and timeliness of candidate priority ranking.

[0119] Estimated benefits of candidate anti-interference action a2 to be evaluated Let a be the linear benefit model weight vector of the candidate anti-interference action a2 to be evaluated. Let be the transpose of the linear payoff model weight vector of the candidate anti-interference action a2 to be evaluated. Let be the context vector for the t-th transmission cycle. This is the parameter matrix corresponding to the candidate anti-interference action a2 to be evaluated. Let be the reward vector corresponding to the candidate anti-interference action a2 to be evaluated. Substitution .

[0120] The upper bound of the payoff confidence interval is the upper limit of the payoff that can be estimated at a given confidence level when statistically estimating the expected payoff of candidate anti-interference actions in the current context based on historical observation data.

[0121] S223, obtain the second action switching cost and the second estimated execution time corresponding to the candidate anti-interference action to be evaluated.

[0122] The second action switching cost refers to a quantitative measure of the additional overhead required to switch from the currently executing anti-interference action to the candidate anti-interference action a2 to be evaluated (e.g., signaling interaction latency during the switching process); the second estimated execution time refers to the estimated time overhead required for the candidate anti-interference action a2 to be evaluated to be selected, completed, and produce an anti-interference effect (e.g., parameter configuration time of the action).

[0123] S224. Calculate the priority of the candidate anti-interference actions to be evaluated based on the estimated revenue, the upper bound of the estimated revenue confidence interval, the switching cost of the second action, and the second estimated execution time.

[0124] In some possible implementations, the priority of the candidate anti-interference actions to be evaluated is calculated based on the estimated benefit, the upper bound of the estimated benefit confidence interval, the second action switching cost, and the second estimated execution time. This includes calculating the priority of the candidate anti-interference actions to be evaluated according to the following formula:

[0125] in, Let a2 be the priority score of the candidate anti-interference action to be evaluated in the t-th transmission cycle. To estimate the revenue, To estimate the upper bound of the return confidence interval, Let be the context vector of the communication session in the t-th transmission cycle. for The transpose of , Here is the parameter matrix of the candidate anti-interference action a2 to be evaluated. To explore the coefficient, The switching cost of the second action corresponding to the candidate anti-interference action a2 to be evaluated. The second estimated execution time is the candidate anti-interference action a2 to be evaluated. They are respectively The corresponding penalty coefficient.

[0126] Priority of the above-mentioned candidate anti-interference actions to be evaluated The calculation formula is as follows: The first term, estimated benefit, drives the system to prioritize actions with good historical performance, fully reusing verified experience. The second term, the upper bound of the estimated benefit confidence interval, provides a structured incentive for exploration behavior. For actions with few historical executions or novel current contexts, this term automatically increases, encouraging the system to explore in areas of high uncertainty. For frequently used actions whose estimated benefits have converged, this term automatically decreases, reducing ineffective exploration. The third term, cost penalty, explicitly incorporates action switching overhead and execution time into the priority ranking, mechanistically suppressing the selection of high-cost actions. Thus, adaptive decision-making automatically achieves a statistically optimal balance between utilizing known high-quality actions and exploring unknown potential high-quality actions without manual parameter tuning or pre-set exploration strategies.

[0127] In some possible implementations, the priority of the candidate anti-interference actions to be evaluated is calculated based on the estimated benefit, the upper bound of the estimated benefit confidence interval, the switching cost of the second action, and the second estimated execution time. This includes: determining the second interference exposure risk corresponding to the candidate anti-interference actions to be evaluated; and calculating the priority of the candidate anti-interference actions to be evaluated according to the following formula:

[0128] in, The priority score of the candidate anti-interference action a2 to be evaluated in the t-th transmission cycle. To estimate the revenue, To estimate the upper bound of the return confidence interval, The context vector of the communication session in the t-th transmission cycle. for The transpose of , Here is the parameter matrix of the candidate anti-interference action a2 to be evaluated. To explore the coefficient, The switching cost of the second action corresponding to the candidate anti-interference action a2 to be evaluated. The second estimated execution time is the candidate anti-interference action a2 to be evaluated. To assess the second interference exposure risk corresponding to the candidate anti-interference action a2, They are respectively The corresponding penalty coefficient.

[0129] Among them, the second interference exposure risk refers to the quantitative measure of the degree of risk that the communication signal will be intercepted, tracked or targeted by the interference source when the candidate anti-interference action a2 to be evaluated is executed (e.g., the low interception characteristic index of the transmission waveform corresponding to the candidate anti-interference action a2 to be evaluated).

[0130] Priority of the above-mentioned candidate anti-interference actions to be evaluated The calculation formula adds a penalty item for interference exposure risk. Interference exposure risk quantification measures the potential risk of communication signals being intercepted, tracked, or specifically suppressed by the interference source when executing the candidate anti-interference action to be evaluated. It is applicable to adversarial scenarios where the interference source possesses intelligent tracking or reactive interference capabilities. While ensuring current transmission reliability, it guides the system to prioritize more covert or less likely to escalate interference by explicitly penalizing high-exposure-risk actions, achieving a balance between immediate anti-interference effectiveness and long-term link security. This allows adaptive decision-making to automatically achieve a statistically optimal balance between utilizing known high-quality actions and exploring unknown potential high-quality actions without manual parameter tuning or pre-set exploration strategies. It also considers latency feasibility and anti-interference covertness, driving the priority ranking of the candidate security action set to continuously optimize as the electromagnetic environment evolves.

[0131] In some possible implementations, when the candidate anti-interference action a2 to be evaluated is a joint action, the priority of the joint action is determined as follows: the joint action is decomposed into frequency domain atomic actions, waveform atomic actions, and redundant atomic actions. Frequency domain atomic actions include operating frequency selection, subband selection, or subcarrier group selection; waveform atomic actions include spreading parameter selection, modulation order selection, coding redundancy selection, interleaving depth selection, or symbol scatter mapping strength selection; and redundant atomic actions include short packet redundant transmission, repeated transmission, or dual-link replication. The atomic priorities corresponding to each of the frequency domain atomic actions, waveform atomic actions, and redundant atomic actions are calculated separately. ; Obtain the cooperative gain values ​​corresponding to frequency domain atomic actions, waveform atomic actions, and redundant atomic actions during historical cooperative execution; Perform conflict screening on combinations of frequency domain atomic actions, waveform atomic actions, and redundant atomic actions through a compatibility table. Conflict screening includes screening out atomic action combinations that have radio frequency resource conflicts, protocol parameter conflicts, or joint execution times exceeding the delay budget of the next transmission cycle; For atomic action combinations that pass the conflict screening, sum the atomic priority of the frequency domain atomic actions, the atomic priority of the waveform atomic actions, the atomic priority of the redundant atomic actions, and the cooperative gain value, and deduct the joint handover cost to obtain the joint action priority of the joint action.

[0132] In practical applications, the atomic priority of frequency domain atomic actions is calculated based on sub-vectors in the context vector related to candidate frequency band energy, window occupancy rate, prediction overlap, and handover time. The atomic priority of waveform atomic actions is calculated based on sub-vectors in the context vector related to signal-to-interference-plus-noise ratio margin, block error rate, interference bandwidth, and processing overhead. The atomic priority of redundancy atomic actions is calculated based on sub-vectors in the context vector related to retransmission count, end-to-end delay, link dependence, and remaining resources.

[0133] The collaborative gain value is determined as follows: the actual total gain obtained by frequency domain atomic actions, waveform atomic actions and redundant atomic actions in different combinations during the historical execution process is recorded; the difference between the actual total gain and the sum of the atomic gains when each atomic action is executed alone is taken as the collaborative gain value of the corresponding combination; for atomic action combinations that have not appeared in the history, the collaborative gain value is zero or a preset default value.

[0134] For example, the priority of joint actions is calculated according to the following formula:

[0135] in, To determine the joint action priority of the candidate anti-interference action a2 to be evaluated, Frequency domain atomic actions Atom priority, Waveform Atom Actions Atom priority, For redundant atomic actions Atom priority, Frequency domain atomic actions Waveform Atomic Actions and redundant atomic actions The collaborative gain value generated during historical collaborative execution. The joint switching cost corresponding to joint action a2, This is the penalty coefficient for joint handover costs.

[0136] By decomposing the high-dimensional joint action space into three dimensions—frequency domain, waveform, and redundancy—and calculating atomic priorities separately, the computational complexity is reduced from the product level of the number of actions in each domain to the summation level. The introduction of cooperative gain values ​​captures the additional benefits or offsetting effects generated by the cooperation between actions in different domains, compensating for the inability of simple atomic priority summation to reflect the interaction effects of actions. Simultaneously, a compatibility table filtering mechanism applies hard constraints to infeasible combinations from three dimensions: RF resources, protocol parameters, and latency budget, ensuring that the joint actions ultimately entering the ranking are feasible at both the physical and protocol layers. Thus, within the limited computational latency budget of adaptive decision-making, the optimization efficiency and output quality of the multi-domain joint anti-interference strategy are effectively improved, balancing the requirements of action space coverage, evaluation accuracy, and output feasibility.

[0137] S225, the candidate anti-interference actions in the candidate security action set are sorted and updated in descending order of priority, and the candidate anti-interference actions with higher priority are selected first in the next transmission cycle.

[0138] Adaptive decision-making updates the set of candidate security actions for the next transmission cycle while executing the current anti-interference action in the fast path decision. During the update, based on the estimated benefits and upper bounds of the benefit confidence intervals of each candidate action, combined with the penalty deduction for action switching costs and expected execution times, the priority of candidate actions is dynamically calculated and sorted. This allows the set of candidate security actions to continuously adapt to changes in the time-varying interference environment and service requirements, breaking the limitations of a fixed pre-configured action set. The security action set can continuously self-optimize with changes in the channel environment and interference modes, thereby improving the accuracy and long-term robustness of decisions in subsequent transmission cycles while maintaining the real-time performance of anti-interference communication.

[0139] In some possible implementations, adaptive decision-making for updating the candidate security action set for the next transmission cycle further includes: identifying unverified candidate anti-interference actions in the candidate security action set that have never been actually executed or have not been used for a preset duration; generating shadow evaluation results of the unverified candidate anti-interference actions using spectrum observation data of the current communication session and the calibrated link model without transmitting actual signals, wherein the shadow evaluation results include predicted benefit values; adjusting the priority of the unverified candidate anti-interference actions based on the shadow evaluation results, and retaining the unverified candidate anti-interference actions that still meet the admission conditions after adjustment in the candidate security action set.

[0140] Furthermore, using the spectrum observation data of the current communication session and the calibrated link model, shadow evaluation results of unverified candidate anti-interference actions are generated, including: determining the action domain type to which the unverified candidate anti-interference action belongs, including frequency domain actions, waveform actions, and redundant actions; based on the action domain type, extracting the link quality parameters of the corresponding action domain from the spectrum observation data of the current communication session, and inputting the link quality parameters into the calibrated link model to obtain the predicted transmission performance index corresponding to the unverified candidate anti-interference action; and determining the predicted gain value of the unverified candidate anti-interference action based on the difference between the predicted transmission performance index and the current actual transmission performance index.

[0141] Here, the link model is defined as a pre-established model used to describe the mapping relationship between link quality parameters and transmission performance indicators, such as the signal-to-interference-plus-noise ratio (SINR)-block error rate (BER) mapping model, waveform configuration-BER performance curve model, or redundant transmission success rate model.

[0142] For example, without verifying that the candidate anti-interference action is a frequency domain action, the received power, noise power, and interference power of the candidate frequency band are extracted from the spectrum observation data of the current communication session; the predicted signal-to-interference-plus-noise ratio (SINNR) of the candidate frequency band is estimated based on the received power, noise power, and interference power; the predicted SINNR is input into the SINNR-block error rate mapping model to obtain the predicted transmission success rate; and the difference between the predicted transmission success rate and the actual transmission success rate of the current operating frequency band is used as the predicted benefit value.

[0143] For example, without verifying that the candidate anti-interference action is a waveform action, the measured signal-to-interference-plus-noise ratio (SIR) of the current operating frequency band is extracted from the spectrum observation data of the current communication session; the measured SIR is input into the block error rate performance curve model under different waveform configurations to obtain the predicted block error rate; and the difference between the predicted block error rate and the actual block error rate under the current waveform configuration is used as the predicted benefit value.

[0144] For example, without verifying that the candidate anti-interference action is a redundant action, the single transmission success rate, the number of redundant replications, and the inter-link correlation coefficient of the current link are extracted from the spectrum observation data of the current communication session; the single transmission success rate, the number of redundant replications, and the inter-link correlation coefficient are input into the redundant transmission success rate model to obtain the comprehensive transmission success rate; the difference between the comprehensive transmission success rate and the single transmission success rate is used as the predicted revenue value.

[0145] Furthermore, based on the shadow evaluation results, the priority of unverified candidate anti-interference actions is adjusted, including: obtaining the predicted revenue value corresponding to the unverified candidate anti-interference action; weighting and fusing the predicted revenue value with the estimated revenue of the unverified candidate anti-interference action based on historical actual execution data to obtain a mixed revenue estimate; replacing the original estimated revenue with the mixed revenue estimate and participating in the priority calculation of the unverified candidate anti-interference action; wherein, the weight of the predicted revenue value in the weighted fusion is lower than the weight of the estimated revenue based on historical actual execution data.

[0146] It is understood that, in addition to the predicted revenue value, the shadow assessment result may also be one or more of the following parameters: predicted reliability, predicted revenue confidence interval, predicted action hold time, predicted resource cost, predicted switching cost, or predicted interference exposure risk. This invention does not limit these parameters.

[0147] The admission criteria include one or more of the following: reliability criteria, latency budget constraints, and protocol compatibility criteria. The reliability criteria confirm that the confidence lower bound of the estimated benefit of unverified candidate anti-interference actions is higher than a preset minimum reliability threshold. The latency budget constraints confirm that the expected execution time of the unverified candidate anti-interference actions is less than the latency budget of the next transmission cycle. The protocol compatibility criteria confirm that the unverified candidate anti-interference actions are compatible with the current protocol stack, RF channel configuration, and service type. When the above admission criteria are met, the unverified candidate anti-interference actions are marked as pending verification actions and retained in the candidate security action set. Unverified candidate anti-interference actions that do not meet the admission criteria are temporarily removed from the candidate security action set, and shadow evaluation is re-triggered in subsequent transmission cycles. After the pending verification action is actually executed and the link quality verification result is obtained, its pending verification mark is cleared, and its parameter matrix and benefit vector are updated with the actual benefit corresponding to the link quality verification result.

[0148] By utilizing current spectrum observation data and calibrated link models, predicted benefits are derived from three dimensions: frequency domain, waveform, and redundancy. These benefits are then incorporated into priority calculations with low weights, providing opportunities for new actions to enter the candidate pool while ensuring that current transmission security is not affected. Without transmitting actual signals or occupying service resources, zero-risk pre-assessment and orderly introduction of unverified candidate actions are achieved, effectively alleviating the cold start dilemma of online learning and accelerating the convergence and optimization of anti-interference strategies.

[0149] In some embodiments, the latency-aware anti-interference communication method further includes: collecting service link quality indicators of the communication session and determining the link quality verification result based on the service link quality indicators; if the link quality verification result shows that the link quality is higher than the reliability threshold, extending the action holding time of the current anti-interference action or increasing the priority of the current anti-interference action in the current security action set; if the link quality verification result shows that the link quality is lower than the reliability threshold, shortening the action holding time of the current anti-interference action, reducing the priority of the current anti-interference action, or removing the current anti-interference action from the current security action set.

[0150] Among them, the service link quality index refers to a set of multi-dimensional parameters collected from the receiving or transmitting end of the communication session after the anti-interference action is executed, used to measure the current transmission quality of the communication link. These parameters include one or more of the following: signal-to-interference-plus-noise ratio (SINR), block error rate, retransmission count, end-to-end latency, queue length, throughput, transmission success rate, and control packet timeout count. The link quality verification result refers to a binary judgment conclusion obtained by comparing the collected service link quality index with a preset reliability threshold, used to confirm whether the currently executed anti-interference action has effectively restored the link quality.

[0151] When the link quality verification result shows that the link quality is higher than the reliability threshold, it indicates that the current anti-interference action is effective, and positive incentives are implemented: extend the retention time of the current anti-interference action so that it remains effective in multiple subsequent transmission cycles and reduces unnecessary action switching; or increase the priority of the current anti-interference action in the set of security actions so that it will be selected first in similar interference scenarios in the future.

[0152] When the link quality verification result shows that the link quality is lower than the reliability threshold, it indicates that the current anti-interference action has failed to effectively combat the current interference. Negative adjustments should be made: shorten the holding time of the current anti-interference action to accelerate its failure and free up decision space; reduce the priority of the current anti-interference action to reduce its probability of being selected in the future; or directly remove the current anti-interference action from the current set of security actions to prevent invalid actions from continuing to occupy the candidate space and mislead subsequent decisions.

[0153] The actual link feedback after the anti-interference action is executed is directly applied to the dynamic adjustment of the safety action set, so that the safety action set is always composed of effective actions that have been verified in practice, thus ensuring the quality of candidate actions for fast path decision-making from the source.

[0154] Combination Figure 7 As shown, this embodiment of the invention provides a latency-aware anti-interference communication system 70, including a service state determination module 701 and an anti-interference response strategy generation module 702. The service state determination module 701 is configured to determine the remaining latency budget and fast state urgency of a communication session in the current transmission cycle. The anti-interference response strategy generation module 702 is configured to determine and execute an anti-interference response strategy for the communication session based on the remaining latency budget and fast state urgency to achieve anti-interference communication. The anti-interference response strategy includes executing only fast path decisions, or executing fast path decisions and adaptive decisions in parallel. Fast path decisions are used to select anti-interference actions from the current set of security actions and execute them in the current transmission cycle. Adaptive decisions are used to update the candidate set of security actions for the next transmission cycle. The updated candidate set of security actions serves as the current set of security actions for the next transmission cycle.

[0155] Combination Figure 8 As shown, this embodiment of the invention provides a time-delay-aware anti-interference communication device 80, including a processor 800 and a memory 801, and may also include a communication interface 802 and a bus 803. The processor 800, communication interface 802, and memory 801 can communicate with each other via the bus 803. The communication interface 802 can be used for information transmission. The processor 800 can call logical instructions in the memory 801 to execute the time-delay-aware anti-interference communication method of the above embodiment.

[0156] Furthermore, the logic instructions in the aforementioned memory 801 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium.

[0157] The memory 801, as a computer-readable storage medium, can be used to store software programs and computer-executable programs, such as the program instructions / modules corresponding to the methods in the embodiments of the present invention. The processor 800 executes functional applications and data processing by running the program instructions / modules stored in the memory 801, thereby implementing the time-delay-aware anti-interference communication method in the above method embodiments.

[0158] The memory 801 may include a program storage area and a data storage area. The program storage area may store the operating system and application programs required for at least one function; the data storage area may store data created based on the use of the terminal device. Furthermore, the memory 801 may include high-speed random access memory and may also include non-volatile memory.

[0159] Processor 800 is a circuit with signal processing capabilities. In one implementation, processor 800 can be a circuit with instruction read and execute capabilities, such as a central processing unit (CPU), microprocessor, graphics processing unit (GPU) (which can be understood as a type of microprocessor), or digital signal processor (DSP). In another implementation, processor 800 can achieve certain functions through the logical relationships of hardware circuits. These logical relationships of hardware circuits are fixed or reconfigurable. For example, processor 800 can be a hardware circuit implemented as an ASIC or a programmable logic device (PLD), such as an FPGA. In reconfigurable hardware circuits, the process of the processor loading a configuration document and configuring the hardware circuit can be understood as the processor loading instructions to achieve the functions of some or all of the above modules. Furthermore, it can also be a hardware circuit designed for artificial intelligence, which can be understood as a type of ASIC, such as a neural network processing unit (NPU), tensor processing unit (TPU), or deep learning processing unit (DPU). The processor 800 is used to execute related programs to implement the functions required by the units in the time-delay-aware anti-interference communication device of the present invention, or to execute the time-delay-aware anti-interference communication method of the present invention.

[0160] As can be seen, each module in the above device can be one or more processors (or processing circuits) configured to implement the above methods, such as: CPU, GPU, NPU, TPU, DPU, microprocessor, DSP, ASIC, FPGA, or a combination of at least two of these processor types.

[0161] Furthermore, the modules in the above devices can be integrated in whole or in part, or they can be implemented independently. In one implementation, these modules are integrated together as a system-on-a-chip (SOC). The SOC may include at least one processor for implementing any of the above methods or for implementing the functions of the modules of the device. The at least one processor may be of different types, such as CPU and FPGA, CPU and artificial intelligence processor, CPU and GPU, etc.

[0162] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention. Clearly, those skilled in the art can make various alterations and modifications to the invention without departing from its spirit and scope. Thus, if these modifications and modifications of the invention fall within the scope of the claims and their equivalents, the invention is also intended to include these modifications and modifications.

Claims

1. A time-delay-aware anti-interference communication method, characterized in that, include: Determine the remaining latency budget and fast state urgency of the communication session in the current transmission cycle; Based on the remaining latency budget and the urgency of the rapid state, determine and implement the anti-interference response strategy for the communication session to achieve anti-interference communication; The anti-interference response strategy includes performing only fast path decision, or performing fast path decision and adaptive decision in parallel. Fast path decision is used to select anti-interference action from the current set of security actions and execute it in the current transmission cycle. Adaptive decision is used to update the set of candidate security actions for the next transmission cycle. After the candidate security action set is updated, it becomes the current set of security actions for the next transmission cycle.

2. The method according to claim 1, characterized in that, Determine the remaining latency budget for the service, including: obtaining the maximum tolerable latency, consumed transmission latency, queue waiting latency, and policy execution estimated latency of the communication session in the current transmission cycle; determining the security protection margin based on the statistical deviation between the actual end-to-end latency and the estimated end-to-end latency of multiple historical transmission cycles; and obtaining the remaining latency budget for the service by subtracting the consumed transmission latency, queue waiting latency, policy execution estimated latency, and security protection margin from the maximum tolerable latency for the service. Determining the urgency of the fast state includes: constructing a fast state vector for the communication session in the current transmission cycle, wherein the fast state vector consists of at least two of the following: the normalized value of the queue growth rate, the link quality degradation rate, the proportion of abnormal service packet transmission, and the proportion of ready actions in the current set of security actions; assigning corresponding weights to each component of the fast state vector, and summing the weighted components to obtain the urgency of the fast state.

3. The method according to claim 1, characterized in that, Adaptive decision-making includes first adaptive decision-making and second adaptive decision-making; Based on the remaining latency budget and the urgency of the rapid state, determine the anti-interference response strategy for the communication session, including: By combining the latency threshold and the urgency threshold, threshold determination is made on the remaining latency budget and the urgency of the fast state of the service to determine the latency-urgency state interval of the communication session in the current transmission cycle. Based on the time delay-urgency state interval, the decision mode of the anti-interference response strategy is determined. The decision mode includes a single decision mode and a combined decision mode. The single decision mode is to execute only the fast path decision, and the combined decision mode is to execute the fast path decision and the adaptive decision in parallel. When the decision-making mode is a combined decision-making mode, the first adaptive decision or the second adaptive decision is determined to be executed based on the adaptive decision level corresponding to the delay-urgency state interval. The first adaptive decision updates the set of candidate security actions for the next transmission cycle within the range of the first candidate action, and the second adaptive decision updates the set of candidate security actions for the next transmission cycle within the range of the second candidate action. The range of the second candidate action is greater than the range of the first candidate action.

4. The method according to claim 3, characterized in that, Fast path decision is used to select anti-interference actions from the current set of safety actions, including: Based on the short-window spectrum situation diagram of the communication session, a migration judgment value is calculated to characterize whether strong interference is migrating to the current operating frequency range. If the migration decision value is greater than or equal to the decision threshold, anti-interference actions that meet the time delay constraints and reliability constraints are selected from the current set of security actions to generate a set of executable actions. Calculate the fast selection score corresponding to each anti-interference action in the set of executable actions, and select the target anti-interference action from the set of executable actions based on the fast selection score, as the anti-interference action to be executed in the fast path decision output.

5. The method according to claim 4, characterized in that, Based on the short-window spectrum situation diagram of the communication session, a migration judgment value is calculated to characterize whether strong interference is migrating to the current operating frequency range, including: Based on the spectral centroid movement trajectory of the interference block in the short-window spectral situation diagram, a trajectory fitting algorithm is used to predict the predicted centroid and predicted bandwidth of the interference block in future scheduling cycles. Determine the current operating frequency range of the communication session, as well as the predicted interference range consisting of the predicted centroid and the predicted bandwidth, and calculate the predicted overlap between the current operating frequency range and the predicted interference range. The migration decision value is obtained by weighting and summing the predicted overlap, the normalized super-noise energy of the current operating frequency range, the window occupancy rate of the frequency unit in the short window spectrum situation diagram, and the link quality deterioration rate.

6. The method according to claim 4, characterized in that, Calculate the fast selection score for each anti-interference action in the set of executable actions, including: Obtain the actual success benefit of each anti-interference action to be evaluated in the set of executable actions during a preset time period; Determine the rule matching degree between the anti-interference action to be evaluated and the current interference type and current link state, and obtain the first action switching cost and the first estimated execution time corresponding to the anti-interference action to be evaluated; By weighting and summing the success benefits and rule matching degree as positive factors, and the first action switching cost and the first estimated execution time as negative factors, we can obtain the rapid selection score of the anti-interference action to be evaluated.

7. The method according to claim 1, characterized in that, Adaptive decision-making is used to update the set of candidate security actions for the next transmission cycle, including: The short-window spectrum situation diagram of the communication session, service link quality indicators, and historical action benefits are jointly encoded into a context vector; Based on the context vector, the estimated benefit and the upper bound of the confidence interval of the estimated benefit are determined according to the parameter matrix and benefit vector corresponding to each candidate anti-interference action to be evaluated in the candidate security action set. Obtain the second action switching cost and the second estimated execution time corresponding to the candidate anti-interference action to be evaluated; The priority of the candidate anti-interference actions to be evaluated is calculated based on the estimated revenue, the upper bound of the estimated revenue confidence interval, the switching cost of the second action, and the second estimated execution time. Candidate anti-interference actions in the candidate security action set are sorted and updated in descending order of priority. Among them, the candidate anti-interference actions with higher priority are selected first in the next transmission cycle.

8. The method according to claim 7, characterized in that, Based on the estimated revenue, the upper bound of the estimated revenue confidence interval, the switching cost of the second action, and the second estimated execution time, the priority of the candidate anti-interference actions to be evaluated is calculated, including calculating the priority of the candidate anti-interference actions to be evaluated according to the following formula: ； Among them, P t(a2) U represents the priority score of candidate anti-interference action a2 to be evaluated in the t-th transmission cycle. t(a2） To estimate the revenue, To estimate the upper bound of the confidence interval for returns, x t Let x be the context vector of the communication session in the t-th transmission cycle. t T For x t The transpose of A a2 Let C be the parameter matrix of a2, δ be the exploration coefficient, and C be the parameter matrix of a2. sw(a2) T is the switching cost of the second action corresponding to a2. exec(a2) Let a2 be the second estimated execution time, and μ1 and μ2 be C, respectively. sw(a2) and T exec(a2) The corresponding penalty coefficient.

9. A time-delay-aware anti-interference communication system, characterized in that, include: The service status determination module is configured to determine the remaining latency budget and fast status urgency of the communication session in the current transmission cycle. The anti-interference response strategy generation module is configured to determine and execute the anti-interference response strategy for the communication session based on the remaining latency budget and the urgency of the rapid state, so as to achieve anti-interference communication. The anti-interference response strategy includes performing only fast path decision, or performing fast path decision and adaptive decision in parallel. Fast path decision is used to select anti-interference action from the current set of security actions and execute it in the current transmission cycle. Adaptive decision is used to update the set of candidate security actions for the next transmission cycle. After the candidate security action set is updated, it becomes the current set of security actions for the next transmission cycle.

10. A time-delay-aware anti-interference communication device, comprising a processor and a memory storing program instructions, characterized in that, The processor is configured to execute the delay-aware anti-interference communication method as described in any one of claims 1 to 8 when executing the program instructions.

Images