A low-latency high-reliability unmanned control communication system

Abstract

The application discloses a low-delay high-reliability unmanned control communication system, which comprises a session establishment module, an interlocking encoding module, a group sending module and a tail risk evaluation module.The session establishment module establishes a control session context, which contains a control sequence number mapping rule, a time limit window function and an encoding seed.The interlocking encoding module generates a control block according to a control period and obtains a main segment and an interlocking segment by segmentation.The interlocking segment is coupled and generated by finite field operation of a second segment of the current control block and a first segment of the next control block.The group sending module forms an encoding group according to the control sequence number mapping rule and sends the encoding group.The tail risk evaluation module constructs a quantile summary based on a one-way delay sample, performs peak threshold extreme value modeling, and generates a tail risk level.The application cooperates interlocking encoding, tail risk evaluation and preemption replacement, and considers low delay, high reliability and instruction time limit consistency in a complex network.

Description

Technical Field

[0001] This application relates to the field of communication control technology, and in particular to a low-latency, high-reliability unmanned control communication system. Background Technology

[0002] As unmanned platforms such as drones and unmanned vehicles operate on a large scale in inspection, emergency response, and logistics scenarios, control communication is evolving from short-range dedicated links to multi-access converged networks oriented towards 5G / 6G, private networks, and satellites. It relies on edge computing, network slicing, cross-domain forwarding, and multi-link backup to achieve remote and cross-regional unmanned control. This type of service is a typical closed loop: the control end issues small packet instructions according to the control cycle and collects telemetry and status. Control and backhaul often share the same bearer and scheduling queue. The resource allocation, queuing, switching, confirmation feedback, and retransmission strategies on the network side jointly determine the end-to-end latency, jitter, and reliability.

[0003] Existing methods often pursue lower average latency or improve delivery rate through redundancy and retransmission, but they neglect two types of problems: First, under congestion and uncertain scheduling, the highly reliable mechanism is prone to tail delay explosion. Even a very small number of timeout packets can cause instructions to lag, phase margin to decrease, and trigger closed-loop instability. Second, when the link is restored after a short-term disconnection, the cached historical instructions may arrive out of order and exceed the "freshness window". Reliable delivery may instead cause expired instructions to be executed, duplicate actions and semantic inconsistencies, which may lead to jitter or even security risks. Summary of the Invention

[0004] To address the above problems, embodiments of the present invention provide a low-latency, highly reliable unmanned control communication system, the system comprising: A session establishment module establishes a control session context, which includes control sequence number mapping rules, time window functions, and encoding seeds. An interlocking coding module generates control blocks according to a control cycle and segments them to obtain main segments and interlocking segments. The interlocking segments are generated by coupling the second segment of the current control block and the first segment of the next control block through finite field operations. The group transmission module forms and transmits encoded groups according to the control sequence number mapping rule; the tail risk assessment module constructs quantile summaries based on one-way delay samples and performs peak over-threshold extreme value modeling to generate tail risk levels. The strategy scheduling module selects one of the following based on the tail risk level: single-path forwarding, dual-path replication forwarding, or preemptive replacement forwarding. The dependency clearing module determines the decodeable sequence number range based on the decoding window summary reported by the terminal during preemption replacement forwarding, and clears the encoded groups in the queue whose control sequence numbers are outside the range and whose interlocked fragment references the failed control block. A replacement generation module, which generates a replacement control block from the terminal's most recent state vector and sends it via an interlocked encoding module and a group transmission module; The window decoding module is a terminal that performs sliding decoding based on the time-sensitive window function, discards groups outside the window, and restores the control block execution within the decodeable sequence number range.

[0005] Furthermore, the interlocking encoding module divides the control block into a first segment and a second segment, with the main segment being the first segment; the interlocking segment is the result of the addition operation of the second segment and the first segment of the next control block in a finite field; the window decoding module jointly solves the second segment of the corresponding control block based on the main segment and the interlocking segment in the adjacent control sequence number encoding group.

[0006] Furthermore, the control sequence number mapping rule maps the transmission time slot sequence to the control sequence number one by one. The transmission time slot sequence is generated by the time base shared by the control terminal and the unmanned platform terminal. The group transmission module determines the control sequence number corresponding to the coded group based on the position of the transmission time slot to which the coded group belongs in the transmission time slot sequence.

[0007] Furthermore, the tail risk assessment module performs threshold screening on the one-way time delay samples to obtain the excess quantity samples, and updates the parameters of the generalized Pareto distribution on the excess quantity samples online. Based on the updated distribution, the conditional tail bound is calculated; the tail risk level is determined by the relationship between the conditional tail bound and the risk boundary corresponding to the time window function.

[0008] Furthermore, the quantile summary adopts the Greenwald–Khanna quantile summary structure to maintain the ordered summary of the one-way time delay sample and outputs the high quantile estimate; the tail risk assessment module performs consistency judgment on the high quantile estimate and the conditional tail boundary, and determines the tail risk level based on the judgment result.

[0009] Furthermore, when the policy scheduling module selects dual-path replication and forwarding, it replicates the coded packets with the same control sequence number into a primary packet and a backup packet. The primary packet enters the first scheduling queue and is sent through the first forwarding path, while the backup packet enters the second scheduling queue and is sent through the second forwarding path. The first scheduling queue and the second scheduling queue maintain independent queuing states and one-way delay sample sets, respectively.

[0010] Furthermore, the interlock dependency table is constructed based on the clearing module. The interlock dependency table records the next control number referenced by the interlock segment of the coded group corresponding to each control number. When the decodeable number range determined by the decoding window digest does not contain the next control number, the coded group corresponding to the control number is added to the clearing set and removed from the queue.

[0011] Furthermore, the replacement generation module performs a one-step state prediction on the terminal's most recent state vector to obtain a predicted state vector, and inputs the predicted state vector and the reference state vector into the incremental control update law to generate a replacement control block; the interlocking encoding module performs segmentation and interlocking operations on the replacement control block according to the encoding seed to obtain the main segment and interlocking segment corresponding to the replacement control block.

[0012] Furthermore, when the window decoding module detects an inconsistency between the interlock dependency and the received encoded packet set, it generates a resynchronization request. The resynchronization request carries a decoding window summary and the most recently executed control sequence number. After the network node receives the resynchronization request, the session establishment module updates the encoding seed in the control session context and updates the timed window function. The interlock encoding module and the window decoding module perform interlock encoding and sliding decoding based on the updated control session context. A low-latency, high-reliability unmanned control communication system further includes: a decorrelation routing module, which extracts an over-threshold indication sequence consistent with threshold screening based on the one-way delay sample sets maintained by the first scheduling queue and the second scheduling queue, calculates the common over-threshold ratio and the time delay correlation to form a path correlation index, and performs reselection or remapping on the queue domain binding relationship of the second forwarding path when the path correlation index meets the correlation enhancement condition, so that the second forwarding path and the first forwarding path fall into different congestion domains. At the same time, the updated dual-path binding relationship is written into the dual-path replication forwarding configuration of the policy scheduling module, so that the primary and backup packets with the same control sequence number enter different scheduling queues and forwarding path combinations.

[0013] The technical effects and advantages of the low-latency, high-reliability unmanned control communication system provided by this invention are as follows: This invention achieves a balance between low latency, high reliability, and instruction timeliness consistency in complex networks through interlocked coding, tail risk assessment, and preemptive replacement. It constructs quantile summaries using unidirectional latency samples and performs extreme value modeling to establish a tail risk level. This drives the policy scheduling module to switch between single-path and dual-path replication and preemptive replacement, shifting scheduling decisions from "average level" to "tail risk" constraints, reducing the disturbance of long-tail queuing to closed-loop control. When tail risk increases, the system, through preemptive replacement forwarding in conjunction with a dependency clearing module, clears coded packets in the queue whose interlocked dependencies have failed and are outside the decodeable sequence number range, preventing lagging packets from continuously accumulating in the queue and dragging down the tail latency distribution. The current control block is coupled with the next control block to generate interlocked segments. The window decoding module solves these segments jointly using adjacent control sequence numbers, combining a timeliness window function with decodeable sequence number range constraints. This makes it difficult for packets outside the window or with unsatisfactory interlocked dependencies to form reproducible control blocks, reducing the risk of expired execution due to out-of-order delivery. Attached Figure Description

[0014] Figure 1 This is a flowchart of a low-latency, high-reliability unmanned control communication system in Embodiment 1; Figure 2 This is a schematic diagram of queue clearing based on decoding window digest and interlock dependency table in Example 1; Figure 3 This is a flowchart of a low-latency, high-reliability unmanned control communication system in Example 2. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: Please see Figure 1 As shown, an embodiment of the present invention provides a low-latency, high-reliability unmanned control communication system, including a control terminal, a network node, and an unmanned platform terminal. The system includes: Session establishment module: Establishes a control session context, which includes control sequence number mapping rules, time window functions, and encoding seeds; Interlocking coding module: Generates control blocks according to the control cycle and segments them to obtain main segments and interlocking segments. Interlocking segments are generated by coupling the second segment of the current control block and the first segment of the next control block through finite field operations. Group transmission module: Forms and transmits coded groups according to the control sequence number mapping rules; Tail risk assessment module: Constructs quantile summaries based on one-way delay samples and performs peak over-threshold extreme value modeling to generate tail risk levels; Policy scheduling module: Select one of the following based on the tail risk level: single-path forwarding, dual-path replication forwarding, or preemptive replacement forwarding; Dependency clearing module: During preemption replacement forwarding, it determines the decodable sequence number range based on the decoding window summary reported by the terminal, and clears the encoded groups in the queue whose control sequence numbers are outside the range and whose interlocked fragment references the failed control block; Replacement generation module: Generates a replacement control block from the terminal's most recent state vector and sends it through the interlocked encoding module and the group transmission module; Window decoding module: The terminal slides the decoding according to the time-sensitive window function, discards the groups outside the window, and restores the control block execution within the decodeable sequence number range.

[0017] In this embodiment, the interlocking coding module performs segmentation and interlocking operations on the control blocks generated by the control terminal according to the control cycle to form coded groups. The control block can be understood as a set of control information to be issued within the cycle (such as the representation of control quantities such as attitude, speed and heading), and its internal structure can be organized into a continuous data sequence according to a predetermined format. To facilitate interlocking coding, the interlocking coding module first divides each control block into a first segment and a second segment according to a preset segmentation rule. The segmentation rule can be based on byte boundaries, field boundaries, or fixed length. Regardless of the rule used, the first segment and the second segment maintain a consistent parsing method between the control terminal and the unmanned platform terminal. The interlocking coding module defines the first segment as the main segment, which is directly sent with the coded group of the corresponding control sequence number as the explicit part of the control block under that control sequence number.

[0018] The construction of interlocked segments depends on the relationship between control blocks with adjacent control numbers: For control block k, the interlocking encoding module takes the second segment of the control block and performs a finite field addition operation with the first segment of the control block with control number k+1 to obtain the interlocked segment of control number k. Finite field addition refers to the addition performed after mapping the segmented data to finite field elements. In implementation, the operation is usually performed on a symbol-by-symbol basis: The interlocking encoding module first splits the second segment and the first segment of the next control block into symbol sequences according to the same symbol partitioning method and uses the same finite field representation method, such as based on a binary extended field. The representation is as follows: then, addition is performed on the symbols corresponding to each pair of positions, and the interlocked segments are concatenated. If a finite field representation with a feature of 2 is used, the addition can be equivalent to bitwise XOR. If other finite field representations are used, the interlocked coding module stores the operation rules of addition and addition inverse, which can be implemented by looking up a table or polynomial representation to ensure that the operation of the control end and the terminal end is consistent. The interlocked segments obtained in this way, together with the main segments, constitute the encoded group payload part of the control sequence number k. The interlocked segments themselves do not carry the plaintext of the first segment of the next control block separately, but instead solidify the dependency relationship with the next control block in the operation structure.

[0019] The window decoding module of the unmanned platform terminal buffers the received encoded packets according to the control sequence number, and jointly solves the second segment of the corresponding control block by combining the main segment and interlocked segment of adjacent control sequence numbers. Specifically, after receiving the encoded packet of control sequence number k, the terminal can directly obtain the first segment of control block k, i.e., the main segment; when the terminal simultaneously obtains the encoded packet of control sequence number k+1, it can extract the first segment of control block k+1 from it. At this time, the window decoding module performs a finite field inverse operation on the interlocked segment of control sequence number k to obtain the second segment of control block k: in the finite field, the second segment = interlocked segment - next control segment. The first segment of the block is subtraction, which is the inverse operation of addition. If a finite field with a characteristic of 2 is used, the inverse operation is consistent with the addition form and can be implemented using the same bitwise XOR. The window decoding module concatenates the obtained second segment with the previously obtained first segment according to the segmentation rules to restore the complete control block k, and hands it over to the subsequent processing flow. Since the restoration of the second segment depends on the main segment of k+1, the terminal cannot independently restore the complete control block with only an expired control sequence number k encoded group, thus forming a decoding constraint on adjacent sequence numbers. This constraint is determined by the construction method of the interlocked segment and does not depend on additional filtering rules.

[0020] An example (used only to illustrate the operational relationship, not limiting the actual data bit width or field size): Assuming a finite field addition equivalent to bitwise XOR is used, the first segment of control sequence number k is 0x3C, the second segment is 0xA5, and the first segment of control sequence number k+1 is 0xF0. Then, the interlocked segment = 0xA5 ⊕ 0xF0 = 0x55. The terminal receives the main segment 0x3C of control sequence number k and the interlocked segment 0x55. After receiving the main segment 0xF0 of control sequence number k+1, it calculates the second segment = 0x55 ⊕ 0xF0 = 0xA5, thus concatenating it with the first segment 0x3C to restore control block k. The joint solution process corresponding to this example is that the window decoding module uses the main segments of adjacent control sequences as decoupling factors of the interlocked segments to restore the second segment of the target control block.

[0021] In this embodiment, the system establishes a control sequence number mapping rule for each control session. This rule maps the transmission time slot sequence to the control sequence number one-to-one. This is used to identify which control cycle the coded packet belongs to in a unified way when network replication and forwarding, preemption and clearing, or out-of-order arrival occur. The control sequence number is an incrementing index within the control session, used to identify the temporal order of control blocks. The transmission time slot sequence is a sequence of discretized transmission opportunities occupied by the control session. Each transmission time slot represents one logical transmission opportunity, and its boundary is given by the shared time base.

[0022] The shared time reference is jointly held by the control terminal and the unmanned platform terminal. It can come from the synchronization results of the same time source or from time alignment information provided by the network side. In order for the control terminal and the terminal to reach a consistent judgment without relying on the sequence number carried in each packet, the control session context stores the usage of the time reference, including: the method for determining the time slot start point, the method for dividing the time slot boundaries, and the correction strategy for short-term drift. In specific implementation, the control terminal selects a reference time as the "zero point" of the session time slot sequence during the session establishment phase and writes the expression of the reference time (such as the offset description relative to the shared time reference) into the control session context. After receiving the control session context, the terminal uses the same zero point and boundary division method to generate a local transmission time slot sequence. In this way, both parties only need to indicate "which transmission time slot they are currently in" at any given time to obtain a consistent sequence number result.

[0023] The core of the control sequence number mapping rule is the definite relationship between position and sequence number: each time slot position in the transmission time slot sequence corresponds to a unique control sequence number, and this correspondence remains unchanged during the session. In implementation, the control terminal assigns a sequence position index to each transmission time slot, and the sequence position index increases sequentially from zero point to time slot. The control sequence number mapping rule specifies the mapping relationship between the control sequence number and the sequence position index. For example, the same increment value can be directly used, or a session-level offset can be introduced in the mapping to align the control sequence number with the session establishment time. Since the terminal generates the sequence position index according to the same zero point and the same time slot boundary, the terminal can reproduce the same mapping locally, ensuring consistency in the judgment of "which time slot corresponds to which control sequence number".

[0024] When forming coded packets, the group transmission module does not rely on external instructions to temporarily specify the control sequence number. Instead, it finds the position of the coded packet in the transmission time slot sequence based on the transmission time slot to which the coded packet belongs, and determines the corresponding control sequence number accordingly. The transmission time slot to which the coded packet belongs refers to the logical time slot to which the coded packet is submitted to the wireless side transmission queue. It is calculated by the group transmission module before the packet enters the queue: the group transmission module reads the local time value under the current shared time base, aligns it with the session zero point, divides the current time slot position index according to the time slot boundary, and then obtains the control sequence number according to the control sequence number mapping rule. If queuing occurs on the network side and the actual air interface transmission crosses the time slot boundary, the group transmission module still uses the enqueuing time slot as the transmission time slot to which the packet belongs, ensuring that subsequent modules (such as the dependency clearing module and the window decoding module) use the same semantic sequence number source when judging interlock dependencies and decodeable sequence number intervals.

[0025] To avoid inconsistencies in time slot boundaries between the two ends due to slight drift in the shared time base, in this embodiment, the terminal can perform time slot attribution verification after receiving the encoded packet: the terminal calculates the time slot position index to which the packet should belong from the receiving time and performs a consistency check with the control sequence number mapping result calculated locally for the packet; when the check result shows a continuous deviation, the terminal adjusts the local zero point or boundary calculation method through the correction strategy in the control session context, thereby restoring the consensus between the two parties on the same transmission time slot sequence; this verification and correction does not require adding an extra control sequence number field to each packet, and the sequence number is still determined by the time slot position.

[0026] For example: Assume the session zero point is a certain reference time, the transmission time slots are divided according to fixed boundaries, and the control sequence number mapping rule stipulates that "the 0th transmission time slot corresponds to control sequence number 0, the 1st transmission time slot corresponds to control sequence number 1, and so on"; when the group transmission module calculates locally that it is currently in the 5th transmission time slot, it maps the time slot position index to control sequence number 5, and combines the main segment corresponding to control sequence number 5 with the interlocked segment to form an encoded group for transmission; the terminal calculates based on the same zero point and boundary, and the receiving end can also determine that the group belongs to control sequence number 5, so that in the window decoding module, it establishes an adjacency relationship with the group related to control sequence number 6, and completes the joint solution with the interlocked segment.

[0027] In this embodiment, the tail risk assessment module performs tail risk modeling on the one-way delay samples within the control session and converts the modeling results into tail risk levels for use by the policy scheduling module and the dependency clearing module. The one-way delay sample refers to the one-way time difference sample between when the coded packet enters the transmission path on the sending side and when it can be confirmed to have arrived on the receiving side. To reduce the impact of network jitter and backhaul uncertainty on the measurement, the sending side and the receiving side are aligned on the same time axis based on a shared time reference. The sending side records the time stamp of each coded packet entering the transmission queue, and the receiving side records the reception time stamp before the coded packet enters the window decoding module. The tail risk assessment module forms a one-way delay sample based on the difference between the two. The tail risk assessment module saves the samples in groups according to the control session to avoid the aliasing of queue states of different service flows.

[0028] The tail risk assessment module first performs threshold screening. The goal of threshold screening is not to describe the average level, but to focus on long-tail events that may trigger control instability. To this end, the tail risk assessment module maintains a threshold generation rule in the control session context: this rule can be derived from the high quantile estimate of historical statistics, the boundary derivation of the time-delay window function, or a combination of both. Once the threshold is determined, the tail risk assessment module compares each newly arrived one-way delay sample, and only samples that exceed the threshold are retained as over-threshold samples. For each over-threshold sample, the tail risk assessment module further constructs an excess amount sample, which is the difference between the over-threshold sample and the threshold. The excess amount sample only expresses "how much it exceeds" and no longer carries the absolute size of the basic threshold, making subsequent model updates more robust to threshold drift.

[0029] After obtaining the excess quantity samples, the tail risk assessment module updates their parameters online using the generalized Pareto distribution. The generalized Pareto distribution is used to characterize the tail shape of the excess quantity after the peak exceeds the threshold. Its parameters can reflect the changes in tail thickness and scale. The online update is not an offline batch processing, but rather a continuous absorption of new excess quantity samples during the operation of the control session to update the current distribution parameters. In implementation, the tail risk assessment module maintains a set of current parameter states for each control session and simultaneously maintains a sample weight decay mechanism, so that recent samples have a greater impact on the parameters and the impact of long-term samples gradually weakens, thereby tracking changes in tail shape caused by congestion, handover, and radio fading. The parameter update can be completed using a recursive maximum likelihood approximation or recursive moment estimation method. The module only needs to store a small amount of cumulative values ​​to complete the update, avoiding the storage of a large number of historical samples on network nodes.

[0030] Based on the updated distribution parameters, the tail risk assessment module calculates the conditional tail bound. The conditional tail bound is an estimated description of the upper tail bound that the one-way delay may still reach under the condition that the threshold has already been exceeded. During the calculation, the tail risk assessment module substitutes the risk level description and the current distribution parameters into the tail quantile derivation relationship of the distribution to obtain a corresponding exceedance quantile value, and adds it to the threshold to form the conditional bound for the tail of the one-way delay. To ensure that the bound can be consistently interpreted within the control session, the risk level description and the threshold generation rules are stored together in the control session context, so that the conditional tail bound output by the tail risk assessment module at any time is consistent with the same set of session constraints.

[0031] The tail risk level is determined by the risk boundary relationship between the conditional tail bound and the time-limited window function. The time-limited window function is a constraint in the control session context, which describes the effective window of a control block corresponding to a certain control sequence number in time, and the relationship between the window boundary and the sequence number. The tail risk assessment module maps the conditional tail bound to the time scale of the time-limited window function to form a risk boundary relationship judgment: when the conditional tail bound falls within the range allowed by the time-limited window function, the tail risk level remains at a low level; when the conditional tail bound approaches or crosses the boundary of the time-limited window function, the tail risk level increases. This level is based on the tail shape estimated by the model, which can reflect in advance the trend of "a very small number of timeout events" becoming more frequent or more serious, so that the strategy scheduling module can complete the scheduling mode switch before the risk increases, and the dependency clearing module can clear queue groups that will cause interlocking dependencies to fail when needed.

[0032] For example: Suppose a control session obtains a threshold of several time units according to the threshold generation rule. If a one-way delay sample observed in a certain instance exceeds the threshold, it forms an excess sample. The tail risk assessment module recursively updates the current distribution parameters after several consecutive excess samples, calculates the excess quantile estimate at the selected risk level, and synthesizes it with the threshold to form a conditional tail bound. If the time scale corresponding to the conditional tail bound is close to the upper edge of the effective window given by the time-sensitive window function, the tail risk assessment module will increase the tail risk level. The strategy scheduling module will then enter preemptive replacement forwarding. The dependency clearing module will clear the coded groups in the queue whose interlocked dependencies can no longer be satisfied according to the decodeable sequence number interval, to avoid long-tail queuing dragging expired interlocked groups into the execution window.

[0033] In this embodiment, when processing one-way time-delay samples, the tail risk assessment module maintains two mutually verifying quantities: one is the conditional tail bound obtained based on threshold screening and exceedance samples, and the other is the high quantile estimate obtained based on quantile summaries. The two quantities represent the "tail upper bound trend inferred by the model" and the "nonparametric high quantile position of the sample distribution," respectively. The tail risk assessment module links the two through consistency judgment to determine the tail risk level and avoid misjudgment caused by relying solely on a certain statistical caliber when there are short-term fluctuations or sparse samples.

[0034] Quantile summaries are structures that compress the ordered distribution of one-way delay samples. The tail risk assessment module does not store all historical samples, but maintains a summary table sorted in ascending order of sample values. Each entry in the summary table corresponds to a representative value and is accompanied by two counts: one describing the minimum rank increment of the representative value in the sort, and the other describing the rank uncertainty that the representative value is allowed to cover. These two types of counts together provide an error bound, enabling the estimation of any quantile point within a given error bound without storing the full set of samples. The structure used here is the Greenwald–Khanna quantile summary structure: whenever a new one-way delay sample arrives, the tail risk assessment module inserts the sample into the appropriate position in the summary table according to its size, and updates the counts of adjacent entries. When the length of the summary table grows to the point where it does not match the error bound, a compression process is performed, merging adjacent entries that are "mergeable in rank intervals" into one entry. During merging, the representative value is retained and the corresponding count is accumulated, thereby controlling the structure size. Since insertion and compression only involve local entries, the tail risk assessment module can run continuously online under the normal computing resources of network nodes.

[0035] The high quantile estimate is calculated from the summary table. The tail risk assessment module accumulates the "minimum rank increment" in the summary table from small to large based on the rank position of the target quantile, such as the quantile position near the tail of the distribution. When the accumulated rank crosses the target rank interval, the representative value of the current entry is taken as the estimated value of the quantile. Since the summary table entries also carry rank uncertainty, the tail risk assessment module obtains an estimated rank interval when outputting the high quantile estimate. This interval corresponds to the error bound. The subsequent consistency determination will use the information of this interval instead of just looking at the single-point estimate.

[0036] Consistency determination occurs when both the high quantile estimate and the conditional tail bound are available. The conditional tail bound is derived from the tail risk assessment module, which updates the parameters of the generalized Pareto distribution on the excess quantity sample online and maps the risk level description to tail quantiles to synthesize the conditional tail bound. Both have the same time scale and express the magnitude of the one-way delay in the tail region. When the tail risk assessment module performs consistency determination, it introduces the error bound of the quantile summary: the high quantile estimate is expanded into an allowable interval, determined by the estimated rank interval and the neighborhood entries of the representative value, and then the relative position of the conditional tail bound is compared with this allowable interval; if the conditional tail bound falls within the allowable interval, the two quantities are considered consistent; if the conditional tail bound is consistently higher than the allowable interval, the two quantities are considered consistent. If the upper limit of the allowable interval is reached, it indicates that the risk trend given by the tail model has exceeded the tail position reflected by the sample quantiles, and the tail risk assessment module increases the tail risk level. If the conditional tail boundary is consistently lower than the lower limit of the allowable interval, it indicates that the number of samples exceeding the threshold is insufficient or the threshold screening is too conservative, resulting in a tight extrapolation of the model. The tail risk assessment module reduces the tail risk level or maintains the level unchanged and waits for more samples exceeding the threshold to enter, according to the rules in the control session context. Here, "consistent" does not depend on the exact number threshold. A stability criterion can be given by the control session context, such as changing the level only when the consistency ratio within the sliding observation window reaches a certain condition, in order to avoid frequent switching triggered by a single accidental measurement error.

[0037] To link consistency determination with the time-bound window function, the tail risk assessment module maps both the high quantile estimate and the conditional tail boundary to the risk boundary relationship of the time-bound window function before outputting the tail risk level. The time-bound window function provides the time boundary that the control block can be accepted by the terminal window decoding module. The tail risk assessment module uses the conclusions of "whether the high quantile estimate approaches the boundary" and "whether the conditional tail boundary crosses the boundary" as additional criteria for consistency determination: when both indicate that they are approaching or crossing the boundary, the consistency determination tends to increase the tail risk level; when the two are inconsistent and the quantile summary error boundary is wide, the tail risk assessment module prioritizes maintaining the level and continues to accumulate samples to avoid over-reliance on extreme value model extrapolation when the sample size is insufficient.

[0038] For example: Suppose that the quantile summary output of a certain stage of unidirectional time-delay samples has a high quantile estimate of approximately 12 (within the same time unit), and the allowable interval roughly covers 11-13; if the calculated conditional tail bound of the same stage is 12.5, then the conditional tail bound falls within the allowable interval, the consistency is determined to be consistent, and the tail risk level remains unchanged; if the conditional tail bound rises to 16 in subsequent stages, while the high quantile estimate of the quantile summary remains stable around 12 and the allowable interval does not expand significantly, then the conditional tail bound continues to be higher than the upper edge of the allowable interval, and the consistency is determined to be... If there is inconsistency, the tail risk assessment module will raise the tail risk level and hand it over to the strategy scheduling module to trigger preemptive replacement forwarding. At the same time, the queue clearing module will clear queue groups whose interlocking dependencies can no longer be satisfied based on the decoding window summary reported by the terminal. In this way, the combination of high quantile estimation and conditional tail bound is not "calculating an extra indicator", but forming an executable closed-loop judgment chain: quantile summary provides a robust anchor point for the tail position of the sample, extreme value modeling provides extrapolation of the long tail deterioration trend, and the two constrain the transition conditions of tail risk level through consistency judgment.

[0039] In this embodiment, after the tail risk level enters the dual-path replication and forwarding state, the policy scheduling module replicates the coded packets corresponding to the same control sequence number into the primary packet and the backup packet, and sends them to the first scheduling queue and the second scheduling queue respectively. The control sequence number adopts the sequence number value obtained by the control sequence number mapping rule in the control session context, ensuring that the primary packet and the backup packet still point to the same control sequence number in the replication, out-of-order and retransmission environment, without introducing additional sequence number interpretation discrepancies. The coded packets adopt the packet structure formed by the interlocked coding module and the group transmission module, which includes the primary segment and the interlocked segment. The primary packet and the backup packet are consistent in packet load, and only differ in forwarding path and scheduling queue affiliation.

[0040] Dual paths are two distinct forwarding paths maintained by network nodes within the control session context. Each forwarding path is defined by a set of path elements, such as the outgoing interface, next-hop selection rules, forwarding tunnel or forwarding identifier, and a scheduling queue instance bound to that path. When the policy scheduling module enters dual-path replication forwarding, it first determines the replication point location: the replication point is set before the entry of the sending queue on the network node, ensuring that the replication action occurs before queuing, guaranteeing that the primary packet and the backup packet are affected by the queuing process of their respective queues, thus forming statistical independence between the path and the queue. Subsequently, the policy scheduling module writes the replicated primary packet into the first scheduling queue and the replicated backup packet into the second scheduling queue. The two queues connect the first forwarding path and the second forwarding path respectively, forming a one-to-one correspondence between the queue and the path, preventing the primary packet from drifting between the two paths and disrupting statistical attribution.

[0041] The first and second scheduling queues maintain independent queuing states. In this embodiment, the queuing state is a set of observable queue operations, including but not limited to: representations of queue occupancy, head-of-queue waiting time, dequeueing rhythm, and congestion level related to the queue. These representation values ​​are updated by network nodes when primary or backup packets are enqueued or dequeued, and are stored separately for each queue instance. Cross-queue merging updates is prohibited. In this way, when the policy scheduling module needs to switch between "continue dual-path replication" and "switch to preemptive replacement" later, it can determine which path's tail risk is worsening based on the queuing states of the two queues, rather than mixing the two paths into a single average indicator.

[0042] Meanwhile, the first and second scheduling queues maintain independent sets of one-way delay samples. The definition of a one-way delay sample is as follows: for each coded packet entering the queue, an enqueue time stamp is recorded, and for the terminal side, a reception time stamp before the packet enters the window decoding module is recorded. The two are aligned based on a shared time reference to form a one-way delay sample. To achieve separate maintenance, when the main packet is dequeued and submitted to the first forwarding path, the network node binds the enqueue time stamp of the packet to the path identifier. When the terminal sends back the reception time stamp of the packet or its equivalent acknowledgment information, the tail risk assessment module assigns the sample to the sample set corresponding to the first scheduling queue according to the path identifier. The sample collection process for backup packets on the second scheduling queue and the second forwarding path is similar. If both the main packet and the backup packet arrive at the terminal, the terminal side only confirms the reception time stamp of one of the packets with the network node. The confirmation selection follows the reception order and decodeable sequence number interval rules of the window decoding module to ensure that the sample collection is consistent with the actual executable link of the terminal. Unacknowledged duplicate packets do not participate in the sample set update to avoid miscounting "duplicate delivery" as "shorter delay".

[0043] For example: After the coded packet with control sequence number k is formed, the policy scheduling module is in a dual-path replication and forwarding state, and the replication point generates a primary packet and a backup packet. The primary packet enters the first scheduling queue and is sent through the first forwarding path, while the backup packet enters the second scheduling queue and is sent through the second forwarding path. If the terminal receives the primary packet first and includes it in the window decoding module for processing, it sends back an acknowledgment and the primary packet corresponding to the received time stamp, and the one-way delay sample set of the first scheduling queue is updated once. If the backup packet arrives subsequently but is identified by the window decoding module as a duplicate packet with the same control sequence number and does not enter the decoding processing link, then the update of the sample set of the second scheduling queue is not triggered. Conversely, if the queuing deteriorates in the first forwarding path, and the primary packet arrives late while the backup packet arrives first, then the sample update naturally falls into the sample set of the second scheduling queue. Through this "queue-path attribution" sample maintenance method, the tail risk level output by the tail risk assessment module can reflect which specific path is generating the long tail, and the policy scheduling module then decides whether to maintain dual-path replication and forwarding or switch to the process of dependency clearing and replacement generation. The logic before and after is closed and the terminology is consistent.

[0044] like Figure 2 As shown, in this embodiment, the dependency clearing module maintains an interlock dependency table on the network node side and links it with queue management. The interlock dependency table is a data structure built according to the control session dimension. The table entries are indexed by the control sequence number and record the next control sequence number referenced by the interlock segment of the coded group corresponding to that control sequence number. The control sequence number comes from the control sequence number mapping rule in the control session context. The coded group is formed by the interlock encoding module and the group sending module. The group contains a main segment and an interlock segment. The next control sequence number referenced by the interlock segment corresponds to the interlock segment. Adjacent entries in the lock coding relationship, i.e., the interlocked segment with control sequence number k, are generated by coupling the second segment of control block k with the first segment of control block k+1. Therefore, for an entry with control sequence number k, its next control sequence number is recorded as k+1. The dependency clearing module does not need to resolve this reference relationship from the payload, because this reference relationship is determined by the interlock coding rules and is fixed in the control session context. The module only needs to read its control sequence number k when enqueuing the coded group, and fill in the next control sequence number k+1 of the reference according to the rules, and write this mapping into the interlock dependency table.

[0045] The updates to the interlock dependency table are consistent with queue operations. Specifically, when the policy scheduling module submits an coded group to a scheduling queue, the dependency cleanup module performs two tasks within the same transaction boundary: First, it binds the queue identifier (e.g., queue node pointer or queue sequence number) of the coded group to the control sequence number k, forming a "sequence number - queue location" index; second, it writes the dependency relationship "k references k+1" in the interlock dependency table. In this way, during subsequent cleanup, it is not necessary to traverse the group payload one by one in the queue, but to directly locate the queue element to be removed through the interlock dependency table and the "sequence number - queue location" index. When a group is dequeued, removed, or preempted, the dependency cleanup module synchronously deletes the corresponding table entries and indexes to prevent dangling references.

[0046] The decoding window summary is generated by the window decoding module on the unmanned platform terminal side and reported to the network node side, reflecting the current decoding window status of the terminal. The decoding window summary contains at least two types of information: one type describes the position of the most recently executed or confirmed control sequence number of the terminal; the other type describes the range of decodable sequence numbers based on that position, that is, the range of control sequence numbers that the terminal can complete sliding decoding and restore the control block under the constraints of the currently received coded packet set and interlocking dependency relationship. This summary can be expressed in the form of a range, such as the description of the start and end sequence numbers, or it can be expressed in a more compact form, such as the availability summary of several sequence numbers within the window. However, regardless of the expression used, the network node side will eventually reduce it to a range of decodable sequence numbers for the dependency clearing module to make inclusion judgment.

[0047] Upon receiving a new decoding window digest, the dependency clearing module first confirms the control session identifier corresponding to the digest based on the control session context and selects the corresponding interlock dependency table accordingly. Then, the module parses the decodeable sequence number range from the digest and performs a consistency check on each entry in the interlock dependency table: For each entry, with control sequence number k and the referenced next control sequence number k+1, it checks whether the decodeable sequence number range contains the next control sequence number. This check means that if the terminal wants to complete the joint solution for control sequence number k, it must have the availability of the main fragment for control sequence number k+1 within its decoding window. If the decodeable sequence number range does not contain k+1, the interlock dependency of control sequence number k cannot be satisfied on the terminal side. Even if the encoded group eventually arrives, it cannot form a reproducible control block in the current window. For such entries, the dependency clearing module adds the encoded group corresponding to control sequence number k to the clearing set. The clearing set is a set of location information that can be directly mapped to queue elements, usually generated by the "sequence number - queue location" index, ensuring that subsequent removal operations can be completed within the queue using normal operations.

[0048] After the cleanup set is generated, the dependency cleanup module performs removal according to the queue dimension. If the coded group may be distributed in different scheduling queues, such as the first scheduling queue and the second scheduling queue in the dual-path replication forwarding state, the module performs the removal process for each queue separately: locking the queue, locating the queue element according to the cleanup set, unlinking from the queue list or queue array and updating the corresponding queuing state, and deleting the interlocked dependency table entries and indexes. To avoid inconsistencies in state caused by concurrency with dequeueing, removal and dequeueing are performed using the same queue mutual exclusion mechanism or the same queue transaction sequence. If it is found that the target group has been dequeued or has been covered by the replacement generation module, the target is skipped and only the table entries are cleaned up to ensure that the interlocked dependency table and the actual queue state converge and are consistent.

[0049] For example: Suppose that at a certain moment, the decodeable sequence number range obtained by the decoding window digest reported by the terminal is [200, 208]; there is an entry in the interlock dependency table with control sequence number 198, which records that the next control sequence number it references is 199; there is also an entry with control sequence number 207, which records that the next control sequence number it references is 208; there is also an entry with control sequence number 208, which records that the next control sequence number it references is 209; the dependency clearing module checks whether the decodeable sequence number range contains the referenced sequence number: for 198→199, since 199 is not in [200, Within 208, it is determined that the interlock dependency cannot be satisfied within the terminal window. The coded group corresponding to control sequence number 198 is added to the clearing set and removed from the queue. For 207→208, since 208 is within the interval, it is retained. For 208→209, since 209 is not within the interval, the coded group corresponding to control sequence number 208 is added to the clearing set and removed. After this processing, the interlock fragment reference relationship of the coded groups retained in the queue is consistent with the currently decodeable sequence number interval of the terminal. Subsequent sliding decoding will not be dragged down by groups whose references fall outside the window.

[0050] In this embodiment, the replacement generation module participates in control block generation when the system enters the preemptive replacement forwarding state. Its goal is not to reuse lagging encoded packets in the multiplexing queue, but to generate a control block that is more consistent with the current time based on the terminal's current decodeable sequence number range. This block is then handed over to the interlocked encoding module to form a new encoded packet. To ensure that the replacement control block is consistent with the control session context, the replacement generation module uses the control sequence number mapping rules in the control session context to determine the control sequence number corresponding to the replacement control block, and uses the same time window function to constrain the executable time domain of the replacement control block. The replacement control block is then segmented and interlocked by the interlocked encoding module according to the same encoding seed, ensuring its compatibility with the existing main segment and interlocked segment structure.

[0051] The terminal's most recent state vector comes from the state feedback of the unmanned platform terminal. The state feedback contains a set of measured or estimated state variables of the terminal at the most recent moment. This set of state variables may include control-related variables such as position, velocity, attitude, angular velocity, energy state, and load state. The specific value dimensions are determined by the control session context and maintain a fixed semantic order to form the state vector. In order to enable the network node side to directly use the state vector, the state vector returned by the terminal is consistent with the input dimension of the control update law on the control end side, and uses a time stamp aligned with the shared time base to make "most recent" have a deducible meaning. After receiving the state vector, the replacement generation module first performs a validity check: it checks whether the control session identifier corresponding to the state vector is consistent with the current control session context, and determines whether the time stamp of the state vector falls within the acceptable feedback window based on the time window function. After passing the check, it is used as the starting state for one step of state prediction.

[0052] One-step state prediction refers to advancing the terminal's most recent state vector along the control session time by one control cycle to obtain the predicted state vector. In implementation, the replacement generation module maintains a prediction model consistent with the terminal's dynamics. This prediction model can be in the form of discrete state transitions: the predicted state vector is obtained by applying a state transition mapping to the terminal's most recent state vector; the mapping parameters can be preset in the control session context or issued by the control terminal during session establishment or update; to adapt to network-side implementation, the prediction step usually adopts a single mapping calculation without multi-step iterations, thus keeping the computational load within a predictable range; if the terminal's feedback includes execution feedback or disturbance estimation of control inputs, the replacement generation module can use it as an external input to the prediction model, making the predicted state vector closer to the actual evolution of the terminal; after the prediction is completed, the replacement generation module uses the predicted state vector as the current state input for the subsequent incremental control update law.

[0053] In this embodiment, the reference state vector represents the set of target state quantities that the control terminal expects the terminal to reach or maintain. It has the same dimension and semantic order as the terminal's most recent state vector. The reference state vector can come from the control terminal's task planning output, trajectory tracking output, or attitude maintenance output, and can be updated with the control sequence number. To ensure that the replacement control block is aligned with the control sequence number, the replacement generation module reads the reference state vector corresponding to the current control sequence number according to the control sequence number mapping rule. If the reference state vector is issued in advance by the control terminal and cached on the network node side, the replacement generation module also checks its validity boundary according to the time window function before use to avoid using reference information that lags behind the current control sequence number.

[0054] The incremental control update law is used to generate replacement control blocks from the predicted state vector and the reference state vector. The control update law is a rule that maps "state deviation" to "control increment". The replacement generation module does not directly output the absolute control quantity, but rather the incremental control quantity relative to the previously executed control block. This ensures continuity of the replacement control block when coupled with the terminal execution link. In implementation, the replacement generation module first calculates the state deviation vector, which is obtained by subtracting the reference state vector from the predicted state vector along their corresponding dimensions. Then, the state deviation vector is input into the incremental control update law mapper to obtain the control increment vector. Finally, the control increment vector is... The control increment vector is synthesized with the control quantity reference of the most recently executed control block to form a replacement control block. The control quantity reference of the most recently executed control block can be indirectly determined by the "most recently executed control sequence number" carried by the terminal in the decoding window digest. Based on this, the network node retrieves the corresponding control block or its digest from the session context cache. If the system adopts a representation that only caches control increments, the synthesis process is completed in the increment domain to maintain expression consistency. Through the chain of "prediction-deviation-increment-synthesis", the replacement control block obtained by the replacement generation module is more consistent with the current state of the terminal in terms of value and maintains the same rhythm as the session sequence number progression.

[0055] After the replacement control block is generated, the interlocking coding module performs segmentation and interlocking operations on it according to the coding seed to obtain the main segment and interlocking segment corresponding to the replacement control block. This process is consistent with that of ordinary control blocks: the interlocking coding module divides the replacement control block into a first segment and a second segment according to the segmentation rules, with the first segment serving as the main segment; the interlocking segment is obtained by performing finite field operations on the second segment of the replacement control block and the first segment of the control block corresponding to the next control sequence number. The control block corresponding to the next control sequence number can be a successor control block already generated by the control end, or a successor replacement control block generated simultaneously by the replacement generation module in the preemptive replacement forwarding state; regardless of the source of the successor control block, its segmentation rules and coding seed are consistent with the current session context, thereby ensuring that the reference relationship of the interlocking segments still conforms to the structural constraint of "control sequence number references next control sequence number" recorded in the interlocking dependency table; the interlocking coding module assembles the main segment and the interlocking segment into a coding group and then hands it over to the group sending module for enqueueing and sending, so that the replacement control block enters the same transmission, copying, clearing, and decoding link as the original control block.

[0056] For example: Assume the terminal's most recent state vector contains two dimensions, denoted as position error and velocity error. The replacement generation module uses a one-step prediction to advance it to the time corresponding to the next control sequence number to obtain the predicted state vector. The reference state vector also gives the target (position error and velocity error) at that time. The replacement generation module calculates the state deviation and obtains the control increment through the incremental control update law, for example, obtaining a control increment vector in the form of (acceleration increment, steering angle increment). This vector is then synthesized with the control reference of the most recently executed control block to obtain the replacement control block. The interlocking coding module then segments the replacement control block and interlocks it with the first segment of the control block of the next control sequence number to obtain the main segment and the interlocked segment, which then enters the subsequent transmission and window decoding process. Through this process, the generation of the replacement control block does not depend on the lagging historical packets in the queue, but is driven by the terminal state feedback, and maintains consistency with the constraints of the interlocking coding structure, control sequence advancement, and time window function.

[0057] In this embodiment, the window decoding module maintains a set of received coded packets on the terminal side. This set is indexed by control sequence number and records whether the main segment and interlocked segment of the coded packet corresponding to each control sequence number are available. The interlocked dependency is determined by the adjacent coupling rule of the interlocked coding module: the interlocked segment of control sequence number k references the main segment of the next control sequence number k+1. When the window decoding module performs a dependency check on each sequence number to be decoded during sliding decoding, the interlocked dependency is inconsistent with the set of received coded packets. In this embodiment, it can be identified by two types of operable criteria: The first type is that the decodeable sequence number range obtained by the decoding window digest reduction includes control sequence number k, but the state of "interlocked segment of k is available while main segment of k+1 is missing" exists in the set of received coded packets for a long time, which makes it impossible to solve the second segment of k by interlocking operation; The second type is that the interlocked dependency is formally complete and both the interlocked segment and the next main segment are available, but after decoupling according to the coding seed in the current control session context, the recovered second segment does not match the control block format consistency check and shows continuous anomalies on adjacent control sequences, indicating that the current coding seed and the terminal-side received set are no longer convergent.

[0058] When the above inconsistency is detected, the window decoding module generates a resynchronization request. The resynchronization request carries a decoding window summary and the most recently executed control sequence number. The decoding window summary is used to enable the network node to reduce the currently decodeable sequence number range of the terminal and its advancement position. The most recently executed control sequence number serves as an anchor point for the terminal to confirm execution, avoiding the fallback interpretation of the executed control block after resynchronization. After the network node receives the resynchronization request, the session establishment module updates the encoding seed and the validity window function within the corresponding control session context. The encoding seed is switched using the method of "starting from the new effective starting point control sequence number," and the effective starting point is determined by advancing backward based on the most recently executed control sequence number. The validity window function synchronizes the reference origin with the effective starting point. Point alignment is maintained to ensure consistent sequence number progression with the control sequence number mapping rules, enabling the terminal to redetermine the discard range outside the window and the decodeable sequence number interval. The updated control session context is simultaneously distributed to the interlocked encoding module and the terminal-side window decoding module. The interlocked encoding module performs segmentation and interlocking operations on subsequent control blocks or replacement control blocks according to the updated encoding seed, starting from the effective start point. The window decoding module resets the sliding decoding process with the most recently executed control sequence number as the anchor point, recalculates the decodeable sequence number interval according to the updated time window function, and re-interlocks and decodes the received encoded packet set that falls on or after the effective start point according to the new encoding seed. Packets outside the window are discarded or ignored, thereby restoring convergent decoding progression.

[0059] For example: If the terminal has recently executed control sequence number 300, and the decodeable sequence number range reduced by the decoding window digest includes 301, but the interlocked fragment of 301 exists in the received encoded packet set while the main fragment of 302 is continuously missing, the window decoding module determines the inconsistency and reports a resynchronization request (carrying the decoding window digest and 300); the network node determines the new effective starting point based on 300 and updates the encoding seed and the time-limited window function; the interlocked encoding module generates interlocked fragments according to the new encoding seed starting from the effective starting point; after the window decoding module is reset according to the new context, it re-forms the received encoded packet set that satisfies the interlock dependency based on the newly arrived packets, and the sliding decoding continues.

[0060] Example 2: like Figure 3As shown, this embodiment further improves upon the design of Embodiment 1. The difference is that in actual operation of Embodiment 1, it was found that the first and second forwarding paths share a congestion domain during dual-path replication forwarding. This causes the unidirectional delay samples of the first and second scheduling queues to exhibit synchronous increases and common threshold aggregation in the tail interval. This results in a high correlation between the arrival delays of the primary and backup packets under the same control sequence number. The backup path fails to provide effective compensation when long-tail events occur, and fails to stably suppress tail delay risks and avoid persistent gaps in the decoding window when the tail risk level is in the dual-path replication forwarding stage. Based on this, a low-latency, high-reliability unmanned control communication system further includes: De-correlation routing module: Based on the one-way delay sample sets maintained by the first and second scheduling queues respectively, extract the over-threshold indication sequence consistent with the threshold screening, calculate the common over-threshold ratio and time delay correlation to form the path correlation index, and when the path correlation index meets the correlation enhancement condition, perform reselection or remapping on the queue domain binding relationship of the second forwarding path, so that the second forwarding path and the first forwarding path fall into different congestion domains. At the same time, write the updated dual-path binding relationship into the dual-path replication forwarding configuration of the policy scheduling module, so that the primary and backup packets with the same control sequence number enter different scheduling queues and forwarding path combinations.

[0061] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

[0062] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present application, based on the technical solution and concept of the present application, should be covered within the scope of protection of the present application.

Claims

1. A low-latency, high-reliability unmanned control communication system, characterized in that, The system includes: A session establishment module establishes a control session context, which includes control sequence number mapping rules, time window functions, and encoding seeds. An interlocking coding module generates control blocks according to a control cycle and segments them to obtain main segments and interlocking segments. The interlocking segments are generated by coupling the second segment of the current control block and the first segment of the next control block through finite field operations. A group transmission module, which forms and transmits encoded groups according to a control sequence number mapping rule; The tail risk assessment module constructs quantile summaries based on one-way time delay samples and performs peak over-threshold extreme value modeling to generate tail risk levels. The strategy scheduling module selects one of the following based on the tail risk level: single-path forwarding, dual-path replication forwarding, or preemptive replacement forwarding. The dependency clearing module determines the decodeable sequence number range based on the decoding window summary reported by the terminal during preemption replacement forwarding, and clears the encoded groups in the queue whose control sequence numbers are outside the range and whose interlocked fragment references the failed control block. A replacement generation module, which generates a replacement control block from the terminal's most recent state vector and sends it via an interlocked encoding module and a group transmission module; The window decoding module is a terminal that performs sliding decoding based on the time-sensitive window function, discards groups outside the window, and restores the control block execution within the decodeable sequence number range.

2. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, The interlocking encoding module divides the control block into a first segment and a second segment. The main segment is the first segment; the interlocking segment is the result of the addition operation of the second segment and the first segment of the next control block in a finite field; the window decoding module solves the second segment of the corresponding control block by jointly solving the main segment and the interlocking segment in the adjacent control sequence number encoding group.

3. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, The control sequence number mapping rule maps the transmission time slot sequence to the control sequence number one by one. The transmission time slot sequence is generated by the time base shared by the control terminal and the unmanned platform terminal. The group transmission module determines the control sequence number corresponding to the coded group based on the position of the transmission time slot to which the coded group belongs in the transmission time slot sequence.

4. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, The tail risk assessment module performs threshold screening on the one-way delay samples to obtain the excess quantity samples, and updates the parameters of the generalized Pareto distribution on the excess quantity samples online. Based on the updated distribution, the conditional tail bound is calculated; the tail risk level is determined by the relationship between the conditional tail bound and the risk boundary corresponding to the time window function.

5. The low-latency, high-reliability unmanned control communication system according to claim 4, characterized in that, The quantile summary uses the Greenwald–Khanna quantile summary structure to maintain an ordered summary of the one-way time delay samples and outputs the high quantile estimate; the tail risk assessment module performs a consistency judgment on the high quantile estimate and the conditional tail boundary, and determines the tail risk level based on the judgment result.

6. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, When the policy scheduling module selects dual-path replication and forwarding, it replicates the coded packets with the same control sequence number into a primary packet and a backup packet. The primary packet enters the first scheduling queue and is sent through the first forwarding path, while the backup packet enters the second scheduling queue and is sent through the second forwarding path. The first scheduling queue and the second scheduling queue maintain independent queuing states and one-way delay sample sets, respectively.

7. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, The dependency clearing module constructs an interlock dependency table, which records the next control number referenced by the interlocked segment of the encoded group corresponding to each control number. When the decodeable number range determined by the decoding window digest does not contain the next control number, the encoded group corresponding to the control number is added to the clearing set and removed from the queue.

8. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, The replacement generation module performs a one-step state prediction on the terminal's most recent state vector to obtain a predicted state vector, and inputs the predicted state vector and the reference state vector into the incremental control update law to generate a replacement control block. The interlocking encoding module performs segmentation and interlocking operations on the replacement control block based on the encoding seed to obtain the main segment and interlocking segment corresponding to the replacement control block.

9. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, When the window decoding module detects an inconsistency between the interlock dependency and the set of received encoded packets, it generates a resynchronization request. The resynchronization request carries a decoding window summary and the most recently executed control sequence number. After the network node receives the resynchronization request, the session establishment module updates the encoding seed in the control session context and updates the time window function; the interlocked encoding module and the window decoding module perform interlocked encoding and sliding decoding according to the updated control session context.

10. The low-latency, high-reliability unmanned control communication system according to claim 1, characterized in that, Also includes: The decorrelation routing module extracts over-threshold indication sequences consistent with threshold screening based on the one-way delay sample sets maintained by the first and second scheduling queues, calculates the common over-threshold ratio and time delay correlation to form a path correlation index, and performs reselection or remapping of the queue domain binding relationship of the second forwarding path when the path correlation index meets the correlation enhancement condition, so that the second forwarding path and the first forwarding path fall into different congestion domains. At the same time, the updated dual-path binding relationship is written into the dual-path replication forwarding configuration of the policy scheduling module, so that the primary and backup packets with the same control sequence number enter different scheduling queues and forwarding path combinations.

Images