A congestion-aware routing method for deep space networks

Abstract

The application discloses a congestion-aware routing method for a deep space network. In order to solve the problem of redundant inter-satellite forwarding and transmission efficiency reduction caused by the lack of remote link queue state awareness in a distributed long-delay deep space network, an exogenous mapping load element is maintained for each contact at each node. When the node perceives the loop risk caused by inconsistent queue states through the quasi-loop data packet, the delay increment interval of unknown remote traffic is solved by using the locally known routing information without explicit information exchange, and the exogenous mapping load is updated accordingly. Then, the load is mapped to the subsequent route estimation time cost calculation as an available capacity penalty term, so as to actively suppress the selection of contacts with potential congestion risk in the forward routing stage. The application can improve the remote traffic awareness capability of the long-delay distributed network, reduce the redundant inter-satellite relay, and improve the end-to-end transmission efficiency and reliability of the deep space network without increasing the signaling overhead.

Description

Technical Field

[0001] This invention belongs to the field of communication technology, and in particular relates to a congestion-aware routing method for deep space networks. Background Technology

[0002] In terrestrial internet or traditional low-latency networks with stable topologies, nodes can typically obtain state samples of non-adjacent links within a very short timescale through continuous data streams, explicit signaling interactions, or high-frequency ACK feedback, thus forming accurate network state awareness. However, deep space links have physical characteristics such as high latency, limited bandwidth, and time-varying intermittent connectivity, making the real-time queue information interaction mechanism relied upon by traditional routing ineffective in such scenarios. Therefore, Contact Graph Routing (CGR), as a predictive dynamic routing framework, calculates the optimal routing path through a pre-computed node visibility plan (i.e., the contact plan).

[0003] In early deep space, transmission distance and inter-satellite visibility were the main factors affecting routing efficiency. Routing algorithms focused on achieving reliable end-to-end transmission as early as possible within a limited routing window. However, with the increase in data volume and topology scale, if only propagation delay and link establishment waiting time are considered as routing costs, while ignoring the excessively long transmission delays caused by limited bandwidth and queuing delays caused by external traffic, this single-factor routing decision-making approach will lead to problems such as premature congestion on specific routes, low transmission efficiency, and insufficient resource utilization. Therefore, achieving network congestion awareness in the absence of real-time global queue information synchronization and with limited single-satellite computing power becomes crucial for improving routing efficiency.

[0004] Existing distributed routing protocols for deep space networks typically rely on information from local transit packets to guide capacity management and route switching. However, they still lack effective awareness of the queue status of non-adjacent remote links. For example, the Delay Tolerant Networking Working Group's recommendation standard entitled "SCHEDULE-AWARE BUNDLEROUTING" (document number: CCSDS 734.3-B-1) submitted to the International Space Data Systems Advisory Committee (CCSDS) proposes pre-aware routing in space environments where the topology is stable but link establishment changes over time. Its route switching is affected by local backlog traffic waiting to be forwarded to the next hop. Subsequently, the paper "Schedule-Aware BundleRouting: Analysis and Enhancements," published in the International Journal of Satellite Communications and Networking, vol. 39, no. 3, pp. 237–249, 2021, proposed Unibo-CGR. This paper introduced queuing delay calculated based on the capacity occupancy of remote contacts along the forwarding path using local traffic data. This enhanced the local nodes' awareness of remote contact availability and improved transmission efficiency. However, the inconsistency in the status of the same contact queue held by different nodes in a distributed network increased uncontrollable loop behavior. Coupled routes with visibility between forwarding nodes will blindly forward packets to coupled routes because they cannot know the availability degradation of each other's downstream contacts, resulting in large-scale traffic loop oscillations. To address this, Unibo-CGR uses an anti-loop mechanism (AL) based on adding an extension block to the packet header to record the nodes traversed historically, detecting the risk of loops in each candidate route. This mechanism avoids high-risk loop paths containing already visited nodes during the route feasibility check phase, unless there are no other candidate routes available.

[0005] The AL loop prevention mechanism only passively avoids loops when it detects that a data packet is about to become a loop. While this limits the uncontrollable loop behavior of a single data packet, it is still a passive loop avoidance strategy that operates on a single data packet. Subsequent routing calculations cannot detect the performance degradation and congestion of the remote link caused by unused local traffic. The problem of data packets being blindly forwarded to seemingly feasible but actually overloaded coupled routes still exists. This is especially frequent in coupled downstream paths with transmission bottlenecks, causing a large number of data packets to undergo frequent and ineffective route switching between relay satellites, severely wasting scarce inter-satellite link resources. At the same time, because the storage space, available bandwidth, and energy consumption of deep space satellite nodes are subject to extremely strict physical constraints, excessively redundant forwarding behavior not only fails to alleviate congestion on the remote link but also further exacerbates local network paralysis, leading to a decrease in end-to-end transmission efficiency and a degradation of overall link utilization and transmission performance. Summary of the Invention

[0006] Objective: To address the problem of reduced transmission efficiency in existing deep space networks due to the susceptibility of queue state information to failure, which hinders effective synchronization of remote queue state information, this invention discloses a congestion-aware routing method for deep space networks. This method uses loop prediction information as the source of information for remote queue state and potential congestion risk awareness, updating the externally mapped load as an unknown traffic assessment indicator. This solves the problem of redundant inter-satellite forwarding and its associated transmission efficiency caused by the lack of remote queue state awareness and passive loop avoidance in deep space networks.

[0007] The above objectives are achieved through the following technical solutions:

[0008] This invention proposes a congestion-aware routing method for deep space networks. The method utilizes quasi-circular packets as an implicit detection of the remote queue status, and can estimate the remote network load without explicit information exchange.

[0009] Furthermore, the method specifically includes the following steps:

[0010] S1. A unified contact plan is implemented on each node of the deep space network, and a dynamically evolving external mapping load overhead item is introduced and maintained on each node for each contact in the contact plan. This item represents the unknown traffic overhead from distant sources that may occupy the contact resources but have not passed through the local node. The external mapping load is an overall mapping constructed for the predetermined capacity of the same contact by non-local nodes in the deep space network, and the initial value of the external mapping load item is set to 0. The contact plan is maintained by the mission center based on orbit calculations and distributed to each node of the deep space network. Each contact in the contact plan represents a visibility link establishment opportunity in the space network. The specific information of each contact includes: sending node, receiving node, link establishment time, link termination time, and link rate.

[0011] S2. Each node enables a forward-looking loop prevention mechanism. When a data packet is expected to form a loop when forwarded according to the current routing result, and the loop prevention mechanism prevents the data packet from actually being forwarded to the loop path, the local node identifies the data packet as a potential loop packet.

[0012] S3. When the local node detects a potential loop packet, it estimates the delay interval caused by unknown traffic from the remote end based on the network-wide consistent response plan, and performs external source mapping load update: the expected loop formation event is used as a reverse observation signal for remote potential congestion risk perception, the estimated arrival time of the optimal loop path and the non-loop optimal path to the destination node is extracted, and the delay introduced by the remote unknown traffic overhead on the loop path is calculated. The estimated value of this delay satisfies the following non-zero bounded interval: In the formula, The delay is unknown. For local nodes; and The non-loop forwarding paths obtained from local node routing are respectively with the optimal loop path The estimated time to reach the destination node; The ideal arrival time under conditions of no backlog of flow interference; The queuing delay growth granularity leading to route switching is preset; then, the median of a non-zero bounded interval is taken as the estimated value, and a capacity-aware confidence adjustment mechanism is introduced for the optimal loop path. Any contact on , its first Secondary external source mapping load increment The calculation is derived from the update event triggered by the simulated ring packet, and the calculation formula is as follows: In the formula, For contact Link speed; This represents the current remaining capacity percentage of the contact.

[0013] S4. In the subsequent routing calculation phase, the external source mapping load of each contact is mapped as a penalty term for available capacity and integrated into the queuing delay calculation of the corresponding forwarding hop, increasing the estimated time cost of selecting contacts with congestion risk: When mapping the multi-time-based accumulated external source mapping load to the capacity penalty term, a capacity-aware time-decrease mechanism is introduced, which decays based on the update time of the increment of each external source mapping load. At that time, contact External mapping load The attenuation formula is as follows: In the formula, express The corresponding update event trigger timestamp; This represents the capacity-aware depreciation rate factor, which is based on the current percentage of actual remaining capacity at the time of contact. Adjustments are made, and the calculation formula is as follows: In the formula, This is the sensitivity coefficient. To accelerate the adjustment of the capacity threshold for the decay rate, and These are the upper and lower bounds of the set attenuation rate, respectively.

[0014] Beneficial effects:

[0015] 1. This invention eliminates the need for explicit state synchronization signaling between satellite nodes. Instead, it transforms the generation of ring packets under the original AL anti-loop mechanism into implicit detection of queue state inconsistencies, thereby sensing potential congestion risks in remote links and effectively saving deep space bandwidth resources.

[0016] 2. By forward-suppressing the switching of potentially congested paths, subsequent data packets tend to choose links with no or low risk of congestion loops. In deep space networks, this prompts packets to be forwarded to the deep space backbone links more quickly, making every effort to avoid traffic oscillations between relay nodes caused by state inconsistencies.

[0017] 3. Existing anti-loop mechanisms and extended block basic strategies are reused. A linear superposition mechanism is used during external source mapping load updates, which minimizes the computational overhead of each contact's single external source mapping load update and mapping compared to the basic routing algorithm, making it suitable for spaceborne computing environments with limited computing resources. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall process of the congestion-aware routing method for deep space networks according to the present invention;

[0019] Figure 2 This is a schematic diagram of a coupled routing model;

[0020] Figure 3 This is a schematic diagram of a deep-space Earth-Moon network scenario.

[0021] Figure 4 A comparison chart of the average number of hops per package for end-to-end delivery;

[0022] Figure 5 This is a comparison chart of average end-to-end delivery delay per package. Detailed Implementation

[0023] The specific implementation schemes of the present invention are further described below with reference to the accompanying drawings and specific application scenarios:

[0024] A congestion-aware routing method for deep space networks according to the present invention includes the following steps:

[0025] 1. Initialization of contact plan and external mapping load.

[0026] The mission center is responsible for maintaining a unified contact plan based on orbital calculations and distributing a consistent contact plan to all nodes in the deep space network. Each contact in the contact plan represents a visibility link establishment opportunity in the space network. Specific information includes the sending node, receiving node, link establishment time, link termination time, and link rate. The total contact capacity can be obtained by multiplying the link establishment duration by the link rate. The contact queuing delay is obtained by dividing the predetermined capacity by the link rate, but it does not necessarily affect the actual first byte transmission time of the data packet at the next hop. This value depends on the latest of the packet's last byte arrival time at the current node and (the link establishment time of the next hop contact + queuing delay). Based on this, CGR performs iterative calculations at each hop to obtain the estimated Projected Bundle Arrival Time (PBAT) for the route. This invention, based on this, correlates the delay caused by unknown traffic with the load, achieving a unified mapping of the availability of the same contact across different paths.

[0027] After the Contact mission is completed, each satellite node of the Deep Space Network independently constructs an attribute structure for each Contact and manages the Contact. Maintain a global external source mapped load status metric This characterizes the amount of unknown traffic reserved on the contact by a coupled node that has a connection with a remote node but does not pass through the local node. External mapped load. Its initial value is 0.

[0028] 2. Recognition of simulated loops

[0029] Each satellite node runs a routing algorithm based on Unibo-CGR locally and enables the AL loop prevention mechanism. When the routing result of a data packet forms a loop, and the loop prevention mechanism prevents actual forwarding to the loop, the local node identifies the data packet as a pseudo-loop packet and triggers an external source mapping load update event.

[0030] 3. External Mapping Load Update Method

[0031] like Figure 2 In the coupled routing model shown, local modeling nodes are used. From the perspective of, For any relay node on the downstream coupling path. Let be the final destination node. Assume that under no-queue conditions, from... arrive The optimal path is ,from to the destination node The optimal path is Each path contains one or more contacts, but does not include contacts leading to coupled nodes.

[0032] When the ring packet triggers an update event, let:

[0033] 1) The non-loop forwarding paths selected from the candidate routing table (i.e., candidate paths excluding loop paths) contain... Calculated The same contact, used to ensure Optimal route to destination and The non-loop forwarding route corresponds to this.

[0034] 2) By Identified loop path (i.e., through) The optimal path (but one that will form a loop) includes... Calculated The same contact, used to ensure Paths abandoned due to congestion and The currently identified loop paths are the same.

[0035] When effective real-time explicit inter-satellite information exchange cannot be supported, this invention addresses the limitations of node-based communication. The local system utilizes local state and loop routing metrics to assess unknown remote queuing delays caused by simulated loop events. Perform interval estimation.

[0036] set up and Local nodes The PBAT metric for the corresponding path obtained from the routing; This represents the ideal arrival time under conditions of no backlog of traffic interference. In the event of a quasi-loop event, the node... Based on locally known spatiotemporal information, the actual latency increment caused by unknown traffic from a distant location is deduced. The following non-zero bounded intervals are satisfied: in, for Preset choose And give up The difference in arrival times between routes to the destination node, which leads to the granularity of queuing delay growth during route switching, can typically be configured as a single queuing increment generated by a single data packet at the backbone contact, which acts as a bottleneck. To maintain risk neutrality in decision-making within the highly uncertain and information-isolated deep space environment, this invention extracts the arithmetic mean of the aforementioned non-zero interval as... The estimated value of the signalless measurement.

[0037] Furthermore, in order to adapt to occasional loop false alarms caused by non-congestion factors such as index symmetry, this invention calculates each specific contact... The The second update increments the external mapping load. At this time, a capacity-aware confidence adjustment mechanism based on the cosine function is introduced. For the optimal loop-forming path... Any contact on , its first Secondary external source mapping load increment The calculation is derived from the update event triggered by the simulated ring packet, and the calculation formula is as follows: in, For contact Link speed, This represents the current percentage of actual remaining capacity for that contact. This mechanism prevents a sharp drop in confidence during periods of ample link resources and encourages a more cautious approach to estimated unknown loads at distant locations when remaining capacity is reduced, particularly when it is nearing depletion. It also prevents limited remaining capacity from being completely disabled due to externally mapped loads.

[0038] Furthermore, regarding algorithm time complexity control, each contact external mapping load contains numerous update components, all of which need to be decayed according to their respective update timestamps. To reduce the complexity of managing multi-time-based decay, this invention utilizes the principle of linear superposition to record the time of the last update of the contact external mapping load. Before superimposing new update components in each update, the existing external mapping load is decayed according to the time elapsed since the last update, thus achieving... The time complexity is constant. Additionally, updates are performed on the path contained within a single loop. The complexity of a single external mapping load update event with one contact is... .

[0039] 4. External source mapping load capacity awareness time-decrease and penalty term mapping

[0040] Considering the dynamic changes and time-switching characteristics of deep space network traffic, and to prevent historically expired congestion misjudgments from interfering with subsequent normal routing selection, this invention introduces a capacity-aware time-dependent decay mechanism based on time-linear superposition for externally mapped loads. At the current moment... ,touch Global external mapping load The attenuation calculation formula is as follows: in, express The corresponding update event trigger timestamp. To enable the decay process to adaptively match the dynamic congestion changes of the contact, this invention introduces a Sigmoid nonlinear adjustment function to adjust the decay rate factor. Based on the current actual remaining capacity percentage of the contact. Dynamic sensitivity adjustment is performed. The calculation formula is as follows: in, This is the sensitivity coefficient. This serves as a baseline for triggering accelerated capacity decay. When the available capacity ratio approaches... At this point, the decay rate will be adjusted more sensitively. When the actual remaining capacity is low and faces the risk of exhaustion and high congestion, the decay rate will approach the set lower bound. At this point, the externally mapped load decays slowly locally, causing the route to adopt a more persistent avoidance strategy for this contact. Lower bound of decay rate. The reciprocal of represents the longest queuing delay determined by the duration of the bottleneck contact in the routing, which in deep space networks is typically an inter-object backbone link with the lowest transmission rate and the longest distance to transmission. Based on this, the more pronounced the bottleneck characteristics of the backbone link, the greater the gain that this invention can generate. This design ensures that expired historical data does not lead routing decisions to be overly pessimistic about the contact to the point of complete abandonment.

[0041] The updated external source mapping load for each contact is incorporated into routing decisions through queuing delay. During capacity-based queuing delay calculation, the external source mapping load that has undergone capacity-aware time-decrease is included as unknown traffic occupying that contact, serving as a penalty for available capacity within the pre-defined capacity. This appropriately delays the PBAT metric calculated locally for routes abandoned by remote nodes due to congestion risk. After a ring-forming event occurs, all subsequent packets will have their routing selection probability adjusted based on the actual first-byte transmission time of the next hop, prioritizing routes that can be forwarded faster to the Deep Space Backbone or other low-load relay nodes without congestion risk.

[0042] A specific application scenario is described below, and the internal interaction process is shown in the appendix. Figure 1 As shown:

[0043] This invention targets distributed, long-latency deep space networks, and is most effective in topologies with coupled candidate paths. Triggered by a pseudo-loop event under an enabled AL loop prevention mechanism, the load state of distant non-adjacent links is obtained based on local loop and candidate route information. This invention establishes an implementation scheme for externally mapped load updates and mapping models. Based on this, risk overhead is dynamically injected during the forward routing phase to adjust the corresponding hop count queuing delay, dynamically controlling the probability of encountering loop congestion risks. This invention can be applied to deep space networks such as Earth-Moon and Earth-Mars networks, and can also be introduced as a congestion detection scheme in a fully distributed architecture into other distributed routing frameworks and scenarios where state synchronization cannot be effectively implemented.

[0044] To illustrate how this invention can be used in a real network, based on the appendix... Figure 3 The following describes a deep-space Earth-Moon network data transmission scenario, which includes two ground stations, one mission center, two Earth-side relays, two Earth-side relays, and one lunar lander. The deep-space backbone link between Earth and the Moon uses the lowest data rate and the longest propagation delay to simulate a challenging deep-space link.

[0045] In this application scenario, the mission planning system generates a contact plan covering 86,400 seconds based on the visibility window information of each spacecraft's orbit and ground station. Each relay satellite runs the Unibo-CGR routing framework (with the AL anti-loop mechanism enabled) and loads the congestion-aware routing method proposed in this invention.

[0046] (1) Scene parameters and external mapping load update

[0047] Taking a conditional Poisson traffic flow of 1 MB packets with a lifespan of 15000 seconds transmitted from the lunar lander to the ground mission control center as an example. The network backbone relay link bandwidth is configured at 50 kBps, at which point the link capacity is close to saturation. When a ring-forming packet triggers a ring-forming event:

[0048] 1) The local node automatically captures the loop-forming event, reads the current local time and the estimated time difference to the destination node for the two paths. The queuing delay growth granularity is set according to packet size and backbone link speed, and the queuing delay estimate caused by unknown traffic overhead at the far end on the local optimal loop that triggers the loop-forming event is obtained through an update algorithm.

[0049] 2) Calculate the capacity-aware time-degradation rate based on the average connection establishment time of the backbone link, which acts as the bottleneck. Set the sensitivity coefficient and sensitivity threshold for adjusting the attenuation rate based on the sensitivity test in this scenario, and attenuate historical externally mapped loads.

[0050] 3) Based on the current remaining volume percentage of the contact, calculate the current updated confidence of the contact using the capacity-aware confidence adjustment formula, and add it to the external mapping load.

[0051] (2) Mapping of available capacity penalty terms with forward avoidance control

[0052] When a deep-space relay node routes its subsequent data packets:

[0053] 1) During the estimated arrival time of the destination node in the route calculation phase, read the external source mapping load, external source mapping load update event and current time of each contact included in the route, attenuate the external source mapping load according to the attenuation calculation formula, and add it as the estimated capacity overhead of unknown traffic to the calculation of queuing delay factor.

[0054] 2) According to the iterative calculation formula, the externally mapped load will dynamically adjust the PBAT index of the downstream path that contains the congestion risk contact, so that it rises appropriately.

[0055] 3) This congestion-aware routing mechanism prevents a large number of traffic flows from blindly forwarding to congested relay links that have been abandoned by coupled nodes. While introducing traffic awareness to improve transmission efficiency, it effectively overcomes the risk of transmission degradation caused by sparse and mismatched state information in deep space communication.

[0056] The simulation performance of this invention, baseline pre-defined sensing routes, and baseline Unibo-CGR routes in this Earth-Moon application scenario is compared, and the results are as follows: Figure 4 and Figure 5 As shown, this invention effectively reduces the number of inter-satellite redundant forwarding hops caused by inconsistent queue status information and simultaneously optimizes end-to-end delivery latency.

Claims

1. A congestion-aware routing method for deep space networks, characterized in that, Includes the following steps: S1. A unified contact plan is implemented on each node of the deep space network, and a dynamically evolving external mapping load is introduced and maintained on each node for each contact in the contact plan. This load is used to characterize the unknown traffic overhead from distant sources that may occupy the contact resources but have not passed through the local node. The external mapping load is a general mapping constructed for the predetermined capacity of the same contact by non-local nodes in the deep space network, and the initial value of the external mapping load item is set to 0. The contact plan is maintained by the mission center based on orbit calculations and distributed to each node of the deep space network. Each contact in the contact plan represents a visibility link establishment opportunity in the space network. The specific information of each contact includes: sending node, receiving node, link establishment time, link loss time, and link rate. S2. Each node enables a forward-looking loop prevention mechanism. When a data packet is expected to form a loop when forwarded according to the current routing result, and the loop prevention mechanism prevents the data packet from actually being forwarded to the loop path, the local node identifies the data packet as a potential loop packet. S3. When a local node detects a potential loop packet, it estimates the delay interval caused by unknown traffic from the remote end based on a consistent contact plan across the entire network, and performs an external source mapping load update: the expected loop event is used as a reverse observation signal for remote potential congestion risk perception; the estimated arrival time of the destination node for the optimal loop path and the non-loop optimal path is extracted; the delay interval introduced by the overhead of unknown traffic from the remote end on the loop path is calculated; the median of the non-zero bounded interval is taken as the estimated value; a capacity-aware confidence adjustment mechanism is introduced; the capacity-aware confidence adjustment mechanism is to weight the estimated value with confidence based on the real-time actual remaining capacity percentage of the contact, and combined with the link rate of the contact, update the increment of the external source mapping load of each contact on the loop path; S4. In the subsequent routing calculation phase, the penalty term for mapping the external source mapping load of each contact to the available capacity is incorporated into the queuing delay calculation of the corresponding forwarding hop: When mapping the external source mapping load accumulated at different times to the penalty term of the available capacity, a capacity-aware time-decrease mechanism is introduced. The capacity-aware time-decrease mechanism is based on the update time of the multiple external source mapping load increments contained in each contact, and the decay rate factor is adjusted according to the actual remaining capacity percentage of the contact, increasing the estimated time cost of selecting contacts with congestion risk. The time delay interval caused by the unknown remote traffic overhead on the loop path in step S3 is estimated by taking the median of the non-zero bounded interval, specifically satisfying the following non-zero bounded interval: In the formula, The delay is unknown. For local nodes; and The non-loop forwarding paths obtained from local node routing are respectively with the optimal loop path The estimated time to reach the destination node; The ideal arrival time under conditions of no backlog of flow interference; This is the preset granularity of queuing delay growth that leads to route switching.

2. The congestion-aware routing method for deep space networks according to claim 1, characterized in that: The capacity-aware confidence adjustment mechanism described in step S3 specifically addresses the optimal loop-forming path. Any contact on , its first Secondary external source mapping load increment The calculation is derived from the update event triggered by the simulated ring packet, and the calculation formula is as follows: In the formula, For contact Link speed; This represents the current percentage of actual remaining capacity at this contact. Pi is a constant.

3. The congestion-aware routing method for deep space networks according to claim 2, characterized in that: The capacity-aware time-decrease mechanism described in step S4 specifically refers to the mechanism at the current moment... At that time, contact External mapping load The attenuation formula is as follows: In the formula, Indicates the first Secondary external source mapping load increment The corresponding update event trigger timestamp; This represents the capacity-perceived time-depreciation rate factor, which is based on contact Current actual remaining capacity percentage Adjustments are made, and the calculation formula is as follows: In the formula, This is the sensitivity coefficient. To accelerate the adjustment of the capacity threshold for the decay rate, and These are the upper and lower bounds of the set attenuation rate, respectively. is the base of the natural logarithm.

Images