Deterministic transmission method for point cloud video stream in intelligent fusion identification network

Abstract

The invention provides a deterministic transmission method of a point cloud video stream in an intelligent fusion identification network. The method comprises the steps that a scheduling plane of a DT-CNST architecture in the intelligent fusion identification network collects demand information of holographic service flow in the intelligent fusion identification network and obtains calculation, storage and network resource distribution conditions in the intelligent fusion identification network, and a control plane allocates resources for the holographic service flow according to the collected information through a CTS model, transmitting a resource allocation result to a data plane; the data plane executes the holographic service according to the resource allocation result; the scheduling plane captures a point cloud frame of a holographic service through a camera, converts the point cloud frame into a periodic PCV flow through a CTS model, deploys a BCQF mechanism through a switch, and achieves deterministic transmission of the PCV flow through models of a sending queue, a buffering queue and a receiving queue. According to the invention, high bandwidth and ultra-low delay transmission requirements of the point cloud video stream can be met, and stable holographic communication experience can be provided for users through dynamic identifier management and resource optimization.

Description

Technical Field

[0001] The present invention relates to the field of computer application technologies, and particularly to a deterministic transmission method for point cloud video streams in an intelligent fusion identification network. Background Art

[0002] With the rise of the sixth-generation mobile communication technology (6G), the Metaverse, and the Computing Power Network (CPN), holographic-type communication (HTC) is gradually becoming an important trend in future communication methods. Holographic communication technology uses volumetric video technology to bring users an immersive experience. Holographic communication features large bandwidth, low latency, and strong computing power, and requires the network to have the ability to adaptively schedule diverse resources such as computing, forwarding, and storage.

[0003] Point cloud video is an emerging medium composed of 3D data, which can provide a highly immersive and interactive user experience. The frames in a point cloud video stream are called point cloud frames, which consist of hundreds of thousands of points with coordinate and color information. These points together construct a three-dimensional space, allowing viewers to observe and experience the video content from different angles. In a real-time transmitted point cloud video, a point cloud frame (PCF) usually contains approximately 200,000 to 1,000,000 points, and the data size ranges from 4MB to 20MB. To meet the transmission requirements of such high-quality videos, a point cloud video with 30 frames per second requires a transmission rate of 1 - 10 Gbps, which undoubtedly poses higher requirements for network bandwidth.

[0004] The introduction of the intelligent fusion identification network provides strong support for real-time holographic communication systems. Through multi-identifier fusion and intelligent parsing capabilities, the intelligent fusion identification network can dynamically manage the mapping between identifiers and resources in the network, optimize the data transmission path, and reduce transmission latency.

[0005] A real-time volumetric video transmission method in the prior art includes technologies for transmitting three-dimensional volumetric video data through a network to perform real-time rendering, display, and interaction at the receiving end. The disadvantages of this method include: this method focuses on efficient data compression, and some focus on data stream transmission, ignoring how to ensure precise control and consistency during the transmission process. In an uncertain network environment, data fluctuations and losses may lead to a decline in image quality and further increase latency. This instability makes it difficult for the prior art to be widely applied in actual holographic communication systems.

[0006] A point cloud compression (PCC) method in the prior art includes: while storing and transmitting a point cloud model containing a large amount of three-dimensional point data through an encoding method, key geometric and attribute information is retained as much as possible to achieve real-time three-dimensional data exchange and processing under limited bandwidth and storage conditions. The disadvantages of this method include: the data size of each frame of traditional two-dimensional video is the same, while a point cloud frame PCF consists of thousands of point clouds, and even in the same PCV stream, the data size of each PCF varies greatly. The variation in the data size of each frame will cause fluctuations in the time consumed by the PCC algorithm to process the PCF, and this fluctuation in the calculation time makes it difficult to achieve a bounded calculation delay. Summary of the Invention

[0007] An embodiment of the present invention provides a deterministic transmission method for a point cloud video stream in an intelligent fusion identification network to effectively improve the transmission efficiency of the point cloud video stream.

[0008] To achieve the above objective, the present invention adopts the following technical solutions.

[0009] A deterministic transmission method for a point cloud video stream in an intelligent fusion identification network, in which a deterministic transmission DT-CNST architecture is set up in the intelligent fusion identification network. The DT-CNST architecture includes a data plane, a control plane, and a scheduling plane. The method includes:

[0010] The scheduling plane collects the demand information of the holographic service flows in the intelligent fusion identification network, obtains the distribution of computing, storage, and network resources in the intelligent fusion identification network, and transmits the collected information to the control plane;

[0011] The control plane, through the traffic shaper CTS model, optimizes the PPO-greedy resource trading algorithm according to the collected information through a hybrid probability strategy to allocate computing time slots, computing power sizes, forwarding time slots, and storage resources for the holographic service flows, and transmits the resource allocation results to the data plane;

[0012] After the configuration is completed, the data plane executes the holographic service according to the resource allocation results;

[0013] The scheduling plane captures the point cloud frames of the holographic service through a camera, converts the point cloud frames into a periodic holographic video PCV stream through the CTS model, and the scheduling plane uses the switch to deploy the BCQF mechanism to achieve the deterministic transmission of the PCV stream through the models of the sending queue, buffer queue, and receiving queue.

[0014] Preferably, the control plane passes the CTS to optimize the PPO-greedy resource trading algorithm according to the collected information, and allocates computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flow, and transmits the resource allocation result to the data plane, including:

[0015] Construct a CNS model, set the constraint conditions and resource trade-off relationships of the CNS model. The CNS model allocates computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flow according to the collected information, the constraint conditions, and the resource trade-off relationships through the HPG-RT algorithm, obtains a preliminary resource allocation result, optimizes the preliminary resource allocation result through the PPO-greedy resource trading algorithm, obtains the resource allocation result of the holographic service flow, and transmits the resource allocation result to the data plane.

[0016] Preferably, the constructing of the CNS model and setting the constraint conditions and resource trade-off relationships of the CNS model include:

[0017] Construct a CNS model, set that the CNS model contains 7 constraint conditions and 3 resource trade-off relationships. The constraint conditions specifically include:

[0018] 1) Forwarding period and super-period size constraints. The minimum period size depends on the transmission delay of packets in a single queue between two adjacent nodes, and the maximum period size is the greatest common divisor of the periods of all data flows, as shown in formula (13):

[0019]

[0020] Among them, Q size is the maximum number of packets that a queue can accommodate, D proc is the maximum processing delay, D prog is the maximum propagation delay in the network, the preset value Q size depends on the capabilities of the switch and is greater than N pkt , is the minimum period, is the maximum period, N pkt represents the maximum number of packets allowed to be transmitted within a single time slot;

[0021] As shown in formula (14), the size of the forwarding period C net is within the range of and and is also one of the common divisors of all periods.

[0022]

[0023] represents a positive integer and is a mathematical symbol

[0024] As shown in formula (15), the super cycle HC net represents the number of forwarding cycles in a complete cycle, and its size should not be less than the least common multiple of the periods of all data streams;

[0025] HC net = LCM(S.periods) #(15)

[0026] where j ∈ [0, NC net - 1] and NC net NC net is the number of cycles included in a super cycle, S refers to the traffic set, and S.periods refers to the set of periods of all flows;

[0027] 2) Network resource constraint represents the state of data stream s on link e(m, n) during the j-th forwarding cycle i , and the specific formula is shown in formula (16):

[0028]

[0029] should satisfy the periodicity shown in formula (17) and the continuity shown in formula (18).

[0030]

[0031] where:

[0032]

[0033] e(m, n) refers to the link between node v m and node v n , s i represents the i-th PCV flow, s i .FC net represents the number of forwarding cycles occupied by flow s i , and are both network cycles, representing the k-th and j-th cycles;

[0034] 3) Calculation cycle size constraint, as shown in formula (20), the calculation cycle C com is an integer multiple of the forwarding cycle size:[[]]

[0035]

[0036] 4) Computational power constraint: As shown in Equation (21), the sum of the computational power occupied by the data flow does not exceed the computational power of the edge computing node;

[0037]

[0038] where, denotes the state of s i in the computing cycle If s i is assigned to then g otherwise it is 0. is the computational power of the edge computing node, and s i .R com is the computational power allocated to s i ;

[0039] 5) Computational delay constraint: The computational delay of point cloud filtering and compression is as shown in Equation (22):

[0040]

[0041] denotes the number of computing cycles required for the filtering and compression of the flow si, including the filtering delay and the compression delay. C com is the size of the computing cycle, denotes the filtering delay of the flow si, denotes the compression delay of the flow si, and ΔT denotes the maximum variance of the processing and computing delay of the flow si;

[0042] The computational delay of the data flow s i is as shown in Equation (23):

[0043]

[0044] where the offset of s i satisfies:

[0045] s i .off × C net ≤ s i .period #(24)

[0046] L com denotes the overall computational delay of the flow, denotes the decompression delay, and s i .off denotes the number of offset network cycles of the flow si, which means that the data will wait for off network cycle times before being sent after processing;

[0047] 6) Storage delay and resource constraint: The storage delay of the flow s i in the switch SWm needs to satisfy Equation (25):

[0048]

[0049] Combined with formula (10), the range refers to the range of the storage time of flow si in the switch, and the range is formalized as formula (26):

[0050]

[0051] Denote the traffic set passing through switch sw m of

[0052] Denote the time that flow si needs to be stored in the switch.

[0053] For any two flows stored in switch SWm, the storage delay needs to satisfy the condition shown in formula (27):

[0054]

[0055] 7) MTP delay and jitter constraints. The constraint requires that the MTP delay shown in formula (28) must be less than the set upper limit of the MTP delay.

[0056]

[0057] Denote the maximum motion-to-imaging MTP delay of flow si

[0058] ddl mtp Denote the maximum allowable MTP delay of flow si

[0059] The jitter is constrained as shown in formula (29).

[0060] s i .J mtp ≤s i .period.#(29)

[0061] s i .J mtp Denote the maximum motion-to-imaging MTP jitter of flow si

[0062] The resource trade-off relationship specifically includes:

[0063] 1) The trade-off between network resources and computing resources. The relationship between network resources and computing resources is shown in formula (30);

[0064]

[0065] R com The network resources allocated to flow si

[0066] Compression ratio of the α compression algorithm

[0067] 2) Trade - off between network resources and storage resources. The relationship between network resources and storage resources is shown in formula (31);

[0068]

[0069] 3) Trade - off between computing resources and storage resources. The relationship between computing resources and storage resources is shown in formula (32);

[0070]

[0071] Preferably, the CNS model allocates computing time slots, computing power, forwarding time slots, and storage resources for holographic service flows through the HPG - RT algorithm according to the collected information, the constraint conditions, and the resource trade - off relationship, obtains a preliminary resource allocation result, and optimizes the preliminary resource allocation result through the PPO - greedy resource trading algorithm to obtain the resource allocation result of the holographic service flow, including:

[0072] Based on the constraint conditions and the resource trade - off relationship of the CNS model, define Sched(s i ) as whether the flow can be scheduled. The objective function of the CNS model is shown in formula (33), and the constraint conditions are shown in formula (34);

[0073]

[0074] Among them, S suc is the number of successfully scheduled flows;

[0075] Use the CNS resource trade - off HPG - RT algorithm based on hybrid PPO - Greedy to solve the objective functions shown in formula (33) and formula (34) to obtain the number of successfully scheduled data flows maxS suc , The first step of the HPG - RT algorithm uses the PPO algorithm to schedule computing resources and network resources for each data flow. The PPO algorithm generates rewards and the next state through an agent, and the trajectory is saved in the trajectory memory. When the network needs to be updated, the trajectory will be sampled in small batches and enter the new actor network to generate new action probabilities, and calculate the ratio ρ t (θ),, generate V π by the evaluation network according to the input batch current state and batch next state t and V traget (st t+1 ), calculate V π (st t) and V traget (st t+1 ) The error between them, and the evaluation loss is obtained according to this error. After obtaining the action loss and the evaluation loss, the action network and the evaluation network are updated;

[0076] The second step of the HPG-RT algorithm further maximizes the number of successfully scheduled data streams maxS based on the result of the PPO algorithm through the PPO-greedy resource trading algorithm suc , allocates more optimal computing power and storage cycle settings resources for each stream through resource trade-off, generates a gating control list of BCQF while maximizing the number of scheduled streams, and obtains the resource allocation result of the holographic service flow by integrating the calculation results of the first and second steps of the HPG-RT algorithm.

[0077] Preferably, the scheduling plane captures the point cloud frame of the holographic service through a camera, and converts the point cloud frame into a periodic holographic video PCV stream through the CTS model, including:

[0078] CTS stores the collected PCFs in the polygon file format PLY, and calculates the data size of the PCF through the number of vertices and the number of bytes occupied by a single vertex in the PLY format. The specific calculation is shown in Equation (1);

[0079] DS pcf = Numofvertices×28Bytes + 334Bytes.#(1)

[0080] Among them, 28 bytes is the data size of a single vertex, and 334 bytes is the size of the PLY file header;

[0081] The PCFs are shaped into the same data size through the redundant vertex filtering Draco algorithm, and the calculation formula of the compression ratio α of the redundant vertex filtering algorithm is shown in Equation (2);

[0082]

[0083] For any stream s in the Draco algorithm i , the allocated computing cycle should be greater than the sum of the maximum filtering and compression delays, as shown in Equation (3):

[0084]

[0085] represents the number of computing cycles required for the filtering and compression of stream si, including the filtering delay and the compression delay, C com is the size of the computing cycle, represents the filtering delay of stream si, Denote the compression delay of stream \(s_i\), and \(\Delta T\) denote the maximum variance of the processing and computing delay of stream \(s_i\);

[0086] The calculation method is shown in formula (4):

[0087]

[0088] Denote the filtering delay of stream \(s_i\), \(\max(\text{vertices})\) denote the maximum number of vertices owned by the point cloud frame file of stream \(s_i\), and \(\text{sample}(\text{vertices})\) denote the number of vertices that need to be sampled in the point cloud frame file of stream \(s_i\);

[0089] Among them, \(\text{sample}(\text{vertices})\) is the number of sampled vertices, and \(\tau\) vertex refers to the execution time required for filtering a vertex;

[0090] The compression time and decompression time using the Draco algorithm are calculated as shown in formula (5):

[0091]

[0092] Among them, \(\alpha\) is the compression ratio, and \(s\) i .R com is the computing resource allocated for data stream \(s\) i ; Compression computing delay, Decompression computing delay, For the filtered point cloud frame use the Draco algorithm for compression, set the compression ratio to \(\alpha\), and allocate the computing resource as \(s\) i .R com , For the compressed point cloud frame use the Draco algorithm for decompression, and allocate the computing resource as \(s\) i .R com ;

[0093] Through the above design, CTS shapes the PCFs into a PCFs stream with the same frame size and equal frame time intervals. The period of the stream and the compressed frame data size are shown in formula (6):

[0094]

[0095] s i .period is the period of stream \(s_i\), and \(s\) i .fps hardware is the frame rate of the hardware device that generates stream \(s_i\), The size of the compressed point cloud frame in the stream si, the compression ratio α, DS pfs (sample(vertices)) The size of the number of sampled vertices.

[0096] Preferably, the scheduling plane uses a switch to deploy the BCQF mechanism to achieve deterministic transmission of the PCV stream through models of a sending queue, a buffer queue, and a receiving queue, including:

[0097] The BCQF network transmission mechanism manages data streams through models of a sending queue, a buffer queue, and a receiving queue, and divides time into multiple forwarding cycles C net , and the number of data packets that can be forwarded in each forwarding cycle is shown in Equation (7):

[0098]

[0099] where BW is the link bandwidth, GD is the protection bandwidth within a forwarding cycle, MTU represents the maximum transmission unit, and IFG represents the frame interval, obtaining the compressed stream s i The number of forwarding cycles s occupied by the PCFs of i .FC net As shown in Equation (8):

[0100]

[0101] When a data packet enters a switch supporting BCQF, it is directed to a specified port. Only the data packets in the PCV stream can enter the stream storage area. The stream storage area sets a buffer and a corresponding selective enqueue gating for each PCV stream. The enqueueing of data packets is managed by the gating list of the selective enqueue gating. Only within the allowed cycle can the data packets enter the sending queue, the buffer queue, or the receiving queue;

[0102] The buffered data packets are sent using the sending queue. The receiving queue and the buffer queue are used to receive and buffer data packets. Only when the buffer queue is full will the remaining data packets enter the receiving queue. The selective transmission gating sets a gating after each queue. With the help of the forwarding gating, the sending queue, the buffer queue, or the receiving queue can switch roles. The stream s i The storage cycle on the switch SWn is defined as Equation (9):

[0103]

[0104] where represents the set of streams passing through the switch sw n ;

[0105] The stream s iThe upper and lower bounds of the network latency of a data packet in BCQF are shown in Equations (10) and (11) as follows:

[0106]

[0107] s i .path is shown in Equation (12) as follows

[0108] s i .path = {sw0, …, sw m , …, sw M}# (12).

[0110] As can be seen from the technical solutions provided by the embodiments of the present invention above, the present invention designs a Deterministic Transmission based on Computing, Network, and Storage Resource Trading (DT-CNST) mechanism to ensure that the point cloud video stream in real-time holographic communication can be transmitted in a deterministic and low-latency manner. The combination of the intelligent characteristics of the Zhirong identification network and the DT-CNST transmission mechanism can not only meet the high-bandwidth and ultra-low-latency transmission requirements of the point cloud video stream, but also ensure the bounded end-to-end MTP latency through dynamic identification management and resource optimization, providing users with a stable holographic communication experience.

[0111] Additional aspects and advantages of the present invention will be given in part in the following description, and these will become apparent from the following description or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0112] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention, and those of ordinary skill in the art can obtain other drawings based on these drawings without creative efforts.

[0113] Figure 1 FIG. [ID] is an architecture diagram of a DT-CNST transmission mechanism provided by an embodiment of the present invention;

[0114] Figure 2 FIG. [ID] is a processing flowchart of a method for deterministic transmission of a point cloud video stream in a Zhirong identification network provided by an embodiment of the present invention;

[0115] Figure 3A flowchart of the operation of a CTS (Computing-based Traffic Shaper) provided by an embodiment of the present invention;

[0116] Figure 4 A BCQF model diagram provided by an embodiment of the present invention;

[0117] Figure 5 A schematic diagram of packet group forwarding between switches provided by an embodiment of the present invention.

[0118] Figure 6 A scheduling framework diagram of a CNS model based on PPO provided by an embodiment of the present invention. Detailed implementation manners

[0119] The following details the implementation manners of the present invention. Examples of the implementation manners are shown in the accompanying drawings, where the same or similar reference numerals represent the same or similar elements or elements with the same or similar functions from beginning to end. The implementation manners described below by referring to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be construed as a limitation to the present invention.

[0120] Those skilled in the art of the present technology can understand that, unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "including" used in the description of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups. It should be understood that when we say an element is "connected" or "coupled" to another element, it can be directly connected or coupled to other elements, or there may also be intermediate elements. In addition, the "connection" or "coupling" used herein may include wireless connection or coupling. The phrase "and / or" used herein includes any unit and all combinations of one or more related listed items.

[0121] Those skilled in the art of the present technology can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as the general understanding of those of ordinary skill in the art to which the present invention belongs. It should also be understood that terms such as those defined in a general dictionary should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted with an idealized or overly formal meaning unless defined as here.

[0122] For ease of understanding of the embodiments of the present invention, the following will further explain with several specific embodiments as examples in conjunction with the accompanying drawings, and each embodiment does not constitute a limitation to the embodiments of the present invention.

[0123] In an embodiment of the present invention, the DT-CNST transmission mechanism combines computing, network, and storage resource trading to achieve deterministic transmission of point cloud video streams in holographic communication, while providing a stable and fast frame rate for real-time holographic communication on mobile devices. An architecture of the DT-CNST transmission mechanism provided by an embodiment of the present invention is as shown in Figure 1 shown. The architecture of DT-CNST adopts the SDN (Software Defined Network) method and is divided into three planes: the data plane, the control plane, and the scheduling plane.

[0124] On the data plane, the 3D camera is responsible for capturing and generating PCFs (point cloud frames). Subsequently, based on computing CTS, the PCFs are converted into a periodic holographic video stream PCV (Point Cloud Video) through filtering and compression operations. PCV is video data composed of continuous point cloud frames. Each frame of point cloud represents three-dimensional space data at a certain time point. The present invention adopts the BCQF (Buffered Cyclic Queuing and Forwarding) mechanism to ensure the deterministic network transmission of PCFs, avoiding transmission conflicts and delays.

[0125] The control plane uses a centralized controller to collect heterogeneous resource information in the network, sends the heterogeneous resource information to the scheduling plane for processing, and obtains flow tables and configuration files. The scheduling plane distributes these flow tables and configuration files to switches in the network to guide the switches to process and forward data packets. The control plane ensures the reasonable allocation and optimization of network resources and provides necessary routing and scheduling information for the transmission of holographic video streams.

[0126] The scheduling plane uses the CNS (Computing-Network-Storage Resource Tradeoff) model to optimize the allocation of computing, network, and storage resources by establishing new scheduling constraints and resource trading relationships. In addition, the scheduling plane uses an algorithm that combines the advantages of PPO (Proximal Policy Optimization) and the greedy algorithm, namely the HPG-RT (Hybrid PPO-Greedy Resource Trading) algorithm, to maximize the number of scheduled PCV streams and improve resource utilization. Through the above scheduling mechanism, the real-time performance and stability of holographic video communication are ensured.

[0127] A processing flow chart of a method for deterministic transmission of point cloud video streams in an intelligent fusion identification network provided by an embodiment of the present invention is as shown in Figure 2 shown and includes the following processing steps;

[0128] Step S10: The scheduling plane collects the requirements of real-time holographic service flows and the distribution of resources such as computing, storage, and network.

[0129] Step S20: The scheduling plane allocates computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flow according to the collected information. The PPO algorithm and the greedy resource optimization algorithm are used in this part. The control plane sends the algorithm results to the data plane.

[0130] Step S30: After the configuration is completed in the data plane, the holographic service starts to execute.

[0131] Step S40: The scheduling plane captures the point cloud frames of the holographic service through a camera and converts the point cloud frames into a periodic holographic video PCV stream through the CTS model.

[0132] Step S50: The scheduling plane deploys the BCQF mechanism using a switch to achieve deterministic transmission of the PCV stream through the models of the sending queue, buffer queue, and receiving queue.

[0133] The above Step S20 includes: establishing a CNS model to optimize the allocation of computing, network, and storage resources using scheduling constraints and resource trading relationships, using the HPG-RT algorithm to handle complex scheduling scenarios through DRL (Deep Reinforcement Learning), generating an initial allocation strategy for computing and network resources, and then further optimizing resource allocation through the greedy resource optimization algorithm to improve scheduling efficiency and resource utilization.

[0134] To maximize the schedulable data streams and optimize the efficient utilization of resources, the present invention establishes a CNS model to study the balance among computing, network, and storage resources. The CNS model contains 7 constraint conditions and 3 resource trade-off relationships. The constraint conditions specifically include:

[0135] 1) Forwarding period and super-period size constraints. The definition of the period size is similar to that in TSN and DetNet. The minimum period size depends on the transmission delay of packets in a single queue between two adjacent nodes, and the maximum period size is the greatest common divisor (GCD) of the periods of all data streams, as shown in formula (13):

[0136]

[0137] where Q size is the maximum number of packets that a queue can accommodate, D proc is the maximum processing delay, D prog is the maximum propagation delay in the network. The preset value Q size depends on the capabilities of the switch and is greater than Npkt 。

[0138] As shown in formula (14), the forwarding cycle size should be within and , and it is also one of the common divisors of all cycles.

[0139]

[0140] As shown in formula (15), the hypercycle represents the number of forwarding cycles in a complete cycle, and its size should not be less than the least common multiple (LCM) of all data flow cycles.

[0141] HC net = LCM(S.periods) #(15)

[0142] Therefore, the forwarding cycle can be represented by c j or cycle j , where j ∈ [0, N Cnet −1] and

[0143] 2) Network resource constraint. The network resource constraint analyzes the occupancy status of network resources and specifies the forwarding cycle allocation of data flow s i . represents the status of data flow s on the jth forwarding cycle i on link e(m, n), and the specific formula is shown in formula (16):

[0144]

[0145] should satisfy the periodicity shown in formula (17) and the continuity shown in formula (18).

[0146]

[0147] Where:

[0148]

[0149] 3) Computation cycle size constraint. The computing node uses TDM to divide the computing time into multiple equal computing cycles. The filtering and compression delay should be an integer multiple of the computing cycle size. To simplify scheduling, as shown in formula (20), the computing cycle size should be an integer multiple of the forwarding cycle size:

[0150]

[0151] Usually, C com is set to Cnet , so the calculation period can also be represented by , where j ∈ [0, N C com −1] and

[0152] 4) Computational power constraint. When the sum of the computational powers occupied by the data streams shown in Equation (21) does not exceed the computational power of the edge computing node, multiple data streams can be calculated simultaneously in one calculation period.

[0153]

[0154] Among them, represents the state of s i in the calculation period . If s i is assigned to then g otherwise it is 0. is the computational power of the edge computing node, and s i .R com is the computational power allocated to s i .

[0155] 5) Computational delay constraint. The computational delay constraint analyzes whether the computational delays of the point cloud frames in the data stream are equal, and ensures that the computational delay is less than the period of the data stream under the condition of allocating sufficient computational power and calculation period. CTS offsets the computational fluctuation ΔT by introducing a calculation period, and the computational delays of point cloud filtering and compression are shown in Equation (22):

[0156]

[0157] Since the data sizes of the compressed point cloud frames are equal, the decompression delays of each frame are also equal. Therefore, the computational delay of the data stream s i is shown in Equation (23):

[0158]

[0159] Among them, the cheap amount of s i satisfies:

[0160] s i .off × C net ≤ s i .period #(24)

[0161] 6) Storage delay and resource constraint. The storage resource is used for the storage area, and the storage delay is set to buffer the stream until it is allowed to be sent to avoid stream conflicts. The storage delay of the stream s i in the switch SWm needs to satisfy Equation (25):

[0162]

[0163] Combined with formula (10), the range can be formalized as formula (26):

[0164]

[0165] To separate the PCV flow. The storage delay of any two flows stored in switch SWm needs to meet the condition shown in equation (27):

[0166]

[0167] 7) MTP delay and jitter constraint. The constraint requires that the MTP delay shown in equation (28) must be less than the set upper limit of MTP delay.

[0168]

[0169] At the same time, to avoid unstable phenomena in the HTC system, the jitter is constrained as shown in equation (29).

[0170] s i .J mtp ≤s i .period.#(29)

[0171] The resource trade - off relationship specifically includes:

[0172] 1) Trade - off between network resources and computing resources. The relationship between network resources and computing resources is shown in equation (30). The more computing resources are allocated, the less network resources are consumed, and vice versa. For example, providing more computing resources can provide higher computing power within the same computing cycle, enabling a higher compression ratio and thus reducing the required network resources.

[0173]

[0174] 2) Trade - off between network resources and storage resources. The relationship between network resources and storage resources is shown in formula (31). The storage cycle of a flow depends on the forwarding cycles of other flows. The less network resources are allocated to a flow, the greater the probability that the flow requires more storage cycles, and vice versa.

[0175]

[0176] 3) Trade - off between computing resources and storage resources. The relationship between computing resources and storage resources is shown in equation (32). The more computing cycles are consumed, the fewer storage cycles can be allocated, and vice versa.

[0177]

[0178] After completing the 7 constraint conditions and 3 resource trade-off relationships of the CNS model, define Sched(s i ) as whether the flow can be scheduled. The optimization goal of the CNS model is to maximize the number of successfully scheduled flows, as shown in Equation (33), and the constraint conditions are shown in Equation (34).

[0179]

[0180] Among them, S suc is the number of successfully scheduled flows.

[0181] By establishing the CNS model, the optimization goal is determined to be maxS suc . The present invention proposes a CNS resource trade-off algorithm based on hybrid PPO-Greedy (HPG-RT). The first step of the HPG-RT algorithm uses the PPO algorithm to calculate resources and network resources for each data flow scheduling. The second step further maximizes the number of successfully scheduled data flows max S suc based on the results of the PPO algorithm through the Greedy algorithm.

[0182] A PPO-based CNS model scheduling framework provided by an embodiment of the present invention is as Figure 6 shown. The substitution of the CNS model scheduling should extract actions and probabilities through the action network, and the agent generates rewards and the next state. The trajectory is saved in the trajectory memory. When the network needs to be updated, the trajectory will be sampled in small batches and enter the new actor network to generate new action probabilities. The ratio ρ t (θ) can be calculated. At the same time, through the evaluation network, V π (st t ) and V traget (st t+1 ) can be generated according to the input batch current state and batch next state. The advantage can be generated according to the scheduling situation, and the evaluation loss can be calculated by the mean square error (MSE) of the error between V π (st t ) and V traget (st t+1 ). After obtaining the action loss and the evaluation loss, the action network and the evaluation network are updated. The pseudo-code of the PPO-based CNS model is shown in Algorithm 1.

[0183]

[0184] The PPO algorithm establishes an action network for the agent and can obtain the forwarding period and computing period based on the network. However, the settings of the computing power and storage period of the flow have not yet reached the optimal solution. Therefore, the embodiments of the present invention propose a greedy-based optimization method to allocate more optimal CNC resources for each flow through resource trade-off, and generate a gating control list for BCQF while maximizing the number of scheduled flows.

[0185] The pseudocode of the greedy resource optimization algorithm is shown in Algorithm 2. In line 1, the action network can generate the effect of each flow using the forwarding period. In line 2, the remaining computing power of each edge switch can be obtained from the agent. In line 3, the algorithm iterates over the data stream j. In lines 4 and 5, the algorithm calculates the maximum allowable storage delay and the minimum required computing power In lines 6 to 7, the algorithm iterates over the computing power from the minimum to the larger value and then updates the required forwarding period. In lines 8 to 14, the algorithm iterates over the storage period. If all constraints are satisfied, the flow can be scheduled; otherwise, it cannot. In lines 15 to 16, the scheduling result of the flow is updated and recorded. In line 18, the algorithm generates a gating table based on the scheduling result.

[0186]

[0187]

[0188] The above step S50 includes: CTS captures PCFs through a 3D camera by filtering and compressing, converts the PCFs into periodic PCV streams, and uses time-division multiplexing technology to ensure that all PCFs are processed within the specified time, thereby achieving the consistency of computing delay and providing a basis for the determinism of network transmission.

[0189] The working process of a CTS provided by the embodiments of the present invention is as Figure 3 shown. The collected PCFs are in the PLY (Polygon File Format) file format. The format element type table is shown in Table 1. In this format, each vertex contains attributes such as (x, y, z, red, green, blue, alpha) without face information. Therefore, the data size of the PCF can be calculated by the number of vertices and the number of bytes occupied by a single vertex. The specific calculation is shown in Equation (1).

[0190] DS pcf = Numofvertices × 28Bytes + 334Bytes. #(1)

[0191] Among them, 28 bytes is the data size of a single vertex, and 334 bytes is the size of the PLY file header.

[0192] The goal of redundant vertex filtering is to shape the PCFs into the same data size. Since different PCV streams correspond to different user facial features, each user needs to filter redundant vertices specifically. In addition, the main information in real-time communication is concentrated in the user's facial area, and the filtering algorithm can preset the number of vertices to be retained for each user during system debugging.

[0193] Element types in Table 1 PLY file

[0194]

[0195]

[0196] PCC further reduces the data size of PCFs to reduce bandwidth requirements. After redundant vertex filtering, the data sizes of PCFs in the same PCV stream are equal, so that the compression and decompression times in the same stream are also equal. In the present invention, the Draco algorithm is used as the PCC algorithm, and the calculation formula of its compression ratio α is shown in Equation (2).

[0197]

[0198] The Draco algorithm provides five different compression levels (cl1 to cl5), and each level corresponds to a different compression ratio range. For example, cl1 corresponds to the compression ratio range (0, 0.2), cl2 corresponds to [0.2, 0.4), and so on.

[0199] As Figure 3 shown, during the process of vertex filtering and PCFs generation, calculation fluctuations still exist. Therefore, CTS assigns a specific calculation cycle for each stream and delays the sending time of PCFs to eliminate the influence of calculation fluctuations. Assume that the hardware frame rate of a stream is set to 25 FPS. Theoretically, the time interval between every two PCFs is 40 ms. However, due to calculation fluctuations, the actual time interval is 40 ms ± △T (△T is less than 40 ms), and the data sizes of PCF0 and PCF1 are also different. CTS shapes the data sizes of PCF0 and PCF1 into consistency through redundant vertex filtering and the PCC algorithm, and limits the upper bound of the calculation delay to include multiple calculation cycles to compensate for calculation fluctuations. For example, assume that the generation time of PCF0 is later than the ideal generation time, while the generation time of PCF1 is earlier than the ideal generation time. CTS assigns 3 calculation cycles for this stream, and each calculation cycle is 8 ms. Within these 3 calculation cycles, both PCF0 and PCF1 can complete the filtering and compression operations. Therefore, PCF0 should be sent in calculation cycle 3, and PCF1 should be sent in calculation cycle 8. The time interval between the two PCFs is 5 calculation cycles, which is equal to 40 ms.

[0200] For any stream si , the allocated computing period should be greater than the sum of the maximum filtering and compression delays, as shown in formula (3):

[0201]

[0202] The calculation method is as shown in formula (4):

[0203]

[0204] Among them, sample(vertices) is the number of sampled vertices, and τ vertex refers to the execution time required for filtering a vertex

[0205] The compression time is related to the compression algorithm, compression level, and PCFs data size. In this system, the compression time and decompression time are calculated as shown in formula (5):

[0206]

[0207] Among them, α is the compression ratio, and s i .R com is the computing resource allocated for the data stream s i

[0208] Through the above design, CTS can shape the captured user's PCFs into a PCFs stream with the same frame size and equal frame time intervals, thereby converting the PCV stream into a periodic stream. The period of the stream and the size of the compressed frame data are as shown in formula (6):

[0209]

[0210] The above step S60 includes: The BCQF network transmission mechanism manages the data stream through the model of the sending queue SQ (Sending Queue), buffering queue BQ (Buffering Queue), and receiving queue RQ (Receiving Queue), ensuring the deterministic transmission of PCFs. Through special queue conversion operations, BCQF avoids packet loss and delay during transmission, thus ensuring the real-time and stability of the holographic video stream.

[0211] ​To address the situation where the PCV flow lacks strict periodicity, leading to non-deterministic transmission, the present invention filters excessive vertices in PCFs based on computed CTS to maintain a constant data size, and then compresses the PCFs using the PCC algorithm to reduce the network bandwidth requirements. During the filtering and compression process, CTS divides the computation time into multiple cycles and employs time-division multiplexing technology to allocate specific time slices for computation, ensuring that the computation latency of all PCFs in the same PCV flow is within an upper bound range.

[0212] Since a PCF is different from a data packet and consists of frames composed of multiple MTUs (Maximum Transmission Unit), traditional traffic shaping mechanisms cannot achieve deterministic transmission of PCFs. To achieve deterministic transmission at the frame granularity, the present invention proposes a BCQF mechanism that includes a flow storage area, a three-queue model, and special queue switching operations.

[0213] In BCQF, time is divided into forwarding cycles C net , and the number of data packets that can be forwarded in each forwarding cycle is shown in Equation (7):

[0214]

[0215] where BW is the link bandwidth, GD is the protection bandwidth within a forwarding cycle, MTU represents the maximum transmission unit, and IFG represents the inter-frame gap. Therefore, it can be concluded that the number of forwarding cycles s i occupied by the PCFs of the compressed flow s i .FC net is as shown in Equation (8):

[0216]

[0217] According to Equation (8), it can be seen that some PCFs require multiple forwarding cycles to complete transmission. The BCQF model is as shown in Figure 3 , and BCQF focuses on the deterministic transmission of frames, achieving the allocation of sufficient forwarding cycles for each flow while avoiding conflicts between flows.

[0218] A BCQF model diagram provided by an embodiment of the present invention is as shown in Figure 4 . BCQF consists of a storage area for each PCV flow, a three-queue model, and a selection transmission gating. When a data packet enters a switch supporting BCQF, it will be directed to a specified port, and only the data packets in the PCV flow can enter the flow storage area. The flow storage area sets a buffer and a corresponding selection enqueue gating for each PCV flow. The enqueueing of data packets is managed by the gating list of the selection enqueue gating, and the data packets can only enter the three-queue model within the allowed cycles.

[0219] The three-queue model uses a transmission queue to send buffered data packets, and a buffer queue and a reception queue to receive and buffer data packets. The three queues will sequentially switch roles from the transmission queue to the buffer queue and then to the reception queue. At the same time, there are always two queues receiving data packets, so it is necessary to clarify the reception order of the reception queue and the buffer queue. As Figure 4 shown, flow s1 requires 3 cycles to complete data packet forwarding, and its frame arrives at SWn in cycle 1. In cycle 1, only 0.6 cycles of data packets arrive, and the data packets for the remaining 2.4 cycles will arrive in subsequent cycles. The 0.6-cycle data packets in cycle 1 are the first data packets of flow s1 and need to enter the reception queue. In subsequent cycles, the 1-cycle data packets that arrive need to enter the buffer queue (the RQ of the previous cycle) first. Only when the buffer queue is full will the remaining data packets enter the reception queue. Selective transmission gating sets gating after each queue, and data packets can only be sent when the gating state is enabled. With the help of forwarding gating, the three queues in the three-queue model can switch roles. For other low-priority queues, the gating state is always enabled to forward non-PCV flow data when the transmission queue is empty.

[0220] Figure 5 This is a schematic diagram of data packet group forwarding between switches provided by an embodiment of the present invention. Different from the deterministic transmission of single data packets in DetNet, PCF is a frame composed of thousands of data packets, and it is difficult to control its single-hop transmission within one cycle using BCQF. When a PCF arrives at a switch, there may be three situations as Figure 5 shown. Therefore, BCQF introduces a buffer queue to buffer the data packets in the PCF and sends the PCF through a scheduling cycle to avoid flow conflicts.

[0221] Flow conflicts may also occur when multiple PCV flows are forwarded in the network. To solve this problem, the present invention proposes a flow storage area. As Figure 5 shown, the PCF of flow s2 arrives at SWn from cycle 2 to cycle 3, but since flow s1 is still transmitting, flow s2 cannot enter the three-queue model. After the reception queue becomes idle in cycle 4, flow s2 enters the reception queue for storage. By setting the storage cycle for each flow, flow conflicts can be avoided. The storage cycle of flow s i on switch SWn is defined as in Equation (9):

[0222]

[0223] where represents the set of flows passing through switch SWn.

[0224] Therefore, flow s iThe upper and lower bounds of the network latency of a data packet in BCQF are shown in Equations (10) and (11) as follows:

[0225]

[0226] s i .path is shown in Equation (12) as follows

[0227] s i .path = {sw0, …, sw m , …, sw M}#(12)

[0228] In summary, the DT-CNST transmission mechanism designed in the embodiments of the present invention can ensure the deterministic and low-latency transmission of PCV streams, meet the strict requirements of real-time communication, and solve the problems faced in the field of real-time holographic communication. CTS ensures the periodicity of PCV streams through its periodic generation and processing, laying a foundation for the consistency of network transmission. The BCQF mechanism realizes deterministic transmission at the frame granularity, further improving the reliability and efficiency of data transmission. The CNS model optimizes the resource allocation strategy by introducing seven innovative scheduling constraints and three resource trade-off relationships, thus maximizing the number of schedulable PCV streams. The introduced HPG-RT provides intelligent scheduling for the calculation and forwarding cycles of each PCV stream, and realizes the optimal allocation of resources through the combination of deep reinforcement learning and greedy algorithm.

[0229] The DT-CNST mechanism of the present invention provides an efficient and reliable transmission solution for real-time holographic communication through a series of innovative technologies, significantly improving the transmission quality of holographic video streams and the user experience.

[0230] The present invention establishes a CNS model and an HPG-RT algorithm, combines the advantages of PPO and Greedy algorithms, first generates the calculation and forwarding cycles for each PCV stream, and then uses the Greedy algorithm to optimize the actual CNS resource allocation to maximize the number of schedulable PCV streams, thereby effectively reducing the algorithm complexity and improving the scheduling efficiency.

[0231] The present invention proposes a computation-based traffic shaper, which converts PCV streams into periodic streams by filtering redundant vertices and applying point cloud compression technology. At the same time, time-division multiplexing technology is used to divide the computation time into multiple computation cycles, ensuring the periodicity of PCV streams and providing a basis for achieving bounded computation latency and deterministic transmission.

[0232] The present invention proposes a buffered circular queue and a forwarding mechanism. Through a three-queue model of a sending queue (SQ), a buffered queue (BQ), and a receiving queue (RQ), deterministic transmission at the frame granularity is achieved. The BCQF mechanism divides the sending time into forwarding cycles through TDM technology and introduces a storage area to avoid flow conflicts, ensuring deterministic transmission of the PCF in the network, thereby improving the performance and user experience of the holographic communication system.

[0233] Those of ordinary skill in the art can understand that the drawings are only schematic diagrams of an embodiment, and the modules or processes in the drawings are not necessarily essential for implementing the present invention.

[0234] From the description of the above embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus a necessary general hardware platform. Based on such an understanding, the technical solution of the present invention, in essence, or the part that makes contributions to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present invention.

[0235] Each embodiment in this specification is described in a progressive manner. The same or similar parts among the embodiments can be referred to each other, and the key points of each embodiment are the differences from other embodiments. In particular, for the apparatus or system embodiments, since they are basically similar to the method embodiments, they are described relatively simply. For the relevant parts, reference can be made to the partial description of the method embodiments. The apparatus and system embodiments described above are only illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they can be located in one place, or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0236] The above is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present invention should be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A deterministic transmission method for point cloud video streams in an intelligent fusion identification network, characterized in that Set up a deterministic transmission DT-CNST architecture in the Zhirong Identification Network. The DT-CNST architecture includes a data plane, a control plane, and a scheduling plane. The method includes: The scheduling plane collects the demand information of the holographic service flows in the Zhirong Identification Network, obtains the distribution of computing, storage, and network resources in the Zhirong Identification Network, and transmits the collected information to the control plane; The control plane, through the traffic shaper CTS model, optimizes the PPO-greedy resource trading algorithm according to the collected information through a hybrid probability strategy to allocate computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flows, and transmits the resource allocation results to the data plane; After the configuration is completed, the data plane executes the holographic service according to the resource allocation results; The scheduling plane captures the point cloud frames of the holographic service through a camera, converts the point cloud frames into a periodic holographic video PCV stream through the CTS model, and the scheduling plane uses the switch to deploy the BCQF mechanism to achieve deterministic transmission of the PCV stream through the models of the sending queue, buffer queue, and receiving queue.

2. The method according to claim 1, wherein The control plane, through CTS, optimizes the PPO-greedy resource trading algorithm according to the collected information through a hybrid probability strategy to allocate computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flows, and transmits the resource allocation results to the data plane, including: Construct a CNS model, set the constraint conditions and resource trade-off relationships of the CNS model. The CNS model allocates computing time slots, computing power, forwarding time slots, and storage resources for the holographic service flows through the HPG-RT algorithm according to the collected information, the constraint conditions, and the resource trade-off relationships, obtains the preliminary resource allocation results, optimizes the preliminary resource allocation results through the PPO-greedy resource trading algorithm, obtains the resource allocation results of the holographic service flows, and transmits the resource allocation results to the data plane.

3. The method according to claim 2, wherein The constructing of the CNS model and setting the constraint conditions and resource trade-off relationships of the CNS model include: Construct a CNS model, and set that the CNS model contains 7 constraint conditions and 3 resource trade-off relationships. The constraint conditions specifically include: 1) Forwarding period and super-period size constraints. The minimum period size depends on the transmission delay of the data packets in a single queue between two adjacent nodes, and the maximum period size is the greatest common divisor of the periods of all data streams, as shown in formula (13): Among them, Q size is the maximum number of data packets that a queue can accommodate, D proc is the maximum processing delay, D prog is the maximum propagation delay in the network, and the preset value Q size depends on the capabilities of the switch and is greater than N pkt , is the minimum period, is the maximum period, N pkt represents the maximum number of data packets allowed to be transmitted within a single time slot; As shown in formula (14), the forwarding period C net is in the range of and and is also one of the common divisors of all periods; represents a positive integer and is a mathematical symbol; As shown in formula (15), the hypercycle HC net represents the number of forwarding cycles in a complete cycle loop, and its size should not be less than the least common multiple of the periods of all data streams; HC net = LCM(S.periods)#(15) where \(j\in[0,N_C net - 1]\) and N_C net , \(N_C net is the number of periods in a hyper - period, \(S\) refers to the set of flows, and \(S.\text{periods}\) refers to the set of periods of all flows; 2) Network resource constraint, Indicates the j-th forwarding cycle The upstream data flow s i The state on the link e(m,n), and the specific formula is shown in Equation (16): It should satisfy the periodicity shown in Equation (17) and the continuity shown in Equation (18); Where: e(m,n) refers to the link between node v m and node v n , and s i represents the i-th PCV flow, and s i .FC net represents the number of forwarding cycles occupied by flow s i , and and are both network cycles, representing the k-th and j-th cycles; 3) Calculate the cycle size constraint. As shown in Equation (20), calculate the cycle C com whose size is an integer multiple of the forwarding cycle size: 4) Computing power constraint. As shown in formula (21), the sum of the computing power occupied by the data streams does not exceed the computing power of the edge computing node; Among them, represents s i in the calculation period the state, if s i is assigned to then g otherwise 0, is the computing power of the edge computing node, s i .R com is the computing power assigned to s i ; 5) Computing delay constraint. The computing delay of point cloud filtering and compression is as shown in formula (22): Indicates the number of computing cycles required for the filtering and compression of flow si, including the filtering delay and the compression delay, C com is the computing cycle size, Indicates the filtering delay of flow si, Indicates the compression delay of flow si, and ΔT represents the maximum variance of the processing and computing delay of flow si; Data stream s i The computational latency is shown in Equation (23): where s i has an offset that satisfies: s i .off×C net ≤s i .period#(24) L com Indicates the overall computational delay of the stream Indicates the decompression delay, s i .off indicates the number of offset network cycles of stream si, which means that the data will wait for off network cycle times before being sent after processing; 6) Storage delay and resource constraints, flow s i The storage delay in switch SWm needs to satisfy formula (25); Combined with formula (10), the range refers to the range of the storage time of the flow si in the switch, and the range is formalized as formula (26): Indicates the traffic set passing through switch sw m ​ Indicates the time that the flow si needs to be stored in the switch; For any two flows stored in the switch SWm, the storage delay needs to meet the condition shown in formula (27): 7) MTP delay and jitter constraint. The constraint requires that the MTP delay shown in formula (28) must be less than the set MTP delay upper limit; Indicates the maximum motion-to-imaging MTP delay of stream si ddl mtp Indicates the maximum allowable MTP delay of stream si The jitter is constrained as shown in formula (29); s i .·J mtp ≤s i .period.#(29) s i .J mtp Indicates the maximum motion of stream si to the imaging MTP jitter The specific resource trade - off relationships include: 1) The trade - off between network resources and computing resources. The relationship between network resources and computing resources is shown in Equation (30); R com Network resources allocated to flow si The compression ratio of the α compression algorithm 2) The trade - off between network resources and storage resources. The relationship between network resources and storage resources is shown in Equation (31); 3) The trade - off between computing resources and storage resources. The relationship between computing resources and storage resources is shown in Equation (32); 4. The method according to claim 3, characterized in that The described CNS model allocates computing time slots, computing power, forwarding time slots, and storage resources for holographic service flows through the HPG - RT algorithm according to the collected information, the described constraints, and the resource trade - off relationships, obtains a preliminary resource allocation result, and optimizes the preliminary resource allocation result through the PPO - greedy resource trading algorithm to obtain the resource allocation result of the holographic service flow, including: Based on the constraint conditions and the resource trade-off relationship of the CNS model, define Sched(s i ) as whether the flow can be scheduled. The objective function of the CNS model is shown in Equation (33), and the constraint conditions are shown in Equation (34); Among them, S suc is the number of successfully scheduled flows; The HPG-RT algorithm for CNS resource trade-off based on the hybrid PPO-Greedy is adopted to solve the objective functions shown in Eqs. (33) and (34), and the number of successfully scheduled data streams maxS is obtained. suc In the first step of the HPG-RT algorithm, the PPO algorithm is used to schedule computing resources and network resources for each data stream. The PPO algorithm generates rewards and the next state through an agent, and the trajectory is saved in the trajectory memory. When the network needs to be updated, the trajectory will be sampled in batches and enter a new actor network to generate new action probabilities, and the ratio ρ t (θ) is calculated. Then, the evaluation network generates V π (st t ) and V traget (st t+1 ) according to the input batch of the current state and the batch of the next state. The error between V π (st t ) and V traget (st t+1 ) is calculated through the mean square error MSE, and the evaluation loss is obtained according to this error. After obtaining the action loss and the evaluation loss, the action network and the evaluation network are updated. The second step of the HPG-RT algorithm further maximizes the number of successfully scheduled data streams maxS based on the results of the PPO algorithm through the PPO-greedy resource trading algorithm suc , allocates more optimal computing power and storage cycle setting resources for each stream through resource trade-off, generates a gating control list for BCQF while maximizing the number of scheduled streams, and obtains the resource allocation result for the holographic service stream by integrating the calculation results of the first and second steps of the HPG-RT algorithm.

5. The method according to any one of claims 1 to 4, characterized in that The described scheduling plane captures the point cloud frames of holographic services through a camera and converts the point cloud frames into a periodic holographic video PCV flow through the CTS model, including: CTS stores the collected PCFs in the polygon file format PLY, and calculates the data size of the PCF according to the number of vertices and the number of bytes occupied by a single vertex in the PLY format. The specific calculation is shown in Equation (1); DS pcf = Numofvertices × 28 Bytes + 334 Bytes. #(1) Among them, 28 bytes is the data size of a single vertex, and 334 bytes is the size of the PLY file header; The PCFs are shaped into the same data size through the redundant vertex filtering Draco algorithm. The calculation formula for the compression ratio α of the redundant vertex filtering algorithm is shown in Equation (2); For any flow s, the computing period allocated by the Draco algorithm i , should be greater than the sum of the maximum filtering and compression delays, as shown in Equation (3): Indicates the number of computing cycles required for the filtering and compression of stream si, including the filtering delay and the compression delay, C com is the computing cycle size, Indicates the filtering delay of stream si, Indicates the compression delay of stream si, and ΔT represents the maximum variance of the processing and computing delay of stream si; The calculation method is as shown in formula (4): Indicates the filtering delay of stream si, max(vertices) represents the maximum number of vertices in the point cloud frame file of stream si, and sample(vertices) represents the number of vertices to be sampled from the point cloud frame file of stream si; where sample(vertices) is the number of sampled vertices, and τ vertex refers to the execution time required to filter a vertex; Compression time using the Draco algorithm And decompression time The calculation formula is as shown in Equation (5): where α is the compression ratio, s i .R com is the computing resource allocated for the data stream s i , compression computing delay, decompression computing delay, For the filtered point cloud frame compress it using the Draco algorithm, set the compression ratio to α, and allocate the computing resource as s i .R com , For the compressed point cloud frame decompress it using the Draco algorithm, and allocate the computing resource as s i .R com ; Through the above design, CTS shapes the PCFs into a PCFs flow with the same frame size and equal frame time intervals. The period and compressed frame data size of the flow are shown in Equation (6): s i .The period of stream si, s i .fps hardware .The frame rate of the hardware device that generates stream si .The size of the compressed point cloud frame in stream si, α compression rate, DS pfs .(sample(vertices)) The size of the number of sampled vertices 6. The method according to claim 5, wherein The described scheduling plane uses the switch to deploy the BCQF mechanism to achieve deterministic transmission of the PCV flow through the models of the send queue, buffer queue, and receive queue, including: The BCQF network transmission mechanism manages data streams through the models of a sending queue, a buffer queue, and a receiving queue, and divides time into multiple forwarding cycles C net , and the number of data packets that can be forwarded in each forwarding cycle is shown in Equation (7): where BW is the link bandwidth, GD is the protection bandwidth within a forwarding cycle, MTU represents the maximum transmission unit, and IFG represents the inter-frame gap, and the compressed stream s is obtained i The forwarding cycles occupied by the PCFs of i .FC net As shown in Equation (8): When a data packet enters a switch supporting BCQF, it is directed to a specified port. Only the data packets in the PCV flow can enter the flow storage area. The flow storage area sets a buffer and corresponding selective enqueue gating for each PCV flow. The enqueueing of data packets is managed by the gating list of the selective enqueue gating. Only within the allowed period can the data packets enter the send queue, buffer queue, or receive queue; Buffered data packets are sent using a transmission queue. A reception queue and a buffer queue are used for receiving and buffering data packets. Only when the buffer queue is full will the remaining data packets enter the reception queue. Selective transmission gating sets a gate after each queue. With the help of forwarding gating, the transmission queue, buffer queue, or reception queue can switch roles. Flow s i Storage cycle on switch SWn Define as Equation (9): Among them represents the flow set through switch sw n ; Flow s i The upper and lower bounds of the network latency of a data packet in BCQF are shown in Equations (10) and (11) as follows: s i .path is as shown in Equation (12) s i .path = {sw0,…,sw m ,…,sw M}#(12).

Images