Online SVC Multicast Method Based on UAV Relay in NOMA Network

Abstract

This invention proposes an online SVC multicast method based on UAV relay in NOMA networks, aiming to maximize video reception quality at the base station edge. The method first constructs a visual graph model to characterize the coupling between UAV deployment and multicast group association. Based on this model, the video reception quality maximization problem is modeled as a clique-based nonlinear integer programming problem, decoupling the UAV-multicast group association and subchannel allocation subproblems. The former is transformed into a maximum weight clique problem with a vertex limit, and an improved maximum weight clique algorithm based on branch and bound is used to determine the UAV-multicast group association mode. For the latter, a rule-matching strategy is adopted to obtain the optimal resource allocation strategy with less computational cost. Simulation results show that this invention outperforms existing benchmark schemes in terms of aggregated peak signal-to-noise ratio (PSRN), spectral efficiency, and adaptability.

Description

Technical Field

[0001] This invention belongs to the field of communication network technology, specifically an online SVC multicast method based on UAV relay in NOMA networks. Background Technology

[0002] With the development of network technology, real-time video services (such as video conferencing, live sports broadcasts, etc.) have been integrated into people's lives. Omdia pointed out in its report "Network Traffic Forecast: 2019–24" that by 2025, video is expected to account for more than 75% of the total traffic of wireless networks. The surge in real-time video traffic has put enormous pressure on network resource allocation. Compared with unicast, multicast[1] does not use bandwidth resources that are limited by the number of access users.

[0003] Scalable video coding (SVC)[2][3] encodes video into a base layer and multiple enhancement layers. The device can adjust the number of decoding layers and reconstruct the complete video according to the network environment and decoding capabilities. Due to its flexibility and adaptability, SVC multicast has become a promising video multicast quality enhancement scheme. Each SVC video layer under Orthogonal Multiple Access (OMA) is transmitted on a different orthogonal channel. Through power domain multiplexing, Non-Orthogonal Multiple Access (NOMA)[4] can provide services to multiple terminals on the same channel. When the transmitter sends different video layers in a non-orthogonal manner, the receiver can run Successive Interference Cancellations (SIC)[5] to demodulate the signal according to the strength of the received signal power.

[0004] In the edge area of ​​a base station, the communication link between the user and the base station is generally a non-line-of-sight communication link. Relying solely on the base station for multicast is insufficient to guarantee that edge users receive high-quality video. Due to their mobility, drones can be temporarily deployed to the edge area covered by the base station, and their flight altitude and line-of-sight link [6] help reduce the resource consumption of video transmission. By combining NOMA and SVC, drones deployed in the edge area of ​​a base station can share the spectrum with the base station and perform signal superposition in the power domain, thereby improving the fairness and resource utilization of the multicast service at the edge of the base station.

[0005] Many challenges remain to be addressed in using drones to improve video quality at the edge of base stations:

[0006] (1) Drone-Base Station Collaboration

[0007] Most existing drone video distribution is cache-enabled, that is, it provides offline services to ground users by carrying a cache

[15]

[16] . The video cache needs to be updated regularly to improve the cache hit rate

[10] . To support online services, one approach is to establish a link to the base station via mmWave [7][8]. However, the normal communication distance of mmWave is 150 meters

[17] , which is much smaller than the coverage radius of macro base stations, making it difficult for drones at the edge of the base station to establish a stable connection. Most existing layered multicast schemes are geared towards ground networks. Some researchers have explored a NOMA-enabled SVC multicast scheme in which the ground base station sends the base layer and enhancement layer [2]. Under this framework, some researchers have proposed a joint power allocation and sub-packet scheme to maximize the aggregated multicast rate while satisfying power and rate constraints [9].

[0008] (2) Layered video decoding in multi-drone scenarios.

[0009] Macro base stations transmit the base layer and small base stations transmit the enhancement layer, which is a common layered video multicast strategy

[18] . The base layer has higher requirements for transmission rate than the enhancement layer, which means that macro base stations need to occupy more bandwidth resources to transmit the base layer. UAVs can transmit the base layer with less resources thanks to line-of-sight links. After receiving the base layer and enhancement layer, user equipment decodes them in order of power strength and eliminates interference. UAV deployment must take into account the receiving order of the video layer and the signal receiving rate of the user end, but existing work has not addressed these issues. Some researchers have studied a full-duplex NOMA system with multiple UAVs working together, which improves the aggregate throughput of the system through dynamic user clustering, UAV placement and power allocation

[12] . Other researchers have proposed a joint optimization strategy of power allocation and UAV trajectory to improve the user receiving rate

[13] .

[0010] (3) Selection of association between drones and multicast groups.

[0011] Associating each multicast group with a unique drone can avoid co-channel interference, but it cannot unleash the advantages of drones. Allowing a multicast group to associate with multiple drones helps improve service flexibility, but the interference caused by drone signal superposition may offset the benefits of spectrum reuse. Most existing works consider unicast applications. Reference

[10] studies the SVC hierarchical caching scheme in multi-drone networks, minimizing user access latency through the joint design of hierarchical cache placement, drone deployment, and user association. Reference

[11] studies uplink transmission in drone-assisted cellular networks, and designs a multi-agent Q-learning algorithm to determine drone deployment and association schemes, minimizing the power consumption of users and drones as much as possible. Summary of the Invention

[0012] Consider a scenario where multiple drones act as relays to provide online video services to users in the edge area of ​​a base station. For this scenario, this invention designs a NOMA-based air-to-ground cooperative SVC video multicast framework to maximize the aggregated video reception quality of multicast groups in the base station edge area. The main technical contributions include:

[0013] First, a visual graph model is constructed to characterize the coupling relationship between decision variables (drone deployment, multicast group association) and video decoding (overlay coding and SIC). The decision corresponding to each clique in the graph satisfies the requirements of overlay coding and SIC.

[0014] Second, based on the graphical model of the design, the problem of maximizing multicast group video reception quality is transformed into a clique-based spectrum allocation problem, which belongs to an integer linear programming problem. For ease of processing, the joint optimization problem is decoupled into two sub-problems: UAV deployment-association optimization and spectrum allocation.

[0015] Second, the first subproblem is transformed into a special maximum weight clique problem, which is solved using an improved branch-and-bound algorithm. For the second subproblem, the solution strategy is dynamically selected based on the number of multicast groups and sub-channels.

[0016] Simulation results show that the proposed scheme outperforms typical benchmark schemes in terms of aggregated PSNR and spectral efficiency. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of an online SVC video multicast scenario using drone relay;

[0018] Figure 2 This is a schematic diagram of power domain superposition and spectrum division;

[0019] Figure 3 This is a schematic diagram illustrating the impact of drone deployment on interference cancellation;

[0020] Figures 4(a) and 4(b) are schematic diagrams of signal superposition and interference cancellation under different correlation modes, wherein:

[0021] Figure 4(a) shows the association pattern between a multicast group and a single drone.

[0022] Figure 4(b) shows the association mode of multicast groups and multiple drones;

[0023] Figure 5 It is a graph model about drone deployment-association patterns.

[0024] Figures 6(a) to 6(d) These are groups of different sizes, including:

[0025] Figure 6(a) shows all the cliques with a single vertex.

[0026] Figure 6(b) shows all the cliques with two vertices.

[0027] Figure 6(c) shows all three-vertex cliques.

[0028] Figure 6(d) shows all the cliques with four vertices;

[0029] Figures 7(a) to 7(c) This represents the total PSNR value for different numbers of drones (multicast group number N=4), where:

[0030] Figure 7(a) shows three drones.

[0031] Figure 7(b) shows four drones.

[0032] Figure 7(c) shows five drones;

[0033] Figures 8(a) to 8(c) This represents the average PSNR value under different numbers of multicast groups (number of drones α = 6), where:

[0034] Figure 8(a) shows three multicast groups.

[0035] Figure 8(b) shows four multicast groups.

[0036] Figure 8(c) shows five multicast groups;

[0037] Figures 9(a) to 9(b) This refers to the drone deployment and multicast group association in Proposed-2 (number of drones α = 6, number of multicast groups N = 5), where:

[0038] Figure 9(a) shows the location where the drone was placed.

[0039] Figure 9(b) shows the association patterns of UAV 1, UAV 6 and the multicast group;

[0040] Figures 10(a) to 10(b) This refers to the drone deployment and multicast group association in Proposed-1 (number of drones α = 6, number of multicast groups N = 5), where:

[0041] Figure 10(a) shows the location where the drone was placed.

[0042] Figure 10(b) shows the association patterns of UAV 1, UAV 6 and the multicast group. Detailed Implementation

[0043] 1 Overview

[0044] This invention studies an online SVC video multicast method based on UAV relay in Non-Orthogonal Multiple Access (NOMA) radio access networks, aiming to maximize video reception quality at the base station edge. The UAV carries communication equipment as an aerial base station, and is simply referred to as a "UAV".

[0045] This invention constructs a visual graph model to characterize the coupling between UAV deployment and multicast group association. Based on this model, the problem of maximizing video reception quality is modeled as a clique-based nonlinear integer programming problem, which is solved by coupling two subproblems: UAV-multicast group association (Problem 1) and subchannel allocation (Problem 2).

[0046] Problem 1 is transformed into a maximum weight clique problem with a limited number of vertices. An improved maximum weight clique algorithm based on branch and bound is developed to determine the UAV-multicast group association pattern.

[0047] For problem 2, design a rule matching strategy to obtain the optimal resource allocation strategy with less computational cost.

[0048] Simulation results show that the proposed scheme outperforms the benchmark scheme in terms of aggregated peak signal-to-noise ratio (PSRN), spectral efficiency, and adaptability.

[0049] 2 System Model

[0050] like Figure 1 As shown, multiple drones are deployed as relays at the coverage edge of the macro base station. There are three different types of links: BS-to-UAV (B2U) links, BS-to-Device (B2D) links, and UAV-to-Device (U2D) links. Users requesting the same video stream belong to a multicast group. A multicast group can be associated with one or more drones; that is, multiple drones can provide services to users within a multicast group. Figure 1 For example, multicast group 1 is associated with drone 2, and multicast group 2 is associated with both drone 1 and drone 2. The Mobile Edge Computing (MEC) controller determines drone placement (i.e., deployment location) and its association with multicast groups by accessing global information, and allocates spectrum resources to each multicast group. Key symbols are listed in Table 1.

[0051] Table 1 Important Symbols

[0052] Table 1 Important symbols

[0053]

[0054] Based on Scalable Video Coding (SVC), each video is encoded into a base layer and an enhancement layer. Each multicast group is assigned a different orthogonal sub-channel. Figure 2 For example, suppose Figure 1 Multicast group 1 was allocated 4 sub-channels (sub-channel 1 to sub-channel 4), and multicast group 2 was allocated 5 sub-channels (sub-channel 5 to sub-channel 9). The corresponding spectrum resources correspond to the shaded and unshaded parts.

[0055] The former (shaded area) represents the spectrum resources used by B2U communication, that is, the spectrum resources occupied by the base station transmitting video to the drone (base layer relay).

[0056] The latter (the unshaded portion) represents the spectrum resources shared by U2D communication (drones sending the base layer of video to users via U2D links (base layer broadcast)) and B2D communication (base stations sending the enhancement layer of video to users via B2D links (enhancement layer broadcast)).

[0057] 2.1 Communication model

[0058] refer to Figure 3 Let l j,k =(x j ,y j ,z j,k ) represents a candidate drone deployment location, where For the index of the projection position on the XY plane, It is the height index (z) corresponding to j. j,k (Height). Placed at l j,k The line-of-sight communication probability between the UAV and ground device i is defined as

[14] .

[0059]

[0060] in Representative in l j,k The horizontal distance from the drone to user i (i.e., the user's ground equipment). o1 and o2 are constants determined by the environment. j,k The path loss model from the drone to ground device i is defined as

[0061]

[0062] Where c1 is the bandwidth frequency, and c2 is the speed of light. η LoS and η NLoS These represent additional path loss models for line-of-sight and non-line-of-sight links, respectively.

[0063] The azimuth and elevation angles of the directional antenna equipped on the UAV are denoted as θ1 and θ2, respectively. It is assumed that the half-power beamwidths of the antenna at the azimuth and elevation angles are equal. The antenna gain at azimuth (θ1, θ2) is

[19]

[0064]

[0065] in G0 represents the antenna gain outside the beamwidth of the directional antenna. This is because in practical applications... The antenna gain is simplified to G0 = 0

[20] .

[0066] Based on equations (2) and (3), in l j,k The channel gain from the drone to the ground device i within the coverage area is From base station m to ground equipment The channel gain is calculated as g m,i =30+35lg(h) m,i

[21] , of which This represents the horizontal distance from base station m to ground device i.

[0067] 2.2 UAV Deployment Model

[0068] The receiving end performs interference cancellation sequentially according to the power of the received signal. Figure 3 Multicast group 1 is associated with both drone 1 and drone 2. During the reception of signals from drone 1 and the base station, users covered by drone 1 may experience interference from the signal from drone 2. The following explanation, using users covered by drone 1 as an example, illustrates how to coordinate the placement of adjacent drones to ensure these users can correctly and sequentially decode the signals from drone 1 and the base station.

[0069] Constraints for decoding drone signals:

[0070] exist Figure 3 In the diagram, point A at the edge of drone 1's coverage area is the closest point to base station m. Assume drone 1 is deployed at location l. j,k Drone 2 was deployed in l j′,k′ The signal gain from drone 1 to point A is the signal gain at its minimum to the coverage edge, and is expressed as...

[0071]

[0072] R j,k Indicates deployment at location l j,k The drone 1 covers the ground radius;

[0073] The coordinates of the base station m are (x m ,y m ,z mThe channel gain from macro base station m to point A is the maximum signal gain to the coverage edge of UAV 1, expressed as...

[0074]

[0075] in, From macro base station to drone 1 (l j,k The horizontal distance is γ, where γ is the signal path fading index. s and p m These represent the transmission power of the drone and the base station, respectively. If the signal power received at point A from the base station is less than the signal power from drone 1, then the signal power from the base station within the coverage area of ​​drone 1 will be less than the signal power from drone 1. Only when condition (4) is met can users covered by drone 1 decode / reconstruct the signal from the drone and perform interference cancellation.

[0076]

[0077] Constraints for decoding base station signals:

[0078] Because of spectrum reuse, users covered by drone 1 will be subject to interference from drone 2. Figure 3 In the diagram, point B at the edge of UAV 1's coverage area is the closest point to UAV 2 within the coverage area. The channel gain from UAV 2 to point B is the maximum channel gain to the coverage area of ​​UAV 1, denoted as...

[0079]

[0080] in This represents the horizontal distance between the two drones. Point C, at the edge of drone 1's coverage area, is the point farthest from the macro base station within its coverage range. The channel gain from the macro base station to point C is the minimum channel gain to the coverage area of ​​drone 1, denoted as...

[0081]

[0082] If the interference signal power received at point B from UAV 2 is less than the signal power received at point C from the base station, then all interference signals within the coverage area of ​​UAV 1 are less than the base station signal. Therefore, for UAV 1... j,k and l j′,k′ For the two drones, only if inequality (5) holds can the user under the coverage of drone 1 decode and reconstruct the signal from the base station.

[0083]

[0084] 2.4 Video Layer Decoding

[0085] Constraints (4) and (5) ensure that users covered by a drone can receive both the base layer and the enhancement layer signals from the drone and the base station. Placed in l j,k The set of users in the multicast group n covered by the drone is represented as Binary variable q j,k,n =1 indicates that the multicast group n is deployed in l j,k The drone is associated, otherwise q j,k,n =0. Receive l j,k The signal sent by the drone, user It will be subject to interference from base stations and other drones associated with multicast group n. Assume each sub-channel has a bandwidth of e, and the number of sub-channels allocated to multicast group n is b. n User (Ground Equipment) Decoding from position l j,k The reachability rate of the drone signal is expressed as a function of... and b n The function, i.e.

[0086]

[0087] Where σ 2 This represents the average background noise power.

[0088] Constraint (5) guarantees that, except for position l j,k Besides the drone itself, the interference signal strength from other drones is lower than the signal strength from base station m. Users decoding signals from the base station will experience interference from signals from other distant drones. (This is related to ground equipment.) The achievable rate of decoding base station signals is about and b n The function, i.e.

[0089]

[0090] When multicast group n requests video layer a, the minimum bit rate that the ground equipment within the group supports for normal decoding is λ. a,n The conditions under which the base layer (a=1) and enhancement layer (a=2) can be correctly received and decoded are as follows:

[0091]

[0092] and

[0093]

[0094] The prerequisite for the enhancement layer to be correctly received / decoded and reconstructed with the base layer is that the base layer is correctly received. Only when both (8) and (9) are true can the receiver decode the enhancement layer.

[0095] Next, the superposition coding of UAV and base station signals in the power domain and interference cancellation are explained in (4) and (5). Figure 1 For example, consider two possible association patterns:

[0096] 1) Each multicast group is associated with a unique drone base station (see Figure 4(a)): Multicast group 1 is associated only with drone 2, and users within the coverage area of ​​drone 2 receive a stronger signal from drone 2 than from the base station. Basic layer signal X 1,1 and enhancement layer signal X 2,1 The signal is transmitted simultaneously through the transmission power of UAV 2 and the transmission power of the base station, respectively. As long as the reception rate of the ground equipment covered by UAV 2 meets (4) and (5), X can be decoded by interference cancellation. 1,1 and X 2,1 Obtain the base layer and enhancement layer

[0097] 2) A multicast group is associated with multiple drone base stations (see Figure 4(b)): Multicast group 2 is associated with both drone 1 and drone 2. Users in multicast group 2 covered by drone 1 are farther from drone 2. These users receive stronger signals from drone 1 than from the base station. They also receive interference from drone 2, which is weaker than the base station signal. (Base layer signal X) 1,2 The signal is propagated through the transmission power of UAV 1, while the enhancement layer signal X... 2,2 The propagation depends on the base station. If the user reception rate under the coverage of UAV 1 meets (4) and (5), they can decode X through SIC. 1,2 and X 2,2 No further processing of interference signals is required.

[0098] 3. The problem of maximizing video quality

[0099] 3.1 Graph Model Construction

[0100] As discussed in Section 2.3, drone deployment and multicast group association affect the superposition coding and interference cancellation of drone and base station signals. This section constructs a visual graph model to characterize the coupling between different decision variables, providing support for drone deployment and multicast group association.

[0101] By applying the mean-shift clustering algorithm

[22] to all users within the multicast group, a set of candidate locations for drones is obtained. Among them l j =(x j ,y j () represents a candidate position of the drone on the XY axis.

[0102] make Let represent an undirected graph where each vertex corresponds to a candidate decision for drone placement associated with a multicast group, and must satisfy (4). Thus, the vertex set can be represented as:

[0103]

[0104] in, A collection of multicast group indexes.

[0105] When located in l j,k and l j',k' When both drones are associated with multicast group n, the constraint corresponding to (5) is reformulated as follows:

[0106]

[0107] Regardless of the association mode used, a drone hovering at a planar position index j can only select a unique altitude index k, corresponding to...

[0108] l j,k =l j',k' ,j=j′(11b)

[0109] v j,k,n and v j′,k′,n′ There exists an edge between them. Therefore, the set of edges is represented as...

[0110]

[0111] represent A clique is a subset of vertices in an undirected graph, where every pair of vertices is connected. Each clique can be mapped to a set of decision variables containing the "drone placement-multicast group association pattern".

[0112] 3.2 Case Analysis

[0113] To facilitate understanding of graph generation and clique selection, this specific implementation provides an example scenario where multicast group 1 and multicast group 2 exist, and the candidate locations of the UAV include l 1,1 l 1,2 l 2,2 Assume that the channel gain from the base station to the ground and the channel gain from each UAV candidate location to the ground satisfy conditions (i) to (v).

[0114] (i)

[0115] (ii)

[0116] (iii)

[0117] (iv)

[0118] (v) From (i), (ii), and (iii), the vertex set is:

[0119]

[0120] Because the channel gain condition under the coverage of the UAV (corresponding to (iv)) satisfies (11a), position l 1,1 and l 2,2 A drone can be associated with a multicast group simultaneously, meaning that vertex pairs (v 1,1,2 v 2,2,2 ) and (v 1,1,1 v 2,2,1 There exists an edge at each position. Since condition (v) is not satisfied (10a), position l 1,2 and l 2,2 Drones cannot be associated with the same multicast group simultaneously, which determines the vertex pair (v 1,2,2 v 2,2,2 ), vertex pairs (v 1,2,1 v 2,2,1 There is no connection. 1,1 and l 1,2 The flight altitudes are different, which does not satisfy (11b). Therefore, the vertex pair (v) 1,2,1 v 1,1,2 ), (v 1,2,1 v 1,1,1 ), (v 1,2,2 v 1,1,1 ), (v 1,2,2 v 1,1,2 There are no connections between any of them. Therefore, the set in this scenario is...

[0121]

[0122] Figure 5 Representative based on and The constructed graph model. Next, the cliques are classified according to the number of vertices, see... Figures 6(a) to 6(d) This is to explain the impact of the number of available drones (denoted as α) on the selection of a cluster.

[0123] 1) The size of the group is The situation (see Figure 6(a)): If the group {v 1,1,1 If selected, a drone is placed in l 1,1 , serving multicast group 1.

[0124] 2) Group size is Case (see Figure 6(b)): If α = 1, candidate scheme {v 1,1,1 ,v 1,1,2} represents in l 1,1 Deploy a drone to serve multicast groups 1 and 2. If α = 2, the other choice is {v}. 1,1,1 ,v 2,2,2} represents placing a drone in l 1,1 Service multicast group 1, place another drone in l 2,2 , serving multicast group 2.

[0125] 3) The group size is Case (see Figure 6(c)): If α = 2, the clique {v 1,1,1 ,v 1,1,2 ,v 2,2,2} indicates that a drone is placed in l 1,1 Associated multicast groups 1 and 2; another drone is placed at location l 2,2 , associated multicast group 2.

[0126] 4) The group size is Case (see Figure 6(d)): If α = 2, the clique {v 1,1,1 ,v 1,1,2 ,v 2,2,2 ,v 2,2,1} represents placing a drone on l 1,1 And associated with multicast groups 1 and 2; another drone was placed in l 2,2 , serving multicast groups 1 and 2.

[0127] Based on the above examples, the selection of a group must take into account the number of available drones. For instance, the strategy corresponding to the group in Figure 6(d) requires 2 drones. If the number of drones α = 1, the group in Figure 6(d) cannot be selected.

[0128] 3.3 Problem Modeling

[0129] The problem of maximizing video quality is transformed into a cluster-based spectrum partitioning problem, which essentially involves finding a cluster to determine the drone placement and association patterns and to determine the number of sub-channels for each multicast group.

[0130] Define q j,k,n Used to determine v j,k,n Is it in the selected group? Inside

[0131]

[0132] 0-1 variable u 1,n,i (u 2,n,iThis represents whether user i in multicast group n receives the base layer (enhancement layer). The aggregated PSNR of the video received in multicast group n is represented as a value about b. n and The function, i.e.

[0133]

[0134] right After applying equation (14) to the multicast group, the video quality maximization problem is modeled as follows:

[0135]

[0136]

[0137]

[0138]

[0139]

[0140]

[0141]

[0142]

[0143]

[0144]

[0145] Constraint (15a) means that if vertex pair (v j,k,n ,v j',k',n' There is no edge between the two vertices, meaning they do not belong to the same clique. Constraint (15b) embodies the principle that a UAV is allowed to serve multiple multicast groups. Constraint (15b) includes a symbolic function sgn(·) used to compute the clique. The number of drones to be deployed must be less than or equal to the number of available drones, α. When placed in l j,k When the drone is not associated with any multicast group, Otherwise, it is 1. In (15c) and (15d), Represent a sufficiently large constant to guarantee

[0146]

[0147] in u 1,n,i =1(u 1,n,i In the case of (=0), user i can (cannot) receive / decode the base layer. θ in (15e) and (15f) is also a sufficiently large constant to ensure...

[0148]

[0149] in u 1,n,i When u = 1, 2,n,i =1(u 2,n,i =0) indicates that user i can (cannot) receive and decode the enhancement layer. The sum of the number of sub-channels allocated to each multicast group should not exceed the total number B held by the base station, guaranteed by (15g).

[0150] Due to the strong coupling between UAV placement, association patterns, and sub-channel allocation, the solution... The problem involves high computational complexity. Therefore, decoupling the problem and designing adaptive algorithms is unavoidable.

[0151] 4. Algorithm Design Based on Clique

[0152] For ease of handling, The problem is decoupled into two sub-problems: 1) the association between UAV deployment and multicast groups, and 2) the multicast group spectrum allocation. The former is solved using an improved maximum weight clique search algorithm based on branch and bound, while the latter uses a lightweight algorithm based on rule matching to determine the sub-channel allocation.

[0153] 4.1 Sub-problem 1: Drone-Multicast Group Association

[0154] The weight of a vertex is defined as

[0155]

[0156] Reflecting vertex v j,k,n Contribution to video quality improvement. The drone placement and multicast group association subproblem is transformed into a maximum weight clique problem with vertex number constraints, described as follows:

[0157]

[0158] st(15a),(15b),(15c),(15d),(15e),(15f),(15h),(15i)

[0159] As can be seen from the analysis in Section 3.2, the selection of the clique is limited by the number of UAVs and is not equivalent to the required number of UAVs, which makes it impossible to use the maximum clique or maximum weight clique algorithm

[23]

[24] to solve the problem. Another possible approach is the maximum k-clique or maximum k-weighted clique algorithm

[25] . Both algorithms output cliques of size k, but... The size of the output clique is unknown. Considering these special characteristics, an improved maximum weight clique algorithm based on branch and bound was developed, and the implementation details are shown in Algorithm 1.

[0160] The algorithm's input includes:

[0161] The largest weighted clique discovered so far,

[0162]

[0163] The current set of candidate vertices being processed.

[0164] Γ(v j,k,n ): with vertex v j,k,n The set of all connected vertices.

[0165] In the initial stage, and Set to empty, for All vertices in the array.

[0166]

[0167]

[0168] Before the group search, The weight is input into the bounding function `get_upper_bound()` to obtain the weight of the maximum weight clique, denoted as `t`, which serves as the upper bound of the maximum weight clique in the subgraph (line 4). A subgraph is a graph whose node set and edge set are subsets of the node set and edge set of a given graph, respectively. If The upper bound of the weight is not greater than the current largest weighted group. So in The recursive search on terminates and returns. (Lines 5-6). Conversely, from... Choose the vertex with the largest weight. Used to obtain the group The required number of drones. If this number exceeds α, skip the search at that point; otherwise, add the point to the list. Then Choice and All vertices of the connection are used as the new candidate set. exist This algorithm is executed recursively (lines 13-14). If from... The group that returned from China The weight is greater than Then update and from Remove from Continue searching for vertices that meet the criteria until... If empty, return

[0169] According to (13), and The corresponding optimal decision is During the search process, the bounding procedure prunes branches that do not meet the upper bound to reduce the size of the entire search tree. On the other hand, unnecessary computations are filtered out by the number of drones required to observe a cluster. In the worst case, the computational complexity of searching all vertices is...

[0170] 4.2 Sub-problem 2: Sub-channel allocation

[0171] Given the output of Algorithm 1 The multicast group spectrum partitioning subproblem is modeled as

[0172]

[0173] st(15c),(15d),(15e),(15f),(15g),(15i)

[0174] user The minimum rates required to receive the base layer and enhancement layer are expressed as follows:

[0175]

[0176] and

[0177]

[0178] Once the minimum receive rate requirements for both the base layer and enhancement layer decoding are met, there is no need to add more subchannels. The maximum number of subchannels required for multicast group n is expressed as...

[0179]

[0180] Considering the impact of the number of multicast groups N and the number of sub-channels B on computational complexity, we design a rule matching strategy.

[0181] a) When B≤2N, the PSNR obtained for each multicast group (by...) The decision is to perform a fast sorting process and allocate a sub-channel to each of the top BN multicast groups with the highest PSNR values.

[0182] b) When At this time, all multicast group users can receive both base layer and enhancement layer video. In this case, multicast group n is assigned... There are 1 sub-channels, and the complexity is O(1).

[0183] c) When In this context, an improved knapsack algorithm is used to determine the spectrum partitioning. N multicast groups are considered as N types of items, each with B items. These items need to be placed in a knapsack with a capacity of B. The b-th item in the nth type... n The weight of the item is b n Its profit comes from Decision. For the remaining b sub-channels, the maximum PSNR of the first n multicast groups is denoted as F(n,b). If Each subchannel is assigned to a multicast group n. Rule (24) is used to filter out an inappropriate number of subchannels to reduce unnecessary computation. By recursively solving (24) for the maximum PSNR of the first n multicast groups each time, a spectrum allocation strategy that maximizes the aggregated PSNR of the multicast groups is finally obtained.

[0184]

[0185] The proposed strategy aims to achieve depth matching between B and N and the algorithm. When B ≤ 2N, the solution... The complexity is O(n 2 );when When the computational complexity is O(1), when... At this time, the time complexity of iterating through all cases in each recursive call is O(B). The worst-case time complexity of the entire loop is O(NB). 2 Without proper classification, simple cases become complex, increasing the computational burden. The proposed algorithm can obtain the optimal solution at a lower cost.

[0186] 5. Experimental Design and Results Analysis

[0187] Table 2 Experimental Environment Parameter Settings

[0188] Table 2Experimental parameters

[0189]

[0190]

[0191] Simulation experiments were designed on the MATLAB platform to verify the effectiveness of the proposed solution. A scenario was considered where a macro base station collaborates with multiple UAVs. We considered improving the video quality for users within the base station's coverage edge (a ring-shaped area 750–900 m from the base station). Detailed parameter settings are shown in Table 2.

[0192] The proposed methods are divided into three categories, named Proposed-1, Proposed-2, and Proposed-3, and the differences are shown in Table 3. Three benchmark schemes were selected, as shown in Table 4, and compared with the proposed schemes.

[0193] Table 3 shows the proposed scheme settings.

[0194] Table 3Implementation of proposed methods

[0195] Classification of proposed methods Drone placement and multicast group association strategies Spectrum allocation strategy Proposed-1 Algorithm 1 Strategies in Section 4.2 Proposed-2 Algorithm 1 Spectrum average division Proposed-3 Fixed flight altitude Strategies in Section 4.2

[0196] Table 4. Benchmark Scheme Settings

[0197] Table 4 Implement of baselines

[0198]

[0199] 5.1 Impact of the Number of Drones on PSNR

[0200] In Figures 7 and 8, the aggregated PSNR of Proposed-1 increases rapidly as B increases from 5 to 11 and then plateaus after B=11. Benefiting from the drone placement-multicast group association strategy, the proposed scheme can adapt to resource-constrained situations, with the inflection point of growth located to the left. In Baseline-1, each multicast group is associated with only one drone, limiting the performance improvement even with increasing B. Proposed-1 controls interference between drones within an acceptable range, ensuring that the benefits of resource reuse outweigh the impact of interference. Compared to Baseline-2, the NOMA overlay coding in Proposed-1 allows the base station to share spectrum resources with drones for signal transmission without consuming additional spectrum resources. Under any B, the aggregated PSNR value of Proposed-1 is higher than that of OMA. In Baseline-3, the base station needs to send both the base layer and enhancement layer frequencies to the drone, which then forwards them to the user. When B<9, the aggregated PSNR is [value missing]. Proposed-1 can still provide basic video services to some users even when B<8.

[0201] Figures 7(a) to 7(c) The three subgraphs illustrate the impact of different numbers of drones on the total PSNR received by the user. The more drones there are, the greater the improvement in total PSNR brought by the proposed solution. In Proposed-1, the drones are relatively dispersed, resulting in less interference between drones associated with the same multicast group. This allows for selection of the drone placement height based on the maximum coverage radius, consistent with the decision to directly select the maximum coverage height in Proposed-3. Figure 7(a) and 7(b)The proposed solution is similar to the PSNR aggregation in Proposed-3. More drones help improve resource utilization but increase co-channel interference. Adjustable flight altitude helps reduce interference between drones.

[0202] In Figure 7(a), the vertical axis is named Total PSNR (dB); the horizontal axis is named Number of available subchannels.

[0203] In Figure 7(b), the vertical axis is named Total PSNR (dB); the horizontal axis is named Number of available subchannels.

[0204] In Figure 7(c), the vertical axis is named Total PSNR (dB), and the horizontal axis is named Number of available subchannels.

[0205] 5.2 Impact of Multicast Group Number on PSNR

[0206] Figures 8(a) to 8(c) The impact of the number of multicast groups on the average PSNR of users is shown in Figure 8(a). As shown in Figure 8(a), the average PSNR of the proposed scheme at N=5 is approximately twice that of OMA and about 8.4% higher than other NOMA benchmark schemes. In Table 5, the total PSNR received by multicast group 2 under the Proposed-1 scheme is about 7.9% higher than that under Proposed-2. It can be seen that the more multicast groups there are, the more obvious the advantage of the proposed scheme compared to other schemes. In Table 6, the total PSNR received by multicast group 5 under Proposed-1 is 3.68 times that under Proposed-2. The average spectrum allocation in Proposed-2 leads to unsatisfactory video quality for some users in these multicast groups. Meanwhile, other multicast groups receive excess spectrum resources, and the received video quality does not improve significantly. Proposed-1 can dynamically allocate the number of sub-channels, better match the spectrum resource needs of each multicast group, and avoid resource waste.

[0207] Table 5 shows the total PSNR received by each multicast group (α=5, N=4, B=12).

[0208] Table 5Aggregate PSNR received by each multicast group (α=5, N=4, B=12)

[0209] Multicast Group 1 Multicast Group 2 Multicast Group 3 Multicast Group 4 mean variance Proposed-1 909.74 625.29 946.75 503.54 746.33 17799.57 Proposed-2 909.74 579.5 946.75 503.54 734.88 23503.98 Proposed-3 769.78 658.2 719.53 473.92 655.35 16712.75 Baeseline-1 419.88 394.92 454.44 207.34 369.145 12231.49 Baeseline-2 122.69 366 501.59 269.72 315 25483.41 Baeseline-3 855.92 579.5 894 503.54 708.24 38264.07

[0210] Table 6 shows the total PSNR received by each multicast group (α=6, N=5, B=12).

[0211] Table 4Aggregate PSNR received by each multicast group (α=6, N=5, B=12)

[0212] Multicast Group 1 Multicast Group 2 Multicast Group 3 Multicast Group 4 Multicast Group 5 mean variance Proposed-1 979.72 671 984.62 592.4 702.66 786.08 33657.2 Proposed-2 979.72 671 984.62 592.4 190.8 683.71 107407 Proposed-3 839.76 701.5 757.4 562.78 731.4 718.56 10230.32 Baeseline-1 419.88 366 454.44 207.34 222.6 334.052 12839.46 Baeseline-2 131.68 109.34 537.12 355.44 386.86 304.088 32860.86 Baeseline-3 921.76 0 929.76 0 0 370.304 257117.47

[0213] 5.3 Visualization of UAV Deployment

[0214] Figure 9(a) shows the drone placement when N=5 and α=6 in the Proposed-3 scheme. Since the drone altitude is fixed in the Proposed-3 scheme, the altitude that covers the largest area is always chosen. If two drones placed too close together serve the same multicast group, the interference from signal superposition will offset the benefits of spectrum reuse. For example, to manage interference, drones 1 and 6 are associated with different multicast groups. Drones 2 and 3 are farther apart, and interference is less when sharing the spectrum, so they are allowed to be associated with the same multicast group. Figure 9(b) shows the association patterns of drones 1 and 6 with multicast groups in Figure 9(a), where drone 1 is associated with multicast groups 1, 2, and 5, and drone 6 is associated with multicast groups 3 and 4. Figure 10(a) shows the drone placement when N=5 and α=6 in Proposed-1. Unlike proposed-3, drones associated with the same multicast group in Proposed-1 can reduce co-channel interference by adjusting their altitude. Thus, each drone serves multiple multicast groups on different sub-channels, as shown in Figure 10(b). Proposed-1, by flexibly adjusting the placement of the drone, can still achieve a higher total PSNR than other solutions.

[0215] 6. Summary

[0216] This paper proposes an online SVC multicast method based on UAV relay and NOMA, maximizing the overall video service quality within multicast groups through efficient resource management. Since the joint optimization problem of UAV deployment, the association pattern between UAVs and multicast groups, and spectrum allocation is an integer nonlinear programming problem, it is decoupled into two sub-problems: UAV deployment-multicast group association and spectrum allocation, to find suitable solutions. This chapter designs a maximum weight clique search algorithm with UAV constraints and a lightweight spectrum allocation algorithm based on rule matching to determine UAV deployment, multicast group association patterns, and spectrum allocation. Simulation results using real data demonstrate that the proposed scheme significantly outperforms other schemes.

[0217] References

[0218] [1]Harte,L.Introduction to Data Multicasting[M].America:AlthosPublishing,2008.

[0219] [2]Hao Zhu,Yang Cao,Tao Jiang,and Qian Zhang,Scalable NOMA multicastfor SVC streams in cellular networks[J].IEEE Transactions on Communications,2018,66(12):6339-6352.

[0220] [3]Schierl T,Stockhammer T,Wiegand T.Mobile video transmission usingscalable video coding[J].IEEE Transactions on Circuits and Systems for VideoTechnology,2007,17(9):1204-1217.

[0221] [4]Ding Z,Liu Y,Choi J,et al.Application of Non-orthogonal MultipleAccess in LTE and 5GNetworks[J].IEEE Communications Magazine,2017,55(2):185-191.

[0222] [5]Mahady I A,Bedeer E,Ikki S,et al.Sum-rate maximization of NOMAsystems under imperfect successive interference cancellation[J].IEEECommunications Letters,2019,23(3):474-477.

[0223] [6]Shen H,Ye Q,Zhuang W,et al.Drone-Small-Cell-Assisted ResourceSlicing for 5G Uplink Radio Access Networks[J].IEEE Transactions on VehicularTechnology,2021,70(7):7071-7086.

[0224] [7]A.Razavieh et al.,"Extremely-Low Threshold Voltage FinFET for 5GmmWave Applications,"in IEEE Journal of the Electron Devices Society,vol.9,pp.165-169,2021,doi:10.1109 / JEDS.2020.3046953.

[0225] [8]K.-M.Chen,Y.-H.Pan and T.-S.Lee,"Low-Complexity Beam Selection forHybrid Precoded Multi-User mmWave Communications,"2018IEEE InternationalConference on Signal Processing,Communications and Computing(ICSPCC),Qingdao,China,2018,pp.1-5,doi:10.1109 / ICSPCC.2018.8567785.

[0226] [9]M.N.Dani,D.K.C.So,J.Tang and Z.Ding.Resource Allocation forLayered Multicast Video Streaming in NOMA Systems[J].IEEE Transactions onVehicular Technology,2022,71(11):11379-11394.

[0227]

[10] B.Zeng,C.Zhan,Y.Yang,et al.Access Delay Minimization for ScalableVideos in Cache-Enabled Multi-UAV Networks[C].Proceedings of IEEE GlobalCommunications Conference(GLOBECOM),Rio de Janeiro,Brazil,2022,389-394.

[0228]

[11] L.Wang,H.Zhang,S.Guo and D.Yuan,"Deployment and Association ofMultiple UAVs in UAV-Assisted Cellular Networks With the Knowledge ofStatistical User Position,"in IEEE Transactions on Wireless Communications,vol.21,no.8,pp.6553-6567,Aug.2022,doi:10.1109 / TWC.2022.3150429.

[0229]

[12] M.Katwe,K.Singh,P.K.Sharma,et al.Dynamic User Clustering andOptimal Power Allocation in UAV-Assisted Full-Duplex Hybrid NOMA System[J].IEEE Transactions on Wireless Communications,2022,21(4):2573-2590.

[0230]

[13] H.Zeng,X.Zhu,Y.Jiang,et al.UAV-Ground BS Coordinated NOMA withJoint User Scheduling,Power Allocation and Trajectory Design[C].Proceedingsof IEEE International Conference on Communications(ICC),Dublin,Ireland,2020,1-6.

[0231]

[14] A.Al-Hourani,S.Kandeepan and S.Lardner.Optimal LAP Altitude forMaximum Coverage[J].IEEE Wireless Communications Letters,2014,3(6):569-572.

[0232]

[15] Zhang T,Wang Z,Liu Y,et al.Caching placement and resourceallocation for cache-enabling UAV NOMA networks[J].IEEE Transactions onVehicular Technology,2020,69(11):12897-12911.

[0233]

[16] S.Chai and V.K.N.Lau.Mixed-Timescale Request-Driven UserAssociation,Trajectory and Radio Resource Control for Cache-Enabled Multi-UAVNetworks[J].IEEE Transactions on Signal Processing,2022,70(10):4997-5011.

[0234]

[17] Rappaport T S,Maccartney G R,Samimi M K,et al.WidebandMillimeter-Wave Propagation Measurements and Channel Models for FutureWireless Communication System Design[J].IEEE Transactions on Communications,2015,63(9):3029-3056.

[0235]

[18] Ghermezcheshmeh M,Shah-Mansouri V,Ghanbari M.Analysis andperformance evaluation of scalable video coding over heterogeneous cellularnetworks[J].Computer Networks,2019,148:151-163.

[0236]

[19] C.A.Balanis,Antenna Theory:Analysis and Design[M].4th ed,Hoboken,NJ,USA:Wiley,2016

[0237]

[20] H.He,S.Zhang,Y.Zeng,et al.Joint Altitude and BeamwidthOptimization for UAV-Enabled Multiuser Communications[J].IEEE CommunicationsLetters,2018,22(2):344-347.

[0238]

[21] Ye Q,Zhuang W,Zhang S,et al.Dynamic radio resource slicing for atwo-tier heterogeneous wireless network[J].IEEE Transactions on VehicularTechnology,2018,67(10):9896-9910.

[0239]

[22] D.Comaniciu and P.Meer.Mean shift analysis and applications[C].Proceedings of the Seventh IEEE International Conference on Computer Vision,Kerkyra,Greece,1999:1197-1203.

[0240]

[23] Ostergard P. A Fast Algorithm for the Maximum Clique Problem. Discrete Applied Mathematics, 2002, 120(1 - 3): 197–207.

[0241]

[24] Jiang Hua. Research on Exact Algorithms for the Maximum Clique Problem [D]. Huazhong University of Science and Technology, 2017.

[0242]

[25] Wu J, Li C M, Jiang L, et al. Local search for diversified top-k clique search problem [J]. Computers & Operations Research, 2020, 116: 104867.

[0243]

[26] T.M. Nguyen, W. Ajib and C. Assi. A Novel Cooperative NOMA for Designing UAV-Assisted Wireless Backhaul Networks [J]. IEEE Journal on Selected Areas in Communications, 2018, 36(11): 2497 - 2507.

[0244]

[27] D. Zhai, H. Li, X. Tang, et al. Height Optimization and Resource Allocation for NOMA Enhanced UAV-Aided Relay Networks [J]. IEEE Transactions on Communications, 2021, 69(2): 962 - 975.

[0245]

[28] Tong, Z., Shen, H., Shi, N., Wang, T., and Bai, G., Joint resource optimization for nonorthogonal multiple access-enhanced scalable video coding multicast in unmanned aerial vehicle-assisted radio-access networks [J]. ETRI Journal,7, 2022.

[0246]

[29] Ioannis-Prodromos Belikaidis,Andreas Georgakopoulos,EvangelosKosmatos,Valerio Frascolla,and Panagiotis Demestichas,Management of 3.5-GHzspectrumin 5G dense networks:A hierarchical radio resource management scheme,IEEE Veh.Technol.Mag.13(2018),no.2,57–64。

Claims

1. An online SVC multicast method based on UAV relay in a NOMA network. In the online SVC video multicast scenario using UAV relay, multiple UAVs are deployed as relays at the coverage edge of a macro base station, with three different types of links: The ground base station to drone B2U link is used by the base station to send the SVC base layer to the drone; The ground base station to ground equipment B2D link is used by the base station to send the SVC enhancement layer to the user; The U2D link between drones and ground devices is used by drones to send the base layer of SVC to users. Users requesting the same video stream belong to the same multicast group; A multicast group is associated with one or more drones, meaning that one or more drones provide services to users within a multicast group; the mobile edge computing (MEC) controller determines the deployment location of drones and the association mode between drones and multicast groups by accessing global information, and allocates spectrum resources to each multicast group; Its characteristic is that the steps of the online SVC multicast method include: Step 1) Construct a visual graph model to depict the coupling between drone placement and multicast group association; Step 2) Based on the visual graph model, the problem of maximizing video reception quality is modeled as a nonlinear integer programming problem P1 based on clique; Step 3) Decouple problem P1 into the UAV placement-multicast group association subproblem and the subchannel allocation subproblem; Step 4) Solve the two sub-problems: Step 4.1) Transform the drone placement-multicast group association subproblem into the maximum weight clique problem in the undirected graph of the visualization graph model with a limit on the number of vertices, and then use the improved maximum weight clique algorithm based on branch and bound to determine the drone placement-multicast group association mode. Step 4.2) Use a rule matching strategy to solve the sub-channel allocation problem and obtain the optimal resource allocation strategy with low computational cost; In step 1), a visual graph model is constructed to characterize the coupling between different decision variables, providing support for drone deployment and multicast group association: By applying a clustering algorithm to all users within all multicast groups, a set of candidate drone locations is obtained. ,in For the index of the projection position on the XY plane, This represents a candidate position of the drone on the XY axis; make This represents an undirected graph where each vertex corresponds to a candidate decision for drone placement and multicast group association, and must satisfy the following conditions: Then, the vertex set is represented as ， in, A set of multicast group indexes, Is with The height index corresponding to the location; This represents the maximum signal gain from the macro base station m to the edge of the drone's coverage area; p represents the minimum signal gain from the drone to the edge of coverage. s p represents the transmission power of the drone. m This indicates the base station's transmission power; When any two are located and When all drones are associated with multicast group n, and The corresponding constraints are re-expressed as （A） This represents the maximum channel gain within the coverage area of ​​the second drone compared to the first drone. This represents the minimum channel gain from the macro base station m to the coverage area of ​​the first drone; A drone hovering at a planar position index j can only choose a unique altitude index k, corresponding to （B） The first drone corresponds to the vertex if and only if (A) or (B) is true. The vertex corresponding to the second drone There exists an edge between them; therefore, the set of edges is represented as ， represent A clique is a subset of vertices in an undirected graph, where every pair of vertices is connected. Each group They are all mapped to a set of decision variables containing "drone placement-multicast group association"; In step 2), the problem of maximizing video quality is transformed into a cluster-based spectrum partitioning problem. Specifically, a cluster is found to determine the drone placement and multicast group association, and the number of sub-channels in each multicast group is determined. Define q j,k,n Used to determine vertex v j,k,n Is it in the selected group? Inside ， 0-1 variable u 1,n,i u 2,n,i These represent whether user i in multicast group n has received the base layer and enhancement layer respectively, with 0 indicating receipt and 1 indicating no receipt; The aggregate signal-to-noise ratio (PSNR) of the video received by multicast group n is expressed as a function of subchannel b. n and The function, i.e. ， Then for Multicast group usage function Subsequently, the video quality maximization problem P1 was modeled as follows: ， st （a） （b） （c） （d） （e） （f） （g） （h） （i） q j,k,n Corresponding to the first drone and its vertex v j,k,n The judgment, Corresponding to the second drone and its vertex The judgment; This indicates the minimum bit rate that the ground equipment within the multicast group can support for normal decoding when the multicast group n requests the base layer. This indicates the minimum bit rate at which the ground equipment within the multicast group supports normal decoding when the multicast group n requests the enhancement layer. Indicates user Decoding from position l j,k The achievable rate of the drone signal; Indicates in the user The achievable rate of the signal from the decoding base station m; Constraint (a) represents if vertex pair ( , There is no edge between these two vertices; they do not belong to the same clique. Constraint (b) represents a drone being allowed to serve multiple multicast groups; constraint (b) contains a symbolic function. Used to calculate groups The number of drones to be deployed must be less than or equal to the number of available drones. When placed When the drone is not associated with any multicast group, Otherwise, it is 1; In constraints (c) and (d), Represent a sufficiently large constant to guarantee ， in u 1,n,i =1、u 1,n,i When =0, user i can or cannot receive and decode the base layer, respectively; In constraints (e) and (f) It is a sufficiently large constant to ensure ， in u 1,n,i When u = 1 2,n,i =1 or u 2,n,i =0 indicates that user i can or cannot receive and decode the enhancement layer; Constraint (g) ensures that the sum of the number of sub-channels allocated to each multicast group does not exceed the total number B held by the base station; In step 3), the problem It is decoupled into the UAV-multicast group association subproblem and the subchannel allocation subproblem; In step 4.1), the sub-problem of drone-multicast group association is solved: The weight of a vertex reflects the vertex v j,k,n The vertex weight is defined as the contribution to improving video quality. ， The drone placement-multicast group association subproblem is transformed into a maximum weight clique search problem P1.1 with a vertex number constraint, described as follows: ， st(a), (b), (c),(d),(e),(f),(h),(i) An improved maximum weight clique algorithm based on branch and bound is used to search for the maximum weight clique to solve problem P1.1; The algorithm's input includes: : The largest weighted clique discovered so far, The current set of candidate vertices being processed. : with vertex The set of all connected vertices; initial stage, and Set to empty, for All vertices in the array; Before executing the algorithm, The weights are input into a bounding function to obtain the maximum weight clique, denoted as t, which serves as the upper bound of the maximum weight clique in the subgraph. The algorithm flow is as follows: if The upper bound of the weight is not greater than the current largest weighted group. So in The recursive search on terminates and returns. Conversely, from Select the vertex with the largest weight ; Winning Team The required number of drones, if this number exceeds Skip the search at that point; otherwise, add the point to the list. ; Then Choice and All vertices of the connection are used as the new candidate set. ,exist This algorithm is executed recursively. If from The group that returned from China The weight is greater than Then update and from Remove from ; Continue searching for vertices that meet the criteria until... If empty, return ; and The corresponding optimal decision is , Represents the vertex with the largest weight. In the group Inside; In step 4.2), the sub-channel allocation sub-problem is solved: From the output of step 4.1) The subchannel allocation numerator problem P1.2 is modeled as follows: ， s.t. , user The minimum rates required to receive the base layer and enhancement layer are expressed as follows: ， and ， Indicates in The channel gain from the drone to the ground device i within the coverage area; Indicates the distance from base station m to ground equipment Channel gain; Indicates in The channel gain from the drone to the ground device i within the coverage area; Once the minimum receive rate requirements for decoding at the base layer and enhancement layer are met, there is no need to add more subchannels; the maximum number of subchannels required for multicast group n is expressed as... ， Indicates user The required rate for receiving data from the base layer. Indicates user The rate required to receive the enhancement layer; Considering the impact of the number of multicast groups N and the number of sub-channels B on computational complexity, the following matching strategy is adopted: a) when At that time, the PSNR obtained by each multicast group is sorted, and a sub-channel is allocated to each of the top BN multicast groups with the highest PSNR values; b) When When all multicast group users can receive both base layer and enhancement layer video, then multicast group n is assigned... Sub-channels; c) When At that time, an improved knapsack algorithm was used to determine the spectrum partitioning: N multicast groups are considered as N types of items, with B items of each type; these items need to be placed in a knapsack with a capacity of B. In the nth category, the th The weight of each item is Its profit comes from Decide; For the remaining b sub-channels, the maximum PSNR of the first n multicast groups is denoted as: ;if ,but Each sub-channel is assigned to multicast group n; The following formula is used to filter out an inappropriate number of sub-channels. By recursively solving the following formula, the maximum value of the PSNR of the first n multicast groups is calculated each time, and finally the spectrum allocation strategy that maximizes the aggregated PSNR of the multicast groups is obtained. ; 。 2. The online SVC multicast method based on UAV relay in NOMA networks according to claim 1, characterized in that: In step 1), the clustering algorithm is the mean-shift algorithm.

3. The online SVC multicast method based on UAV relay in NOMA networks according to claim 1, characterized in that: In step 4.2), during the clique search process of the improved maximum weight clique algorithm based on branch and bound, the bounding procedure will prune branches that do not meet the upper limit.

Images