A Design Method for Unmanned Aerial Vehicle Transmitters for Cellular Networks

Abstract

This invention discloses a design method for UAV-borne transmitters in non-cellular networks, belonging to the field of wireless communication. The method includes: establishing a multi-user asynchronous transmission model and a fronthaul quantization noise model under non-cellular networks; a central processing node aligning and merging the compressed forwarding signals of each access point according to the arrival time of the target UAV, decomposing the quantity to be detected into the target signal, multi-user asynchronous interference, thermal noise, and fronthaul quantization noise, and establishing an equivalent signal-to-interference-plus-noise ratio (SINR) model for each UAV; constructing a multi-user codebook optimization problem with the objective of maximizing the minimum equivalent SINR among all UAVs; and constructing and training a codebook generation model to generate a multi-user transmission codebook adapted to the current configuration based on network topology information and updating the UAV transmitter configuration. This invention improves the SINR and communication reliability of multi-user transmission at the receiving end, meeting the communication requirements of wide-area access and reliable transmission for multiple UAVs.

Description

Technical Field

[0001] This invention relates to the field of wireless communication technology, and more specifically to a design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks. Background Technology

[0002] When a drone connects to a non-cellular network as an aerial user, it transmits uplink data to multiple widely deployed ground-based distributed access nodes. Through close coordination, these distributed access points can eliminate near-far effects, significantly improving system coverage and multi-user access capabilities. However, drones cover a wide area, and the transmission distances between drones and different ground access points vary significantly. Furthermore, the fixed speed of electromagnetic signals in the air means that signals transmitted at the same time arrive at different receivers at different times, introducing a severe latency inconsistency problem in multi-user communication.

[0003] While traditional cellular systems also experience differences in distance between users, latency can be controlled through mechanisms such as cyclic prefixes. However, in unmanned aerial vehicle (UAV) non-cellular network architectures, the widespread distribution of ground access points and the dispersed locations of UAVs in the air mean that the propagation distance from a UAV to different access points can vary by several kilometers or more. This significantly increases the magnitude of signal transmission latency differences, often exceeding the preset cyclic prefix length of the air interface. This amplified spatial asynchrony severely undermines the synchronization foundation of multi-user communication, causing symbol misalignment in UAV signals at the receiving end, and consequently severely impacting system performance.

[0004] Specifically, the different transmission delays of different drone users will cause their optimal sampling points to no longer coincide. When the signal sampling point deviates from the optimal sampling point, the sampled signal will face complex inter-symbol interference. This causes the inter-user interference to worsen from single-symbol interference under the synchronization assumption to multi-symbol interference. This complex asynchronous superposition effect will severely degrade signal detection performance. However, existing multi-user transmitters, such as transmitters based on sparse spreading codes and transmitters based on Walsh codes, are mostly designed based on the ideal synchronous transmission assumption and fail to consider the asynchronous interference characteristics caused by delay inconsistencies in drone non-cellular networks.

[0005] Meanwhile, unlike the centralized antenna access of traditional cellular networks, the transmission latency of each drone user varies at different non-cellular network access points; even at the same access point, different drone users arrive at different times. This heterogeneous latency further complicates the structure of multi-user asynchronous interference signals. Therefore, it is necessary to design transmitters to match the actual needs of drone communication systems in real-world wide-area drone asynchronous communication scenarios.

[0006] On the other hand, in actual deployments of non-cellular networks, the fronthaul link between the access point and the central processing node is usually limited by bandwidth, transmission rate, and resource overhead, making it impossible to ideally transmit all lossless received signals. Therefore, limited fronthaul capacity introduces compression errors or quantization noise, further affecting the multi-user signal combining effect and the final signal-to-interference-plus-noise ratio at the central processing node. Summary of the Invention

[0007] Based on the above analysis, the present invention aims to disclose a design method for a UAV-borne transmitter for non-cellular networks; to improve the signal-to-interference-plus-noise ratio and communication reliability of multiple users at the receiving end, and to meet the communication requirements of wide-area access and reliable transmission for multiple UAVs.

[0008] This invention discloses a design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks, comprising: Establish a multi-user asynchronous transmission model; in a non-cellular network including distributed access points and a central processing node, during the asynchronous transmission of uplink communication between each UAV and each access point, multi-user asynchronous interference originates from the difference in propagation distance from each UAV to different access points, as well as the heterogeneous relative delay introduced when the central processing node aligns and merges according to the arrival time of the target UAV. A compression forwarding mechanism is used to quantize and compress the received signals at each access point, and a fronthaul quantization noise model is established. The central processing node aligns and merges the compressed forwarding signals of each access point according to the arrival time of the target UAV, decomposes the quantity to be detected into the target signal, multi-user asynchronous interference, thermal noise and forward quantization noise, and establishes the equivalent signal-to-interference-plus-noise ratio model for each UAV. To maximize the minimum equivalent signal-to-interference-plus-noise ratio among all UAVs, a multi-user codebook optimization problem is constructed. Build and train a codebook generation model so that the model can generate a multi-user transmission codebook adapted to the current configuration based on the input network topology information and update the UAV transmitter configuration.

[0009] Furthermore, the fronthaul quantization noise model adopts an additive complex Gaussian noise model, and the quantization noise variance is jointly determined by the fronthaul capacity and the average received power of the access point; the average received power is obtained by superimposing the thermal noise power on the product of the large-scale fading gain and the transmit power of all UAVs.

[0010] Furthermore, the multi-user codebook optimization problem is to maximize the minimum equivalent signal-to-interference-plus-noise ratio (SINR) among all UAVs, with constraints including codeword normalization, upper limit of transmit power, and the number of quantization bits and quantization noise variance determined by the fronthaul capacity.

[0011] Furthermore, the number of quantization bits is determined by the fronthaul capacity. for: ; Quantization noise variance for: ; in, The available fronthaul capacity from distributed access points to the central processing node. The number of symbol blocks contained in each frame. The length of a complex digital character, For the first Average received power of each access point; , This represents the total number of access points.

[0012] Furthermore, the first The equivalent SINR of a drone in a fronthaul architecture is represented as follows: ; in, For the target signal power, This represents the total asynchronous interference power. Thermal noise power, This is the noise power for forward quantization.

[0013] Furthermore, the target signal for the decomposition of the quantity to be detected is represented as: ; in, For the first Target signal items of a drone; For the central processing node to the first The drone, the first The merging weight of signals from each access point For the first The drone to the first Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first The first drone digital character individual character symbols, For the first The drone in the first Data symbols sent by a symbol block, For symbol period, A unit energy pulse shaping function; The number of symbol blocks contained in each frame. The total number of access points. The length of the complex digital character; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the coherent combining result of the gains of multiple access points under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Target signal power of a drone Approximate value.

[0014] Furthermore, the multi-user asynchronous interference of the quantity to be detected is represented as: ; in, For the first The drone is the first The drone in the first Asynchronous interference of one symbol; For the central processing node to the first The drone, the first The merging weight of signals from each access point , The first The, the The drone to the first Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first The first drone digital character individual character symbols, For the first The drone in the first Data symbols sent by a symbol block, For symbol period, For the first The drone is relative to the first The drone in the first The relative propagation delay of each access point A unit energy pulse shaping function; The number of symbol blocks contained in each frame. The total number of access points. The length of the complex digital character.

[0015] Furthermore, the thermal noise of the decomposition of the detectable quantity is expressed as: ; in, For the first The drone in the first Thermal noise received by each symbol; For the central processing node to the first The drone, the first The merging weight of signals from each access point For the first The drone to the first Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first At each access point Additive white Gaussian noise at any given time; For the first The first symbol The transmitter reference time corresponding to each chip With the The drone to the first Propagation delay between access points sum; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the incoherent superposition result of the thermal noise of each access point after merging and weighting under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Thermal noise power of a drone Approximate value.

[0016] Furthermore, the forward quantization noise of the decomposition of the quantity to be detected is expressed as: ; in, For the first The drone in the first Each symbol is subject to forward quantization noise; For the central processing node to the first The drone, the first The merging weight of signals from each access point For the first The drone to the first Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character, "Indicates conjugate, For the first Access points Forward quantization noise at any given time; For the first The first symbol The transmitter reference time corresponding to each chip With the The drone to the first Propagation delay between access points sum; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the incoherent superposition result of the quantization noise of each access point after merging and weighting under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Forward quantization noise power of a drone Approximate value.

[0017] Furthermore, constructing and training the codebook generation model includes: Construct a codebook generation model that includes a heterogeneous graph attention task encoding network and a policy network; In this task coding network, the drones and access points in the non-cellular network are modeled as two types of nodes in a heterogeneous graph, and the communication link between the drone and the access point is modeled as edges in a bipartite graph. The policy network takes task embedding, current codebook state and latency information as input to generate new codebooks or codebook adjustments, and performs normalization processing on the codewords of each UAV. The codebook generation model is trained using a meta-learning framework with the equivalent SINR under the minimum forward architecture as the optimization objective. When configuring a new network, the codebook generation model uses learned initialization parameters and task embeddings to quickly generate an adapted multi-user launch codebook and update the UAV transmitter configuration.

[0018] Compared with traditional methods, the present invention has the following technical advantages: This invention discloses a UAV transmitter design method for non-cellular networks. By modifying the multi-user codebook structure embedded in the transmitter, it changes the mapping between digital signals and physical resources. Combined with the positional relationship between the UAV and the ground access point, propagation delay, channel fading, and fronthaul capacity constraints, it generates a multi-user codeword sequence adapted to non-cellular asynchronous communication scenarios, thereby improving the signal-to-interference-plus-noise ratio (SIR) and communication reliability of the multi-user receiver. Furthermore, the method further considers the fronthaul quantization noise introduced by the compression forwarding mechanism and incorporates it into a unified model of target signals, asynchronous interference, thermal noise, and quantization noise. This allows the central processing node to quickly generate a multi-user codebook adapted to the current non-cellular asynchronous communication environment based on pre-acquired configuration information of UAV users, ground access points, and fronthaul links. This codebook is then updated with the UAV terminal transmitter configuration for subsequent data transmission, effectively meeting the communication requirements of wide-area access and reliable transmission for multiple UAVs. Attached Figure Description

[0019] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 This is a flowchart of the design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks in an embodiment of the present invention. Figure 2 This is a schematic diagram of a drone communicating with a non-cellular network in an embodiment of the present invention; Figure 3 This is a schematic diagram of the architecture of an unmanned aerial vehicle (UAV) transmitter system for non-cellular networks in an embodiment of the present invention. Detailed Implementation

[0020] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and, together with the embodiments of the present invention, serve to illustrate the principles of the present invention.

[0021] One embodiment of the present invention discloses a design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks, such as... Figure 1 As shown, it includes the following steps: S1. Establish a multi-user asynchronous transmission model. In a non-cellular network including distributed access points and a central processing node, during the asynchronous transmission of uplink communication between each UAV and each access point, multi-user asynchronous interference originates from the difference in propagation distance from each UAV to different access points, as well as the heterogeneous relative delay introduced when the central processing node aligns and merges the data according to the arrival time of the target UAV. S2. Employ a compression forwarding mechanism to quantize and compress the received signals at each access point, and establish a fronthaul quantization noise model. S3. The central processing node aligns and merges the compressed forwarding signals of each access point according to the arrival time of the target UAV, decomposes the quantity to be detected into target signal, multi-user asynchronous interference, thermal noise and forward quantization noise, and establishes the equivalent signal-to-interference-plus-noise ratio model of each UAV. S4. To maximize the minimum equivalent signal-to-interference-plus-noise ratio among all UAVs, construct a multi-user codebook optimization problem. S5. Build and train the codebook generation model so that the model can generate a multi-user transmission codebook adapted to the current configuration based on the input network topology information and update the UAV transmitter configuration.

[0022] Specifically, S1 includes: S1-1. Perform topology modeling of UAVs and access points in a non-cellular network environment; like Figure 2 As shown, in a non-cellular network including distributed access points (APs) and a central processing node (CPU), multiple drones simultaneously communicate uplink with various access points. Due to the wide distribution of drones in the air, their propagation distances to different access points vary significantly, causing signals transmitted at the same time to arrive at different receivers at different times, thus introducing severe latency inconsistency.

[0023] To quantitatively characterize this asynchronous property, we first model the network topology. Assume there are a total of... Taiwanese drones and Communication with the access point, the first The spatial coordinates of the access points are as follows , , No. The spatial coordinates of the drone are as follows , ;No. The drone to the first The transmission distance between the access points is denoted as . It can be obtained from Euclidean distance: ; According to the laws of electromagnetic wave propagation in free space, the first The drone to the first The propagation delay between access points is denoted as . , can be represented as: ; in, The speed at which electromagnetic waves propagate in free space.

[0024] Construct a distance matrix by combining the distances and delays between all drones and all access points. and delay matrix ; , .

[0025] S1-2, Model the communication channel of the UAV in a non-cellular network environment; No. The drone to the first Channel coefficients between access points It can be decomposed into large-scale fading components and small-scale fading components. ; in, The large-scale fading power gain is determined by topological location, path loss, and shadowing fading. These are the random coefficients for small-scale fading, characterizing the random variations in fast fading.

[0026] For the The drone and the first The path loss of a link between access points can be represented using a logarithmic distance model as follows: ; in, For reference distance Path loss at the location, The path loss exponent depends on the propagation environment, such as free space. =2, City Macrocell ≈3~4.

[0027] Further considering log-normal shadowing fading, the large-scale fading power gain It can be represented as: ; in The standard deviation of shadow fading. For the first The drone to the first A standard normal random variable with shadow fading between access points; Small-scale fading can be modeled as Rayleigh or Rice fading based on the actual propagation environment.

[0028] A large-scale fading matrix is ​​constructed by combining the large-scale fading power gains between all UAVs and all access points. , its first Line number Column elements are ; ; The large-scale fading matrix With delay matrix Together, they serve as key input parameters for subsequent codebook optimization and task encoding.

[0029] S1-3, Perform baseband transmission signal modeling; No. The drone uses a length of The complex codeword is spread and transmitted along its data symbols. This codeword is denoted as: The codewords satisfy the normalization constraint. .

[0030] The codebook matrix is ​​composed of the codewords of all drones. : ; set up For the first The drone in the first The data symbols sent at each symbol time, , The number of symbol blocks contained in each frame. For the first The transmit power of each drone, The maximum launch power of the UAV is constrained.

[0031] For pulse shaping function, If the symbol period is , then the th The baseband transmission signal of a drone is modeled as follows: ; This formula shows that: each data symbol Coded Spread spectrum Each chip has a spacing of [number] chips. Through pulse shaping function Mapped to a continuous-time waveform.

[0032] S1-4. Perform asynchronous signal reception modeling; Because different drones travel different distances to the same access point, their signals experience different delays when reaching the same access point. Therefore, the... The asynchronous superimposed signals received by each access point are modeled as follows: ; in, For the first Additive white Gaussian noise at each access point, with variance of: .

[0033] S1-5, Analysis of Heterogeneous Relative Delay and Multi-Symbol Crosstalk; When the central processing node aligns and merges the signals from each access point according to the arrival time of the target UAV, the signals of non-target UAVs at each access point are offset relative to their own transmitter reference time. This offset is the heterogeneous relative delay. ; , The first , The drone to the first Propagation delay between access points.

[0034] Due to the existence of this heterogeneous relative delay, the sampling point of the non-target UAV signal deviates from the optimal position in the merging window, causing the interference between users to deteriorate from single-symbol interference under the synchronization assumption to multi-symbol crosstalk, thus generating multi-user asynchronous interference.

[0035] Through steps S1-1 to S1-5 above, the multi-user asynchronous transmission model established in this embodiment can simultaneously characterize the asynchronous arrival process of multiple UAV signals at multiple access points, the multi-AP alignment and merging process at the central processing node, and the multi-symbol interference mechanism caused by propagation delay differences between different UAVs. Compared with traditional transmitter designs based on ideal synchronization assumptions, this embodiment accurately models the signal transmission process in a real wide-area UAV asynchronous communication scenario, providing an accurate physical layer foundation for subsequent codebook optimization.

[0036] In non-cellular networks, access points typically do not independently perform final data detection. Instead, they transmit locally received samples to the central processing node (CPU) via a fronthaul link for joint processing. Since the fronthaul link capacity between the access point (AP) and the CPU is limited, this embodiment of the invention employs a compress-and-forward (CF) mechanism to compress and transmit the locally received signals from the AP.

[0037] Specifically, S2 includes: S2-1, Fronthaul capacity allocation; make This indicates the available fronthaul capacity from the access point (AP) to the central processing node (CPU) (in bits per second, or equivalent to the total number of bits per frame). Indicates the first The number of quantization bits allocated by each access point for each complex sample. When each frame contains There are 1 symbol block and the codeword length is 1 At that time, it is advisable to: ; The fronthaul capacity is evenly distributed across all multisampling points within each frame. The larger the value, the higher the quantization accuracy; The smaller the value, the higher the degree of compression.

[0038] S2-2, Perform additive quantization noise modeling (AQNM); The forward quantization noise model adopts an additive complex Gaussian noise model; Each access point receives samples. The CF forwarding signal is obtained after quantization and compression. Under the Additive Quantization Noise Model (AQNM), the forwarding signal can be equivalently represented as: , ; in The noise is the forward quantization noise, and it is assumed to be independent of the received signal and thermal noise.

[0039] S2-3, Calculation of quantization noise variance; The quantization noise variance is determined by both the fronthaul capacity and the average received power at the access point; the average received power is obtained by superimposing the thermal noise power on the product of the large-scale fading gain and the transmit power of all UAVs. If the first The average received power of the access points is The quantization noise variance is determined by both the fronthaul capacity and the average received power at the access point. ; ; in, For large-scale fading power gain, This represents the variance of thermal noise.

[0040] Quantization noise variance With fronthaul capacity There is a clear monotonic relationship: Current transmission capacity As the quantization noise variance increases, As the number of steps decreases, the system gradually degenerates into a near-lossless forward transmission. Current transmission capacity Quantization noise variance under constraints Increased forward propagation quantization noise has become a significant factor affecting the detection performance of the central processing node, and must be considered in conjunction with the transmitter codebook design phase.

[0041] Therefore, this embodiment incorporates the fronthaul capacity constraints in non-cellular networks into the transmitter codebook design process, considering the compression forwarding from the access point to the central processing node and the impact of quantization noise, making the codebook optimization objective more consistent with the actual deployment conditions of non-cellular networks. By jointly modeling the quantization noise variance with the codebook matrix, delay matrix, and large-scale fading matrix, the transmitter design is no longer limited to the ideal lossless fronthaul assumption, but achieves joint optimization of multi-user codebooks under the constraint of limited fronthaul capacity.

[0042] Specifically, S3 includes: S3-1, Alignment and Merging of Multiple Access Points and Construction of Detectable Quantities; After receiving the CF forwarding signals from all access points (APs), the central processing node processes the signals for the first... The drone performs codeword matching filtering and multiple access point (AP) merging.

[0043] To reflect the arrival time alignment in multiple access point (AP) merging, let the first... The first symbol The transmitter reference time corresponding to each chip is: ; For the target drone The central processing node is at the Each access point (AP) branch should read the CF forwarding sampling time aligned with the arrival time of the drone, i.e.: ; No. The drone in the first The quantity to be detected at each symbol time should be written as a weighted sum of the matched filter outputs after aligned sampling from multiple access points (APs): ; in, For the first The AP targets the first The combined weight of each drone, which can be taken as... or its normalized form This gives strong links a higher weight in the merging process.

[0044] S3-2, Decomposition of the quantity to be detected; CF forwarding signal And substituting the AP's received signal into the above equation. Because: ; Therefore, the quantity to be detected can be decomposed into: ; in, Indicates the first Target signal items of a drone; Indicates the first The drone is the first Asynchronous interference generated by multiple users from a single drone; This represents the equivalent noise after thermal noise is combined by multiple APs; This represents the equivalent noise after the forward quantization noise is combined by multiple APs.

[0045] S3-3, Target signal power calculation; Because each AP branch follows Aligned sampling ensures that the target signal returns to the same transmitter reference time in each AP branch. Its specific form is: ; If we expand further ,but: ; Under ideal sampling, codeword normalization, and large-scale channel modeling conditions, the target signal power can be approximated as the coherent combination result of multiple AP gains: .

[0046] S3-4, Calculation of total asynchronous interference power; Because the CPU has the first When merging drones, the first one is used. The arrival times of the drones at each AP, therefore the first... The drone in the first The interference sampling time in each AP branch is offset relative to its own transmitter reference time. ,Right now: ; Expand We can obtain: ; As can be seen from the above formula, interference between drones is not a direct superposition at a uniform moment, but rather occurs specifically targeting the drone. In a multi-AP alignment and merging window, the relative delay on each AP branch is considered. This determines its sampling position and codeword correlation. The total asynchronous interference power experienced by a drone can be denoted as: ; Under large-scale channel modeling, can be , Codeword cross-correlation and latency difference The calculations are performed together to reflect the total impact of asynchronous interference on each AP after being combined by the central processing node.

[0047] S3-5, Thermal noise power calculation; The equivalent noise resulting from the AP-side thermal noise after matched filtering by the central processing node and merging with multiple APs is denoted as […]. It can be written as: ; Under ideal sampling, codeword normalization, and large-scale channel modeling conditions, the first The thermal noise power of a drone can be approximated as the incoherent superposition of the thermal noise from each access point after weighting and combining: .

[0048] S3-6, Calculation of front-pass quantization noise power In the CF fronthaul architecture, the equivalent noise term after quantization noise is merged by the central processing node also needs to be considered. Its form is: ; Under ideal sampling, codeword normalization, and large-scale channel modeling conditions, the first Forward quantization noise power of a drone It can be approximated as the incoherent superposition of the quantization noise of each access point after being combined and weighted: .

[0049] Therefore, the target signal power Asynchronous interference power Thermal noise power and CF quantization noise power All of these are equivalent quantities after the merging of multiple APs, and together they constitute the numerator and denominator of the subsequent SINR expression.

[0050] S3-7, Equivalent signal-to-interference-plus-noise ratio expression; No. The equivalent SINR of a drone in the CF fronthaul architecture can be expressed as: .

[0051] In large-scale channel modeling, the target signal power, interference power, thermal noise power, and quantization noise power can all be derived from the codebook matrix. Delay matrix Large-scale fading matrix (Its elements are) ), fronthaul capacity and noise variance Calculated. Specifically, large-scale fading. It can replace the averaging effect of instantaneous channel power, thereby reducing the transmitter's dependence on instantaneous channel estimation.

[0052] Specifically, in S4, the multi-user codebook optimization problem is to maximize the minimum equivalent signal-to-interference-plus-noise ratio (SINR) among all UAVs. The constraints include codeword normalization, upper limit of transmit power, and the number of quantization bits and quantization noise variance determined by the fronthaul capacity.

[0053] Based on the equivalent signal-to-interference-plus-noise ratio (SINR) model established by S3, and with the objective of maximizing the minimum equivalent SINR among all UAVs, the specific form of the multi-user codebook optimization problem is as follows: ; ; ; The optimization problem described above takes into account the asynchronous correlation between codewords, the merging gain of multiple access points, and the quantization noise introduced by the limited fronthaul capacity. Therefore, it can more accurately describe the performance target of the asynchronous communication codebook design for UAVs in non-cellular networks.

[0054] Specifically, S5 includes: S5-1. Construct a codebook generation model that includes a heterogeneous graph attention task encoding network and a policy network; In the overall architecture of the model, Task coding network: Extracts topology and channel features of non-cellular networks to generate global task embeddings; Policy network: Generates new codebooks or adjusts codebook size based on task embedding.

[0055] Among them, the task coding network models the drones and access points in the non-cellular network as two types of nodes in a heterogeneous graph, and models the communication link between the drone and the access point as edges of a bipartite graph. The characteristics of a drone node include its location, maximum latency, and current codeword status. Access point node characteristics include access point location, average received power, and fronthaul capacity; Edge features include transmission distance Delayed transmission Large-scale decay And quantization noise parameters related to fronthaul compression.

[0056] The network calculates the importance weights of different links based on the aforementioned node and edge features, and obtains the global task embedding through multi-layer message passing. It is used to characterize the topology, channel characteristics, asynchronous delay features, and fronthaul capacity constraints in the current non-cellular configuration.

[0057] Policy networks with task embedding The current codebook state and latency information are used as inputs to generate a new codebook or codebook adjustment amount, and normalization processing is performed on the codewords of each UAV to ensure that they meet the requirements. .

[0058] In the aforementioned heterogeneous graph attention task coding network, this embodiment utilizes the topological relationship between the UAV and the access point to construct a heterogeneous graph structure. The UAV and access point are modeled as different types of nodes, and distance, propagation delay, large-scale fading, and fronthaul correlation parameters are used as edge features. This enables the codebook generation model to perceive the spatial structure and link differences of the non-cellular network. Through the graph attention mechanism, the importance weights of different links are adaptively calculated, allowing the model to fully extract network topology and channel features, generating task embeddings that characterize the global properties of the current non-cellular configuration. This provides highly discriminative environmental awareness information for the policy network to generate an adapted codebook.

[0059] S5-2. The above codebook generation model is trained using a meta-learning framework; During the training phase, multiple non-cellular configuration tasks are sampled from the task space. Each task consists of the UAV location, access point location, delay matrix T, large-scale fading matrix B, and fronthaul capacity. Composition; The model uses the equivalent SINR under the minimum forward architecture in S4 as the optimization objective, so that the generated codebook can simultaneously suppress asynchronous multi-user interference and forward quantization noise.

[0060] Based on the aforementioned meta-learning framework, a codebook optimization solution method based on graph attention networks and meta-learning is adopted. This method can quickly generate multi-user codebooks adapted to the current environment under different UAV locations, access point locations, latency distributions, and fronthaul capacities. By performing meta-training in a diverse task space, the model learns good parameter initialization, enabling it to generate high-performance codebooks with only a small number of gradient updates when facing new non-cellular network configurations. Compared with traditional scene-by-scene optimization methods, this significantly reduces computational complexity and time overhead, meeting the real-time update requirements of transmitter codebooks in multi-UAV wide-area access scenarios.

[0061] S5-3. Perform online deployment and configuration updates of the codebook generation model; When faced with a new non-cellular network configuration, the model uses the learned initialization parameters and task embeddings to quickly generate a multi-user transmit codebook that adapts to the current configuration. If necessary, it further improves the SINR under the minimum forward pass architecture through a small number of gradient updates and updates the transmitter configuration of the UAV terminal for subsequent data transmission.

[0062] The UAV-borne transmitter design method applied in this embodiment provides a system architecture for a UAV-borne transmitter system without cellular networks, as follows: Figure 3 As shown. The system includes Unmanned aerial vehicle launcher and Each UAV transmitter uses a distributed access point and a multi-user asynchronous transmission mechanism. The transmitter sequentially undergoes symbol modulation, resource mapping, and pulse shaping. Processing: Information bitstream Symbol sequence is obtained after symbol modulation. The resource mapping module configures the codebook based on the codebook configuration signal from the meta-learning codebook optimizer of the graph structure task encoding. The symbols are mapped to the corresponding transmission resources, then pulse-shaped filtered before being transmitted to the wireless channel; each UAV simultaneously transmits its own location information. Report to the codebook optimizer.

[0063] Because the propagation distances between each drone and different access points vary, the signals have independent propagation delays. After wireless channel transmission, the data is asynchronously superimposed at the same access point. Each access point... Receive superimposed signals and superimpose thermal noise. , forming a received signal Each access point will share its location information. The received signal is compressed and quantized via the fronthaul link before being transmitted to the central processing node. The central processing node aligns and merges the compressed and forwarded signals from each access point according to the arrival time of the target UAV, and performs joint detection processing.

[0064] Meanwhile, the meta-learning codebook optimizer for graph structure task coding is based on the location of each UAV. Location of each access point The network topology information is used to generate a multi-user transmission codebook adapted to the current configuration. The data is then distributed to each UAV transmitter to update the resource mapping configuration, thus forming a closed-loop control system of "topology awareness—codebook generation—signal transmission—joint reception". Through this architecture, the system can achieve wide-area access and reliable transmission for multiple UAVs, taking into account multi-user asynchronous interference, thermal noise, and forward quantization noise.

[0065] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks, characterized in that, include: Establish a multi-user asynchronous transmission model; in a non-cellular network including distributed access points and a central processing node, during the asynchronous transmission of uplink communication between each UAV and each access point, multi-user asynchronous interference originates from the difference in propagation distance from each UAV to different access points, as well as the heterogeneous relative delay introduced when the central processing node aligns and merges according to the arrival time of the target UAV. A compression forwarding mechanism is used to quantize and compress the received signals at each access point, and a fronthaul quantization noise model is established. The central processing node aligns and merges the compressed forwarding signals of each access point according to the arrival time of the target UAV, decomposes the quantity to be detected into the target signal, multi-user asynchronous interference, thermal noise and forward quantization noise, and establishes the equivalent signal-to-interference-plus-noise ratio model for each UAV. To maximize the minimum equivalent signal-to-interference-plus-noise ratio among all UAVs, a multi-user codebook optimization problem is constructed. Build and train a codebook generation model so that the model can generate a multi-user transmission codebook adapted to the current configuration based on the input network topology information and update the UAV transmitter configuration.

2. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 1, characterized in that, The fronthaul quantization noise model adopts an additive complex Gaussian noise model, and the quantization noise variance is jointly determined by the fronthaul capacity and the average received power of the access point; the average received power is obtained by superimposing the thermal noise power on the product of the large-scale fading gain and the transmit power of all UAVs.

3. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 1, characterized in that, The multi-user codebook optimization problem is to maximize the minimum equivalent signal-to-interference-plus-noise ratio (SINR) among all UAVs. The constraints include codeword normalization, upper limit of transmit power, and the number of quantization bits and quantization noise variance determined by the fronthaul capacity.

4. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 3, characterized in that, The number of quantization bits determined by the fronthaul capacity for: ; Quantization noise variance for: ; in, The available fronthaul capacity from distributed access points to the central processing node. The number of symbol blocks contained in each frame. The length of a complex digital character, For the first Average received power of each access point; , This represents the total number of access points.

5. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 1, characterized in that, No. The equivalent SINR of a drone in a fronthaul architecture is represented as follows: ； in, For the target signal power, This represents the total asynchronous interference power. Thermal noise power, This is the noise power for forward quantization.

6. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 5, characterized in that, The target signal of the decomposition of the quantity to be detected is represented as: ； in, For the first Target signal items of a drone; For the central processing node to the first The first drone, the first The combining weight of signals from each access point For the first The drone to the Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first The first drone digital character individual character symbols, For the first The drone in the first Data symbols sent by a symbol block, For symbol period, A unit energy pulse shaping function; The number of symbol blocks contained in each frame. This represents the total number of access points. The length of the complex digital character; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the coherent combining result of the gains of multiple access points under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Target signal power of a drone Approximate value.

7. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 5, characterized in that, The multi-user asynchronous interference of the decomposition of the quantity to be detected is represented as follows: ； in, For the first The drone is the first The drone in the first Asynchronous interference of one symbol; For the central processing node to the first The first drone, the first The combining weight of signals from each access point , The first The, the The drone to the Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first The first drone digital character individual character symbols, For the first The drone in the first Data symbols sent by a symbol block, For symbol period, For the first The drone is relative to the first The drone in the first The relative propagation delay of each access point A unit energy pulse shaping function; The number of symbol blocks contained in each frame. This represents the total number of access points. The length of the complex digital character.

8. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 5, characterized in that, The thermal noise of the decomposition of the detectable quantity is expressed as: ； in, For the first The drone in the first Thermal noise received by each symbol; For the central processing node to the first The first drone, the first The combining weight of signals from each access point For the first The drone to the Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character. For the first At each access point Additive white Gaussian noise at any given time; For the first The first symbol The transmitter reference time corresponding to each chip With the The drone to the Propagation delay between access points sum; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the incoherent superposition result of the thermal noise of each access point after merging and weighting under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Thermal noise power of a drone Approximate value.

9. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 5, characterized in that, The forward quantization noise of the decomposition of the quantity to be detected is expressed as: ； in, For the first The drone in the first Each symbol is subject to forward quantization noise; For the central processing node to the first The first drone, the first The combining weight of signals from each access point For the first The drone to the Complex channel coefficients for each access point For the first The first drone digital character The conjugate of a code character, "Indicates conjugate, For the first Access points Forward quantization noise at any given time; For the first The first symbol The transmitter reference time corresponding to each chip With the The drone to the Propagation delay between access points sum; In the equivalent signal-to-interference-plus-noise ratio (SINR) model, the incoherent superposition result of the quantization noise of each access point after merging and weighting under ideal sampling, codeword normalization, and large-scale channel modeling conditions is used as the first... Forward quantization noise power of a drone Approximate value.

10. The design method for an unmanned aerial vehicle (UAV) transmitter for non-cellular networks according to claim 1, characterized in that, Building and training the codebook generation model includes: Construct a codebook generation model that includes a heterogeneous graph attention task encoding network and a policy network; In this task coding network, the drones and access points in the non-cellular network are modeled as two types of nodes in a heterogeneous graph, and the communication link between the drone and the access point is modeled as edges in a bipartite graph. The policy network takes task embedding, current codebook state and latency information as input to generate new codebooks or codebook adjustments, and performs normalization processing on the codewords of each UAV. The codebook generation model is trained using a meta-learning framework with the equivalent SINR under the minimum forward architecture as the optimization objective. When configuring a new network, the codebook generation model uses learned initialization parameters and task embeddings to quickly generate an adapted multi-user launch codebook and update the UAV transmitter configuration.

Images