Search

CN-122028202-A - Unmanned aerial vehicle phase shifter switch and power control method based on graph neural network

CN122028202ACN 122028202 ACN122028202 ACN 122028202ACN-122028202-A

Abstract

The invention relates to an unmanned aerial vehicle phase shifter switch and power control method based on a graph neural network, which comprises the steps of constructing a communication system model, constructing a phase shifter disabling vector of each unmanned aerial vehicle and an optimization problem P1 of transmitting power of each E-UAV by taking the maximum energy efficiency of the system as a target, converting the problem P1 into a continuous optimization problem P2, modeling the communication system into an abnormal graph comprising four types of nodes, mapping initial characteristics of each node into a node embedding vector of a unified dimension, embedding the initial node of a current frame and the node of a previous frame, updating the embedding of a receiving node, feeding back the updated receiving node embedding to a corresponding E-UAV node, updating the embedding of the E-UAV node, and outputting the transmitting power of the E-UAV and all the phase shifter disabling decisions. The invention improves the energy efficiency of the system and can simultaneously ensure the QoS requirement of heterogeneous service.

Inventors

  • ZHENG FUCHUN
  • LAI LIFENG

Assignees

  • 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院)

Dates

Publication Date
20260512
Application Date
20260128

Claims (10)

  1. 1. The unmanned aerial vehicle phase shifter switch and power control method based on the graph neural network is characterized by comprising the following steps of: Step S10, a system modeling and optimization problem construction step, namely constructing a network auxiliary full duplex NAFD communication system model, wherein the system comprises T transmitting access points T-AP, R receiving access points R-AP, U URLLC unmanned aerial vehicles U-UAVs and E eMBB unmanned aerial vehicles E-UAVs; Step S20, a problem transformation and heterogeneous diagram construction step, in which a binary phase shifter deactivation variable is relaxed into a continuous variable, a problem P1 is transformed into a continuous optimization problem P2, the communication system is modeled into a different composition comprising four types of nodes of a T-AP node, an R-AP node, an E-UAV node and a U-UAV node based on a HGNN framework, wherein a pilot sequence sent by a transmitting node is used as an initial node characteristic, and a receiving node stacks an original pilot signal received by the receiving node, separates a real part from an imaginary part and is used as an initial node characteristic; step S30, a distributed node embedding updating and decision step utilizing time correlation, wherein the initial characteristics of each node are mapped into node embedding vectors with uniform dimension by utilizing a multi-layer perceptron MLP; The method comprises the steps of utilizing time correlation to embed an initial node of a current frame and a node of a previous frame, updating the embedding of a receiving node through node embedding update MLP, feeding back the updated receiving node embedding to a corresponding E-UAV node, aggregating the received embedding information by the E-UAV node and combining the node embedding of the previous frame, updating the embedding of the E-UAV node through the node updating MLP, outputting the transmitting power of the E-UAV through power control MLP based on the updated final node embedding, and controlling the MLP to output the phase shifter deactivation decision of all UAVs through a phase shifter; step S40, an unsupervised training step, namely constructing a comprehensive loss function comprising a negative system energy efficiency item and a plurality of constraint violation penalty items, and updating all MLP parameters through a gradient descent method.
  2. 2. The unmanned aerial vehicle phase shifter switching and power control method based on the graph neural network according to claim 1, wherein the step S10 specifically comprises: Step S101, constructing a system model and a quantization model, and letting Representing a set of APs, where And Respectively T-AP and R-AP, and the UAV set is made to be Wherein For the total number of unmanned aerial vehicles, And Representing a set of E-UAVs and U-UAVs, respectively, wherein each AP is equipped with a set of E-UAVs and U-UAVs A uniform planar array of antennas is formed, And (3) with Respectively representing the number of antenna units of the AP end antenna array in the x and y directions, and adopting the antenna array with the functions of Fully connected hybrid analog-digital architecture of individual radio frequency chains, wherein , For mode selection parameters, tx is transmit mode, rx is receive mode, each UAV Which belongs to the whole UAV set Equipped with a single unit of A UPA consisting of a root antenna, a single radio frequency chain and a low resolution ADC/DAC, And (3) with The antenna unit numbers of the UAV terminal antenna array in the x and y directions are respectively represented, and the UAV terminal antenna array Is denoted as antenna set The radio frequency chain is connected to all antennas through phase shifters, antenna activation/deactivation is achieved, and beamforming is achieved entirely in the analog domain.
  3. 3. The unmanned aerial vehicle phase shifter switching and power control method based on the graph neural network of claim 2, wherein the step S10 further comprises: step S102, modeling the distortion caused by the finite resolution ADC/DAC by using Bussgang model Representation applied to UAV Signal vector of the position The quantization operators of (a) are: ,(1) Wherein the method comprises the steps of , Representing the ADC/DAC selection variable, Representing the quantization of the ADC, Representing DAC quantization with distortion factor of Wherein Representing resolution, quantization noise according to Bussgang decomposition And (3) with Uncorrelated, will Circularly symmetric complex Gaussian noise approximately independent and distributed, i.e Wherein As the variance of the noise is the value of the variance of the noise, Is in combination with Identity matrix with same dimension and resolution of Is modeled as the power consumption of an ADC/DAC Wherein For UAVs Is a function of the ADC/DAC power consumption, Is the sampling frequency.
  4. 4. The unmanned aerial vehicle phase shifter switch and power control method based on the neural network of claim 3, wherein the step S10 further comprises a step S103 of establishing an AP With UAV Channel model between: ,(2) Wherein the method comprises the steps of , Indicating either the UL or UL selection variable, Represents DL, and Which means that the UL is such that, Is the rice factor (rice factor) and, And Which represent DL and UL channels respectively, Representing a complex set, path loss PL is modeled as Wherein Is AP (access point) With UAV The distance between the two plates is set to be equal, And The path loss index and the loss value at the 1 meter reference point, respectively; Representation of Represents the LoS or NLoS selection variable, Representing LoS conditions Representing NLoS condition, AP With UAV The LoS probability of the link between is expressed as: wherein And Is a constant that depends on the environment and, Is the elevation angle; Order the And The elevation angle and the azimuth angle of the propagation path are respectively represented, and the guide vector of the UPA is: , (3) Wherein the method comprises the steps of , Is the carrier wavelength of the light, Is the antenna spacing; Is a horizontal/vertical element index; The LoS component in DL is expressed as Wherein And Respectively from the AP To UAV (unmanned aerial vehicle) Angle of arrival and angle of departure in the horizontal and vertical directions, for UL, Wherein And Representation slave UAV To AP AoA and AoD in horizontal and vertical directions, NLoS component Modeling as having independent co-distributed elements and compliance Is a matrix of (a); T-AP With R-AP The channel between is modeled as: ,(4) Wherein the method comprises the steps of The inter-AP PL is indicated as PL, Is the distance between the APs, Is a small scale debilitating channel with elements independently distributed and compliant, E-UAV With U-UAV The channel between them is given by: wherein Is an E-UAV With U-UAV The PL of the light beam between the two, Is the distance between UAVs, the LoS component is modeled as Wherein Representation slave E-UAV To U-UAV AoA in the horizontal and vertical directions of (C), Representation slave E-UAV To U-UAV AoD in both the horizontal and vertical directions; Since the frame duration is less than the channel coherence time, small scale debilitating components Constant within each frame, but varying between frames, due to temporal correlation, Correlation between successive frames, the first The small scale debilitating component evolution of a frame is expressed as: ,(5) Wherein the method comprises the steps of Is a coefficient of a time-dependent relationship, Is a zero-order Bessel function of the first class, and has the maximum Doppler frequency shift of Wherein Is the carrier frequency of the wave, Is the velocity of the UAV, Is the speed of light; Representing sampling intervals, random terms And (3) with Having the same dimensions, the elements thereof being independently and identically distributed and subject to 。
  5. 5. The unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 4, wherein the step S10 further comprises: step S104, a signal model and a performance index model are constructed: suppose U-UAV T-AP providing maximum reference signal received power in DL Association, similarly, E-UAV With R-AP providing minimum PL in UL Association, E-UAV The transmit signal in UL is: ,(6) Wherein the method comprises the steps of And Representing transmit power and analog transmit beamforming, respectively, and Quantization noise satisfies Wherein Quantizing noise variance for E-UAV end at R-AP The UL received signal at is: ,(7) Wherein the method comprises the steps of Is R-AP Is used for decoding E-UAV Is provided with a hybrid receive beamforming of (a) and (b), Is T-AP Is used for serving U-UAV And the T-AP employs equal power allocation among its associated U-UAVs, DL data symbols Satisfies the following conditions , Is R-AP An additive white gaussian noise vector at the location, and , Is AWGN variance, therefore, R-AP Decoding E-UAV The signal-to-interference-and-noise ratio of the signal is: (8) Wherein the method comprises the steps of Representing interference from other E-UAVs, Is associated with an E-UAV Correlated quantization noise term, and From T-AP to R-AP Is a disturbance of (1); U-UAV the received analog signal at (before passing through the ADC) is: ,(9) Wherein the method comprises the steps of Is a U-UAV AWGN vector at, and ,U-UAV The received signal after the ADC is given by: wherein the U-UAV End quantization noise is And is also provided with , To quantify the noise variance, therefore, in U-UAVs The DL SINR at can be written as: ,(10) Wherein the method comprises the steps of Representing interference from other T-APs serving the remaining U-UAVs, Is the equivalent quantization noise power associated with the desired DL signal, and From E-UAV to U-UAV Is a disturbance of (1); Order the Representing an initial UAV analog beamforming vector selected from a predefined codebook, employing codebook-based analog beamforming at UAV and AP, while AP-side digital beamforming vector is designed using zero-forcing method, at UAV With a single radio frequency chain, PS deactivation is performed by binary vectors Representation of wherein Indicate the first The PS (and antenna) are activated ) Or stop using% ) An effective analog beamforming vector is The number of activated PS is ; For UL long packet transmission, the achievable rate is For DL short packet transmission (URLLC), the achievable rate is approximately Wherein As an inverse of the gaussian Q function, ; And Respectively representing the block length and the target decoding error probability; E-UAV and U-UAV The total power consumption of (a) is respectively And Wherein Is the static circuit power of each radio frequency chain, And The power of each PS and each switch respectively, And Representing the power consumption of the DAC and ADC respectively, And An antenna shifter activation number for the E-UAV and U-UAV ends; Finally, system EE is defined as Wherein Is the total transmission rate of the system and Is the total power consumption of the system.
  6. 6. The unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 5, wherein the step S10 further comprises: Step S105, optimizing the phase shifter disabling vector of each unmanned aerial vehicle and the transmitting power of each E-UAV as: ,(11a) ,(11b) ,(11c) ,(11d) ,(11e) Wherein, the The set of vectors is disabled for the PS of all UAVs, Constraint (11 a) ensures that each E-UAV reaches the lowest UL rate for the UL transmit power variable set of E-UAVs Constraint (11 b) guarantees DL QoS for each U-UAV, with a requirement of size Is within the delay bound The inner successful delivery, constraint (11 c) limits UL transmit power, while constraints (11 d) - (11 e) together define the upper and lower limits of the binary PS activation variables and their number of activations.
  7. 7. The unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 6, wherein the step S20 comprises: step S201, relaxing the binary PS off variable into a continuous variable, i.e . Accordingly, constraints (11 d) and (11 e) are relaxed as (12) After relaxation, the EE maximization problem can be restated as: (13)。
  8. 8. the unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 7, wherein the step S20 further comprises: For each transmitter node Its pilot sequence is used as an initial feature: ,(14) Wherein the method comprises the steps of , Representing AP or UAV selection variables, each transmitting node sequentially transmits its pilot symbols during the pilot phase ,R-AP For the first The received signal of each pilot symbol is: ,(15) Wherein the method comprises the steps of Transmitting the first for E-UAV end The number of pilot symbols is one, Representing the quantization noise of the DAC, Is R-AP AWGN vector at the position, U-UAV For the first The received signal of each pilot symbol is: ,(16) Wherein the method comprises the steps of And Respectively represent U-UAV ADC quantization noise and AWGN vector at; For receiver nodes Using stacked original received pilot signals as initial features, in particular, defining Is the stacked original pilot signal and is constructed As an initial feature, wherein In order to take the real part of the operation, In order to take the imaginary part of the operation, As a set of real numbers, 。
  9. 9. The unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 8, wherein the step S30 comprises: To obtain node embedding with unified dimension D, a multi-layer awareness mechanism based node embedding mapping is applied: ,(17) Wherein the method comprises the steps of Is an MLP for feature mapping; then, using the time correlation, the receiver node Once per frame, i.e.: ,(18) Wherein the method comprises the steps of Representing an MLP for node embedding updates; After updating the receiver node's embeddings, these node embeddings are fed back to the respective E-UAV nodes so that each E-UAV can take into account its impact on other network entities in decision making, in particular E-UAVs Aggregating received embeddings as Wherein Representing aggregated information from the associated receiver node, Is a feedback neighbor set, i.e. E-UAV R-AP and U-UAV nodes of (C-V), and Finally, the aggregated information is fused to update the E-UAV Is embedded in the node: (19) after the final node embeds the updates, each E-UAV is utilized Is embedded to output its UL transmit power: (20) Wherein the method comprises the steps of Is an MLP that outputs (normalized) power coefficients, Is a Sigmoid function, ensure ; Similarly, each UAV is utilized Is embedded into the output PS deactivation decision: (21) Wherein the method comprises the steps of Is an MLP that outputs a relaxed PS disable vector, As an indication function from element to element, Is a predefined threshold, a binarization penalty term is applied to encourage binary output during training Incorporating into the loss function; Explicit CSI is not needed in the unmanned aerial vehicle phase shifter switch and power control method based on the graph neural network, so that the channel estimation overhead is eliminated. To update node embeddings of E-UAVs, node embeddings from R-APs and U-UAVs need to be fed back, thus, incurring the overhead of message delivery, which is in conjunction with the embedment dimension Proportional, therefore, the total signaling overhead for pilot transmission and messaging is 。
  10. 10. The unmanned aerial vehicle phase shifter switching and power control method based on the neural network of claim 1, wherein the step S40 comprises: defining a loss function in equation (21), wherein Representing a trainable parameter set, the penalty comprising a negative EE target, a ReLU-based penalty term, a penalty term that violates UL/DL QoS constraints and transmit power and PS deactivation constraints, and weights corresponding to these terms , , , , A kind of electronic device During offline training, global CSI is required Thus, this assumption is made that a centralized controller gathers for training purposes Parameters are updated by gradient descent: wherein Is the learning rate.

Description

Unmanned aerial vehicle phase shifter switch and power control method based on graph neural network Technical Field The invention relates to the technical field of wireless communication, in particular to an unmanned aerial vehicle phase shifter switch and a power control method based on a graph neural network. Background Cellular connection drone (unmanned AERIAL VEHICLE, UAV) communications provide wide area reliable connectivity using existing cellular infrastructure. In such systems, the Uplink (UL) is typically directed towards enhanced mobile broadband (enhanced mobile broadband, eMBB) (e.g., real-time video) and therefore requires high data rates, while the Downlink (DL) mainly carries ultra-reliable and low-latency communications, URLLC traffic (e.g., command control signals) with strict reliability and delay constraints. To support heterogeneous UL/DL traffic demands, network-assisted full-duplex (NAFD) techniques support UL and DL concurrent transmissions to improve spectral efficiency. However NAFD introduces cross-link interference that is further exacerbated by strong line-of-sight (LoS) propagation in the UAV scenario of cellular connections. Although multi-antenna beamforming techniques may be used to suppress interference, UAVs are limited by size, weight, and energy, it is difficult to deploy large-scale antennas and to afford the power consumption of high-complexity signal processing, thereby affecting their cruising ability. Therefore, how to control power consumption while suppressing interference and meeting quality of service (QoS) requirements of heterogeneous services becomes a key issue in designing an airborne antenna architecture and beamforming scheme of a UAV. Several energy efficiency optimization schemes for eMBB and URLLC coexistence networks have been proposed by the existing studies. However, these approaches often do not adequately consider energy efficient on-board antenna architecture designs suitable for UAVs, and thus are difficult to apply directly to cellular connected UAV scenarios. Furthermore, if these methods are extended to a NAFD-supported UAV network, serious challenges are faced, mainly due to the strong cross-link interference in the system (e.g., eMBB data transmission in UL would severely interfere with URLLC data reception in DL), which constitutes a key performance bottleneck. On the other hand, there have been work to improve the energy efficiency of mimo systems by hybrid beamforming architecture. However, most of these studies do not take into account the power-performance tradeoff associated with the limited resolution analog-to-digital/digital-to-analog converter, which reduces hardware power consumption, but introduces quantization noise that significantly degrades channel estimation and data transmission performance. Furthermore, for systems containing eMBB only or URLLC only, studies have been done to investigate energy efficiency scheme designs based on limited resolution DACs or ADCs, respectively. However, the existing work mostly carries out isolation treatment on two service types, and the attention on heterogeneous service quality requirements is insufficient when the two service types coexist, and meanwhile, the phase shifter on-off control is less combined with the low-resolution converter so as to further reduce the system power consumption. In the method level, the existing scheme depends on centralized optimization, so that higher signaling overhead and calculation burden are brought, frequent re-solving along with network state change is needed, and expandability and instantaneity are limited. Therefore, the UAV airborne antenna architecture and the beam forming scheme which can effectively inhibit interference and control power consumption and simultaneously meet eMBB and URLLC heterogeneous service quality requirements are designed, and the UAV airborne antenna architecture and the beam forming scheme have important research significance and practical value. Disclosure of Invention The invention provides an unmanned aerial vehicle phase shifter switch and a power control method based on a graph neural network, which aim to effectively improve the energy efficiency of a system in a distributed and low-overhead mode and simultaneously guarantee the QoS requirements of heterogeneous services. In order to achieve the above purpose, the present invention provides an unmanned aerial vehicle phase shifter switch and power control method based on a graph neural network, the method comprising the following steps: S10, constructing a network-assisted full duplex NAFD communication system model, wherein the system comprises T transmitting access points T-AP, R receiving access points R-AP, U unmanned aerial vehicles U-UAVs and E unmanned aerial vehicles E-UAVs; Step S20, a problem transformation and heterogeneous diagram construction step, in which a binary phase shifter deactivation variable is relaxed into a continuous variable, a problem P1