Unmanned aerial vehicle dual-polarized antenna multipath interference suppression and polarization mismatch joint optimization algorithm

Abstract

The invention provides an unmanned aerial vehicle dual-polarized antenna multipath interference suppression and polarization mismatch joint optimization algorithm, and relates to the technical field of low-altitude communication, and the algorithm carries out the coupling modeling of a multipath interference problem and a polarization mismatch problem through building a unified dual-polarized channel model which considers multipath propagation and time-varying polarization characteristics; an adaptive polarization state estimation method based on received signal covariance matrix feature analysis is provided; a combined space-time-polarization domain adaptive filter structure is designed; a joint optimization objective function containing a minimum mean square error and a maximum polarization matching degree (MPM) is constructed; a low-complexity adaptive weight updating rule based on a recursive least square (RLS) thought and integrated with polarization constraints is deduced. According to the algorithm, interference suppression and polarization matching are collaboratively optimized through a unified framework, the complexity is reduced by adopting recursive calculation, rapid convergence and strong environmental adaptability are realized, and the low-altitude communication performance of the unmanned aerial vehicle is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of low-altitude communication technology, and in particular to a joint optimization algorithm for multipath interference suppression and polarization mismatch of dual-polarized antennas of unmanned aerial vehicles (UAVs). Background Art

[0002] Unmanned aerial vehicles (UAVs) are increasingly being used in low-altitude areas, encompassing key areas such as urban logistics and distribution, emergency communications, and precision agriculture monitoring. With the increasing complexity of application scenarios and the diversification of mission requirements, UAV communication systems face unprecedented challenges. Dual-polarization antenna technology, by simultaneously transmitting information using two orthogonal polarization channels, horizontal (H) and vertical (V), significantly improves spectrum efficiency and channel capacity, and has become a key enabling technology for high-performance UAV communication systems.

[0003] However, when drones operate in complex low-altitude environments such as urban canyons and hilly areas, their communication link performance is severely constrained by two core factors: multipath interference and polarization mismatch. Multipath interference occurs when wireless signals are reflected, diffracted, and scattered by obstacles such as the ground, buildings, and vegetation, forming multiple propagation paths with different delays, amplitudes, and phases. These paths eventually overlap at the receiving end, causing severe time dispersion and frequency-selective fading of the signal, corrupting the signal waveform and degrading communication quality. Polarization mismatch stems from the inevitable attitude changes (roll, pitch, and yaw) of drones during flight, which prevent the polarization directions of the transmitting and receiving antennas from maintaining ideal alignment. This results in significant attenuation of the received signal energy and may even cause the communication link to be interrupted. These two effects often coexist and couple with each other, forming the main bottleneck for low-altitude drone communications.

[0004] Existing technical solutions generally tend to treat multipath interference suppression and polarization mismatch as two independent issues, failing to fully understand their inherent coupling in complex low-altitude channels. They also lack a unified framework for collaborative optimization that leverages the spatial, temporal, and polarization dimensions of dual-polarized antennas. Consequently, achieving globally optimal communication performance is difficult. Summary of the Invention

[0005] In view of the shortcomings of the existing technology, the purpose of the present invention is to provide a joint optimization algorithm for multipath interference suppression and polarization mismatch of UAV dual-polarization antennas to solve the problems raised in the above-mentioned background technology. The present invention effectively solves the difficult problem of multipath interference and polarization mismatch coupling in UAV low-altitude communication through innovative joint optimization design and low-complexity implementation, and shows significant advantages in communication performance, computing efficiency, environmental adaptability and deployment feasibility, providing key technical support for improving the operation capability of UAVs in complex environments.

[0006] In order to achieve the above-mentioned purpose, the present invention is implemented through the following technical solutions: a joint optimization algorithm for multipath interference suppression and polarization mismatch of dual-polarized antennas of unmanned aerial vehicles, which is composed of multiple functional modules, including a dual-polarized time-varying channel modeling and signal representation module, an adaptive polarization state estimation module, a joint space-time-polarization filter module, a joint optimization criterion and objective function construction module, a low-complexity adaptive weight update module and a polarization branch dynamic merging module. The dual-polarized time-varying channel modeling and signal representation module is used to receive the baseband signal of the dual-polarized antenna array, and describes the multipath propagation, delay spread, and multi-spectrum based on a unified model. The system is configured to eliminate the channel effects of the frequency shift and polarization effects that change with the attitude of the drone; the adaptive polarization state estimation module estimates the main polarization state parameters of the current signal or equivalent polarization channel matrix components in real time based on the second-order statistical characteristics of the received signal; the low-complexity adaptive weight update module iteratively updates the weight vector of the joint filter through an adaptive algorithm to make it approach the optimal solution of the joint optimization objective function and quickly track channel changes; the polarization branch dynamic merging module performs weighted merging of the horizontal (H) and vertical (V) polarization branch output signals processed by the joint filter based on their real-time estimated signal quality to obtain the final optimized output signal.

[0007] Furthermore, the operation of the dual-polarization time-varying channel modeling and signal representation module includes the following:

[0008] Assume that the drone is equipped with an array containing N pairs of orthogonal dual-polarized antenna units (a total of 2N channels). At time t, the horizontal (H) and vertical (V) polarized signal components r received by the nth unit pair are: n,H (t) and r n,V (t) can be expressed as:

[0009]

[0010] Where: s(t)=[s H (t),s V (t)] T is the dual-polarized signal vector sent by the transmitter; L is the number of resolvable multipath paths; τ l is the propagation delay of the lth path; n n (t)=[n n,H (t),n n,V (t)] T is the additive white Gaussian noise vector received by the n-th unit pair; Hn,l(t) is the time-varying polarization channel matrix of the l-th path received by the n-th unit pair.

[0011] Furthermore, H n,l(t) represents the path's fading, phase rotation, angle of arrival, and the polarization conversion relationship between the transmitter and the nth receiver:

[0012]

[0013] where α n,l (t) and φ n,l (t) is the time-varying amplitude and phase of path l to unit n, h XY,n,l (t)(X,Y∈{H,V}) is the time-varying polarization transfer coefficient. The signals of all N unit pairs and M time snapshots (sampling points) are stacked to form a 2NM×1 joint space-time received signal vector x(k), where k is the discrete time index:

[0014]

[0015] And the vector x(k) contains rich information in the spatial domain, time domain and polarization domain.

[0016] Furthermore, the adaptive polarization state estimation module considers the average received signal r(t) of a single unit pair or array in a short time using the covariance matrix of the received signal, wherein the received signal vector r(k)=[r H (k),r V (k)] T , whose polarization covariance matrix is ​​defined as:

[0017]

[0018] In order to adapt to the time-varying channel, the covariance matrix is ​​estimated recursively to obtain the estimated value

[0019]

[0020] Where β (0 < β < 1) is the forgetting factor, which is used to adjust the contribution weights of historical data and current data to the estimation results. When the value is close to or far from 1, the smoothing effect is good but the tracking speed is slow.

[0021] Furthermore, the estimated Perform eigenvalue decomposition:

[0022]

[0023] Where Λ(k) is a diagonal matrix containing eigenvalues ​​λ1(k) ≥ λ2(k), U(k) = [u1(k), u2(k)] is the corresponding eigenvector matrix, the main eigenvector u1(k) corresponds to the main polarization direction of the signal, and the relative sizes of the eigenvalues ​​λ1(k) and λ2(k) represent the degree of polarization.

[0024] Furthermore, the joint space-time-polarization filter module provides a 2NM×1 joint filter weight vector w(k) that acts on the joint space-time received signal vector x(k) to generate a filter output:

[0025] y(k)=wH(k)x(k)

[0026] The filter w(k) structure simultaneously performs weighted processing on signal components from different antenna elements, different time snapshots, and different polarization directions, achieving multipath interference suppression (using space-time differences) and polarization matching (using polarization information) within a unified framework.

[0027] Furthermore, a joint optimization objective function J(w) is constructed by a joint optimization criterion and an objective function construction module. This function minimizes the mean square error (MMSE) between the filter output y(k) and the desired signal d(k) while maximizing the polarization matching of the output signal. The formalized objective function is as follows:

[0028]

[0029] Among them: the first is the standard mean square error (MMSE) term, the goal is to suppress interference and noise so that the output signal approaches the expected signal d(k); the second term is the polarization mismatch penalty term, which is used to guide the filter weight w so that the polarization state of the output signal is consistent with that given by The estimated main polarization direction u1(k) matches; the third term λ r ||w|| 2 is the regularization term, where λ p and λ r is a non-negative weighting coefficient to weigh the importance of each item.

[0030] Furthermore, the low-complexity adaptive weight update module adopts an adaptive update algorithm based on recursive least squares (RLS), and its update rule includes the following steps:

[0031] a) Calculate the gain vector k(k):

[0032]

[0033] b) Calculate the prior error e(k):

[0034] e(k)=d(k)-w H (k-1)x(k)

[0035] c) Update the filter weight w rls (k):

[0036] wrls (k)=w(k-1)+k(k)e(k)

[0037] d) Update the inverse of the covariance matrix P(k):

[0038]

[0039] Where λ is the forgetting factor of RLS (0.95≤λ<1), and the complexity of the RLS algorithm is O((2NM) 2 ).

[0040] Furthermore, the temporary weight w is obtained in each RLS iteration update rls After (k), a correction step is performed to adjust the weights in the direction of satisfying the polarization matching constraint. The adjustment method is one of the following:

[0041] a) Correction based on gradient descent

[0042]

[0043] b) Projection-based correction:

[0044] w(k)=w rls (k)+α(k)(P proj (k)–I)w rls (k)

[0045] where μ pol (k) or α(k) is the adaptive step size or coefficient that controls the polarization correction strength, is the gradient of the polarization mismatch penalty term with respect to the weight, P proj (k) is based on The constructed projection matrix, I is the same as P proj (k) Identity matrices of the same dimension.

[0046] Furthermore, the polarization branch dynamic merging module is based on y H (k) and y V (k) is adaptively combined based on the real-time signal-to-interference-and-noise ratio (SINR) of the two branches. First, the SINR of the two branches is estimated:

[0047]

[0048] where fest(·,·) is the SINR estimation operator, which can be implemented by various standard methods, for example, by calculating the ratio of the desired signal power to the residual error and noise power, i.e. The expected signal d(k) can be obtained through training sequence or decision feedback. Then the normalized combining weight is calculated:

[0049]

[0050] The final optimized output signal is:

[0051]

[0052] Beneficial effects of the present invention:

[0053] 1. This joint optimization algorithm for multipath interference suppression and polarization mismatch for dual-polarized antennas in unmanned aerial vehicles (UAVs) combines multipath interference suppression and polarization mismatch compensation within a unified framework by constructing a unified signal model and a joint optimization objective function. This approach overcomes the suboptimal nature of traditional methods, which often separate these approaches. Simulation results demonstrate that, in complex low-altitude scenarios such as typical urban canyons, the proposed algorithm can reduce the bit error rate (BER) by 1-2 orders of magnitude compared to traditional STAP and polarization diversity methods. Alternatively, it can reduce the required signal-to-noise ratio by 3-5 dB while maintaining the same BER requirement, significantly improving communication reliability and coverage.

[0054] 2. This invention adopts a recursive update mechanism based on the RLS concept and cleverly integrates polarization constraints to control the core computational complexity to O((NM) 2 ) level, which is much lower than the O((NM) of the traditional STAP algorithm 3 Compared with STAP, the computational complexity can be reduced by more than an order of magnitude, significantly reducing the requirements for the UAV's onboard processor capabilities, making it possible to achieve real-time processing on resource-constrained UAV platforms.

[0055] 3. The PC-RLS algorithm employed in this paper inherits the fast convergence advantage of the RLS algorithm. Furthermore, adaptive polarization state estimation and a dynamic merging strategy enable it to rapidly track dynamic changes in low-altitude channels and UAV attitudes. This algorithm converges much faster than LMS and offers lower steady-state error. It can adapt to channel changes within milliseconds, ensuring communication link stability in highly dynamic environments.

[0056] 4. By explicitly incorporating polarization matching into the optimization objective and using the estimated polarization state to guide weight updates, this algorithm is highly robust to polarization mismatch. Even under severe polarization mismatch (e.g., mismatch angle > 60°) or dramatic attitude changes (e.g., roll / pitch angle > 30°), it maintains high output SINR and link success rate, with far less performance degradation than traditional methods. This significantly enhances UAV communication reliability in various flight attitudes.

[0057] 5. This invention does not rely on specific multipath channel models or interference type assumptions. Through adaptive processing, it can cope with a variety of complex low-altitude propagation environments. Whether in urban high-rise buildings, streets, suburban open spaces, or indoor environments, this algorithm demonstrates superior and robust anti-interference performance, demonstrating excellent versatility in all scenarios.

[0058] 6. This invention's lower computational complexity and faster convergence speed translate to lower processor load and energy consumption. The estimated power consumption of this algorithm is significantly lower than that of STAP and comparable ML methods, helping to reduce UAV payload power consumption and extend flight operations.

[0059] 7. The algorithm provided by the present invention mainly relies on the received signal itself for adaptive processing and state estimation, and has low requirements for channel prior knowledge (such as precise arrival angle and interference direction), which is more in line with the situation of incomplete information in practical applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of the joint optimization algorithm for multipath interference suppression and polarization mismatch of the dual-polarized antenna of the UAV of the present invention;

[0061] Figure 2 Schematic diagram of multipath propagation and polarization mismatch of the dual-polarized antenna of the present invention;

[0062] Figure 3 Schematic diagram of the joint filter structure;

[0063] Figure 4 Schematic diagram of the simulation environment for algorithm performance evaluation;

[0064] Figure 5 The figure shows the BER performance comparison of different algorithms in a simulated urban canyon multipath and polarization mismatch environment.

[0065] Figure 6 is the algorithm performance under different polarization mismatch levels;

[0066] Figure 7 Convergence speed comparison of the algorithm provided by the present invention;

[0067] Figure 8 Comparison of the communication link success rate under different UAV posture changes provided by the present invention;

[0068] Figure 9 Comparison of algorithm performance in different multipath environments in the embodiments of the present invention;

[0069] Figure 10 Comparison of the computational complexity of the algorithm provided by this invention and the resource consumption of the drone. DETAILED DESCRIPTION

[0070] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods.

[0071] See also Figures 1 to 10 , the present invention provides the following technical solutions:

[0072] Example 1 Joint Optimization Algorithm for Multipath Interference Suppression and Polarization Mismatch of Dual-Polarized UAV Antennas The algorithm of the present invention is mainly composed of the following functional modules, which work together to achieve the joint optimization goal:

[0073] (1) Dual-polarization time-varying channel modeling and signal representation module: It is responsible for receiving the baseband signal from the dual-polarization antenna array and describing the channel based on a unified model, including multipath propagation, delay spread, Doppler shift, and polarization effects that change with the UAV's attitude.

[0074] (2) Adaptive polarization state estimation module: Based on the second-order statistical characteristics (covariance matrix) of the received signal, it estimates the main polarization state parameters of the current signal or the equivalent polarization channel matrix components in real time, providing a basis for subsequent polarization matching.

[0075] (3) Joint space-time-polarization filter module: A joint filter structure covering all elements of the antenna array (including H and V polarization) and multiple time taps (snapshots) is designed, and its weight vector acts on the space, time and polarization dimensions simultaneously.

[0076] (4) Joint optimization criterion and objective function construction module: Define a joint optimization objective function that comprehensively considers the interference suppression effect (such as the signal-to-interference-and-noise ratio (SINR) of the output signal or the mean square error (MSE) with the desired signal) and the degree of polarization matching (such as the polarization matching factor (PMF)).

[0077] (5) Low-complexity adaptive weight update module: A computationally efficient adaptive algorithm (e.g., improved RLS) is derived to iteratively update the weight vector of the joint filter so that it approaches the optimal solution of the joint optimization objective function and can quickly track channel changes. This update rule needs to embed polarization matching constraints or guidance.

[0078] (6) Polarization branch dynamic merging module: The horizontal (H) and vertical (V) polarization branch output signals processed by the joint filter are weighted and combined according to their real-time estimated signal quality (such as SINR) to obtain the final optimized output signal.

[0079] This embodiment also provides the technical details and working principles of each of the above modules, which are as follows:

[0080] (1) Dual-polarization time-varying channel modeling and signal representation module

[0081] Assume that the UAV is equipped with an array containing N pairs of orthogonal dual-polarized antenna units (a total of 2N channels). At time t, the horizontal (H) and vertical (V) polarized signal components r received by the nth unit pair are: n,H (t) and r n,V (t) can be expressed as:

[0082]

[0083] Where: s(t) = [s H (t),s V (t)]; T is the dual-polarization signal vector sent by the transmitter; L is the number of resolvable multipath paths; τ l is the propagation delay of the lth path; n n (t)=[n n,H (t),n n,V (t)] T is the additive white Gaussian noise vector received by the nth unit pair; H n,l (t) is the time-varying polarization channel matrix of the lth path received by the nth element pair.

[0084] H n,l (t) represents the path's fading, phase rotation, angle of arrival, and the polarization conversion relationship between the transmitter and the nth receiver:

[0085]

[0086] where α n,l (t) and φ n,l (t) is the time-varying amplitude and phase of path l to unit n, h XY,n,l (t)(X,Y∈{H,V}) is the time-varying polarization transfer coefficient. Its value depends not only on the propagation environment but also on the attitude of the UAV relative to the signal source. It directly reflects the polarization mismatch effect. The signals of all N unit pairs and M time snapshots (sampling points) are stacked to form a 2NM×1 joint space-time received signal vector x(k), where k is the discrete time index:

[0087]

[0088] And the vector x(k) contains rich information in the spatial domain, time domain and polarization domain.

[0089] (2) Adaptive polarization state estimation module

[0090] The covariance matrix of the received signal is used to consider the received signal r(t) of a single unit pair (or the average of the array) in a short period of time. Its polarization covariance matrix is ​​defined as:

[0091]

[0092] The characteristic structure of this matrix contains the average polarization state information of the signal. In order to adapt to the time-varying channel, the covariance matrix is ​​estimated recursively:

[0093]

[0094] Where β (0 < β < 1) is the forgetting factor, which is used to adjust the contribution weights of historical data and current data to the estimation results. When the value is close to or far from 1, the smoothing effect is good but the tracking speed is slow.

[0095] 5. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 4 is characterized in that: Perform eigenvalue decomposition:

[0096]

[0097] Where Λ(k) is a diagonal matrix containing eigenvalues ​​λ1(k)≥λ2(k), U(k)=[u1(k),u2(k)] is the corresponding eigenvector matrix, the main eigenvector u1(k) usually corresponds to the main polarization direction of the signal, and the relative size of the eigenvalues ​​λ1(k) and λ2(k) reflects the degree of polarization. 1( k) or the entire matrix to construct polarization matching constraints in subsequent optimization.

[0098] (3) Joint space-time-polarization filter module

[0099] Provide a 2NM×1 joint filter weight vector w(k) and apply it to the joint space-time received signal vector x(k) to generate the filter output:

[0100] y(k)=wH(j)x(k)

[0101] The structure of the filter w(k) allows for simultaneous weighted processing of signal components from different antenna elements, different time snapshots, and different polarization directions, thereby enabling multipath interference suppression (using space-time differences) and polarization matching (using polarization information) within a unified framework.

[0102] (4) Joint optimization criteria and objective function construction module

[0103] A joint optimization objective function J(w) is constructed, which aims to minimize the mean square error (MMSE) between the filter output y(k) and the desired signal d(k), while maximizing the polarization matching of the output signal. A formalized objective function is as follows:

[0104]

[0105] Among them: the first term E[|d(k)-w H x(k)| 2 ] is the standard MMSE term, the goal is to suppress interference and noise, so that the output signal is close to the expected signal d(k) (which can be obtained through training sequence or blind decision feedback), the second term is the polarization mismatch penalty term, which is used to guide the filter weight w so that the polarization state of the output signal is consistent with the estimated main polarization direction u1(k) or The desired polarization state is defined to match.

[0106] This embodiment provides a design solution:

[0107]

[0108] or:

[0109]

[0110] Among them S p Is a selection / mapping matrix. Specifically, it can be a 2×2NM matrix used to extract an equivalent 2×1 weight vector related only to the polarization dimension from the complete space-time-polarization weight vector w. For example, this is achieved by taking a weighted average of the different components of w. The smaller the value of this term, the higher the polarization matching degree. The smaller the value of this term, the higher the polarization matching degree. The third term λ r ||w|| 2 is a regularization term (Tikhonov regularization), which is used to control the size of the weight vector to prevent overfitting and improve numerical stability, λ r is the regularization parameter. p It is a trade-off parameter used to balance the importance between interference suppression performance and polarization matching performance.

[0111] Minimizing J(w) means concentrating the energy of the filter output signal on the main polarization direction of the current channel as much as possible while suppressing interference, thereby jointly solving the multipath interference and polarization mismatch problems.

[0112] (5) Low-complexity adaptive weight update module

[0113] An adaptive update algorithm based on the recursive least squares (RLS) concept is used, and polarization constraints are incorporated. The standard RLS algorithm update rule is: Calculate the gain vector:

[0114]

[0115] Calculate the a priori error:

[0116] e(k)=d(k)-w H (k-1)x(k)

[0117] Update the filter weights:

[0118] w rls (k)=w(k-1)+k(k)e(k)

[0119] Update the inverse of the covariance matrix P(k) = R(k):

[0120]

[0121] Where λ is the forgetting factor of RLS (0.95≤λ<1), and the complexity of the RLS algorithm is approximately O((2NM) 2 ).

[0122] In order to integrate the polarization matching constraint into the RLS update, this embodiment provides a polarization constraint guided RLS method: in each RLS iterative update, a temporary weight w is obtained. rls After (k), an additional correction step is introduced to adjust the weights in the direction that satisfies the polarization matching constraint:

[0123]

[0124] Or use the projection method to project the weight vector into a subspace that is more consistent with the desired polarization state:

[0125] w(k)=w rls (k)+α(k)(P proj (k)-I)w rls (k)

[0126] where μ pol (k) or α(k) is the adaptive step size or coefficient that controls the polarization correction strength, is the gradient of the polarization mismatch penalty term with respect to the weight, P proj (k) is based on The constructed projection matrix is ​​I, which is the identity matrix. This approach explicitly incorporates polarization matching as one of the optimization objectives into the iterative process while maintaining the fast convergence characteristics of RLS, and the increased computational effort is relatively controllable. pol (k) or α(k) can be adjusted according to the estimated polarization mismatch degree or error change rate to obtain better stability and convergence.

[0127] (6) Polarization branch dynamic merging module

[0128] According to y H (k) and y V(k) is adaptively combined based on the real-time signal-to-interference-and-noise ratio (SINR) of the two branches. First, the SINR of the two branches is estimated:

[0129]

[0130] where fest(·) is the SINR estimation operator, and then the normalized combining weight is calculated:

[0131]

[0132] The final optimized output signal is:

[0133]

[0134] The above formula uses square root weights to approximate the effect of MRC combining, but linear weights can also be used directly. This dynamic combining strategy ensures that even when the instantaneous channel quality of a polarization branch is poor, the system can prioritize the branch with better quality, further improving overall communication performance.

[0135] In order to facilitate understanding of the above technical solution, this embodiment also provides a detailed description of the accompanying drawings, which are as follows:

[0136] Figure 1 This is a flowchart of the joint optimization algorithm for multipath interference suppression and polarization mismatch of UAV dual-polarization antennas, which clearly shows the logical relationship and data flow between the algorithm modules;

[0137] Figure 2 The figure shows the typical scenarios faced by drones when communicating in low-altitude complex environments (including buildings and the ground). The red arrow line in the figure represents the signal transmitted by the drone, and the blue square represents the ground receiving station. The signal not only reaches the receiving end through the direct path shown by the blue solid line, but also passes through multiple paths such as the orange dotted line (reflected by the building), the purple dotted line (reflected by the ground) and the brown dotted line (scattered by the building). These paths have different time delays and attenuations, constituting multipath interference. At the same time, the change in the attitude of the drone (schematically represented by the red solid line and the dotted line) causes the polarization direction of its antenna (red ellipse) to deviate from the ideal polarization direction of the ground station receiving antenna, forming a polarization mismatch (there is an angle difference θ between the blue ellipse and the red ellipse). p ), the rotating blue ellipse in the figure represents the actual polarization direction at the receiving end. Multipath effects and polarization mismatch jointly affect the quality of the received signal;

[0138] Figure 3 The figure shows the conceptual structure of the joint space-time-polarization adaptive filter proposed in this invention. The left side shows the input signal, which includes horizontal (H) and vertical (V) polarization components x from N antenna elements (N = 2 in the figure) and M time taps (M = 2 in the figure). H,n(km) and x V,n (km). These signals are respectively related to the corresponding complex conjugate weights w H,n (km) and w V,n (km) are multiplied together (the multiplication is not explicitly shown in the figure; it is implicit in the connection between the weight nodes and the summing node). All weighted signals are summed at the summing node ∑ to obtain the filter output y(k). The key is that all weights w (comprising the vector w(k)) are uniformly adaptively adjusted by the bottom "joint optimization module (PC-RLS)". This module uses the desired signal d(k) (used to calculate the error) and the polarization state estimated from the received signal. (for polarization constraint), and the filter output y(k) (for feedback error), and update the weight vector w(k) according to the PC-RLS algorithm rules to simultaneously optimize interference suppression and polarization matching performance;

[0139] Figure 4 The simulation platform architecture for evaluating the performance of the algorithm of the present invention is demonstrated. The simulation process mainly includes: 1) Simulation environment configuration: setting specific geographical scenes (such as city models), electromagnetic propagation parameters and interference source characteristics. 2) UAV model and trajectory: simulating the flight path of the UAV, real-time attitude changes (roll, pitch, yaw angle) and the characteristics of the dual-polarized antenna it carries. 3) Channel model: Based on the environmental configuration and the state of the UAV, a time-varying multipath channel (ray tracing or statistical models such as Saleh-Valenzuela can be used) and the corresponding time-varying polarization channel matrix H(t) are generated. 4) Algorithm implementation: The generated received signal is input into the joint optimization algorithm module proposed in the present invention and the existing technology algorithm module for comparison (such as traditional STAP, polarization diversity, and machine learning methods) for processing. 5) Receiving end processing: simulate signal acquisition and necessary baseband processing. 6) Performance Evaluation: The output of each algorithm was collected and a series of key performance indicators were calculated, such as bit error rate (BER), output signal-to-interference-and-noise ratio (SINR), algorithm convergence speed (measured by the mean square error (MSE) curve), computational complexity, polarization matching factor (PMF), and performance stability under different attitudes. By comparing these indicators, the superiority of the proposed algorithm over existing technologies was comprehensively evaluated.

[0140] Figure 5The figure shows a comparison of the bit error rate (BER) performance of the joint optimization algorithm proposed in the present invention and three representative existing technologies (traditional STAP, polarization diversity MRC, and machine learning-based methods) in a simulated urban canyon low-altitude communication scenario. The simulation scenario simultaneously takes into account significant multipath interference and polarization mismatch caused by changes in the attitude of the drone. The horizontal axis is the average signal-to-noise ratio (SNR) of the input signal, and the vertical axis is the BER (logarithmic scale). As can be seen from the figure, under the same SNR conditions, the BER of the algorithm of the present invention (blue solid circle) is significantly lower than that of the other three algorithms, especially in the medium and high signal-to-noise ratio areas, where the performance advantage is more obvious. This shows that the present invention can most effectively overcome the challenges of complex low-altitude channels and achieve more reliable communication by jointly optimizing multipath suppression and polarization matching. Although traditional STAP (red dotted square) can suppress some interference, it does not deal with polarization mismatch and its performance is second to none. Polarization diversity (green dotted triangle) can alleviate polarization mismatch but has limited effect under strong multipath. Machine learning-based methods (magenta dotted diamonds) may have better performance, but are usually accompanied by high complexity (not directly reflected in this figure);

[0141] Figure 6 The performance of the joint optimization algorithm of the present invention under different input signal-to-noise ratios (SNRs) and different degrees of polarization mismatch is demonstrated. The X-axis represents the input SNR (dB), the Y-axis represents the polarization mismatch angle between the transmitter and the receiver (0 degrees represents perfect match, 90 degrees represents complete orthogonal mismatch), and the Z-axis represents the improvement (dB) of the output signal signal-to-interference-and-noise ratio (SINR) after algorithm processing relative to the case of using only a single antenna without processing. The surface shows that even in the presence of significant polarization mismatch (for example, the mismatch angle reaches 60 degrees or even greater), the algorithm of the present invention can still provide significant SINR improvement over a wider SNR range. Although the performance decreases slightly with the increase of the mismatch angle, the decrease is much smaller than when no polarization compensation is performed. This proves that the algorithm can effectively combat polarization mismatch, maintain good communication performance under various mismatch conditions, and demonstrates its strong polarization adaptability;

[0142] Figure 7The figure compares the convergence speed performance of the joint optimization algorithm based on polarization constrained RLS (PC-RLS) proposed in the present invention with several other typical adaptive filtering algorithms (standard RLS, LMS, and a simulated STAP-like adaptive algorithm). The horizontal axis is the number of iterations of the algorithm, and the vertical axis is the normalized mean square error (NMSE, usually expressed in a logarithmic scale). The faster the convergence speed, the steeper the curve drops; the lower the steady-state error, the lower the platform the curve eventually reaches. The figure shows that the PC-RLS algorithm (blue solid line) of the present invention has a significantly faster convergence speed than the LMS algorithm (orange dotted line), and compared with the standard RLS algorithm (green dashed line) and the STAP-like adaptive algorithm (red dotted line), while achieving a similar convergence speed, it can achieve a lower steady-state error. This is due to the fast tracking capability of RLS and the more effective handling of interference and polarization mismatch by the joint optimization framework of the present invention. The fast convergence characteristic is crucial for adapting to the rapidly changing channel environment in UAV communications;

[0143] Figure 8 The figure compares the communication link success rate of the algorithm of the present invention and the traditional polarization diversity algorithm under different UAV flight attitudes. The horizontal axis represents the roll angle (Roll) of the UAV, the vertical axis represents the pitch angle (Pitch), and the color and contour lines represent the percentage of the link success rate (the green area represents a high success rate, and the red area represents a low success rate). The left figure is the result of the algorithm of the present invention, and the right figure is the result of the traditional polarization diversity algorithm. It can be seen that when the UAV attitude is close to the horizontal (that is, the roll angle and pitch angle are close to 0 degrees), both algorithms can achieve a higher success rate. However, as the attitude angle (especially the absolute value of the roll angle and pitch angle) increases, the performance of the traditional polarization diversity algorithm ( Figure 8 Right) drops rapidly, and the success rate contour shrinks densely, indicating that it is very sensitive to posture changes. In contrast, the algorithm of the present invention ( Figure 8 The performance of the proposed algorithm (left) shows a much smaller drop-off with attitude angle changes, and the high success rate region (e.g., >90%) covers a significantly larger attitude range than traditional algorithms. This demonstrates that the proposed method, through joint optimization and adaptive polarization processing, can more effectively compensate for polarization mismatch caused by attitude changes, maintaining communication link stability over a wider attitude range.

[0144] Figure 9The anti-interference performance of the proposed algorithm and traditional methods is compared in five typical multipath propagation environments, with the performance metric being the improvement (dB) in the output signal-to-interference-plus-noise ratio (SINR) relative to the input signal-to-interference ratio (SIR = 0 dB). The five axes represent: urban high-rise buildings (strong multipath, short delay spread), urban streets (medium multipath, medium delay spread), suburban open areas (weak multipath, long delay spread), indoor environments (extremely strong multipath, short delay spread), and water / flat land (dominated by specular reflection). As can be seen in the figure, the proposed joint optimization algorithm (solid blue line) achieves the highest SINR improvement in all tested multipath environments, demonstrating its robust adaptability to different multipath characteristics (such as intensity, delay spread, and angular spread). The traditional STAP algorithm (dashed orange line) performs well in environments with large angular spread (such as cities), but performs only moderately well in environments dominated by specular reflection. The polarization diversity algorithm (dashed green line) inherently exhibits weak anti-interference capabilities, resulting in the smallest SINR improvement. The machine learning-based method (red dotted line) performs better, but may not be as stable or optimal as the proposed method. The results show that the proposed algorithm is universal and can effectively suppress interference in various complex low-altitude multipath environments.

[0145] Figure 10 The algorithm of the present invention is compared with other methods from the perspective of computing resource consumption. Assume that the processing involves N antenna elements and M=3 time taps (snapshots). The X axis is the number of antenna elements N.

[0146] 1. Figure 10 Left (computational complexity): The Y axis is giga floating point operations per second (GFLOPS, logarithmic scale). As can be seen, the complexity of the traditional STAP algorithm grows fastest with N (approximately O(N) 3 The complexity of the PC-RLS algorithm and the ML-based reasoning algorithm of the present invention is roughly O(N 2 ), significantly lower than STAP. Polarization diversity (MRC) has the lowest complexity (approximately O(N)). While providing high performance, the algorithm of the present invention keeps the complexity comparable to or better than ML, and significantly lower than STAP.

[0147] 2. Figure 10 Center (Estimated Power Consumption): The Y-axis shows the estimated algorithm runtime power consumption (W). The power consumption trend is similar to complexity, with STAP consuming the highest power and polarization diversity consuming the lowest. The proposed algorithm maintains low power consumption, outperforming both STAP and potentially more power-intensive ML inference modules, thus extending drone flight time.

[0148] 3. Figure 10Right (estimated processing delay): The Y-axis is the estimated time (ms) required to process a single data block. Latency is also related to complexity. The algorithm of the present invention achieves millisecond-level processing delay with its low complexity, which can meet the real-time requirements of high-speed mobile drone scenarios. It is superior to the high-complexity STAP and ML methods that may have large fixed delays. In summary, the algorithm of the present invention effectively controls computational complexity, power consumption, and processing delay while achieving excellent communication performance, demonstrating its practicality and deployment feasibility on resource-constrained drone platforms.

[0149] Example 2 End-to-end joint optimization network based on deep learning

[0150] This embodiment uses the powerful nonlinear mapping and feature extraction capabilities of deep neural networks (DNNs) to build an end-to-end deep learning model that directly maps the received original dual-polarization space-time signal x(k) to the desired output symbol Or the optimized signal y out (k).

[0151] Core ideas:

[0152] 1. Network architecture design: Design a deep neural network that includes convolutional layers (to extract spatiotemporal local features), recurrent layers (such as LSTM / GRU, to process time series dependencies), self-attention mechanisms (to capture long-range dependencies and feature importance), and possibly embedded polarization processing layers (such as using 4×4 convolution kernels to process polarization covariance features). The input of the network is x(k) (which may need to be preprocessed into image or sequence form), and the output is or y out (k).

[0153] 2. Loss function definition: The loss function not only includes the error of symbol recovery (such as cross-entropy loss), but can also explicitly or implicitly include the goals of suppressing interference and matching polarization. For example, auxiliary losses can be added to the intermediate layers to constrain feature representation, or adversarial training can be used to allow the network to learn to generate interference-resistant and polarization-robust signals.

[0154] 3. Training and Inference: Use a large amount of simulated or measured data to train the network offline. Deploy the trained model on the drone for online inference.

[0155] Potential advantages: It may have stronger nonlinear processing capabilities and strong ability to model complex unknown channels, without the need for explicit channel estimation and complex optimization algorithm derivation.

[0156] Potential disadvantages: Requires a large amount of high-quality training data, and the training process is time-consuming and resource-intensive; the model is highly complex, and the computational complexity and power consumption of online inference may still be high; the "black box" nature makes performance difficult to guarantee and explain; the ability to generalize to environmental changes depends on the coverage of the training data.

[0157] Example 3 Joint Channel Estimation and Signal Detection Based on Variational Inference

[0158] This embodiment integrates multipath interference suppression, polarization mismatch compensation, and signal detection into a unified probabilistic graphical model. It uses methods such as variational Bayesian inference to approximate the joint posterior probability distribution, thereby simultaneously estimating channel parameters (including polarization state) and transmitted symbols.

[0159] Core ideas:

[0160] 1. Probabilistic Model Construction: Build a joint probability model p(x, d, H, ...) that describes the received signal x(k), the unknown transmitted symbol d(k), the time-varying channel parameters H(k) (including multipath and polarization information), interference characteristics, and noise. The channel parameters and symbols are usually considered hidden variables.

[0161] 2. Variational Inference: Because directly computing the posterior probability p(d,H|x) is difficult, a parameterized variational distribution q(d,H) is introduced to approximate the true posterior distribution. The variational parameters are optimized by minimizing the KL divergence between q(d,H) and p(d,H|x).

[0162] 3. Iterative update: Usually, iterative methods such as coordinate ascent or expectation propagation are used to update the parameters of the variational distribution, and the symbol estimates and the channel parameters (including polarization parameters) are updated alternately.

[0163] 4. Output: The resulting variational distributions q(d) and q(H) give symbol probability estimates (for decision making) and channel parameter estimates (which can be used for analysis or further processing), respectively.

[0164] Potential advantages: It provides a rigorous probabilistic inference framework that can naturally incorporate prior knowledge and quantify the uncertainty of the estimate; it may achieve performance close to the optimal Bayesian estimation.

[0165] Potential disadvantages: Model construction and derivation are complex; the convergence and computational complexity of iterative algorithms may be high, especially in high-dimensional state spaces; and they are sensitive to the accuracy of model assumptions.

[0166] The above-mentioned Examples 2 and 3 represent different approaches to solving this problem by using two advanced technologies, deep learning and probabilistic inference. They each have their own advantages and disadvantages and may be competitive under specific conditions. However, they differ from the PC-RLS-based joint optimization algorithm proposed in the present invention in terms of implementation complexity, real-time performance, data dependence, and performance guarantee mechanism.

[0167] The basic principles, main features and advantages of the present invention are shown and described above. It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or basic features of the present invention.

[0168] In addition, it should be understood that although this specification is described in terms of implementation methods, not every implementation method contains only one independent technical solution. This narrative method of the specification is only for the sake of clarity. Those skilled in the art should regard the specification as a whole. The technical solutions in each embodiment can also be appropriately combined to form other implementation methods that can be understood by those skilled in the art.

Claims

1. A joint optimization algorithm for multipath interference suppression and polarization mismatch for UAV dual-polarization antennas, characterized by: The optimization algorithm consists of multiple functional modules, including a dual-polarization time-varying channel modeling and signal representation module, an adaptive polarization state estimation module, a joint space-time-polarization filter module, a joint optimization criterion and objective function construction module, a low-complexity adaptive weight update module, and a polarization branch dynamic merging module. The dual-polarization time-varying channel modeling and signal representation module is used to receive the baseband signal of the dual-polarization antenna array and describe the channel including multipath propagation, delay spread, Doppler frequency shift, and polarization effects that change with the attitude of the unmanned aerial vehicle based on a unified model; the adaptive polarization state estimation module estimates the main polarization state parameters of the current signal or equivalent polarization channel matrix components in real time based on the second-order statistical characteristics of the received signal; the low-complexity adaptive weight update module iteratively updates the weight vector of the joint filter through an adaptive algorithm to make it approach the optimal solution of the joint optimization objective function and quickly track channel changes; and the polarization branch dynamic merging module performs weighted merging of the horizontal (H) and vertical (V) polarization branch output signals processed by the joint filter based on their real-time estimated signal quality to obtain the final optimized output signal.

2. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized in that: The operation of the dual-polarization time-varying channel modeling and signal representation module includes the following: Assume that the drone is equipped with an array containing N pairs of orthogonal dual-polarized antenna units (a total of 2N channels). At time t, the horizontal (H) and vertical (V) polarized signal components r received by the nth unit pair are: n,H (t) and r n,V (t) is expressed as: Where: s(t)=[s H (t),s V (t)] T is the dual-polarized signal vector sent by the transmitter; L is the number of resolvable multipath paths; τ l is the propagation delay of the lth path; n n (t)=[n n,H (t),n n,V (t)] T is the additive white Gaussian noise vector received by the n-th unit pair; Hn,l(t) is the time-varying polarization channel matrix of the l-th path received by the n-th unit pair.

3. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 2 is characterized by: H n,l (t) represents the path's fading, phase rotation, angle of arrival, and the polarization conversion relationship between the transmitter and the nth receiver: where α n,l (t) and φ n,l (t) is the time-varying amplitude and phase of path l to unit n, h XY,n,l (t)(X,Y∈{H,V}) is the time-varying polarization transfer coefficient. The signals of all N unit pairs and M time snapshots (sampling points) are stacked to form a 2NM×1 joint space-time received signal vector x(k), where k is the discrete time index: And the vector x(k) contains rich information in the spatial domain, time domain and polarization domain.

4. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized by: The adaptive polarization state estimation module considers the average received signal r(t) of a single unit pair or array in a short time using the covariance matrix of the received signal, wherein the received signal vector r(k)=[r H (k),r V (k)] T , whose polarization covariance matrix is ​​defined as: In order to adapt to the time-varying channel, the covariance matrix is ​​estimated recursively to obtain the estimated value Where β (0 < β < 1) is the forgetting factor, which is used to adjust the contribution weights of historical data and current data to the estimation results. When the value is close to or far from 1, the smoothing effect is good but the tracking speed is slow.

5. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 4 is characterized by: The estimated Perform eigenvalue decomposition: Where Λ(k) is a diagonal matrix containing eigenvalues ​​λ1(k) ≥ λ2(k), U(k) = [u1(k), u2(k)] is the corresponding eigenvector matrix, the main eigenvector u1(k) corresponds to the main polarization direction of the signal, and the relative sizes of the eigenvalues ​​λ1(k) and λ2(k) represent the degree of polarization.

6. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized by: The joint space-time-polarization filter module provides a 2NM×1 joint filter weight vector w(k) that acts on the joint space-time received signal vector x(k) to generate the filter output: y(k)=wH(k)x(k) The filter w(k) structure simultaneously performs weighted processing on signal components from different antenna elements, different time snapshots, and different polarization directions, achieving multipath interference suppression (using space-time differences) and polarization matching (using polarization information) within a unified framework.

7. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized by: It also includes constructing a joint optimization objective function J(w) through a joint optimization criterion and an objective function construction module. This function minimizes the mean square error (MMSE) between the filter output y(k) and the desired signal d(k) while maximizing the polarization matching of the output signal. The formalized objective function is as follows: Among them: the first term E[|d(k)-w H x(k)| 2 ] is the standard mean square error (MMSE) term, the goal is to suppress interference and noise so that the output signal is close to the expected signal d(k); the second term is the polarization mismatch penalty term, which is used to guide the filter weight w so that the polarization state of the output signal is consistent with that given by The estimated main polarization direction u1(k) matches; the third term λ r ||w|| 2 is the regularization term, where λ p and λ r is a non-negative weighting coefficient to weigh the importance of each item.

8. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized by: The low-complexity adaptive weight update module adopts an adaptive update algorithm based on recursive least squares (RLS), and its update rule includes the following steps: a) Calculate the gain vector k(k): b) Calculate the prior error e(k): e(k)=d(k)-w H (k-1)x(k) c) Update the filter weight w rls (k): w rls (k)=w(k-1)+k(k)e(k) d) Update the inverse of the covariance matrix P(k): Where λ is the forgetting factor of RLS (0.95≤λ<1), and the complexity of the RLS algorithm is O((2NM) 2 ).

9. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 8 is characterized in that: In each RLS iteration update, a temporary weight w is obtained rls After (k), a correction step is performed to adjust the weights in the direction of satisfying the polarization matching constraint. The adjustment method is one of the following: a) Correction based on gradient descent b) Projection-based correction: w(k)=w rls (k)+α(k)(P proj (k)–I)w rls (k) where μ pol (k) or α(k) is the adaptive step size or coefficient that controls the polarization correction strength, is the gradient of the polarization mismatch penalty term with respect to the weight, P proj (k) is based on The constructed projection matrix, I is the same as P proj (k) Identity matrices of the same dimension.

10. The UAV dual-polarization antenna multipath interference suppression and polarization mismatch joint optimization algorithm according to claim 1 is characterized by: The polarization branch dynamic merging module is based on y H (k) and y V (k) is adaptively combined based on the real-time signal-to-interference-and-noise ratio (SINR) of the two branches. First, the SINR of the two branches is estimated: where fest(·,·) is the SINR estimation operator, which can be implemented by various standard methods, for example, by calculating the ratio of the desired signal power to the residual error and noise power, i.e. The desired signal d(k) is obtained through training sequence or decision feedback, and then the normalized combining weight is calculated: The final optimized output signal is:

Images