Beamforming optimization method for inductive-sensory integrated systems under external interference environment
By constructing a three-layer Stackelberg game model and a hierarchical alternating optimization framework, the collaborative optimization problem of beamforming in a sensor-integrated system under external interference is solved, achieving a balance between communication and radar sensing performance, reducing computational complexity, and making it suitable for real-time beamforming in dynamic interference environments.
Patent Information
- Application Number
- CN202611262207.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies struggle to effectively describe the master-slave decision-making relationship between malicious jammers and integrated sensing systems under external interference environments, leading to an imbalance between communication performance and radar sensing performance. Furthermore, traditional iterative algorithms have high computational complexity, making it difficult to achieve joint optimization of three-way beamforming.
A three-layer Stackelberg non-cooperative game model is constructed, with the jammer as the leader and the base station and communication user group as followers. The beamforming vector is iteratively updated through a hierarchical alternating optimization framework. The complexity is reduced by using Tinkelbach transform, alternating direction penalty method and weighted minimum mean square error method, so as to achieve collaborative optimization of beamforming.
It improves the system's adaptability to dynamic malicious interference environments, achieves a synergistic balance between communication performance, radar sensing performance, and anti-interference capability, reduces the solution complexity of the three-party beamforming joint optimization, and is suitable for real-time beamforming in dynamic sensing and communication integrated scenarios.
Smart Images

Figure CN123074546A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of wireless communication and radar sensing fusion technology, and in particular to a beamforming optimization method for a sensing-integrated system under external interference environment. Background Technology
[0002] The integrated sensing and communication system utilizes a unified antenna array, radio frequency link, and spectrum resources to simultaneously perform multi-user downlink communication and multi-target radar sensing tasks. The system typically consists of an integrated sensing and communication base station, multiple communication users, and multiple sensing targets. The base station improves user reception quality through communication beamforming and achieves the required gain in the target direction through radar sensing beamforming.
[0003] Existing technologies primarily employ centralized joint optimization or anti-interference beamforming methods. These methods jointly design communication and sensing beams based on communication channel conditions, target directions, and power budgets, using total communication rate, sensing signal-to-interference-plus-noise ratio (SNR), radar beam pattern error, or processing delay as performance indicators. External interference is typically modeled as fixed noise, random interference terms, or interference signals with known directions, and then solved using alternating optimization, weighted minimum mean square error, successive convex approximation, or alternating direction penalty methods. Under this modeling approach, existing technologies establish a joint model of the base station, communication users, and sensing targets under external interference conditions, incorporating external interference as a known term into the communication and sensing received signals, and jointly optimizing the communication and sensing beams under power and performance constraints. While this approach can reduce the impact of external interference to some extent, it still primarily relies on centralized decision-making.
[0004] Existing technologies do not fully consider that malicious jammers can dynamically adjust their jamming direction and power according to the strategies of base stations and communication user groups. Therefore, it is difficult to accurately describe the master-slave decision-making relationship between jammers, base stations, and communication user groups, where "jammers act first, and base stations and user groups respond later." At the same time, the communication beam, sensing beam, and jamming beam are coupled with each other, and optimizing communication performance or sensing performance separately can easily lead to an imbalance in the overall system performance.
[0005] Furthermore, joint optimization problems in multi-antenna, multi-user, and multi-target scenarios are characterized by high dimensionality, non-convexity, and multiple constraints. Traditional iterative algorithms require repeated matrix inversions and updates of auxiliary and dual variables, resulting in significant online computational overhead. Therefore, a sensory-integrated beamforming method is needed that can describe dynamic malicious interference behavior, characterize the master-slave relationship among the three parties, and reduce the complexity of online solutions. Summary of the Invention
[0006] The purpose of this invention is to provide a beamforming optimization method for an integrated sensing system under external interference conditions. This method addresses the technical problems of jointly optimizing three types of beamforming vectors under high-dimensional non-convex multi-constraint conditions in the event of external interference, and the inability to achieve an effective balance between communication performance, radar sensing performance, and anti-interference capability.
[0007] A beamforming optimization method for an integrated sensing system under external interference conditions, comprising the following steps:
[0008] S1: Construct a system model that includes jammers, base stations, multiple communication users and multiple radar targets, and define the objective functions and constraints for the jammers, base stations and communication user groups respectively.
[0009] S2: Construct a three-layer Stackelberg non-cooperative game model with the jammer as the leader and the base stations and communication user groups as followers;
[0010] S3: The game model is solved by using a hierarchical alternating optimization framework. The jamming beamforming vector of the jammer, the radar sensing beamforming vector of the base station, and the communication beamforming vector of the communication user group are updated iteratively until the algorithm converges to obtain the three sets of optimal beams corresponding to the Stackelberg equilibrium point.
[0011] Optionally, in step S1, the objective function of the jammer is defined as: maximizing the ratio of the sum of weighted interference power caused by the downlink communication channel and the radar sensing link to the base station's transmitted signal energy; the constraint condition is: the jammer's transmitted power is less than the maximum budget value of the jammer's transmitted power, and the jammer is in the direction The radiation power generated is less than the radiation power suppression threshold corresponding to that direction.
[0012] Optionally, in step S1, the objective function of the base station is defined as: minimizing the maximum processing delay in the sensing task sent to the MEC; the constraint is: the total radar sensing transmit power of the base station is less than the maximum transmit power of the base station used for radar sensing, and the radar detection accuracy error is less than a preset accuracy threshold.
[0013] Optionally, in step S1, the objective function of the communication user group is defined as: maximizing the total downlink reachable rate; the constraint condition is: the total communication transmit power does not exceed the base station communication power budget.
[0014] Optionally, the three-layer Stackelberg non-cooperative game model constructed in step S2 for:
[0015] )
[0016] in, The player set comprises three decision-making entities: jammer JM, base station BS, and communication user group UG. , Represents each player A set of optional strategies, Represents each player The utility function is defined by the three strategies: the set of interference beamforming vectors, the set of radar sensing beamforming vectors, and the set of communication beamforming vectors. The feasible region includes the corresponding constraints.
[0017] Optionally, the specific steps in step S3 are as follows:
[0018] S3.1: The jammer, acting as the leader, updates the jamming beamforming vector based on the radar sensing beamforming vectors and communication beamforming vector sets of the base station and communication user groups.
[0019] S3.2: The base station and the communication user group act as followers, and under the given interference beamforming vector conditions, update the radar sensing beamforming vector and the communication beamforming vector set respectively.
[0020] S3.3: Repeat steps S3.1-S3.2. When the changes in the three beamforming vectors and the constraint residuals are all less than the preset thresholds, or when the maximum number of iterations is reached, output the current interference beam, radar sensing beam, and communication beam vectors as the three-way equalization strategy.
[0021] Optionally, the interference beamforming vector in step S3 is updated using the Tinkelbach transform and the alternating direction penalty method.
[0022] Optionally, the radar sensing beamforming vector in step S3 is updated using a minimization method and an alternating direction penalty method.
[0023] Optionally, the communication beamforming vector is updated using a weighted minimum mean square error method.
[0024] Because of the adoption of the above technical solution, the present invention has the following advantages:
[0025] 1. In the presence of external malicious interference, this application considers the strategic interaction relationship between the external malicious jammer, the integrated sensor base station, and the communication user group. By constructing a Stackelberg game model with the external malicious jammer as the leader and the integrated sensor base station and the communication user group as followers, this application can describe the master-slave decision-making process in which the jammer first selects the jamming strategy, the base station, and the communication user group, and then responds, thereby improving the system's adaptability to dynamic malicious interference environments.
[0026] 2. This application achieves a coordinated balance between communication performance, radar sensing performance, and system anti-jamming capability. By jointly optimizing the sets of interference beamforming vectors, radar sensing beamforming vectors, and communication beamforming vectors, this application can coordinate the total downlink communication rate of multiple users and the processing delay of multi-target radar under external interference conditions, avoiding system performance biased solely towards the communication or sensing side.
[0027] 3. This application reduces the solution complexity of the three-party beamforming joint optimization problem. For the three sub-problems of jammer, base station, and communication user group, this invention uses Tinkelbach transform and alternating direction penalty method, minimization method and alternating direction penalty method, and weighted minimum mean square error method to solve them respectively. Furthermore, it introduces a deep unfolded network to replace some of the high-complexity matrix operations in the traditional iterative algorithm, thereby improving the real-time update capability of beamforming strategy in dynamic scenarios.
[0028] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description
[0029] The accompanying drawings of this invention are described below.
[0030] Figure 1 This is a schematic diagram of the integrated sensing system of the present invention.
[0031] Figure 2 This is a flowchart of the beamforming optimization method of the present invention.
[0032] Figure 3 This is a comparison chart of the computation time of different solution methods of the present invention.
[0033] Figure 4 This is a performance comparison chart of different solutions and methods of the present invention with the theoretical benchmark method.
[0034] Figure 5 This is a comparison chart showing the differences in base station utility under different solution approaches in the Stackelberg game of this invention. Detailed Implementation
[0035] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the description of the embodiments of the present invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "connected" or "linked" should be interpreted broadly. For example, it can be a fixed connection, a detachable connection, an integral connection, an electrical connection, or a signal connection; it can be a direct connection or an indirect connection through an intermediate medium.
[0036] like Figure 2 The beamforming optimization method for an integrated sensing system under external interference environment is characterized by the following specific steps:
[0037] S1: Construct a system model that includes jammers, base stations, multiple communication users and multiple radar targets, and define the objective functions and constraints for the jammers, base stations and communication user groups respectively.
[0038] S1.1: Construct a system model that includes jammers, base stations, multiple communication users, and multiple radar targets;
[0039] In this embodiment, as Figure 1 The illustrated integrated sensing system includes a sensor equipped with A base station (BS) with integrated antenna and communication capabilities, and individual communication users and One radar target; the The antennas form a uniform linear array ULA. The integrated sensing base station BS transmits an integrated waveform signal, which includes radar signals and downlink communication signals; in order to... While providing communication services to individual communication users, the integrated sensing base station (BS) receives and processes radar echo information from radar targets to achieve sensing functions.
[0040] A set of communication users constitutes the user group UG. The integrated inductive base station BS and the user group UG operate in half-duplex mode. The set of communication users and the set of radar targets are denoted as follows: , Meanwhile, a malicious jammer JM exists near the integrated sensor system. JM is equipped with a... A uniform linear array of array elements emits interference signals to disrupt the downlink communication and sensing links of the integrated sensing base station (BS). It is assumed that the channel links between the BS and each device experience small-scale Rayleigh fading.
[0041] S1.2: Define the objective function and constraints of the jammer:
[0042] The jammer is modeled as an active entity with prior knowledge of the target's direction. This allows the jammer to concentrate its jamming power in a specific direction while minimizing radiation in non-jamming directions, thus enhancing its stealth and avoiding collateral damage. The jammer's transmission is processed by beamforming signals. for:
[0043] (1)
[0044] in, For the jammer beamforming vector, Used to control the spatial radiation direction of interference signals; The jamming signal is emitted by the jammer. The jamming signal satisfies the unit energy condition, and the jammer's transmission power is constrained by the maximum power budget.
[0045] To suppress unwanted radiation in the protection direction, a beam pattern constraint is constructed using a uniform linear array steering vector, and the radiated power in each protection direction is limited to within a preset threshold. Therefore, the objective function of the jammer is defined as maximizing the ratio of the sum of weighted interference power caused to the downlink communication channel and radar sensing link to the base station transmitted signal energy, while satisfying the power constraint and beam pattern constraint.
[0046] (2)
[0047] in, This is the transmitted signal vector from the base station. A positive weighting factor. From jammer to the first Channel vectors for each user This is the channel vector from the jammer to the base station radar sensing receiver. This is the channel vector from the base station to the jammer. This is the conjugate transpose of the vector. The maximum budget constraint for the jammer's transmit power. For the jammer in the first One protection direction The array guide vector on, This is the conjugate transpose of the array guide vector. For the first The radiation power suppression threshold corresponding to each protection direction is used to limit the unwanted radiation power of the jammer in that protection direction. This represents the number of antennas of the jammer.
[0048] S1.3: Define the objective function and constraints of the base station:
[0049] The base station transmits an integrated waveform that simultaneously carries radar sensing signals and downlink communication signals. The composition is as follows:
[0050] (3)
[0051] in, For radar sensing beamforming vectors, For radar sensing signals, For the first Communication beamforming vectors for each communication user To send to the Communication signals of individual communication users The number of communication users.
[0052] The base stations are configured to use the same antenna array for both transmission and reception, and perfect self-interference cancellation is achieved, so the base stations are not affected by their own transmitted signals. Direct interference. Located in the direction Having array response vector and channel gain The The signal reflected back to the base station from the radar target for:
[0053] (4)
[0054] in, Including target reflection coefficient and propagation loss, To receive noise, It is additive white Gaussian noise.
[0055] For each radar target Processing time Defined as task load divided by processing rate:
[0056] (5)
[0057] The target processing latency is determined by the ratio of task load to perception processing rate, and the perception processing rate is determined by the target signal-to-interference-plus-noise ratio. Signal-to-interference-plus-noise ratio of each radar target for:
[0058] (6)
[0059] in, Let Variance be the additive white Gaussian noise at the base station radar sensing receiver, i.e., the received noise power; to describe the squared error between the radar transmit beamformer and the desired beam pattern in different directions, we introduce:
[0060] (7)
[0061] in, For this is the first The weight coefficients of each angle grid point To adjust the desired pattern amplitude to match the scaling factor of the actual beam power scale, To achieve the desired beam pattern at an angle The value at that location, The array guide vector corresponding to that angle is constructed. The pattern error is calculated on a discrete angle grid, and the importance of different directions is adjusted by weights, desired beam values, and scaling factors.
[0062] Based on the above analysis, the objective function of the base station is defined as minimizing the maximum processing latency in the sensing task transmitted to the MEC under power constraints and radar detection accuracy error constraints:
[0063] (8)
[0064] in, For the first Processing time for each radar target , This is the maximum transmit power used by the base station for radar sensing. The total number of discrete angle grid points; This is the preset accuracy threshold.
[0065] S1.4: Define the objective function and constraints for the communication user group:
[0066] The base station communicates with the user group via the downlink, and the communication users The received signal is:
[0067] (9)
[0068] in, From base station to communication user The channel vector, For the base station to send the first Communication beamforming vectors for each communication user For the base station to send the first Communication data symbols of a communication user For communication users Received noise.
[0069] Therefore, users SINR is represented as :
[0070] (10)
[0071] in, For the first The variance of additive white Gaussian noise at the receiver of each communication user, i.e., the received noise power;
[0072] Accordingly, users The downlink achievable rate is:
[0073] (11)
[0074] Therefore, the objective function of the communication user group is defined as maximizing the total downlink reachable rate, with the constraint that the total communication transmit power does not exceed the base station's communication power budget.
[0075] (12)
[0076] in, Budget for base station communication power.
[0077] S2: Construct a three-layer Stackelberg non-cooperative game model with the jammer as the leader and the base stations and communication user groups as followers;
[0078] In this embodiment, the three-layer Stackelberg non-cooperative game model The complete definition is given by the following triples:
[0079] (13)
[0080] in, The player set comprises three decision-making entities: jammer JM, base station BS, and communication user group UG. , Represents each player A set of optional strategies, Represents each player The utility function.
[0081] The three strategies are the set of interference beamforming vectors, radar sensing beamforming vectors, and communication beamforming vectors, respectively. Their feasible regions are determined by the corresponding power constraints and beam pattern constraints. The three objective functions are defined as P1, P2, and P3, respectively.
[0082] S3: The game model is solved by using a hierarchical alternating optimization framework. The jamming beamforming vector of the jammer, the radar sensing beamforming vector of the base station, and the communication beamforming vector of the communication user group are updated iteratively until the algorithm converges to obtain the three sets of optimal beams corresponding to the Stackelberg equilibrium point.
[0083] In this embodiment, due to the mutual coupling between the jamming beamforming vector of the jammer, the radar sensing beamforming vector of the base station, and the communication beamforming vector set of the communication user group, the original optimization problem exhibits non-convexity and multivariate coupling characteristics, making it difficult to directly obtain a closed-form solution. Therefore, this application adopts a hierarchical alternating optimization approach, decomposing the original three-party optimization problem into a jammer beamforming optimization subproblem, a base station radar sensing beamforming optimization subproblem, and a communication user group communication beamforming optimization subproblem. In each iteration, the strategies of other participants are fixed, and the beamforming variables of the current participant are updated accordingly.
[0084] Specifically, the jammer, acting as the leader, updates the jamming beamforming vector based on the responses from the base station and the communication user group. The base station and the communication user group, acting as followers, update the radar sensing beamforming vector and the communication beamforming vector set, respectively, given the external jammer strategy. For the external jammer subproblem, this application employs the Tinkelbach transform and the alternating direction penalty method for solution; for the base station subproblem, this application employs the minimization method and the alternating direction penalty method for solution; and for the communication user group subproblem, this application employs the weighted minimum mean square error method for solution. Through the above hierarchical iterative process, a three-way beamforming strategy satisfying power constraints, sensing performance constraints, and communication performance constraints can be gradually obtained.
[0085] In this Stackelberg game, the jammer with first-mover advantage acts as the leader, while the base stations and user groups act as followers. The solution employs backward induction: the leader first predicts the optimal responses of the followers, and then designs the jamming beamforming vector accordingly. To maximize the destructive effect; subsequently, the followers observed Then, decisions are made simultaneously but not collaboratively—base station radar beam optimization. To minimize perceived latency, user groups optimize communication beamsets. To maximize the total downlink rate, both sides must cope with cross-link interference. When the game reaches the Stackelberg equilibrium, the leader selects a strategy based on rational expectations, and the followers respond optimally afterwards; neither side has an incentive to deviate unilaterally.
[0086] S3.1: Update the interference beamforming vector:
[0087] In a three-party Stackelberg game model, the external jammer acts as the leader, and its strategy variable is the jamming beamforming vector. To solve for the jammer's optimal jamming strategy, this application optimizes the jamming beamforming vector under the conditions of the radar sensing beamforming vector of the fixed base station and the communication beamforming vector set of the communication user group. The jammer's optimization objective is to improve its overall jamming effect on the communication link and radar sensing link while satisfying the transmit power constraint and beam pattern constraint.
[0088] As the leader, the interference machine JM's optimization problem, although fractional in form, only contains decision variables in the numerator. The denominator is influenced by the strategies of the communication user group UG and the base station BS, thus standard fractional programming is not applicable. To decouple the game interaction, the Tinkelbach framework is used to separate the numerator and denominator. Specifically, in each iteration, based on the current strategies of UG and BS, JM reformulates the original problem into a series of convex subproblems using the Tinkelbach framework and updates an auxiliary variable representing the ratio of the numerator to the denominator. The numerator is defined as... The denominator is It is worth noting that when JM optimizes its own strategy, the base station transmits signals. It is predetermined, therefore the denominator Compared to It can be considered a constant, which simplifies the transformation process.
[0089] For a given auxiliary parameter The original problem P1 is equivalent to solving the following parameterized subproblem:
[0090] (14)
[0091] The Tinkerbach algorithm updates iteratively. To approximate the optimal solution: Initialization ; in the outermost layer In the next iteration, solving subproblem (14) yields... Then update the parameters. When the target value Less than the preset tolerance The algorithm converges at this point. The parameterized subproblem (14) is a convex problem with quadratic constraints. To obtain an efficient and distributed solution process, an alternating direction penalty method is adopted. By introducing auxiliary variables... and By decoupling the coupling constraints, problem (14) can be reconstructed as follows:
[0092] (15)
[0093] in, For the first The auxiliary variable corresponding to each protection direction is defined as follows: Used to counter the jammer in the first Variable decoupling is performed on the beam pattern constraints in each protection direction.
[0094] The corresponding augmented Lagrangian function is:
[0095] (16)
[0096] in, , and For Lagrange multipliers, The penalty parameter is used; the alternating direction penalty method solves the problem by alternately updating the following variables:
[0097] In the alternating direction penalty iteration, the interference beamforming vector, power auxiliary variable, and protection direction auxiliary variable are updated sequentially; the closed-form update of the interference beamforming vector is:
[0098] (17)
[0099] in, The coefficient matrix is composed of the quadratic term of the objective function, the power constraint penalty term, and the protection direction constraint penalty term during the update process of the interference beamforming vector. To and An identity matrix with matching dimensions.
[0100] The power auxiliary variable and the protection direction auxiliary variable are obtained through projection respectively:
[0101] (18)
[0102] (19)
[0103] The dual variables and penalty parameters are updated based on the original residuals:
[0104] (20)
[0105] (twenty one)
[0106] (twenty two)
[0107] (twenty three)
[0108] in, and These are the penalty parameter scaling factors corresponding to the power auxiliary variable constraints and the protection direction auxiliary variable constraints, respectively. The original residuals of the equality constraint between the interference beamforming vector and the power auxiliary variable are... This represents the original residual of the equality constraint between the interference beamforming vector and the auxiliary variable of the protection direction.
[0109] To avoid explicit matrix inversion at each layer, the inverse matrix is replaced with a learnable parameterized surrogate consisting of a diagonal approximation, a structure-preserving term, and a residual compensation term. The l-th layer interference beamforming vector is updated as follows:
[0110] (twenty four)
[0111] For the reason A diagonal matrix formed by the reciprocals of its diagonal elements has zero off-diagonal elements. , and For the depth of beamforming deployment network of the jammer The trainable matrix parameters of the layer are used to construct the pair The parameterized approximation, For the first The trainable residual compensation vector of the layer, The layer index is used for the deep unfolded network. The above updates are embedded in the Tinkelbach outer layer iteration until the parameterized objective and constraint residuals satisfy the convergence threshold.
[0112] S3.2: Update radar sensing beamforming vector:
[0113] After updating the jamming beamforming vector of the external malicious jammer, the radar sensing beamforming vector of the base station is further optimized. In this step, the jamming beamforming vector of the external jammer and the set of communication beamforming vectors of the communication user group are fixed, and only the radar sensing beamforming vector of the base station is used as the optimization variable. The optimization objective of the base station is to reduce the maximum processing latency among multiple radar sensing targets while satisfying the radar sensing transmit power constraint and radar beam pattern constraint, thereby improving the processing efficiency of the radar sensing task and the sensing stability under external interference environment.
[0114] (25)
[0115] By introducing a maximum processing delay upper bound variable, and when using fixed jammer and communication user group strategies, the base station problem is transformed into a sensing rate optimization problem with fair weights:
[0116] (26)
[0117] in, To ensure that, under the conditions of fixed jammer and communication user group strategies, the first The equivalent signal-to-interference-plus-noise ratio (SINNR) coefficient corresponding to the first radar target is used to characterize the effect of target channel gain, communication signal interference, external interference, and received noise on the first radar target. The combined impact of radar target perception performance.
[0118] For the non-convex terms in the objective function and pattern constraints, a first-order substitution function is constructed at the current iteration point using the minimax method. The corresponding approximate subproblem is:
[0119] (27)
[0120] in, For the first In the nth minimax iteration The first-order approximate weighting coefficients corresponding to each radar target For the first The array response matrix corresponding to each radar target direction. For the base station array in the target direction The array guide vector on.
[0121] After convex approximation and extended variable transformation, the base station beamforming subproblem can be written as:
[0122] (28)
[0123] in, Let be a positive semidefinite matrix composed of the weighted sum of the response matrices of each radar target array. The positive semi-definite matrix obtained after minimizing and convexizing the radar beam pattern error constraint is used to characterize the convexized beam pattern error constraint.
[0124] After converting the complex variable to its equivalent real number form and introducing an auxiliary variable, we obtain:
[0125] (29)
[0126] in, To extend radar sensing beamforming vector The equivalent real number representation of , Transpose it. For matrix The corresponding equivalent real matrix, For positive semi-definite matrices The matrix obtained by decomposition satisfies , This is the selection matrix used to extract scaling auxiliary components from the extended radar beamforming vector.
[0127] The corresponding augmented Lagrangian function is:
[0128] (30)
[0129] in, Equality constraints , and The corresponding penalty parameters, These are the dual variables (Lagrange multipliers) corresponding to the three equality constraints mentioned above.
[0130] ADPM sequentially updates the radar sensing beamforming vector, power projection variables, pattern constraint variables, and dual variables. The main update formulas are as follows:
[0131] (31)
[0132] (32)
[0133] (33)
[0134] (34)
[0135] (35)
[0136] (36)
[0137] (37)
[0138] in, Auxiliary variables , and Unconstrained intermediate variables before constrained projection.
[0139] The penalty parameters are adaptively adjusted based on the original residuals of each equality constraint:
[0140] (38)
[0141] (39)
[0142] (40)
[0143] in, These are the scaling factors for the corresponding penalty parameters. Constraints The corresponding original residuals are used to measure the degree of violation of the corresponding equality constraints.
[0144] To reduce the overhead of matrix inversion, a learnable agent consisting of diagonally dominant terms, structure-preserving terms, and residual compensation terms is used. The base station beam update for the l-th layer is as follows:
[0145] (41)
[0146] in, The coefficient matrix that needs to be inverted during the update process of the beamforming vector of the base station radar sensing. For the reason A diagonal matrix consisting of the reciprocals of its diagonal elements has zero off-diagonal elements. For base station radar sensing beam depth deployment network, the first The trainable matrix parameters of the layer are used to construct the diagonal-dominated term, the structure-preserving term, and the residual compensation term, respectively, to form a pair. The parameterized approximation, For the layer index of the depth-unfolded network, For the first The trainable residual compensation vector of the layer. The radar sensing beamforming vector that satisfies the power and pattern accuracy constraints is obtained through the above MM-ADPM iteration.
[0147] S3.3: Update the communication beamforming vector:
[0148] After updating the jamming beamforming vectors of the external jammer and the radar sensing beamforming vectors of the base station, the communication beamforming vector set of the communication user group is further optimized. In this step, the jamming beamforming vectors of the external jammer and the radar sensing beamforming vectors are fixed, and only the communication beamforming vectors corresponding to each communication user are used as optimization variables. The optimization objective of the communication user group is to increase the total downlink communication rate of multi-users while satisfying the communication transmit power constraint, thereby improving the overall communication performance of the communication user group under external interference environment.
[0149] To address the nonconvex optimization problem P3 for user groups, the WMMSE transform is employed, converting the nonconvex sum rate maximization problem into an equivalent convex form. For a given set of beamforming vectors... ,user The received signal is given by equation (9). Define a linear receiver. And consider the mean square error: After simple algebraic operations, we can obtain:
[0150] (42)
[0151] For fixed , minimize The minimum mean square error receiver is:
[0152] (43)
[0153] in, The corresponding minimum MSE is:
[0154] (44)
[0155] The minimum mean square error is related to the signal-to-interference-plus-noise ratio (SIR) and achievable data rate as follows:
[0156] (45)
[0157] (46)
[0158] After introducing error weights, the original sum-rate maximization problem is equivalently transformed into a weighted minimum mean square error problem:
[0159] (47)
[0160] When using a fixed communication beam, the receiver and error weights undergo closed-loop updates:
[0161] (48)
[0162] With fixed receivers and weights, the communication beamforming sub-problem is as follows:
[0163] (49)
[0164] The update formula for the communication beamforming vector is obtained from the first-order optimality condition:
[0165] (50)
[0166] in, It is a weighted channel correlation matrix constructed based on the error weights of each communication user, the linear receiver, and the channel vectors from the base station to each communication user.
[0167] The power constraint is satisfied by the Lagrange multiplier or the final layer projection, and the total transmit power relationship is:
[0168] (51)
[0169] in, For communication beamforming update process, the weighted channel correlation matrix and Lagrange multipliers The inverse matrix of the coefficient matrix.
[0170] To avoid matrix inversion and binary search, a learnable parameterized surrogate is used for deep expansion updates:
[0171] (52)
[0172] in, Let be a diagonal matrix consisting of the reciprocals of the diagonal elements of the matrix whose inverse is to be found, with its off-diagonal elements being zero. For communication beamforming depth unfolding network The trainable matrix parameters of the layer are used to construct the diagonal-dominated term, structure-preserving term, and residual compensation term in the surrogate inverse matrix, respectively. Layer index for communication beamforming depth unwrap network For the first Layer for the first Trainable residual compensation vectors introduced by each communication user This is the complex conjugate operator.
[0173] The final layer of the network strictly satisfies the total communication power constraint through power projection:
[0174] (53)
[0175] The receiver, error weights, and communication beamforming vectors are updated alternately until the objective function for the communication user group converges.
[0176] In this embodiment, given a jammer strategy, the base station and communication user group obtain a stable response through the aforementioned MM-ADPM and WMMSE updates; the jammer then updates the leader strategy based on the follower responses. When the changes in the three strategies and the constraint residuals are all less than a preset threshold, or when the maximum number of iterations is reached, the current jamming beam, radar sensing beam, and communication beam are output as the three-way equilibrium strategy.
[0177] The main overhead of traditional algorithms comes from matrix inversion in the jammer, base station, and communication user group subproblems. Their overall complexity per round can be expressed as:
[0178]
[0179] After adopting a deep unfolded parameterized proxy, each layer mainly performs matrix multiplication, reducing the overall complexity to:
[0180]
[0181] in, , and They represent the number of iterations for the traditional subproblem, , and These represent the corresponding number of layers in the expanded network. Let M be the number of jammer antennas and M be the number of base station antennas. This structure reduces the main computation from cubic matrix inversion to quadratic matrix multiplication.
[0182] S4: Simulation Verification:
[0183] To verify the effectiveness of the proposed method, further numerical simulation analysis was conducted. The simulation included a system scenario encompassing an external malicious jammer, an integrated inductive and communication base station, multiple communication users, and multiple radar targets. The performance of theoretical benchmark methods, the proposed deep unfolding method, and other baseline methods was compared in terms of convergence performance, jammer effectiveness, base station radar processing performance, total communication user group rate, and computational complexity.
[0184] In the simulation setup, the integrated inductive and sensory base station (BS) is equipped with 8 antennas, the malicious jammer is equipped with 4 antennas, the number of communication users is 2, and the number of radar targets is 4. The comparison methods include traditional iterative benchmarks, the proposed deep unfolding method, and a reinforcement learning method based on deep deterministic policy gradients. Evaluation metrics include three-party utility, constraint satisfaction, running time, and equilibrium convergence performance. Algorithm 1 is the beamforming optimization algorithm based on deep unfolding networks adopted in this application. It unfolds the traditional iterative solution process into network layers and uses learnable parameterized agents instead of explicit matrix inversion; hereinafter referred to as "Algorithm 1". Algorithm 2 is a deep reinforcement learning algorithm based on deep deterministic policy gradients (DDPG), which learns beamforming policies by constructing DDPG agents for the jammer, base station, and communication user groups respectively; hereinafter referred to as "Algorithm 2".
[0185] Figure 3 The average computation time required by different solutions on JM, BS, and UG was compared and evaluated based on the average results of 10,000 test samples. The results show that the proposed depth-unfold-based designs significantly reduce computational overhead, demonstrating superior performance compared to theoretical benchmark methods. For the JM solution, the theoretical computation time is 49.35 ms, reduced to 2.43 ms in design algorithm 1, and further reduced to 0.35 ms in design algorithm 2. The BS solution shows the most significant speedup: the theoretical time is 110.9 ms, significantly reduced to 11.37 ms in design algorithm 1, and only 0.56 ms in design algorithm 2. Similarly, the UG solution also shows a significant speedup, reducing the computation time from 33.28 ms to 1.34 ms for design algorithm 1 and 0.42 ms for design algorithm 2, respectively. These improvements are directly attributed to the reduction of computationally expensive matrix inversion, which is significantly reduced in traditional methods. Replacing it with a parameterized alternative function reduces the complexity of designing Algorithm 1 to [missing value]. The complexity of Algorithm 2 is reduced to linear complexity. The above analysis highlights the effectiveness of the proposed deep unfolding framework in significantly reducing computational latency, thereby enabling real-time deployment in dynamic and resource-constrained disruptive environments.
[0186] Figure 4The performance retention of the proposed design relative to the theoretical baseline across all three participants is demonstrated. Results show that Design 1 achieves near-optimal performance across all participants, retaining 96.65%, 98.41%, and 91.29% of the baseline utility for the interferator, base station, and user group, respectively. In contrast, Design 2 prioritizes computational efficiency, but with a slight performance decrease, retaining 91.91%, 89.66%, and 86.17% of the baseline performance for the interferator, base station, and user group, respectively. Design 1 achieves a significant improvement in computational speed while maintaining near-optimal performance. This validates its superior trade-off between accuracy and efficiency, making it a highly practical solution for real-time applications.
[0187] Figure 5 This paper presents the convergence trajectories of three methods for base station utility within a Stackelberg game framework, along with the original and transformed objectives. The theoretical baseline performs optimally, achieving the lowest processing latency and highest sensing rate at equilibrium. Design 1 closely approximates the baseline, maintaining near-optimal performance throughout the convergence process, while Design 2 exhibits the most significant degradation. Design 1 embeds the ADPM structure into a learnable deep unfolding framework, preserving the algorithmic fidelity of the original solver and enabling adaptive updates, thus converging to a near-theoretical optimal equilibrium. Design 2, based on model-free deep reinforcement learning, struggles to sustain effective navigation in the high-dimensional policy space in dynamic games, leading to a suboptimal equilibrium. In conclusion, the deep unfolding method achieves an effective balance between optimality and adaptability, enabling the base station to approach theoretically optimal performance while flexibly responding to dynamic interference strategies.
[0188] Simulation results show that the proposed depth unfolding method can significantly reduce the average computation time while maintaining near-traditional iterative benchmark performance, and achieves stable convergence in Stackelberg game iterations. This method balances overall communication rate, radar processing latency, and policy response capabilities under external interference conditions, making it suitable for real-time beamforming in dynamic sensing and communication integration scenarios.
[0189] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.
Claims
1. A beamforming optimization method for an integrated sensing system under external interference environment, characterized in that, The specific steps are as follows: S1: Construct a system model that includes jammers, base stations, multiple communication users and multiple radar targets, and define the objective functions and constraints for the jammers, base stations and communication user groups respectively. S2: Construct a three-layer Stackelberg non-cooperative game model with the jammer as the leader and the base stations and communication user groups as followers; S3: The game model is solved by using a hierarchical alternating optimization framework. The jamming beamforming vector of the jammer, the radar sensing beamforming vector of the base station, and the communication beamforming vector of the communication user group are updated iteratively until the algorithm converges to obtain the three sets of optimal beams corresponding to the Stackelberg equilibrium point.
2. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 1, characterized in that, In step S1, the objective function of the jammer is defined as: maximizing the ratio of the sum of the weighted interference power caused by the downlink communication channel and the radar sensing link to the energy of the base station transmitted signal; The constraints are: the jammer's transmit power is less than the maximum budgeted value of the jammer's transmit power, and the jammer is in the direction of... The radiation power generated is less than the radiation power suppression threshold corresponding to that direction.
3. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 1, characterized in that, In step S1, the objective function of the base station is defined as: minimizing the maximum processing latency in the sensing tasks sent to the MEC; The constraints are: the total transmit power of the base station's radar sensing is less than the maximum transmit power of the base station used for radar sensing, and the radar detection accuracy error is less than the preset accuracy threshold.
4. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 1, characterized in that, In step S1, the objective function of the communication user group is defined as: maximizing the total downlink reachable rate; The constraint is that the total communication transmission power does not exceed the base station's communication power budget.
5. The beamforming optimization method for an integrated sensing system under external interference environment according to any one of claims 1-4, characterized in that, The three-layer Stackelberg non-cooperative game model constructed in step S2 for: ) in, The player set comprises three decision-making entities: jammer JM, base station BS, and communication user group UG. , Represents each player A set of optional strategies, Represents each player The utility function is defined by the three strategies: the set of interference beamforming vectors, the set of radar sensing beamforming vectors, and the set of communication beamforming vectors. The feasible region includes the corresponding constraints.
6. The beamforming optimization method for an integrated sensing system under external interference environment according to any one of claims 1-4, characterized in that, The specific steps in step S3 are as follows: S3.1: The jammer, acting as the leader, updates the jamming beamforming vector based on the radar sensing beamforming vectors and communication beamforming vector sets of the base station and communication user groups. S3.2: The base station and the communication user group act as followers, and under the given interference beamforming vector conditions, update the radar sensing beamforming vector and the communication beamforming vector set respectively. S3.3: Repeat steps S3.1-S3.
2. When the changes in the three beamforming vectors and the constraint residuals are all less than the preset thresholds, or when the maximum number of iterations is reached, output the current interference beam, radar sensing beam, and communication beam vectors as the three-way equalization strategy.
7. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 6, characterized in that, The interference beamforming vector mentioned in step S3 is updated using the Tinkelbach transform and the alternating direction penalty method.
8. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 6, characterized in that, The radar sensing beamforming vector in step S3 is updated using the minimization method and the alternating direction penalty method.
9. The beamforming optimization method for an integrated sensing system under external interference environment according to claim 6, characterized in that, The communication beamforming vector is updated using the weighted minimum mean square error method.
