A method for anti-intermittent sampling and forwarding interference for a communication and perception integrated system

Abstract

The application discloses an anti-intermittent sampling and forwarding interference method for a communication and perception integrated system, and comprises the following steps: constructing a multi-user multi-antenna communication and perception integrated anti-interference system, simultaneously realizing multi-user downlink communication and target perception under the same spectrum resource and hardware platform; establishing a joint model of base station transmitting signals, intermittent sampling and forwarding interference, perception receiving signals and communication receiving signals, and constructing a perception-communication joint state according to real target output, false target output, false target suppression ratio, communication rate, signal-to-interference-and-noise ratio and leakage interference statistics; on this basis, a joint optimization problem of a transmitting beamforming matrix and a non-matching filter is established, and an upper-layer parameter decision method based on transfer learning and a double deep Q network is used to adaptively determine a perception utility weight, a communication utility weight and a real target main peak retention threshold; and an alternating optimization method is used to jointly update the transmitting beamforming matrix and the non-matching filter in a lower layer.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and radar anti-jamming technology, and in particular to an anti-intermittent sampling and forwarding interference method for integrated communication and sensing systems. Background Technology

[0002] With the development of sixth-generation mobile communication technology, integrated communication and sensing has gradually become an important research direction in the field of wireless communication. Integrated communication and sensing systems can simultaneously achieve wireless communication and target sensing using the same spectrum resources, hardware platform, transmission waveform, and array resources. Compared with the traditional separate design of communication and sensing systems, it has advantages such as high spectrum utilization, hardware resource sharing, and significant synergistic gains in communication and sensing.

[0003] In complex electromagnetic warfare environments, improving the anti-jamming capability of integrated communication and sensing systems has become an increasingly prominent research issue in this field. Intermittent sampling-forwarding interference (ISBN) is a typical type of coherent deception interference, usually implemented using digital radio frequency storage technology. This type of interference intermittently samples, stores, and delays the forwarding of signals transmitted by the base station, creating false target components at the sensing receiver that resemble the echo structure of the real target. These false target components manifest as multiple false target peaks in the pulse compression output, thus interfering with the determination of the real target. For integrated communication and sensing systems, ISBN not only reduces the target detection and parameter estimation performance at the sensing end but may also affect the signal-to-interference-plus-noise ratio (SINR) and user communication rate at the communication end. Therefore, anti-jamming design for ISBN requires both suppressing coherent false targets on the sensing side and ensuring multi-user communication rate and quality of service.

[0004] Existing anti-jamming methods for intermittent sampling-forwarding interference mainly fall into two categories: transmit waveform design and receive filtering. These methods can suppress false targets in traditional radar systems, but most are geared towards a single sensing link, focusing primarily on sensing indicators such as false target peak value, sidelobe energy, or mismatched filter output. They do not fully consider multi-user communication rates, inter-user interference, and weak leakage interference feedback at the communication end in integrated communication and sensing scenarios. On the other hand, existing beamforming, resource allocation, and secure transmission optimization methods for integrated communication and sensing are mostly geared towards general noise-type interference, incoherent interference, or ideal channel conditions. They lack dedicated suppression mechanisms for coherent false targets in intermittent sampling-forwarding interference and do not fully consider the joint optimization of the transmit beamforming matrix and the receiver's mismatched filter.

[0005] Furthermore, when intermittent sampling and forwarding interference expands from fixed-delay single-forward to complex forms such as random delay and multiple forwards, the distribution of false targets and the interference structure change. Fixed-model and fixed-parameter methods struggle to simultaneously address false target suppression, communication performance, and the preservation of the true target's main peak. Existing learning-based anti-interference methods mostly focus on channel selection, power control, or resource allocation, making it difficult to directly address coherent false target suppression and communication-aware shared beam optimization. Therefore, it is necessary to provide an anti-intermittent sampling and forwarding interference method for integrated communication-aware systems to achieve a synergistic improvement in perception anti-spoofing capabilities and multi-user communication performance under complex interference scenarios. Summary of the Invention

[0006] The present invention aims to at least partially solve one of the technical problems existing in the related art.

[0007] The purpose of this invention is to provide a method for resisting intermittent sampling and forwarding interference in integrated communication and sensing systems, which solves the problems of false target deception, communication performance degradation, and insufficient adaptive capability in complex interference scenarios in existing integrated communication and sensing systems under intermittent sampling and forwarding interference (ISRJ).

[0008] To achieve the above objectives, this invention provides a method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system, comprising:

[0009] S1. Construct a multi-user, multi-antenna integrated communication and sensing system to simultaneously realize multi-user downlink communication and target sensing under the same spectrum resources and hardware platform;

[0010] S2. Construct a base station transmitted signal model, a source domain fixed-delay single-forward ISRJ model, a target domain random-delay multiple-forward ISRJ model, a real target echo model, a sensed received signal model, and a communication user received signal model, and calculate the false target suppression ratio, the total system communication rate, and the minimum user rate.

[0011] S3. Based on the false target suppression ratio and the total system communication rate, construct the sensing utility and communication utility respectively. Under the constraints of total transmit power, unmatched filter normalization, and real target main peak preservation, establish a joint optimization problem of sensing anti-interference performance and multi-user communication performance.

[0012] S4. The above joint optimization problem is transformed into an upper-level parameter decision problem and a lower-level physical layer optimization problem. The upper-level parameter decision method is based on transfer learning and dual deep Q network, and the lower-level alternating optimization method is used to jointly update the transmit beamforming matrix and unmatched filter, and the state is updated according to the sensing output and communication feedback.

[0013] Preferably, in step S2, the base station transmit signal model is represented as follows:

[0014]

[0015] in, For base station facing A communication user sends a multi-user downlink signal. Represents a multi-user information symbol vector. This represents the multi-user transmit beamforming matrix;

[0016] The source domain fixed-delay single-forward ISRJ model is represented as:

[0017]

[0018]

[0019] in, For source domain fixed-delay single-transmission interference signals, For the source domain interference complex amplitude, For fixed forwarding delay. The source domain interference model corresponds to a single fixed false target scenario; For the jammer to transmit signals, This describes the intermittent sampling process. Type-gated functions, This is the equivalent channel vector from the base station to the intermittent sampling forwarding jammer;

[0020] The target domain random delay multiple forwarding ISRJ model is represented as:

[0021]

[0022] in, To forward the interference signal multiple times with random time delay to the target domain, For the number of reposts, , and The first The complex amplitude, delay, and sampling gating function for each forward;

[0023] The true target echo model is represented as:

[0024]

[0025] in, For the true target echo, For the true target direction, The target complex scattering coefficient, For two-way propagation delay, This is the guide vector for the transmission array;

[0026] The sensing received signal and the communication user received signal model are represented as follows:

[0027]

[0028]

[0029] in, and The source and target domains are respectively sensed and received signals. To sense and receive noise;

[0030] The formula for calculating the false target suppression ratio is:

[0031]

[0032] in, and These are the shift matrices corresponding to the real target range cells and the fake target range cells, respectively. and These are the equivalent effect matrices for the real target and the ISRJ interference, respectively. For the actual target output power, Output total power to the dummy target;

[0033] Total system communication rate The calculation formula is:

[0034]

[0035]

[0036]

[0037] in, For the first Instantaneous signal-to-interference-plus-noise ratio for each communication user For noise power, For the first Channel vectors between each communication user and the ISAC transmitter. The equivalent complex gain of the desired vector, For weak leakage interference at the communication end;

[0038] Minimum User Rate Represented as

[0039] .

[0040] As a preferred option, step S3 specifically involves:

[0041] S31. Construct the perception utility and communication utility based on the false target suppression ratio and the total system communication rate, respectively:

[0042]

[0043]

[0044] in, and These are perception utility and communication utility, respectively. and These are the minimum false target suppression requirements on the sensing side and the minimum total system rate requirements on the communication side, respectively. and These represent the degree to which perception performance and communication performance are met relative to the minimum requirements, respectively.

[0045] S32. Under the constraints of total transmit power, unmatched filter normalization, and real target main peak preservation, establish a joint optimization problem for sensing anti-interference performance and multi-user communication performance:

[0046]

[0047] in, and For the weights of perceived utility and communication utility, satisfy , , ; Maintain a threshold to achieve the true target peak; This represents the maximum permissible transmit power for the base station. - These are the total transmit power constraint, the unmatched filter normalization constraint, and the real target main peak preservation constraint, respectively.

[0048] Preferably, step S4 transforms the aforementioned joint optimization problem into an upper-layer parameter decision problem and a lower-layer physical layer optimization problem. The upper-layer parameter decision method, based on transfer learning and dual deep Q-networks, is used, while the lower-layer method employs an alternating optimization approach to jointly update the transmit beamforming matrix and the unmatched filter. The state is then updated based on the sensing output and communication feedback. Specifically:

[0049] S41. Train DDQN in a source domain fixed-delay single-forward ISRJ scenario to obtain source domain policy network parameters. ;

[0050] S42. Initialize the target domain online Q network using source domain policy network parameters. ;

[0051] S43. The feature importance evaluation method is used to calculate the feature importance distribution of the source domain and the target domain, and the number of frozen layers is determined according to the KL divergence. Only the parameters of the unfrozen layers are updated to obtain the target domain policy network parameters.

[0052] S44. Extract the current perception-communication joint state based on the current target domain policy network parameters;

[0053] S45. Based on the extracted current perception-communication joint state, select an action according to the ε-greedy policy and map the action to perception utility weights. Communication utility weight Maintain a threshold with the true target peak ;

[0054] S46. Based on the selected action and its mapping. , and An alternating optimization algorithm is used to alternately solve the transmit beamforming matrix and the unmatched filter;

[0055] S47. Calculate the reward based on the perception utility, communication utility and communication end leakage interference penalty, obtain the next perception-communication joint state, and store the current perception-communication joint state, action, reward and the next perception-communication joint state into the experience replay pool.

[0056] S48. Sample small batches of data from the experience replay pool, update the target domain online Q network according to the DDQN loss function, and update the target Q network using a soft update method;

[0057] S49. Based on the updated target domain policy network parameters, repeat steps S44-S48 until convergence, and output the target domain policy network parameters, the transmit beamforming matrix, and the unmatched filter.

[0058] Preferably, the number of frozen layers in step S43 The calculation method is as follows:

[0059]

[0060]

[0061] in, The number of transferable layers in the policy network. For adjustment coefficients, To represent the KL divergence between the source and target domains, and They represent the first The importance of each feature in the source and target domains , , To prevent extremely small positive numbers with odd values.

[0062] Preferably, the step of extracting the current sensing-communication joint state in step S44 is as follows:

[0063] S441. Calculate the user's communication observation vector based on the received signal extraction rate, SINR, leakage interference power, and SINR variation:

[0064]

[0065] in, For the first Communication observation vectors of a communication user For the first Individual user communication rate For instantaneous SINR, For communication terminal leakage interference power, The change in SINR between adjacent decision-making periods;

[0066] S442. Compress each user's observations into a system-level communication observation vector. :

[0067]

[0068] in, The total communication rate of the system. For multi-user average SINR, The maximum leakage interference power at the communication end. This represents the average change in SINR.

[0069] S443. Based on the sensing observation vector and the system-level communication observation vector, the current sensing-communication joint state is obtained through concatenation processing:

[0070]

[0071] in, For the first The perception-communication joint state of each decision cycle.

[0072] As a preferred option, the specific method for selecting the action in step S45 is as follows:

[0073]

[0074] in, Action selection for the target domain online Q network, For a set of actions, Let be the action value function for the target domain DDQN.

[0075] Preferably, the transmitted beamforming matrix in step S46 Unmatched filters The specific solution method is as follows: An alternating optimization approach is used: in the first... In each alternation iteration, first fix ,renew ; then in Fixed in each alternating iteration ,renew ,Right now:

[0076]

[0077] .

[0078] As a preferred option, the specific calculation method for the reward in step S47 is as follows:

[0079]

[0080] in, As a reward weight, Indicates the first Statistical data on communication leakage interference during each decision-making cycle;

[0081] Preferably, in step S48, a small batch of data is sampled from the experience replay pool, the target domain online Q-network is updated according to the DDQN loss function, and the target Q-network is updated using a soft update method, specifically as follows:

[0082] S481. Calculate the target value of the empirical sample:

[0083]

[0084] in, As an empirical sample, As a discount factor, Target Q-network parameters for the target domain;

[0085] S482. Based on the target value of the empirical samples, calculate the corresponding DDQN loss function:

[0086]

[0087] in, This refers to a small batch of samples obtained from the experience replay pool;

[0088] S483. Update the target Q network parameters using a soft update method:

[0089]

[0090] in, This is the soft update coefficient.

[0091] Beneficial effects: This invention incorporates real target output, false target output, false target suppression ratio, communication rate, signal-to-interference-plus-noise ratio (SNR), and communication end leakage interference statistics into the state representation, enabling anti-interference decision-making driven by the collaborative efforts of perception observation and communication feedback; This patent jointly optimizes the transmit beamforming matrix and unmatched filter, which can suppress ISRJ coherent false targets while maintaining the main peak of the real target, and also takes into account multi-user communication performance; This patent achieves rapid adaptation from a source domain fixed-delay single-forward scenario to a target domain random-delay multiple-forward scenario through transfer learning, which can improve the adaptive capability under complex ISRJ interference scenarios; This patent adopts a closed-loop hierarchical mechanism that combines upper-layer DDQN parameter decision-making with lower-layer alternating optimization, which can reduce the difficulty of solving the joint optimization problem and improve the robustness of the system in complex electromagnetic countermeasures environments. Attached Figure Description

[0092] Figure 1 This is a model diagram of the multi-user, multi-antenna communication and sensing integrated anti-intermittent sampling and forwarding interference system proposed in this invention.

[0093] Figure 2 This is a system flowchart of the anti-intermittent sampling and forwarding interference method for integrated communication and sensing systems proposed in this invention.

[0094] Figure 3 A comparison chart of normalized pulse compression outputs under ISRJ interference using different methods.

[0095] Figure 4 The curves show the spurious target suppression ratio as a function of the interference-to-information ratio for different methods.

[0096] Figure 5 The curves show the total system communication rate as a function of interference-to-noise ratio for different methods.

[0097] Figure 6 This is a comparison chart of the false target suppression ratios of various methods under different ISRJ modes. Detailed Implementation

[0098] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, embodiments of this invention, and should not be construed as limiting the invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention. In the description of this invention, it should be understood that the terminology used is for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0099] The following is combined Figures 1-6 The present invention describes a method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system. This method involves constructing structured observations on the sensing side and feedback observations on the communication side, jointly optimizing the transmit beamforming matrix and unmatched filter, and combining transfer learning with a dual deep Q network to achieve closed-loop hierarchical anti-interference.

[0100] This patent relates to a multi-user, multi-antenna integrated communication and sensing system. The system includes an integrated communication and sensing base station, K single-antenna communication users, a sensing target, and an ISRJ jammer. The integrated communication and sensing base station is equipped with M transmit antennas, simultaneously performing multi-user downlink communication and target sensing tasks on the same hardware platform and spectrum resources. The ISRJ jammer intermittently samples, stores, and delays the forwarding of the base station's transmitted signals, creating a false target component at the sensing receiver that is coherent with the real target echo, thereby affecting the determination of the real target.

[0101] Example 1: This example provides a method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system, such as... Figure 2 As shown, it includes the following steps:

[0102] Step 1: As Figure 1 As shown, a multi-user, multi-antenna integrated communication and sensing anti-intermittent sampling and forwarding jamming system is constructed. The system includes an integrated communication and sensing base station, K single-antenna communication users, one sensing target, and an ISRJ jammer. The integrated communication and sensing base station is equipped with M transmit antennas, simultaneously completing multi-user information transmission and target sensing tasks under the same hardware platform and spectrum resources. The K communication users are distributed within the base station's coverage area to receive downlink communication signals transmitted by the base station. The sensing target is located within the base station's sensing area and is used to generate real target echoes. The ISRJ jammer intercepts the transmitted signals of the integrated communication and sensing base station, intermittently samples, stores, and delays their forwarding, forming false target interference that is coherent with the real target echo at the sensing receiver.

[0103] Step 2: Regarding the system model, this patent first establishes a transmission signal model for the integrated communication and sensing base station, which is oriented towards... A communication user sends a multi-user downlink signal, and the transmitted signal is represented as follows:

[0104]

[0105] in, Represents a multi-user information symbol vector. This represents the multi-user transmit beamforming matrix. The total transmit power of the integrated communication and sensing base station satisfies the following constraints:

[0106]

[0107] For intermittent sampling and forwarding interference signals, let the equivalent channel vector from the integrated communication and sensing base station to the intermittent sampling and forwarding jammer be... The sampling gate function is The jammer then transmits the following signal:

[0108]

[0109] in, This describes the intermittent sampling process. A type-gated function. Furthermore, in the source domain, the jammer uses a fixed-delay single-forward method, and its forwarded jamming signal is:

[0110]

[0111] in, For the source domain interference complex amplitude, The forwarding delay is fixed. The source domain interference model corresponds to a single fixed fake target scenario. In the target domain, the intermittent sampling forwarding jammer uses a random delay multiple forwarding method, and its forwarding interference signal is:

[0112]

[0113] in, For the number of reposts, , and The first The complex amplitude, delay, and sampling gating function of each forwarding are considered. Compared to the source domain fixed single-false-target scenario, the target domain ISRJ has random delay and multiple false-target characteristics, thus making the interference structure more complex.

[0114] The total echo received by the sensing receiver consists of the real target echo, ISRJ interference echo, and received noise. Let the direction of the real target be... The target complex scattering coefficient is Two-way propagation delay is The guiding vector of the transmission array is Real target echo It can be represented as:

[0115]

[0116] Combining the source and target domain ISRJ interference echoes, the sensed received signals in the source and target domains are as follows:

[0117]

[0118]

[0119] in, and The source and target domains are respectively sensed and received signals. To sense and receive noise.

[0120] Therefore, the false target suppression ratio is defined as the evaluation index on the sensing side as follows:

[0121]

[0122] in, and These are the shift matrices corresponding to the real target range cells and the fake target range cells, respectively. and These are the equivalent action matrices for the real target and the ISRJ interference, respectively. For the actual target output power, The total output power is for the false target. The larger the FSR, the stronger the separability between the real target and the false target, and the better the system's ability to suppress ISRJ false targets.

[0123] Therefore, the observation vector at the sensing end Defined as:

[0124]

[0125] In a communication link, let the first... The channel vector between each communication user and the ISAC transmitter is The equivalent leakage interference generated by the jammer at the user end is Communication noise is Then the first Each user receives the following signal:

[0126]

[0127] in, For the desired signal, To prevent interference from multiple users, For weak leakage interference at the communication end, It is noise.

[0128] Furthermore, to characterize the weak leakage interference experienced by the communication end, the first... Leakage interference power of individual communication user terminals for:

[0129]

[0130] Based on the above received signal model, the first Instantaneous signal-to-interference-plus-noise ratio of individual communication users for:

[0131]

[0132] Furthermore, the total communication rate of the system for:

[0133]

[0134] in, For the first per user rate This refers to the communication bandwidth.

[0135] In addition, to characterize the service guarantee capability of weak users in multi-user communication, a minimum user rate is defined. for

[0136]

[0137] ISRJ primarily functions at the base station sensing receiver, but communication users are also susceptible to weak leakage interference. The receiving state of the communication link is extracted into rate, instantaneous SINR, leakage interference power, and SINR change, and compressed into system-level communication observations, which, together with sensing observations, constitute the joint state input of the upper-layer DDQN. Therefore, the first... Each communication user's communication observation vector is defined based on the received signal extraction rate, SINR, leakage interference power, and SINR variation. for:

[0138]

[0139] in, For the first Individual user communication rate For instantaneous SINR, For communication terminal leakage interference power, The SINR change over adjacent decision periods is used to reflect communication quality fluctuations caused by changes in interference conditions.

[0140] To reduce the dimensionality of the upper-level states while preserving the overall communication state of multiple users, the observations of each user are compressed into system-level communication observation vectors. :

[0141]

[0142] in, The total communication rate of the system. For multi-user average SINR, The maximum leakage interference power at the communication end. This represents the average change in SINR.

[0143] Finally, to simultaneously characterize the anti-interference state of the sensing side and the service quality state of the communication side, the sensing observation vector is concatenated with the system-level communication observation vector, defining the first... The perception-communication joint state for each decision cycle is as follows:

[0144]

[0145] in, For the first The perception-communication joint state of each decision cycle serves as the input for the subsequent upper-layer DDQN.

[0146] Step 3: Due to the false target suppression ratio and the total communication rate of the system Due to differences in physical meaning and numerical scale, this patent transforms it into relative demand satisfaction utility from the perspective of system performance requirements. Let... and Let the minimum false target suppression requirement on the sensing side and the minimum total system rate requirement on the communication side be defined respectively.

[0147]

[0148]

[0149] in, and These represent the degree to which perception performance and communication performance are satisfied relative to the minimum requirements, respectively; the logarithmic function is used to characterize the diminishing marginal returns of performance improvements.

[0150] Therefore, the overall optimization problem can be expressed as:

[0151]

[0152] in, and For the weights of perceived utility and communication utility, satisfy , , ; Maintain a threshold to reflect the true target peak. Constraints — These correspond to the total transmit power constraint, the unmatched filter normalization constraint, and the real target main peak preservation constraint, respectively.

[0153] Step 4: Transform the overall optimization problem into a higher-level parameter decision problem and a lower-level physical layer optimization problem. First, and The problem is coupled with the constraints of perception utility and peak preservation, and the communication utility is affected by multi-user interference and communication leakage interference, resulting in significant non-convexity. Traditional optimization methods cannot directly obtain the global optimum. Secondly, , and The trade-off between sensing anti-interference, communication transmission, and maintaining the main peak of the real target is crucial. If fixed parameters or manual search methods are used, it is difficult to maintain stable performance when the interference structure changes.

[0154] Therefore, the fixed-delay single-forward ISRJ scenario is defined as the source domain, and the random-delay multiple-forward ISRJ scenario is defined as the target domain. The target domain policy network is initialized using a source domain pre-trained strategy, and transfer learning is used to achieve rapid adaptation of the source domain policy to complex interference scenarios in the target domain. The upper layer selects based on the perception-communication joint state. , and The lower layer jointly optimizes the transmit beamforming matrix after the parameters are given. Unmatched filters .

[0155] Upper-layer parameter decision-making is based on transfer learning and a dual-deep Q-network. First, a DDQN is trained in a single-forward ISRJ scenario with a fixed delay in the source domain to obtain the source domain policy network parameters, which are then used to initialize the online Q-network in the target domain. Next, the importance distributions of the state features in the source and target domains are extracted separately, and the feature differences between the source and target domains are measured using KL divergence. The number of frozen layers for fine-tuning the target domain is determined based on these feature differences. In a multi-forward ISRJ scenario with a random delay in the target domain, only the parameters of the unfrozen layers are updated to obtain the target domain policy network parameters. In each decision cycle, the upper-layer DDQN takes the perception-communication joint state as input, selects an action based on the action value function, and maps the selected action to perception utility weights. Communication utility weight Maintain a threshold with the true target peak After the lower-level optimization is completed, the reward is calculated based on the perception utility, communication utility, and communication end leakage interference penalty, and the target domain online Q network and target Q network are updated using experience replay samples.

[0156] First, in the source domain Train DDQN to obtain source domain policy network parameters Then, it is used to initialize the target domain policy network:

[0157]

[0158] This process preserves the learned basic anti-interference rules in the source domain and improves parameter selection capabilities in complex ISRJ scenarios in the target domain. This patent employs a feature importance assessment method to extract the importance distributions of state features in both domains. Let the feature importance distributions of the source and target domains be respectively... and ,in , The difference between the two domains is then expressed by the KL divergence as:

[0159]

[0160] in, and They represent the first The importance of each feature in the source and target domains To prevent extremely small positive numbers with odd values. The larger the value, the more significant the difference in feature distribution between the target domain and the source domain.

[0161] Furthermore, the number of freeze layers during target domain fine-tuning is determined based on the differences between the two domains. Assume the policy network has a total of... There are 1 transferable layer and 1 frozen layer. ,but

[0162]

[0163] in, is the adjustment coefficient. As can be seen from equation (4.3), when the difference between the source domain and the target domain is small, more layers are frozen to preserve the basic anti-interference rules already learned in the source domain; when the difference between the two domains is large, the number of frozen layers is reduced so that more network layers can participate in the training of the target domain.

[0164] After determining the number of frozen layers, the target domain only updates the parameters of the unfrozen layers. The target domain migration fine-tuning problem is represented as follows:

[0165]

[0166] Target domain transfer learning loss function for:

[0167]

[0168] In the formula, The target domain DDQN loss function is... and These represent the parameters corresponding to the unfrozen layers in the target and source domain networks, respectively, with superscripts indicating their positions. Indicates an unfrozen layer; This refers to the parameter offset penalty weight related to domain differences. This patent takes...

[0169]

[0170] in, , To adjust the parameters. When the source domain and target domain are relatively close, A larger value is used to suppress excessive deviation of the target domain parameters from the source domain parameters; when the difference between the two domains is large, This reduces the size of the target domain network, making it more adaptable.

[0171] After completing the target domain migration fine-tuning, the upper-level DDQN is used to select the lower-level optimization parameters in each decision cycle. Let the first... The perception-communication joint state for each decision cycle is as follows: The action set is Then the action selection of the target domain online Q network for

[0172]

[0173] in, This is the action value function for the target domain DDQN. Upper-level actions do not directly output physical layer variables, but rather correspond to lower-level optimization parameters for the current decision cycle. Specifically, actions... Perceived utility weights are determined through a pre-defined mapping. Communication utility weight Maintain a threshold with the true target peak Therefore, the role of the upper-level DDQN is to adaptively determine the weights and threshold parameters in equation (21) based on the current state.

[0174] action Determined through action mapping Then it is handed over to the lower level for solution. .

[0175] After the lower-level optimization is completed, the system calculates the reward based on perception utility, communication utility, and leakage interference level. This patent describes the reward function as follows:

[0176]

[0177] in, As a reward weight, Indicates the first The statistics of communication leakage interference in each decision cycle. Rewards encourage both false target suppression and improved communication performance, while penalizing residual leakage interference.

[0178] To alleviate the overestimation of Q-values ​​in traditional DQN, this patent employs the DDQN update method. Let the target domain target Q-network parameters be... For empirical samples Its target value is

[0179]

[0180] in, Let be the discount factor. The corresponding DDQN loss function is...

[0181]

[0182] in, This is a mini-batch set of samples obtained from the experience replay pool. The target network uses a soft update method.

[0183]

[0184] in, This is the soft update coefficient.

[0185] After the upper-level DDQN completes parameter selection and network updates, it obtains the optimal action for the current decision cycle. And the lower-level optimization parameters are determined by this action. , and The lower layer performs joint optimization of the transmit beamforming matrix W and the unmatched filter h after the parameters are given. Since the two are coupled in perceptual utility and peak preservation constraints, this patent uses an alternating optimization method to solve the problem: in the first... In each alternation iteration, first fix ,renew ; then fixed ,renew ,Right now

[0186]

[0187]

[0188] (1) Unmatched filter update:

[0189] fixed Afterwards, the communication utility item and Unrelated to the target, the unmatched filter is mainly used to enhance the output of the real target and suppress the output of the false target. The equivalent response vector of the real target is defined as...

[0190]

[0191] The equivalent response vector of the sham target is

[0192]

[0193] The corresponding pseudo-target covariance matrix is

[0194]

[0195] To ensure the true target's main peak response and suppress false target output, the unmatched filter is updated using the minimum disturbance energy constraint response criterion:

[0196]

[0197] To satisfy constraints Further normalization to

[0198]

[0199] in, It is the identity matrix. It is a very small positive number, used to improve the numerical stability of matrix inversion.

[0200] (2) Update the transmitted beamforming matrix:

[0201] fixed Subsequently, the transmitted beamforming matrix needs to simultaneously affect both sensing and communication utility. This patent employs an update method combining gradient correction and power projection:

[0202]

[0203] in, For the first The step size of the next iteration. Represents the relationship between complex matrices The Wirtinger gradient is then projected onto the power constraint set:

[0204]

[0205] in, Represents a set The projection operator, with the power constraint set as follows

[0206]

[0207] because Projection can be written as

[0208]

[0209] If the updated The main peak retention constraint is not met. Then reduce the step size. Then repeat equations (4.19) to (4.20) until the feasibility conditions are met.

[0210] The termination condition for alternating optimization is set as follows:

[0211]

[0212] in, This is the convergence threshold.

[0213] The closed-loop hierarchical anti-ISRJ joint optimization algorithm is shown in Table 1:

[0214]

[0215] To verify the effectiveness of this embodiment, the following simulation design was used for verification:

[0216] A multi-user, multi-antenna integrated communication and sensing anti-ISRJ interference scenario was constructed. The system includes an integrated communication and sensing base station, K single-antenna communication users, a stationary sensing target, and an ISRJ jammer. The basic sensing waveform uses LFM signals, with the real target located at range cell 120 and a typical false target located at range cell 140. The source domain is set to fixed-delay single-forward ISRJ, and the target domain is set to random-delay multiple-forward ISRJ. The interference modes include direct forwarding, repeated forwarding, and cyclic forwarding. The experiment compares the matched filtering method, the sensing-only optimization scheme, the no-transfer-learning scheme, and the method of this patent (described below by the proposed algorithm). The false target suppression ratio, the total system communication rate, and the minimum user rate are used as evaluation indicators to verify the anti-interference performance and communication assurance capability of this patent under different interference-to-signal ratios, forwarding times, sampling duty cycles, and interference modes.

[0217] Figure 3 The pulse compression output results under different methods are presented. It can be seen that the matched filtering method exhibits a significant false target peak at range cell 140, with an amplitude close to or even higher than the true target peak, indicating that traditional matched filtering is ineffective in suppressing coherent false targets formed by ISRJ. The sensing-only scheme can reduce the false target peak, but lacks joint constraints on the true target main peak and communication-side performance, thus its suppression effect is limited. The no-transfer-learning scheme further enhances the distinction between true and false targets, but due to the lack of policy transfer from the source domain to the target domain, it still suffers from insufficient adaptation in scenarios with multiple random forwardings in the target domain. In contrast, the proposed algorithm maintains a high true target peak at range cell 120 while significantly reducing the false target peak at range cell 140. This demonstrates that the proposed method can effectively suppress ISRJ false targets while ensuring the main peak's discernibility, showcasing the advantages of joint optimization of the transmit beamforming matrix and the unmatched filter.

[0218] Figure 4 The false target suppression ratio is given under varying interference-to-signal ratio (ISR). As the ISR increases from... As the interference power increases to 20 dB, the false target suppression ratio (SRR) of all methods decreases. This is because the increased interference power leads to an increase in the energy of ISRJ false targets in the pulse compression output, reducing the separability between real and false targets. Under all interference-to-signal ratio (ISR) conditions, the proposed algorithm achieves the highest false target SRR. For example, in the low ISR ratio region, the proposed algorithm maintains a significant gain compared to the non-transfer learning scheme; in the high ISR ratio region, although the performance of all methods decreases, the proposed algorithm still maintains a positive suppression advantage. This result demonstrates that the proposed closed-loop hierarchical mechanism can adaptively adjust the lower-level optimization parameters according to the interference intensity, thereby improving the ISRJ resistance under strong interference conditions.

[0219] Figure 5The overall system rate is presented under varying interference-to-noise ratio (INR). As the interference SNR increases, the overall rate of all four methods decreases significantly. This is because weak leakage interference exists at the user end; increased interference power reduces the user SINR, thus decreasing the overall system communication rate. The proposed algorithm achieves the highest overall rate across the entire INR range. This demonstrates that the proposed method suppresses false targets in the sensing link without sacrificing communication performance. Instead, it participates in upper-level decision-making through low-dimensional feedback on the communication side, enabling lower-level beam optimization to consider multi-user communication quality. Compared to sensing-only schemes, the proposed algorithm shows a more significant rate advantage, validating the necessity of introducing communication constraints and feedback in ISAC anti-interference design.

[0220] Figure 6 The false target suppression ratios (FSRs) of three ISRJ modes—direct forwarding, repeated forwarding, and cyclic forwarding—were compared. It can be seen that the proposed algorithm achieves the highest FSR in all three interference modes, indicating its good adaptability to different forwarding mechanisms. Comparing different modes, the FSR of each method is relatively high in the direct forwarding scenario, while it decreases in the repeated and cyclic forwarding scenarios. This is because repeated and cyclic forwarding increase the structural complexity of the false targets, making their energy distribution more dispersed or persistent, thus increasing the difficulty of suppression. The proposed algorithm maintains its advantage even in complex modes, demonstrating that the closed-loop hierarchical strategy and transfer learning mechanism can improve the algorithm's robustness to different ISRJ modes.

[0221] 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 them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system, characterized in that, include: S1. Construct a multi-user, multi-antenna integrated communication and sensing system to simultaneously realize multi-user downlink communication and target sensing under the same spectrum resources and hardware platform; S2. Construct a base station transmitted signal model, a source domain fixed-delay single-forward ISRJ model, a target domain random-delay multiple-forward ISRJ model, a real target echo model, a sensed received signal model, and a communication user received signal model, and calculate the false target suppression ratio, the total system communication rate, and the minimum user rate. S3. Based on the false target suppression ratio and the total system communication rate, construct the sensing utility and communication utility respectively. Under the constraints of total transmit power, unmatched filter normalization, and real target main peak preservation, establish a joint optimization problem of sensing anti-interference performance and multi-user communication performance. S4. The above joint optimization problem is transformed into an upper-level parameter decision problem and a lower-level physical layer optimization problem. The upper-level parameter decision method is based on transfer learning and dual deep Q network, and the lower-level alternating optimization method is used to jointly update the transmit beamforming matrix and unmatched filter, and the state is updated according to the sensing output and communication feedback.

2. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 1, characterized in that, In step S2, the base station transmit signal model is represented as follows: ； in, For base station facing One communication user sends a multi-user downlink signal. Represents a multi-user information symbol vector. This represents the multi-user transmit beamforming matrix; The source domain fixed-delay single-forward ISRJ model is represented as: ； ； in, For source domain fixed-delay single-transmission interference signals, For the source domain interference complex amplitude, Fixed forwarding delay; To transmit signals for the jammer This describes the intermittent sampling process. Type-gated functions, This is the equivalent channel vector from the base station to the intermittent sampling forwarding jammer; The target domain random delay multiple forwarding ISRJ model is represented as: ； in, To forward the interference signal multiple times with random time delay to the target domain, For the number of reposts, , and The first The complex amplitude, delay, and sampling gating function for each forward; The true target echo model is represented as: ； in, For the true target echo, For the true target direction, The target complex scattering coefficient, For two-way propagation delay, This is the guide vector for the transmission array; The sensing received signal and the communication user received signal model are represented as follows: ； ； in, and The source and target domains are respectively sensed and received signals. To sense and receive noise; The formula for calculating the false target suppression ratio is: ； in, and These are the shift matrices corresponding to the real target range cells and the fake target range cells, respectively. and These are the equivalent effect matrices for the real target and the ISRJ interference, respectively. For the actual target output power, Output total power to the dummy target; Total system communication rate The calculation formula is: ； ； ； in, For the first Instantaneous signal-to-interference-plus-noise ratio for each communication user For noise power, For the first Channel vectors between each communication user and the ISAC transmitter. The equivalent complex gain of the desired vector, For weak leakage interference at the communication end; Minimum User Rate Represented as 。 3. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 2, characterized in that, Step S3 is as follows: S31. Construct the perception utility and communication utility based on the false target suppression ratio and the total system communication rate, respectively: ； ； in, and These are perception utility and communication utility, respectively. and These are the minimum false target suppression requirements on the sensing side and the minimum total system rate requirements on the communication side, respectively. and These represent the degree to which perception performance and communication performance are met relative to the minimum requirements, respectively. S32. Under the constraints of total transmit power, unmatched filter normalization, and real target main peak preservation, establish a joint optimization problem for sensing anti-interference performance and multi-user communication performance: ； in, and For the weights of perceived utility and communication utility, satisfy , , ; Maintain a threshold to achieve the true target peak; This represents the maximum permissible transmit power for the base station. — These are the total transmit power constraint, the unmatched filter normalization constraint, and the real target main peak preservation constraint, respectively.

4. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 3, characterized in that, Step S4 transforms the aforementioned joint optimization problem into an upper-layer parameter decision problem and a lower-layer physical layer optimization problem. Using an upper-layer parameter decision method based on transfer learning and dual deep Q-networks, and an alternating optimization method for the lower layer, the transmit beamforming matrix and unmatched filters are jointly updated. The state is then updated based on the sensing output and communication feedback. Specifically: S41. Train DDQN in a source domain fixed-delay single-forward ISRJ scenario to obtain source domain policy network parameters. ; S42. Initialize the target domain online Q network using source domain policy network parameters. ; S43. The feature importance evaluation method is used to calculate the feature importance distribution of the source domain and the target domain, and the number of frozen layers is determined according to the KL divergence. Only the parameters of the unfrozen layers are updated to obtain the target domain policy network parameters. S44. Extract the current perception-communication joint state based on the current target domain policy network parameters; S45. Based on the extracted current perception-communication joint state, select an action according to the ε-greedy policy and map the action to perception utility weights. Communication utility weight Maintain a threshold with the true target peak ; S46. Based on the selected action and its mapping. , and An alternating optimization algorithm is used to alternately solve the transmit beamforming matrix and the unmatched filter; S47. Calculate the reward based on the perception utility, communication utility and communication end leakage interference penalty, obtain the next perception-communication joint state, and store the current perception-communication joint state, action, reward and the next perception-communication joint state into the experience replay pool. S48. Sample small batches of data from the experience replay pool, update the target domain online Q network according to the DDQN loss function, and update the target Q network using a soft update method; S49. Based on the updated target domain policy network parameters, repeat steps S44-S48 until convergence, and output the target domain policy network parameters, the transmit beamforming matrix, and the unmatched filter.

5. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 4, characterized in that, Number of frozen layers in step S43 The calculation method is as follows: ； ； in, The number of transferable layers in the policy network. For adjustment coefficients, To represent the KL divergence between the source and target domains, and They represent the first The importance of each feature in the source and target domains , , To prevent extremely small positive numbers with odd values.

6. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 5, characterized in that, The extraction step of the current sensing-communication joint state in step S44 is as follows: S441. Calculate the user's communication observation vector based on the received signal extraction rate, SINR, leakage interference power, and SINR variation: ； in, For the first Communication observation vectors of a communication user For the first Individual user communication rate For instantaneous SINR, For communication terminal leakage interference power, The change in SINR between adjacent decision-making periods; S442. Compress each user's observations into a system-level communication observation vector. : ； in, The total communication rate of the system. For multi-user average SINR, The maximum leakage interference power at the communication end. This represents the average change in SINR. S443. Based on the sensing observation vector and the system-level communication observation vector, the current sensing-communication joint state is obtained through concatenation processing: ； in, For the first The perception-communication joint state of each decision cycle.

7. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 6, characterized in that, The specific method for selecting the action in step S45 is as follows: ； in, Action selection for the target domain online Q network, For a set of actions, Let be the action value function for the target domain DDQN.

8. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 7, characterized in that, In step S46, the transmitted beamforming matrix is ​​transmitted. Unmatched filters The specific solution method is as follows: An alternating optimization approach is used: in the first... In each alternation iteration, first fix ,renew ; then in Fixed in each alternating iteration ,renew ,Right now: ； 。 9. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 8, characterized in that, The specific calculation method for the reward in step S47 is as follows: ； in, As a reward weight, Indicates the first The statistics of communication leakage interference in each decision cycle.

10. The method for resisting intermittent sampling and forwarding interference in an integrated communication and sensing system according to claim 9, characterized in that, In step S48, mini-batch data is sampled from the experience replay pool, the target domain online Q-network is updated according to the DDQN loss function, and the target Q-network is updated using a soft update method, specifically as follows: S481. Calculate the target value of the empirical sample: ； in, As an empirical sample, As a discount factor, Target Q-network parameters for the target domain; S482. Based on the target value of the empirical samples, calculate the corresponding DDQN loss function: ； in, This refers to a small batch of samples obtained from the experience replay pool; S483. Update the target Q network parameters using a soft update method: ； in, This is the soft update coefficient.

Images