A Perception-Assisted Distributed Multi-Cell Dynamic Resource Allocation Method

Abstract

This invention relates to the field of resource allocation in wireless communication, specifically to a perception-assisted distributed multi-cell dynamic resource allocation method. This method achieves performance close to centralized multi-cell resource allocation with low inter-station interaction and low channel estimation cost. The technical solution includes: estimating user kinematic parameters and CSI based on dual-function radar signal echoes; determining the user set to be scheduled at the local station based on the estimated CSI; estimating the CSI from the local station to neighboring stations by interactively scheduling user kinematic parameters; constructing a distributed beamforming optimization problem model; and optimizing beamforming using fractional programming combined with semidefinite relaxation and solving for the beamforming vector. This invention is applicable to multi-cell resource allocation.

Description

Technical Field

[0001] This invention relates to the field of resource allocation in wireless communication, and more specifically to a distributed multi-cell dynamic resource allocation method based on perception assistance. Background Technology

[0002] Multi-cell Resource Allocation (MCRA) improves network spectral efficiency by effectively designing user scheduling sets and co-coding. Theoretically, all base stations can transmit CSI (Channel State Information) and other data to a central processing unit (CPU) via wireless backhaul links. In this case, the multi-cell network can be viewed as a large multiple-input multiple-output (MIMO) system to centrally allocate resources, fully leveraging the performance advantages of MCRA. However, as network scale increases, it becomes impractical for base stations to transmit all information back to the CPU for collaborative resource allocation. The main reasons are twofold: First, centralized signal processing is computationally very complex, and a central server with such powerful computing capabilities may not be available. Second, with limited backhaul bandwidth, sharing global channel state information leads to a significant communication burden and high network latency.

[0003] To address the aforementioned two challenges, existing technologies have proposed a signal measurement scheme, SLINR (Signal-to-leakage plus interference plus noise ratio), to replace the traditional SINR (Signal-to-interference plus noise ratio). In this scheme, each base station considers intra-cell interference and power leakage to users in other cells based solely on its own CSI (Collateral Significance), and employs a traffic model to implicitly characterize user scheduling within the cell, thereby achieving distributed multi-cell resource allocation. However, due to the differences between SLINR and SINR, and the fact that traffic models are often irregular in real-world scenarios, this directly leads to ineffective resource allocation. Furthermore, in dynamic networks, frequent channel estimation of the CSI for all users within the system by each base station not only results in significant signaling overhead, but the limited number of orthogonal pilots also restricts the accuracy of CSI estimation. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a perception-assisted distributed multi-cell dynamic resource allocation method, which achieves a performance that approximates that of centralized multi-cell resource allocation with a scheme that has low inter-station interaction and low channel estimation cost.

[0005] The present invention achieves the above objectives by adopting the following technical solution: the present invention provides a distributed multi-cell dynamic resource allocation method based on perception assistance, comprising:

[0006] Step 1: Estimate user kinematic parameters and CSI based on dual-function radar signal echo;

[0007] Step 2: Determine the set of users to be scheduled at this station based on the estimated CSI;

[0008] Step 3: Base stations interactively schedule user motion parameters to estimate the CSI from local to neighboring stations;

[0009] Step 4: Construct a distributed beamforming optimization problem model;

[0010] Step 5: Optimize beamforming by combining fractional programming with semidefinite relaxation and solve for the beamforming vector.

[0011] Furthermore, step 1 specifically includes:

[0012] Each base station uses a sensing-integrated beam to transmit dual-function radar signals to the user terminal, while simultaneously estimating the user's kinematic parameters based on the reflected signals from the user terminal and constructing the corresponding channel state vector for the next time step.

[0013] Estimating the user's kinematic parameters based on the reflected signal from the user terminal and constructing the corresponding channel state vector for the next time step specifically includes:

[0014] At time n-1, base station l, First, a receive filter is used to distinguish the reflected signals from different users. Then, the scheduling user k is estimated based on the matched filtering method. l latency and Doppler shift Final estimated angle Based on the kinematic evolution model, the path loss coefficient predicted at time n is obtained. Doppler shift and angle To construct base station l and serving user k l Direct channel between: Where, N t This indicates the number of transmit antennas at the base station. This represents the set of users served by base station l.

[0015] Furthermore, step 2 specifically includes: constructing the corresponding channel state vector for the next time step based on the base station, and determining the set of users to be scheduled by the base station in the next time step.

[0016] A greedy proportional fairness zero-forcing scheme is used to select the user set for scheduling at time n, specifically including:

[0017] Step 21: Select a set of data in continuous T... s -1 user who was not scheduled at any given time Solving using equal-power zero-forcing beamforming The sum and rate of users

[0018] Step 22, Calculation in It is the historical average rate of user i at the previous n-1 time steps, which is fed back to the serving base station via uplink. and

[0019] Step 23: If R ≥ R′, then Repeat steps 21 and 22 until R < R′; output The user set scheduled by base station l at time n.

[0020] Furthermore, step 3 specifically includes: base stations exchanging kinematic parameters of users to be scheduled by the base station through the backhaul network to assist the base station in estimating its own channel state vectors and those of other neighboring base stations serving users.

[0021] The interaction between base stations via the backhaul network specifically includes:

[0022] Base station l is determined based on its own two-dimensional coordinates Two-dimensional coordinates of the user and movement speed Consider the inherent latency T of the backhaul network d Base station l before the next information transmission time T d By exchanging information with other base stations, base station l estimates its own scheduling of user t from other base stations m. m Direct transmission channel:

[0023] Furthermore, step 4 specifically includes: the base station using the estimated channel state vector to construct a distributed beamforming optimization problem model with virtual proportional fairness and rate maximization under power constraints and perception error constraints.

[0024] The virtual proportional fairness and rate maximization problem model is expressed as:

[0025]

[0026]

[0027]

[0028] in, This represents the precoding vector of the k-th user in cell l at time node n; P represents the user scheduling set of cell l at time node n; l This indicates the maximum transmit power of base station l; User k l The estimation error of i (including angle, time delay, and Doppler shift) is required to be less than a tolerable constant c; additionally, User k l The virtual rate function at time n User k l The average rate in the first n-1 time steps is used as the weighting value.

[0029] Furthermore, optimizing beamforming and solving for the beamforming vector through fractional programming combined with semidefinite relaxation specifically includes: introducing auxiliary variables. and The objective function is equivalent to,

[0030]

[0031] when fixed, but

[0032] The semidefinite problem is represented as:

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] Where G is the matched filter gain. and

[0039] in addition, It is the antenna gain of the base station, a i It is the constant of perception error;

[0040]

[0041]

[0042]

[0043] If the rank-1 constraint is not considered, the optimal beamforming vector can be obtained by eigenvalue decomposition.

[0044] The beneficial effects of this invention are as follows:

[0045] This invention provides a low-interaction, low-cost distributed dynamic multi-cell cooperative user scheduling and precoding design scheme. Its interaction level is extremely low, far below the CSI level, and it will not affect backhaul bandwidth, making it easier to scale to future large-scale heterogeneous network systems. Simultaneously, based on base station sensing capabilities, it estimates user motion parameters and constructs dynamic network channels, thereby reducing the estimation cost of time-varying channels.

[0046] The proportional fairness and rate performance of this invention can be significantly improved compared to scenarios without perceptual assistance, and approach the performance of centralized multi-cell resource allocation. Attached Figure Description

[0047] Figure 1 This is a flowchart of a distributed multi-cell dynamic resource allocation process based on perception assistance provided in an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of a sensor-integrated system model provided in an embodiment of the present invention;

[0049] Figure 3 This is a schematic diagram of the resource allocation framework provided in an embodiment of the present invention;

[0050] Figure 4 This is a performance diagram of resource allocation in a dynamic scenario provided by an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0052] This invention provides a perception-assisted distributed multi-cell dynamic resource allocation method, such as... Figure 1 As shown, it includes:

[0053] Step 1: Estimate user kinematic parameters and CSI based on dual-function radar signal echo;

[0054] Step 2: Determine the set of users to be scheduled at this station based on the estimated CSI;

[0055] Step 3: Base stations interactively schedule user motion parameters to estimate the CSI from local to neighboring stations;

[0056] Step 4: Construct a distributed beamforming optimization problem model;

[0057] Step 5: Optimize beamforming by combining fractional programming with semidefinite relaxation and solve for the beamforming vector.

[0058] In one embodiment of the present invention, step 1 specifically includes:

[0059] Each base station uses an integrated sensing beam to transmit dual-function radar signals to the user terminal, while simultaneously estimating the user's kinematic parameters based on the reflected signals from the user terminal and constructing the corresponding channel state vector for the next time step. The integrated sensing system model is as follows: Figure 2 As shown.

[0060] Step 2 specifically includes: constructing the corresponding channel state vector for the next time step based on the base station, and determining the set of users to be scheduled by the base station in the next time step.

[0061] Step 3 specifically includes: base stations exchanging kinematic parameters of users to be scheduled by the base station through the backhaul network to help the base station estimate its own channel state vectors and those of other neighboring base stations serving users.

[0062] Step 4 specifically includes: The base station uses the estimated channel state vector to construct a distributed beamforming optimization problem model with virtual proportional fairness and rate maximization under power constraints and perception error constraints.

[0063] Taking downstream collaborative multi-point joint transmission as an example, such as Figure 2 As shown, consider a cellular millimeter-wave system with L base stations, whose set is represented as follows: Each base station is equipped with N t Root transmitting antenna and N r A uniform linear array of receiving antennas. This antenna array is used to receive echo signals to actively sense user motion parameters, assuming a sufficiently large number of antennas. Let... Indicates the connection with base station l, Associated user set, k l express The k-th user in the network. Assume that there is always a Loss-of-Stake (LoS) channel between each user and its serving base station, while the inter-cell channels between a base station and users in neighboring cells consist of LoS and multiple channels, such as Ricean fading channels. All base stations are interconnected via backhaul with limited capacity. The transmission time T is divided into N equal-length intervals epochs, each epoch having a duration of ΔT. The backhaul delay is T. d And satisfy T d <ΔT.

[0064] User k l The transmitted signal within epoch n is represented as: The corresponding transmit precoding vector is represented as If user k l If scheduled at epoch n, then otherwise Therefore, user k l The received signal is in, and This indicates the distance from base station m to user k. l The downlink channel, and It is additive white Gaussian noise.

[0065] Proportional fairness is one of the most important transmission metrics in dynamic networks, and the problem of maximizing PFR (Proportional Fairness Rate) can be solved by processing the following problem model at each epoch n:

[0066]

[0067]

[0068]

[0069] Where P l This represents the maximum transmit power of base station l, and the cumulative average rate over the first (n-1) epochs is... and User k l The transmission rate at epochn'.

[0070] The problem model described above is a complex mixed-integer nonlinear programming problem that requires exchanging instantaneous CSI. However, this can lead to significant latency with limited backhaul bandwidth. Furthermore, for mobile users, frequent CSI estimation is required to achieve timely results, resulting in substantial signaling overhead.

[0071] To address the aforementioned issues, this invention develops a novel distributed MCRA framework using base station radar sensing technology. First, this invention employs a SALINR signal metric criterion, which relies solely on CSI originating from the base station for distributed optimization of the beamforming vector. Second, the reflected radar signal is used to allow each base station l to estimate user k. l Kinematic parameters (including angles) distance and speed Under the Loss channel, the acquired kinematic parameters can be used for BSL estimation of the user's channel. Furthermore, it can interact with other base stations to obtain cross-cell LoS channels. The estimated channels are then used for subsequent user scheduling. and beamforming optimization.

[0072] The echo signal received by base station l at epoch n can be modeled as follows:

[0073] in It is the antenna array gain. They represent user j respectively m For base station l, the reflection coefficient, Doppler frequency, and round-trip delay, This represents symmetric complex Gaussian noise. Furthermore, the transmit steering vector a(θ) and the receive steering vector b(θ) are respectively:

[0074]

[0075]

[0076] Since the user angle changes very little within the epoch interval, when k′ l ≠k l At that time, one can obtain and Therefore, base station l with respect to user k l The echo signal is:

[0077]

[0078] in, Using matched filters and angle estimation methods (such as MUSIC), regarding the angle... Delay and Doppler frequency The measured noises are respectively and and

[0079]

[0080] Where G is the matched filter gain, a i ,i={θ,τ,μ} are constants related to system configuration.

[0081] Base station l relies on a state evolution model (such as an extended Kalman filter) to make a one-step prediction, which can obtain the relationship with user k. l Related angles distance speed and reflection coefficient Furthermore, the channel vector can be estimated as

[0082]

[0083] in, It is the path loss coefficient, which depends on and carrier frequency f c .in addition,

[0084] SALINR is an alternative signal measurement metric that can be used as a virtual objective function in distributed multi-cell resource allocation. Specifically, the arithmetic mean of the interference power of base station l to users served by other base stations is introduced as:

[0085]

[0086] Therefore, the virtual PFR maximization problem can be updated to:

[0087]

[0088] in, and It is a SALINR expression.

[0089] While a global CSI is not required compared to the problem model, CSI for users scheduled from one base station to other base stations is still necessary. Furthermore, coupled inter-base station scheduling remains a challenge for interference coordination. To address these challenges, this invention considers resource allocation in the following stages.

[0090] This invention separates user scheduling and beamforming design, employing an enhanced low-complexity proportional fairness zero-forcing greedy (PFZFG) method to determine the scheduling set before beamforming vector optimization. The core idea of ​​PFZFG is to select users with weak channel correlation, thereby obtaining a spatially well-separated scheduling set. This separation helps reduce intra-cell interference in downlink transmission, and combined with SALINR-based precoding, it can mitigate inter-cell interference, thus improving system throughput. Furthermore, it reduces the overlap of beam main lobes towards their respective intended directions, thus further improving the accuracy of active sensing. In addition, the PFZFG method also considers the fairness characteristics of users over continuous periods. If some users have a long duration (e.g., T...),... s If a node has not been scheduled for -1 epochs, it should be scheduled immediately in the current epoch to prevent large motion parameter tracking errors. Specific scheduling algorithms include:

[0091] (1) Select a set of continuous T s -1 user who was not scheduled at any given time Solving using equal-power zero-forcing beamforming The sum and rate of users

[0092] (2) Calculation in It is the historical average rate of user i at the previous n-1 time steps, which is fed back to the serving base station via uplink. and

[0093] (3) If R≥R′, then Repeat steps (2)-(3) until R < R′; output The user set scheduled by base station l at time n.

[0094] Once the set is determined Based on the base station's own coordinates and the user's kinematic parameters, base station l can determine the user's location. coordinates and speed Base station l obtains the user's information by transmitting the above information back. Angle where m≠l Path loss coefficient and Doppler frequency This allows us to estimate the distance from base station l to user t. m Cross-cell channels:

[0095] Based on the above formula, the leakage value can be approximated. Since only the status information of the scheduled users needs to be exchanged, the exchange overhead is low.

[0096] Thus, the beamforming vector optimization problem can be solved in a distributed manner across all base stations. Clearly, the estimated noise variance of the echo signal plays a crucial role in channel estimation and directly affects the effectiveness of the precoding solution. Especially in multi-cell resource allocation, it is difficult to align the beamforming direction with the target user for accurate sensing.

[0097] Therefore, for base stations, this invention considers the virtual PFR maximization problem constrained by power constraints and sensing error constraints as follows:

[0098]

[0099]

[0100]

[0101] To solve the aforementioned non-convex problem, this invention employs a fractional programming approach combined with semidefinite relaxation. Specifically, this involves introducing auxiliary variables... and The objective function is equivalent to

[0102]

[0103] when Fixed, optimal and To handle non-convex perceptual constraints, the semidefinite programming problem can be further represented as:

[0104]

[0105]

[0106]

[0107]

[0108]

[0109] in, and

[0110]

[0111]

[0112] and If we disregard the rank-1 constraint, the semidefinite programming problem is convex and can be easily solved using CVX. Then, we can approximate it using eigenvectors to obtain...

[0113] Simulation results validate the effectiveness of the proposed ISAC-SALINR framework. A three-base station scenario is considered, with coordinates (-100,0), (0,0), and (100,0). Each base station serves 22 users with a speed of 15 m / s. Both the base station and users are represented by f... c =30GHz and bandwidth B=30kHz operation. Additionally, ΔT=20ms, T d =4ms and T s =3. The transmit power is 38dBm, the noise power spectral density is -174dBm / Hz, and the Rice factor ρ = 10dB. Furthermore, a θ =0.1, a τ =6.7×10 -7 a μ =0.5, G=10, c=0.5 and path loss The results of this invention are based on the average results of 10 simulations at different user locations, and the PFR of different schemes running 10 algorithm iterations is compared. This invention also considers lumped weighted minimum mean square error (WMMSE) as a performance comparison scheme.

[0114] Figure 4The PFR obtained over time is plotted. For the case of idea CSI, it is assumed that neither CWMMSE nor SALINR algorithms require awareness, and scheduling is embedded in the optimization of the BF design. The SALINR algorithm of this invention achieves a PFR comparable to CWMMSE. In contrast, over time, the performance of CWMMSE with initial CSI experiences a significant PFR decline due to non-negligible CSI errors. Furthermore, if all users are scheduled, ISAC-SALINR (no scheduling) will use more resources for awareness, leading to a degraded communication performance. In contrast, the PFR obtained by the proposed ISAC-SALINR algorithm is close to that of CWMMSE, validating the effectiveness of our framework. In summary, the method proposed in this invention demonstrates good performance in terms of system performance improvement and resistance to channel fading.

[0115] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for dynamic resource allocation in a distributed multi-cell system based on perception assistance, characterized in that, include: Step 1: Estimate user kinematic parameters and CSI based on dual-function radar signal echo; Step 2: Determine the set of users to be scheduled at this station based on the estimated CSI; Step 3: Base stations interactively schedule user motion parameters to estimate the CSI from local to neighboring stations; Step 4: Construct a distributed beamforming optimization problem model; The base station uses the estimated channel state vector to construct a model for a distributed beamforming optimization problem with virtual proportional fairness and rate maximization under power constraints and sensing error constraints. The virtual proportional fairness and rate maximization problem model is expressed as: in, Indicates the community The precoded vector of the k-th user at time node n; Indicates the community The user scheduling set at time node n; Indicates base station Maximum transmit power; User The estimation error of i is required to be less than a tolerable constant c; additionally, User The virtual rate function at time node n User The average rate over the first n-1 time steps is used as the weighting value; Step 5: Optimize beamforming by combining fractional programming with semidefinite relaxation and solve for the beamforming vector; The optimization of beamforming by fractional programming combined with semidefinite relaxation and the solution of the beamforming vector specifically include: introducing auxiliary variables. and The objective function is equivalent to, ； when fixed, ,but ; The semidefinite problem is represented as: Where G is the matched filter gain. and , Indicates the construction of base stations With service users Direct transmission channel between; in addition, It is the antenna gain of the base station. This indicates the number of transmit antennas at the base station. Indicates the number of receiving antennas. It is the constant of perception error; ； If the rank-1 constraint is not considered, the optimal beamforming vector can be obtained by eigenvalue decomposition.

2. The method for dynamic resource allocation in a distributed multi-cell system based on perception assistance as described in claim 1, characterized in that, Step 1 specifically includes: Each base station uses a sensing-integrated beam to transmit dual-function radar signals to the user terminal, while simultaneously estimating the user's kinematic parameters based on the reflected signals from the user terminal and constructing the corresponding channel state vector for the next time step.

3. The distributed multi-cell dynamic resource allocation method based on perception assistance according to claim 2, characterized in that, Estimating the user's kinematic parameters based on the reflected signal from the user terminal and constructing the corresponding channel state vector for the next time step specifically includes: At time n-1, the base station First, a receive filter is used to distinguish the reflected signals of different users. Then, the scheduling user is estimated based on the matched filtering method. latency and Doppler shift Finally, estimate the angle. Based on the kinematic evolution model, the predicted path loss coefficient at time n is obtained. Doppler frequency shift and angle To build base stations With service users Direct channel between: ,in, This indicates the number of transmit antennas at the base station. Indicates base station The user set of the service Indicates the angle of the base station transmitting antenna array The launch steering vector in the direction.

4. The method for dynamic resource allocation in a distributed multi-cell system based on perception assistance according to claim 1, characterized in that, Step 2 specifically includes: constructing the corresponding channel state vector for the next time step based on the base station, and determining the set of users to be scheduled by the base station in the next time step.

5. The method for dynamic resource allocation in a distributed multi-cell system based on perception assistance according to claim 4, characterized in that, A greedy proportional fairness zero-forcing scheme is used to select the user set for scheduling at time n, specifically including: Step 21: Select a group in succession Users who were not scheduled at any given time , Solving for zero-forcing beamforming with equal power The speed of middle users and ; Step 22, Calculation ,in It is the historical average rate of user i at the previous n-1 time steps, which is fed back to the serving base station via uplink. and ; Step 23, if ,but And repeat steps 21 and 22 until... Output As a base station at time n The user set to be scheduled.

6. The method for dynamic resource allocation in a distributed multi-cell system based on perception assistance according to claim 1, characterized in that, Step 3 specifically includes: base stations exchanging kinematic parameters of users to be scheduled by the base station through the backhaul network to help the base station estimate its own channel state vectors and those of other neighboring base stations serving users.

7. The method for dynamic resource allocation in a distributed multi-cell system based on perception assistance according to claim 6, characterized in that, The interaction between base stations via the backhaul network specifically includes: base station Determined based on its own two-dimensional coordinates Two-dimensional coordinates of the user and movement speed Considering the inherent latency of the backhaul network Base station Before the next information transmission moment If the base station interacts with other base stations, then the base station Estimate the dispatch of users from itself to other base stations m The direct-fire channel.

Images