Joint precoding design method in sensing integration scene

Abstract

The invention discloses a joint precoding design method in a communication and sensing integrated scene. Aiming at a non-cellular cooperative communication and sensing integrated system, the invention provides a joint precoding design scheme based on unsupervised deep learning, and the scheme comprises the steps: building a signal and channel model in a communication and sensing integrated scene; constructing a deeply expanded joint communication sensing precoding design network; the collected training data is utilized to train the deep-expanded joint communication sensing precoding design network; in the deployment stage, the joint communication sensing pre-coding design network is used for outputting optimization algorithm process parameters, and then communication pre-coding vectors and sensing pre-coding vectors under different time slots of the communication and sensing integrated scene are obtained by using the constructed optimization algorithm process. According to the method, deep learning driving is fully utilized, various interferences which are mutually coupled and are difficult to process in the cellular-free cooperative sensing integrated system are inhibited, the problem that the algorithm is complicated in the prior art is solved, and the calculation overhead is greatly reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of cooperative synaesthesia integration of non-cellular large-scale MIMO systems, and in particular relates to a joint precoding design method in a synaesthesia integration scenario. Background Art

[0002] The ISAC technology not only integrates communication and perception capabilities, but also significantly improves spectrum efficiency and reduces economic investment by optimizing the sharing strategy of spectrum and hardware resources. In the process of promoting 6G technology to move towards the intelligent connection of all things, the multi-input multi-output (MIMO) collaborative ISAC system based on a cellular-free architecture plays a key role. In this system, many access points are closely integrated with the central processor, which is uniformly scheduled by the central processor to perform consistent data transmission and reception operations to serve a wide range of user devices. In addition, ISAC technology also shows potential in environmental monitoring and intelligent decision support. The collaboration between access points is not limited to communication, but also includes real-time perception of the surrounding environment, thereby enriching the functions and application scope of the system.

[0003] In such a complex ISAC system, the design of communication and perception precoding is particularly critical. Accurate precoding technology not only ensures the stable transmission and reception of communication signals, but also improves the accuracy and range of perception. The efficiency of precoding design directly affects the system's adaptability in complex environments and its performance in multi-user and multi-task scenarios. Therefore, optimizing the algorithms and strategies for communication and perception beams has become the core of improving the overall performance of the system, achieving efficient resource utilization, and ensuring service quality. Through carefully designed beams, the ISAC system can better coordinate communication and perception functions, providing more stable and reliable services for applications such as the Internet of Vehicles and drones. Summary of the invention

[0004] The purpose of the present invention is to provide a joint precoding design method in a synaesthesia integration scenario, which makes full use of deep learning drive, suppresses various mutually coupled and difficult-to-handle interferences in a non-cellular collaborative synaesthesia integration system, overcomes the problem of complex algorithms in the prior art, and can quickly obtain the optimal joint precoding design in different scenarios.

[0005] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0006] A joint precoding design method in a synaesthesia integration scenario specifically includes the following steps:

[0007] Step 1: Establish a channel model and data transmission model for a non-cellular collaborative interawareness integrated system;

[0008] Step 2: According to the channel model and data transmission model of the non-cellular cooperative interawareness integrated system, a deeply unfolded joint communication perception precoding design network is designed;

[0009] Step 3: Using training data randomly generated according to the channel distribution law, train the deeply expanded joint communication-aware precoding design network;

[0010] Step 4: Use the trained joint communication perception precoding to design the network output optimization algorithm process parameters, and then use the constructed optimization algorithm process to obtain the communication precoding vector and perception precoding vector in different time slots of the synaesthesia integration scenario.

[0011] Furthermore, step 1 specifically includes:

[0012] Step 101: In a non-cellular system, a regional scenario model managed by a central processing unit (CPU) is established, wherein M access nodes (APs) and K users are distributed; and M of the M access nodes (APs) are ul work in uplink mode, and the remaining M dl =MM ul works in downlink mode; K users include K ul uplink demand users and K dl Downlink demand users; each access node AP is equipped with N antennas and connected to the central processing unit CPU through the fronthaul link; at the same time, there is a target with uncertain position in the regional scene model. The cellular system needs to complete the uplink communication with the uplink demand users, the downlink communication with the downlink demand users, and the perception of the target's moving direction and speed at the same time;

[0013] Using the fast fading channel model, the channel conditions set in each coherent time block remain unchanged, and the channel in the system scenario is modeled; let k be any integer in the set [1, K], m be any integer in the set [1, M], and the channel vector from the kth user to the mth access node AP is recorded as a vector with N rows and 1 column Let j be the set [1, K ul ], l is any integer in the set [1, K dl ], the channel from the jth uplink user to the lth downlink user is recorded as a complex number Let u be the set [1, M ul ], i is any integer in the set [1, M dl ], the direct channel without target reflection between the u-th uplink AP and the i-th downlink AP is recorded as an N-row and N-column complex matrix The presence of a target will cause a channel between the uplink AP and the downlink AP that is reflected by the target. The target reflection channel between the u-th uplink AP and the i-th downlink AP is recorded as an N-row N-column complex matrix The CPU cannot know The exact value of the target reflection channel can only be obtained through prior knowledge

[0014] Step 102: The cellular system performs both uplink and downlink communications and perception of the target's moving direction and speed. At this time, each downlink access node AP sends a synaesthesia integrated signal including a downlink communication signal and a perception signal. The signal sent by the i-th downlink access node is in, represents the communication precoding vector of the i-th downlink access node to the l-th downlink user, represents the communication signal to the lth downlink user; represents the perceptual precoding vector of the i-th downlink access node, represents the perception signal of the i-th downlink access node;

[0015] The received signal of the lth downlink user is expressed as:

[0016]

[0017] in, represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal; Indicates the cross-link interference caused by uplink users; represents the noise interference of the lth downlink user;

[0018] According to the desired signal and interference signal received by the lth downlink user, a high-precision downlink communication signal can be demodulated; the communication spectrum efficiency of the lth downlink user is expressed as

[0019]

[0020] Step 103: The signal of the jth uplink user received by the CPU is expressed as:

[0021]

[0022] Among them, H represents the conjugate transpose, v uj represents the maximum ratio combining receiver vector of the jth uplink user’s signal received by the uth uplink AP, Indicates the signal received by the u-th uplink AP; represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal; represents the cross-link interference caused by the downlink access node; represents noise interference; the communication spectrum efficiency of the jth uplink user is expressed as

[0023]

[0024] Step 104: Based on the decoding of the uplink and downlink communication signals in steps 103 and 104, the signal received by the u-th uplink AP is subjected to communication interference elimination and is expressed as in, Indicates the sensing signal that the u-th uplink AP expects to receive; and Respectively represent the residual uplink communication signal interference and the residual downlink communication signal interference after interference elimination; represents noise interference; through the receiver design that maximizes the perceived signal-to-noise ratio, the perceived signal-to-interference-noise ratio of the u-th uplink AP is expressed as definition As an evaluation index of perceptual performance, M ul Indicates the number of access nodes AP working in uplink mode.

[0025] Furthermore, step 2 specifically includes:

[0026] Step 201: designing an input layer of the unfolded joint communication-aware precoding design network;

[0027] The input layer input is the estimated access point and user channel matrix, the channel between the downlink access node and the uplink access node; first establish an M-row M-column (N+N 2 ) page complex matrix Z ln The first N pages are used to store the estimated access point and user channel matrix, and the last N pages are used to store the estimated access point and user channel matrix. 2 The page is used to store the estimated channel matrix between the downlink access node and the uplink access node; then Z ln Each element in is converted to its absolute value, resulting in an M-row and M-column (N+N 2 ) page real matrix Z ln,re , which is a matrix form that is easy for the subsequent network to handle, that is, the output form of the input layer;

[0028] Step 202: designing a precoding layer of the unfolded joint communication-aware precoding design network;

[0029] The precoding layer is composed of L residual modules in series, each of which includes two up-convolution units, one down-convolution unit, and an addition operation; the two up-convolution units use the convolution layer to implement the precoding matrix mapping; the convolution kernel size in the down-convolution unit is 1x1, and its sliding step, zero padding, and number of convolution kernels are set to ensure that the output features of the down-convolution unit and the up-convolution unit have the same dimension; the z output of the input layer is converted to ln,re Input each residual module in turn, and finally the Lth residual module will output an M d1 Row(K dl +1) A real number matrix B with 2N columns Z , and B Z Each element of is between -1 and 1;

[0030] Step 203: designing a constraint layer of the unfolded joint communication-aware precoding design network;

[0031] The output of the precoding layer is B Z The deformation module of the constraint layer is first input to obtain an M dl Row(K dl +1) N columns of imaginary matrix C Z , where B z The first N pages are C Z The real part of the next N pages is C Z The imaginary part of Z The projection module of the constraint layer is input so that C Z satisfy

[0032]

[0033] Among them, c m，i Represents C Z A vector consisting of the elements of all pages in the mth row and ith column of Indicates c m，i The conjugate transpose of AP Indicates the maximum transmission power of each AP; C output by the projection module Z That is, joint precoding that meets power constraints, where C Z All rows, first K dl The three-dimensional matrix composed of the elements of all pages is the communication precoding output by the joint communication-aware precoding design network, C Z The two-dimensional matrix composed of all rows, the last column, and the elements of all pages is the perceptual precoding of the joint communication perceptual precoding design network output;

[0034] Next, C ZThe perceptual constraint module is input into the constraint layer. The perceptual constraint module consists of two convolutional layers and one pooling layer, and outputs a positive real number d Z , which represents the safety distance parameter of the sensing constraint; d Z Will and C Z Together they serve as the output of the constraint layer and are input to the next layer of the network;

[0035] Step 204: designing a loss function calculation layer of the expanded joint communication-aware precoding design network;

[0036] D Z and C Z After being input into the loss function calculation layer together, the loss function L of the joint communication-aware precoding design network is Θ Defined as

[0037]

[0038] in is the penalty factor for communication performance, and They represent the output C when the joint communication perception precoding design network is Θ. Z As the joint precoding, the jth uplink user communication rate and the lth downlink user communication rate are calculated; R sen，Th is the perceived performance threshold that we want to achieve; R sen (Θ) indicates that when the joint communication-aware precoding design network is Θ, the output C Z As the joint precoding, the perceptual performance indicator is calculated.

[0039] Furthermore, step 3 specifically includes the following steps:

[0040] Unsupervised training is used to generate 10,000 sets of Monte Carlo scenarios for the constructed joint communication perception precoding design network, in which the user's location, the location of the access node, the location of the perception target, and the channel between the nodes are randomly generated; the estimated access point and user channel matrix, and the channel between the downlink access node and the uplink access node obtained in each scenario will be input into the constructed joint communication perception precoding design network; each random sampling B mini The input parameter of the group scene data is Θ, and the output is B mini The joint precoding matrix and loss function corresponding to the group scenario; using the Adam optimizer in the deep learning framework PyTorch to train the network will minimize the network parameters Θ mini The average network loss function of the group scene is trained in the direction; Θ is N Iter Iterate training until it converges to the optimal network parameter value Θ * .

[0041] Furthermore, step 4 specifically includes the following steps:

[0042] The estimated access point and user channel matrix and the channel between the downlink access node and the uplink access node obtained in the current coherent time block are input into the joint communication-aware precoding design network with the optimal network parameter value Θ*. The joint communication-aware precoding design network will output the M corresponding to this coherent time. dl Row(K dl +1) The joint precoding matrix C of N columns Z ; where C Z The vector composed of the elements of all pages in the mth row and kth column of represents the communication precoding vector of the mth downlink AP to the kth downlink user in the scenario, C Z The vector composed of all pages in the mth row and the last column of represents the first m The perceptual precoding vector of the downlink AP.

[0043] The joint precoding design method in a synaesthesia integration scenario of the present invention has the following advantages: making full use of deep learning drive, effectively suppressing the mutual coupling interference in the non-cellular collaborative synaesthesia integration system; overcoming the problem of complex algorithms in the prior art, simplifying the precoding design process; greatly reducing the computational overhead and improving the execution efficiency of the algorithm; establishing an accurate signal and channel model, providing a solid foundation for precoding; through unsupervised learning, enabling the algorithm to learn autonomously and adapt to complex communication and perception environments; in the deployment stage, achieving real-time optimization, adapting to communication and perception requirements in different time slots. The present invention further improves system performance and is of great significance to the actual deployment of 6G synaesthesia integration. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is an overall flow chart of a joint precoding design method in a synaesthesia integration scenario of the present invention;

[0045] Figure 2 A comparison diagram of communication performance when different joint precoding methods are adopted in the present invention;

[0046] Figure 3 This is a comparison diagram of the perceived performance when the present invention adopts different joint precoding methods. DETAILED DESCRIPTION

[0047] In order to better understand the purpose, structure and function of the present invention, the joint precoding design method in a synaesthesia integration scenario of the present invention is further described in detail below with reference to the accompanying drawings.

[0048] Assume a non-cellular distributed massive MIMO interawareness integration scenario, where there are M = 10 half-duplex APs, of which the number of uplink APs is M ul =5, the number of downlink APs is M dl =5, each AP is equipped with N = 4 antennas, and there are K = 8 users in the scenario, of which the number of users with uplink requirements is K ul =4, the number of users with downlink demand is K dl =4. All APs and users are randomly distributed in an area with a radius of R = 100m. The maximum uplink transmission power of each user is p UE =100mW, the maximum downlink transmit power of each AP is ρ AP =2W. The path loss exponent of the channel between AP and user is set to α=3.7. Each coherent time block lasts for 200 symbols. The number of updated samples during unsupervised training is B mini =128, a total of N Iter = 2000 training iterations.

[0049] Based on the above-mentioned non-cellular distributed massive MIMO synaesthesia integration scenario, this example provides a joint precoding design method in a synaesthesia integration scenario, which specifically includes the following steps:

[0050] Step 1: Establish a channel model and data transmission model for a non-cellular collaborative interawareness integrated system;

[0051] The step 1 specifically includes:

[0052] Step 101: In a non-cellular system, a regional scenario model managed by a central processing unit (CPU) is established, wherein M access nodes (APs) and K users are distributed; and M of the M access nodes (APs) are ul work in uplink mode, and the remaining M dl =MM ul works in downlink mode; K users include K ul uplink demand users and K dl Downlink demand users; each access node AP is equipped with N antennas and connected to the central processing unit CPU through the fronthaul link; at the same time, there is a target with uncertain position in the scene. The system needs to complete the uplink communication with the uplink demand users, the downlink communication with the downlink demand users, and the perception of the target's moving direction and speed at the same time;

[0053] Using the fast fading channel model, assuming that the channel conditions in each coherent time block remain unchanged, we start to model the channel in the system scenario; let k be any integer in the set [1, K], m be any integer in the set [1, M], and the channel vector from the kth user to the mth access node AP is recorded as a vector with N rows and 1 column In the same way, let j be the set [1, K ul ], l is any integer in the set [1, K dl ], the channel from the jth uplink user to the lth downlink user is recorded as a complex number Let u be the set [1, M ul ], i is any integer in the set [1, M dl ], the direct channel without target reflection between the u-th uplink AP and the i-th downlink AP is recorded as an N-row and N-column complex matrix The presence of a target will cause a channel between the uplink AP and the downlink AP that is reflected by the target. The target reflection channel between the u-th uplink AP and the i-th downlink AP is recorded as an N-row N-column complex matrix The CPU cannot know The exact value of the target reflection channel can only be obtained through prior knowledge

[0054] Step 102: The system simultaneously performs uplink and downlink communications and senses the target's moving direction and speed. At this time, each downlink access node AP sends a synaesthesia integrated signal containing a downlink communication signal and a perception signal. The signal sent by the i-th downlink access node is in, represents the communication precoding vector of the i-th downlink access node to the l-th downlink user, represents the communication signal to the lth downlink user; represents the perceptual precoding vector of the i-th downlink access node, represents the perception signal of the i-th downlink access node;

[0055] The received signal of the lth downlink user is expressed as:

[0056]

[0057] in, represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal;

[0058] Indicates the cross-link interference caused by uplink users; represents the noise interference of the lth downlink user;

[0059] According to the desired signal and interference signal received by the lth downlink user, a high-precision downlink communication signal can be demodulated; the communication spectrum efficiency of the lth downlink user is expressed as

[0060]

[0061] Step 103: The signal of the jth uplink user received by the CPU can be expressed as:

[0062]

[0063] Among them, v uj represents the maximum ratio combining receiver vector of the jth uplink user’s signal received by the uth uplink AP, represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal; represents the cross-link interference caused by the downlink access node; represents noise interference; the communication spectrum efficiency of the jth uplink user is expressed as

[0064]

[0065] Step 104: Based on the decoding of the uplink and downlink communication signals in steps 103 and 104, the signal received by the u-th uplink AP After communication interference elimination, it can be expressed as in, and They represent the residual uplink communication signal interference and the residual downlink communication signal interference after interference elimination respectively. By designing the receiver to maximize the perceived signal-to-noise ratio, the perceived signal-to-interference-to-noise ratio of the u-th uplink AP can be expressed as: definition As an evaluation metric for perceived performance;

[0066] Step 2: Design a deeply expanded joint communication-aware precoding design network;

[0067] The step 2 specifically includes:

[0068] Step 201: designing an input layer of the unfolded joint communication-aware precoding design network;

[0069] The input layer input is the estimated access point and user channel matrix, the channel between the downlink access node and the uplink access node; first establish an M-row M-column (N+N 2 ) page complex matrix Z ln The first N pages are used to store the estimated access point and user channel matrix, and the last N pages are used to store the estimated access point and user channel matrix. 2 The page is used to store the estimated channel matrix between the downlink access node and the uplink access node; then z ln Each element in is converted to its absolute value, resulting in an M-row and M-column (N+N 2 ) page real matrix z ln,re , which is a matrix form that is easy for the subsequent network to handle, that is, the output form of the input layer;

[0070] Step 202: designing a precoding layer of the unfolded joint communication-aware precoding design network;

[0071] The precoding layer consists of L residual modules in series, each of which consists of two up-convolution units, one down-convolution unit, and an addition operation. The two up-convolution units use a convolution layer to implement the precoding matrix mapping. The convolution kernel size in the down-convolution unit is 1x1, and its other architectures (sliding step, zero padding, number of convolution kernels) are set to ensure that the output features of the down-convolution unit and the up-convolution unit have the same dimension; the z output of the input layer is ln , re is input into each residual module in turn, and finally the Lth residual module will output an M dl Row(K dl +1) A real number matrix B with 2N columns Z , and B Z Each element of is between -1 and 1;

[0072] Step 203: designing a constraint layer of the unfolded joint communication-aware precoding design network;

[0073] The output of the precoding layer is B Z The deformation module of the constraint layer is first input to obtain an M dl Row(K dl +1) N columns of imaginary matrix C Z , where B Z The first N pages are C Z The real part of the next N pages is C Z The imaginary part of Z The projection module of the constraint layer is input so that C Z satisfy

[0074]

[0075] Among them, C m，i Represents CZ A vector consisting of the elements of all pages in the mth row and ith column of Indicates c m，i The conjugate transpose of AP Indicates the maximum transmission power of each AP; C output by the projection module Z That is, joint precoding that meets power constraints, where C Z All rows, first K dl The three-dimensional matrix composed of the elements of all pages is the communication precoding output by the joint communication-aware precoding design network, C Z The two-dimensional matrix composed of all rows, the last column, and the elements of all pages is the perceptual precoding of the joint communication perceptual precoding design network output;

[0076] Next, C Z The perceptual constraint module is input into the constraint layer. The perceptual constraint module consists of two convolutional layers and one pooling layer, and outputs a positive real number d Z , which represents the safety distance parameter of the sensing constraint; d Z Will and C Z Together they serve as the output of the constraint layer and are input to the next layer of the network;

[0077] Step 204: designing a loss function calculation layer of the expanded joint communication-aware precoding design network;

[0078] D Z and C Z After being input into the loss function calculation layer together, the loss function L of the joint communication-aware precoding design network is Θ Defined as

[0079]

[0080] in is the penalty factor for communication performance, and They represent the output C when the joint communication perception precoding design network is Θ. Z As the joint precoding, the jth uplink user communication rate and the lth downlink user communication rate are calculated; R sen,Th is the perceived performance threshold that we want to achieve; R sen (Θ) indicates that when the joint communication-aware precoding design network is Θ, the output C Z As the joint precoding, the calculated perceptual performance index;

[0081] Step 3: Using the collected training data, train the deeply unfolded joint communication-aware precoding design network;

[0082] The step 3 specifically includes:

[0083] Step 301: Using the collected training data, train a joint communication-aware precoding design network through unsupervised deep learning;

[0084] Unsupervised training is used to generate 10,000 sets of Monte Carlo scenarios for the constructed joint communication perception precoding design network, in which the user's location, the location of the access node, the location of the perception target, and the channel between the nodes are randomly generated; the estimated access point and user channel matrix, and the channel between the downlink access node and the uplink access node obtained in each scenario will be input into the constructed joint communication perception precoding design network; each random sampling B mini The input parameter of the group scene data is Θ, and the output is B mini The joint precoding matrix and loss function corresponding to the group scenario; using the Adam optimizer in the deep learning framework PyTorch to train the network will minimize the network parameters Θ mini The average network loss function of the group scene is trained in the direction; Θ is N Iter Iterate training until it converges to the optimal network parameter value Θ * ;

[0085] Step 4: Design the network output optimization algorithm process parameters by using the joint communication perception precoding, and then use the constructed optimization algorithm process to obtain the communication precoding vector and the perception precoding vector in different time slots of the synaesthesia integration scenario;

[0086] The step 4 specifically includes:

[0087] Step 401: The estimated access point and user channel matrix and the channel between the downlink access node and the uplink access node obtained in the current coherent time block are input into the joint communication-aware precoding design network with the optimal network parameter value Θ*. The joint communication-aware precoding design network will output the M corresponding to this coherent time. dl Row(K dl +1) The joint precoding matrix C of N columns Z ; where C Z The vector composed of the elements of all pages in the mth row and kth column of represents the communication precoding vector of the mth downlink AP to the kth downlink user in the scenario, C Z The vector composed of all pages in the mth row and the last column of represents the perceptual precoding vector of the mth downlink AP in the scene.

[0088] The above shows the whole process of using the method provided in this example to perform a joint precoding design method in a synaesthesia integration scenario in a non-cellular distributed large-scale MIMO synaesthesia integration scenario.

[0089] Figure 1It is an overall flow chart of the joint precoding design method in the synaesthesia integration scenario.

[0090] Figure 2 It is a comparison chart of communication performance when different joint precoding methods are used, where J-DL represents the joint precoding design scheme we proposed, MRT-CP represents the zero-forcing ZF communication precoding and conjugate sensing precoding scheme, and J-SDP represents the joint precoding method using semidefinite programming (SDP). It reveals that the precoding scheme we proposed is far superior to other schemes in terms of communication performance.

[0091] Figure 3 It is a comparison chart of communication performance when using different joint precoding methods. It shows that our proposed scheme can achieve excellent communication performance while obtaining perceptual performance similar to J-SDP.

[0092] In summary, the present invention aims at the joint precoding design problem in the cooperative synaesthesia integration scenario of the non-cellular large-scale MIMO system, and proposes a joint precoding design method in the synaesthesia integration scenario. The present invention first establishes a channel model and a data transmission model in the non-cellular cooperative synaesthesia integration system scenario; then constructs a deeply expanded joint communication perception precoding design network; then uses the collected training data to train the deeply expanded joint communication perception precoding design network; in the deployment stage, the joint communication perception precoding design network is used to output the optimization algorithm process parameters, and then the constructed optimization algorithm process is used to obtain the communication precoding vector and the perception precoding vector in different time slots of the synaesthesia integration scenario. The present invention makes full use of deep learning drive, suppresses various interferences that are mutually coupled and difficult to handle in the non-cellular cooperative synaesthesia integration system, overcomes the problem of complex algorithms in the prior art, greatly reduces the computational overhead, and is of great significance to the actual deployment of 6G synaesthesia integration.

[0093] It is to be understood that the present invention is described by some embodiments, and it is known to those skilled in the art that various changes or equivalent substitutions may be made to these features and embodiments without departing from the spirit and scope of the present invention. In addition, under the teachings of the present invention, these features and embodiments may be modified to adapt to specific circumstances and materials without departing from the spirit and scope of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed herein, and all embodiments falling within the scope of the claims of this application are within the scope of protection of the present invention.

Claims

1. A joint precoding design method in a synaesthesia integration scenario, characterized in that: The specific steps include: Step 1: Establish a channel model and data transmission model for a non-cellular collaborative interawareness integrated system; Step 2: According to the channel model and data transmission model of the non-cellular cooperative interawareness integrated system, a deeply unfolded joint communication perception precoding design network is designed; Step 3: Using training data randomly generated according to the channel distribution law, train the deeply expanded joint communication-aware precoding design network; Step 4: Use the trained joint communication perception precoding to design the network output optimization algorithm process parameters, and then use the constructed optimization algorithm process to obtain the communication precoding vector and perception precoding vector in different time slots of the synaesthesia integration scenario.

2. The joint precoding design method in the synaesthesia integration scenario according to claim 1 is characterized in that: The step 1 specifically includes: Step 101: In a non-cellular system, a regional scenario model managed by a central processing unit (CPU) is established, wherein M access nodes (APs) and K users are distributed; and M of the M access nodes (APs) are ul work in uplink mode, and the remaining M dl =MM ul works in downlink mode; K users include K ul uplink demand users and K dl Downlink demand users; each access node AP is equipped with N antennas and connected to the central processing unit CPU through the fronthaul link; at the same time, there is a target with uncertain position in the regional scene model. The cellular system needs to complete the uplink communication with the uplink demand users, the downlink communication with the downlink demand users, and the perception of the target's moving direction and speed at the same time; Using the fast fading channel model, the channel conditions set in each coherent time block remain unchanged, and the channel in the system scenario is modeled; let k be any integer in the set [1, K], m be any integer in the set [1, M], and the channel vector from the kth user to the mth access node AP is recorded as a vector with N rows and 1 column Let j be the set [1,K ul ], l is any integer in the set [1,K dl ], the channel from the jth uplink user to the lth downlink user is recorded as a complex number Let u be the set [1,M ul ], i is any integer in the set [1,M dl ], the direct channel without target reflection between the u-th uplink AP and the i-th downlink AP is recorded as an N-row and N-column complex matrix The presence of a target will cause a channel between the uplink AP and the downlink AP that is reflected by the target. The target reflection channel between the u-th uplink AP and the i-th downlink AP is recorded as an N-row N-column complex matrix The CPU cannot know The exact value of the target reflection channel can only be obtained through prior knowledge Step 102: The cellular system performs both uplink and downlink communications and perception of the target's moving direction and speed. At this time, each downlink access node AP sends a synaesthesia integrated signal including a downlink communication signal and a perception signal. The signal sent by the i-th downlink access node is in, represents the communication precoding vector of the i-th downlink access node to the l-th downlink user, represents the communication signal to the lth downlink user; represents the perceptual precoding vector of the i-th downlink access node, represents the perception signal of the i-th downlink access node; The received signal of the lth downlink user is expressed as: in, represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal; Indicates the cross-link interference caused by uplink users; represents the noise interference of the lth downlink user; According to the desired signal and interference signal received by the lth downlink user, a high-precision downlink communication signal can be demodulated; the communication spectrum efficiency of the lth downlink user is expressed as Step 103: The signal of the jth uplink user received by the CPU is expressed as: Among them, H represents the conjugate transpose, v uj represents the maximum ratio combining receiver vector of the jth uplink user’s signal received by the uth uplink AP, Indicates the signal received by the u-th uplink AP; represents the communication signal that the lth downlink user expects to obtain; Indicates the communication interference of other downlink users; represents the interference caused by the channel estimation error between the user and the AP; Indicates the interference caused by the perceived signal; represents the cross-link interference caused by the downlink access node; represents noise interference; the communication spectrum efficiency of the jth uplink user is expressed as Step 104: Based on the decoding of the uplink and downlink communication signals in steps 103 and 104, the signal received by the u-th uplink AP is subjected to communication interference elimination and is expressed as in, Indicates the sensing signal that the u-th uplink AP expects to receive; and Respectively represent the residual uplink communication signal interference and the residual downlink communication signal interference after interference elimination; represents noise interference; through the receiver design that maximizes the perceived signal-to-noise ratio, the perceived signal-to-interference-noise ratio of the u-th uplink AP is expressed as definition As an evaluation index of perceptual performance, M ul Indicates the number of access nodes AP working in uplink mode.

3. The joint precoding design method in the synaesthesia integration scenario according to claim 2 is characterized in that: The step 2 specifically includes: Step 201: designing an input layer of the unfolded joint communication-aware precoding design network; The input layer input is the estimated access point and user channel matrix, the channel between the downlink access node and the uplink access node; first establish an M-row M-column (N+N 2 ) page of the complex matrix Z ln The first N pages are used to store the estimated access point and user channel matrix, and the last N pages are used to store the estimated access point and user channel matrix. 2 The page is used to store the estimated channel matrix between the downlink access node and the uplink access node; then Z ln Each element in is converted to its absolute value, resulting in an M-row and M-column (N+N 2 ) page real matrix Z ln,re , which is a matrix form that is easy for the subsequent network to handle, that is, the output form of the input layer; Step 202: designing a precoding layer of the unfolded joint communication-aware precoding design network; The precoding layer is composed of L residual modules in series, each of which includes two up-convolution units, one down-convolution unit, and an addition operation; the two up-convolution units use the convolution layer to implement the precoding matrix mapping; the convolution kernel size in the down-convolution unit is 1x1, and its sliding step, zero padding, and number of convolution kernels are set to ensure that the output features of the down-convolution unit and the up-convolution unit have the same dimension; the Z output of the input layer is converted into ln,re Input each residual module in turn, and finally the Lth residual module will output an M dl Row(K dl +1) A real number matrix B with 2N columns Z , and B Z Each element of is between -1 and 1; Step 203: designing a constraint layer of the unfolded joint communication-aware precoding design network; The output of the precoding layer is B Z The deformation module of the constraint layer is first input to obtain an M dl Row(K dl +1) N columns of imaginary matrix C Z , where B Z The first N pages are C Z The real part of the next N pages is C Z The imaginary part of Z The projection module of the constraint layer is input so that C Z satisfy Among them, c m,i Represents C Z A vector consisting of the elements of all pages in the mth row and ith column of Indicates c m,i The conjugate transpose of AP Indicates the maximum transmission power of each AP; C output by the projection module Z That is, joint precoding that meets power constraints, where C Z All rows, first K dl The three-dimensional matrix composed of the elements of all pages is the communication precoding output by the joint communication-aware precoding design network, C Z The two-dimensional matrix composed of all rows, the last column, and the elements of all pages is the perceptual precoding of the joint communication perceptual precoding design network output; Next, C Z The perceptual constraint module is input into the constraint layer. The perceptual constraint module consists of two convolutional layers and one pooling layer, and outputs a positive real number d Z , which represents the safety distance parameter of the sensing constraint; d Z Will and C Z Together they serve as the output of the constraint layer and are input to the next layer of the network; Step 204: designing a loss function calculation layer of the expanded joint communication-aware precoding design network; D Z and C Z After being input into the loss function calculation layer together, the loss function L of the joint communication-aware precoding design network is Θ Defined as in is the penalty factor for communication performance, and They represent the output C when the joint communication perception precoding design network is Θ. Z As the joint precoding, the jth uplink user communication rate and the lth downlink user communication rate are calculated; R sen,Th is the perceived performance threshold that we want to achieve; R sen (Θ) indicates that when the joint communication-aware precoding design network is Θ, the output C Z As the joint precoding, the perceptual performance indicator is calculated.

4. The joint precoding design method in the synaesthesia integration scenario according to claim 3 is characterized in that: The step 3 specifically comprises the following steps: Unsupervised training is used to generate 10,000 sets of Monte Carlo scenarios for the constructed joint communication perception precoding design network, in which the user's location, the location of the access node, the location of the perception target, and the channel between the nodes are randomly generated; the estimated access point and user channel matrix, and the channel between the downlink access node and the uplink access node obtained in each scenario will be input into the constructed joint communication perception precoding design network; each random sampling B mini The input parameter of the group scene data is Θ, and the output is B mini The joint precoding matrix and loss function corresponding to the group scenario; using the Adam optimizer in the deep learning framework PyTorch to train the network will minimize the network parameters Θ mini The average network loss function of the group scene is trained in the direction; Θ is trained for a total of N Iter Iterate training until it converges to the optimal network parameter value Θ * .

5. The joint precoding design method in the synaesthesia integration scenario according to claim 4 is characterized in that: The step 4 specifically comprises the following steps: The estimated access point and user channel matrix obtained in the current coherent time block, and the channel input between the downlink access node and the uplink access node have the optimal network parameter value θ * The joint communication-aware precoding design network outputs the M corresponding to this coherence time. dl Row(k dl +1) The joint precoding matrix C of N columns Z ; where C Z The vector composed of the elements of all pages in the mth row and kth column of represents the communication precoding vector of the mth downlink AP to the kth downlink user in the scenario, C Z The vector composed of all pages in the mth row and the last column of represents the perceptual precoding vector of the mth downlink AP in the scene.

Images