A terahertz spectrum environment map construction method based on hybrid active and passive sensing

Abstract

The present invention relates to communication technology and provides a method for constructing a terahertz spectrum environment map based on hybrid active and passive perception, comprising: constructing a virtual communication scene, acquiring and processing a coarse-grained terahertz spectrum environment real map, and training a generative adversarial network for terahertz passive perception based on the map. In a field communication scene, passive perception is periodically performed based on a generative adversarial network to estimate coarse-grained terahertz signal power information and obstacle information. At the initial moment of the current field communication scene or when the estimated coarse-grained obstacle information is updated, the terahertz base station performs active perception, estimates and then updates the coarse-grained obstacle information. Based on the estimated fine-grained obstacle information, a digital twin of the current field communication scene is constructed, the coarse-grained terahertz signal power information is estimated and then updated, and all the information is combined to construct terahertz spectrum environment maps of different granularities.

Description

Technical Field

[0001] The present invention relates to communication technology, and in particular to a method for constructing a terahertz spectrum environment map. Background Art

[0002] Terahertz refers to electromagnetic waves with frequencies between 0.1THz and 10THz.

[0003] Terahertz communication, with its extremely high data transmission rate and wide available communication bandwidth, has been included as a candidate technology for the sixth generation of wireless communication, with the hope that it will play a significant role in future mobile communications. However, factors such as terahertz communication's sensitivity to blocking, sparse communication channels, and extremely narrow directional beams limit its coverage.

[0004] Therefore, the terahertz signal power spectrum density distribution information and real-time obstacle distribution information represented by the terahertz spectrum environment map can be used to adjust the beam direction, use non-direct paths, or reasonably deploy terahertz base stations, etc., so as to enhance terahertz communication coverage. The terahertz spectrum environment map refers to a map that characterizes the spatiotemporal frequency distribution of the signal power spectrum density in the terahertz frequency band and the distribution of obstacles. Therefore, the terahertz spectrum environment map can not only display obstacle information in the communication scene, such as the location and specific shape of obstacles, but also reflect the propagation of terahertz signals in the current communication scene based on the terahertz signal power information therein. Therefore, the terahertz spectrum environment map can apply the obstacle information and terahertz signal power information it provides to the terahertz communication system, thereby helping to improve the coverage of terahertz communication.

[0005] Currently, constructing terahertz spectrum environment maps typically uses densely distributed terahertz monitoring nodes to obtain fine-grained terahertz signal power information. This allows for direct acquisition of coarse-grained obstacle information based on shadow fading and signal attenuation, or for fine-grained obstacle information to be obtained by leveraging the high-precision and high-resolution sensing capabilities of terahertz signals. However, in dynamic scenarios where obstacles may change position, repeated deployment of radio monitoring nodes and multiple monitoring of terahertz signal received power are required, resulting in significant computational overhead and communication latency.

[0006] On July 14, 2022, an invention patent with the publication number CN114900234A and the name “A method and device for constructing a terahertz spectrum environment map” was disclosed. It aims at the dynamically changing communication environment based on the terahertz communication perception integrated system to realize the high-precision construction of the terahertz spectrum environment map, thereby avoiding the repeated deployment of terahertz radio monitoring nodes and the large computational overhead. However, this patent not only requires the periodic occupation of the original communication resources to traverse and implement the communication perception integration in each direction, but also the real map of the terahertz spectrum environment required in the initial stage often requires the occupation of a large number of terahertz radio monitoring nodes, which seriously consumes hardware resources. In addition, the patent also ignores the sparse communication node information commonly used in previous spectrum environment map construction technologies. In addition, the terahertz spectrum environment information at previous moments required for this application will also lead to the long-term accumulation of recursive errors.

[0007] Therefore, when constructing terahertz spectrum environment maps, existing technologies all have problems such as serious consumption of communication / hardware resources, failure to effectively utilize terahertz communication nodes, and long-term accumulation of recursive errors. Summary of the Invention

[0008] The technical problem to be solved by this application is that the communication perception integrated system used in the current technology for constructing terahertz spectrum environment maps is an active perception method, which ignores the passive perception method based on terahertz communication nodes and realizes periodic traversal perception, so that communication / hardware resources are seriously consumed and difficult to be applied in practice; the communication perception integrated system used in the existing technology for constructing low-frequency spectrum environment maps is a passive perception method, which ignores the active perception method based on base stations and terahertz channel characteristics, so that it is difficult to obtain fine-grained terahertz signal power information and obstacle information when directly constructing terahertz spectrum environment maps, and is not suitable for dynamic scenarios where obstacles change position. A method for constructing spectrum environment maps in terahertz bands or lower frequency bands using a generative adversarial network with the assistance of a communication perception integrated system is provided.

[0009] The technical means adopted by the present invention to solve the above technical problems is a method for constructing a terahertz spectrum environment map based on hybrid active and passive sensing, which includes the following steps:

[0010] Passive sensing step: The terahertz base station receives the distributed terahertz signals uploaded by the terahertz communication nodes and periodically inputs the received terahertz signals into a well-trained generative adversarial network to estimate the coarse-grained obstacle information and terahertz signal power information;

[0011] Active sensing step: This step is triggered at the initial moment of the current field communication scene or when the coarse-grained obstacle information estimated in the passive sensing step is updated. The terahertz base station determines the obstacle position based on the obtained coarse-grained obstacle information, first sends a terahertz signal to the obstacle, then receives the corresponding terahertz echo signal, and estimates fine-grained obstacle information based on the terahertz echo signal, thereby updating the coarse-grained obstacle information. Afterwards, the terahertz base station constructs a digital twin of the current field communication scene based on the estimated fine-grained obstacle information, then estimates the fine-grained terahertz signal power information, thereby updating the coarse-grained terahertz signal power information.

[0012] Map construction steps: The terahertz base station combines the latest coarse-grained obstacle information and coarse-grained terahertz signal power information to construct a coarse-grained terahertz spectrum environment map, and combines the latest fine-grained obstacle information and fine-grained terahertz signal power information to construct a fine-grained terahertz spectrum environment map.

[0013] Specifically, the terahertz base station constructs a virtual communication scene including obstacles, terahertz communication nodes and terahertz base stations, wherein the obstacles undergo random changes in position, shape and / or number at random moments in the virtual communication scene. The terahertz base station obtains and processes a coarse-grained terahertz spectrum environment real map of the virtual communication scene, and uses the coarse-grained terahertz spectrum environment real map to train a generative adversarial network for terahertz passive perception. The generative adversarial network is used to receive input terahertz signals and output estimated coarse-grained obstacle information and terahertz signal power information.

[0014] This overcomes the current inability to simultaneously obtain coarse-grained terahertz spectrum environment maps and fine-grained terahertz spectrum environment maps, which easily leads to problems such as insufficient or excessive performance when enhancing terahertz communication coverage.

[0015] In the passive sensing step, the terahertz base station uses the power data uploaded by the terahertz communication node received during regular communication with several terahertz communication nodes in a field communication scenario, and there is no additional communication overhead. Therefore, the present invention only needs to periodically use the terahertz signals passively received by the terahertz communication node to perform coarse-grained passive sensing and use the terahertz signals actively emitted by the terahertz base station on demand to perform fine-grained active sensing, so as to simultaneously obtain the latest coarse-grained and fine-grained terahertz spectrum environment maps in the current scenario without the need for dense deployment of a large number of terahertz radio monitoring nodes and periodic traversal to occupy a large amount of communication resources. In this way, it can effectively solve the problems of serious consumption of communication / hardware resources, failure to effectively utilize terahertz communication nodes, and long-term accumulation of recursive errors in traditional terahertz spectrum environment map construction methods.

[0016] The beneficial effects of the present invention are:

[0017] 1. The latest coarse-grained and fine-grained terahertz spectrum environment maps for the current scenario can be obtained simultaneously, without the need for dense deployment of a large number of terahertz radio monitoring nodes and periodic traversal that consumes a large amount of communication resources. The saved communication resources can enable terahertz communication links to meet users' high-speed data transmission needs with lower latency, while the saved hardware resources can also enable mobile operators to deploy terahertz mobile communication networks at a lower cost. The constructed coarse-grained and fine-grained terahertz spectrum environment maps can effectively support the implementation of different terahertz signal coverage technologies by providing obstacle information of different granularities and dense information on the terahertz signal power spectrum.

[0018] 2. Use a real map of the terahertz spectrum environment based on a virtual communication scenario to implement generative adversarial network training for passive perception, so as to facilitate the rapid implementation of passive perception methods in real communication scenarios with low cost and fast time.

[0019] Given the similarities between virtual and real-world communication scenarios, the coarse-grained obstacle information and terahertz signal power information initially acquired using the trained generative adversarial network in the real-world communication scenario is highly accurate, enabling on-demand implementation of active sensing methods in the real-world communication scenario. Furthermore, the fine-grained obstacle information and terahertz signal power information acquired by active sensing and digital twin methods in the real-world communication scenario can further update the accuracy of the already high-precision coarse-grained information. This not only avoids the deviation of coarse-grained information in the construction of the terahertz spectrum environment map caused by the differences between virtual and real-world communication scenarios, but also avoids the additional real-world data collection, network fine-tuning training, and the corresponding cost and time consumption required by existing technologies to address these differences. Therefore, the present invention uses simulated data for passive perception in virtual communication scenarios and actual data based on active perception in real communication scenarios, which not only ensures the effectiveness and efficiency of the generative adversarial network for hybrid active and passive perception in real communication scenarios, but also avoids consuming additional resources to collect terahertz signal power data and fine-tune the generative adversarial network in real communication scenarios, further saving the cost and time in constructing the terahertz spectrum environment map.

[0020] 3. By using a well-trained generative adversarial network and the received signals periodically obtained by a small number of terahertz communication nodes in field communication scenarios, coarse-grained obstacle information and terahertz signal power information are periodically estimated in a passive sensing manner. This not only eliminates the need to deploy a large number of radio monitoring nodes at the first or subsequent moments, but also utilizes terahertz communication users or terahertz IoT devices in field communication scenarios as terahertz communication nodes, which can effectively save hardware overhead and provide multi-granularity information.

[0021] 4. Determine the obstacle location based on the obtained coarse-grained obstacle information, and then perform active perception of fine-grained obstacle information as needed. This not only utilizes the excellent perception ability of terahertz signals to obtain high-precision and high-resolution obstacle information, but also avoids the additional communication overhead and time delay caused by the existing technology of periodic traversal using the terahertz synaesthesia integrated system based on active perception. That is, it is possible to obtain a fine-grained terahertz spectrum environment map in a field communication scenario with less resource consumption and shorter construction time.

[0022] 5. By integrating the periodically obtained coarse-grained obstacle information, the on-demand fine-grained obstacle information, and the terahertz signal power information of different granularities obtained through digital twins, the latest terahertz spectrum environment estimation maps of different granularities in actual communication scenarios are obtained. Compared with the existing solutions of directly obtaining the spectrum environment map based on the prior obstacle information at the current moment, or recursively obtaining the spectrum environment map based on the prior spectrum environment map at the first moment and the obstacle information obtained through periodic traversal active perception, the spectrum environment maps of different granularities constructed by the present invention are more in line with terahertz communication applications with different granularity requirements under actual dynamic scenarios (such as changes in obstacles), and avoid the long-term accumulation of recursive errors.

[0023] Therefore, the present invention can effectively solve the problems of the prior art in constructing terahertz spectrum environment maps, such as serious consumption of communication / hardware resources, failure to effectively utilize terahertz communication nodes, and long-term accumulation of recursive errors. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic flow diagram of Example 1;

[0025] Figure 2 The actual communication scenario provided in Example 1 is at time t0 or t u‘ 3D schematic diagram of obstacle information at each moment;

[0026] Figure 3 The actual communication scenario provided in Example 1 is at time t0 or t u‘ A bird's-eye view of obstacle information at each moment and a schematic diagram of the corresponding beam direction;

[0027] Figure 4 In the actual communication scenario provided in Example 1, u 3D schematic diagram of obstacle information at each moment;

[0028] Figure 5 In the actual communication scenario provided in Example 1, u A bird's-eye view of obstacle information at all times;

[0029] Figure 6In the actual communication scenario provided in Example 1, u Time relative to t u‘ A top view of the obstacle information and the corresponding beam direction diagram at different times. DETAILED DESCRIPTION

[0030] Both the virtual communication scenario and the real communication scenario include: obstacles in the communication scenario, terahertz communication nodes, and terahertz base stations. Obstacles in the scenario may undergo any changes in position, shape, and number at any moment. Coarse and fine-grained terahertz spectrum environment maps obtained based on hybrid active and passive perception are used to describe the terahertz spectrum environment in the scenario. Coarse and fine-grained terahertz spectrum environment maps can contain terahertz signal power information and obstacle information in two-dimensional communication scenarios, as well as terahertz signal power information and obstacle information in three-dimensional communication scenarios; the embodiments can realize the planar construction of the terahertz spectrum environment map in a communication scenario simplified to two dimensions, and can also realize the three-dimensional construction of the terahertz spectrum environment map in an actual three-dimensional communication scenario.

[0031] Among them, passive perception refers to the technology of wireless communication system to realize the integration of communication perception under software and hardware resource sharing or information sharing by passively receiving communication signals. It can also be called communication perception integration technology based on passive perception. Specifically, the implementation of passive perception technology is mainly through analyzing and processing the received communication signals through direct and reflected propagation mechanisms to obtain information about the target object such as position, distance, speed, etc. Therefore, passive perception technology is generally only implemented by users during downlink communication. Similarly, passive perception technology can be implemented in a centralized manner by a single user, or in a distributed manner by multiple users. The present application realizes passive perception by multiple users, that is, multiple terahertz monitoring nodes in a distributed and downlink communication manner.

[0032] Active perception refers to the wireless communication system actively sending communication signals to achieve communication perception integration under software and hardware resource sharing or information sharing. It can also be called communication perception integration technology based on active perception. Specifically, the implementation of active perception technology is mainly to obtain information about the target object such as position, distance, speed, etc. by analyzing and processing the communication signals sent and the corresponding received signal echoes through the reflection and scattering propagation mechanism. Therefore, active perception technology can be implemented by the base station during downlink communication; it can also be implemented by the user during uplink communication. In addition, active perception technology can be implemented in a centralized manner through a single base station or a single user, or in a distributed manner through multiple base stations or multiple users. The present application implements active perception in a centralized and downlink communication manner through a single base station.

[0033] Example 1

[0034] like Figure 1 As shown, the method for constructing a terahertz spectrum environment map based on hybrid active and passive sensing may include the following steps:

[0035] Step S110: construct a virtual communication scenario including obstacles, terahertz communication nodes and terahertz base stations, obtain and process a coarse-grained real map of the terahertz spectrum environment, and train a generative adversarial network for terahertz passive perception based on it.

[0036] Step S110 includes sub-steps S111 to S115 .

[0037] S111: Divide the virtual communication scene into several grids, randomly place a terahertz base station in a certain grid at each moment, randomly place several obstacles in several grids, and place terahertz communication nodes in the remaining grids.

[0038] The virtual communication scene is constructed by software simulation and is divided into N A =N L *N W *N H grids, where each side of the grid has a length equal to S LWH The perception resolution determined by the grid side length is much lower than the perception resolution of the communication perception integration technology based on active perception in the terahertz frequency band. Therefore, the relevant information about terahertz passive perception obtained in the virtual communication scene is coarse-grained information. The corresponding grid set of the three-dimensional area occupied by the virtual communication scene is where x i is the center position of the ith grid in the virtual communication scene, and also refers to the ith grid, and t∈{1,…,t end} is any time in the virtual communication scene. Assume that the number of obstacles at time t is ≥1, and the corresponding grid set of the three-dimensional area occupied by them is Wherein, the superscript d represents an obstacle, is the center position of the ith grid in the obstacle area, and also refers to the ith grid in the obstacle area that is completely covered by (part of) the obstacle, and N DA (t) is the total number of grids occupied by the obstacle area. Assume the base station position at time t is x a (t), which occupies only one grid, and there are no obstacles or terahertz communication nodes in the grid area. The superscript a represents the terahertz base station. Therefore, the corresponding grid set C(t) of the three-dimensional area that can be used for the placement of terahertz communication nodes is the set D(t) and the element x a The absolute complement of the union of (t) in the set A(t), that is, C(t)=A(t)\(D(t)∪x a (t)), and marked as Wherein, the superscript c represents the terahertz communication node, is the center position of the i-th grid in the area occupied by the terahertz communication node, and the center position is used to place the terahertz communication node. It also refers to the i-th grid in the area, and N CA (t) is the total number of grid cells in the area occupied by the terahertz communication nodes. In addition, at time t and time t+1, the three-dimensional area A(t) and the three-dimensional area A(t+1) are different due to different information such as base stations and obstacles.

[0039] Sub-step S112: Calculate the time from the first moment to the tth moment in the virtual communication scene end The terahertz signal power information of each grid at the moment, and the obstacle information is counted, where t end At a preset moment, each grid has only one type of power information, a, d, or c.

[0040] In an embodiment, a grid of terahertz communication nodes is placed The power spectrum density of the terahertz signal at the grid The terahertz signal power information at the location can be obtained by Frequency f and time t are jointly determined. The power spectrum density of the received signal from the terahertz base station is recorded as Grid The power spectrum density at all locations within the grid is considered to be equal to the center of the grid. The power spectral density at , and the frequency f is the signal bandwidth range within the terahertz transmission window [f min ,f max ] unit frequency within .

[0041] In an embodiment, the virtual communication scene from the first moment to the tth moment can be obtained based on the ray tracing method. end The terahertz signal power information at the moment may include: controlling the grid position x in the virtual communication scene aThe terahertz base station (t) transmits a ray simulating a terahertz signal, wherein the ray can realize functions such as reflection and scattering, and controls each terahertz communication node in the area occupied by the terahertz communication node in the virtual communication scene to receive the ray omnidirectionally. Specifically, each ray is set at an equal angle within the base station beam width. For example, the angle interval can be set to 0.25° by default, and the ray is allowed to interact with the physical environment during the forward process to establish a signal propagation path based on the ray tracing algorithm. At the same time, the number of times the beam is reflected, scattered, etc. during the propagation process is specified. After the signal propagation path is established, the receiving power and signal power spectrum density at each terahertz communication node can be calculated in combination with the sending and receiving distance, water vapor density, obstacle material, etc. Therefore, according to the ray tracing method, the virtual communication scene from the 1st moment to the tth moment can be obtained. end The terahertz signal power information of each grid in the time set C(t) is recorded as

[0042] Grid in obstacle area The obstacle information at The grid x where the terahertz base station is located a The terahertz signal power information at (t) is recorded as Ψ(x a (t),t)=P, where P is the signal power. and x a The obstacle information and power information at all positions within (t) are set to be equal to the value at the center position. end The obstacle information of each grid in the time set D(t) can be counted as

[0043] Sub-step S113: Based on the obtained time from the first moment to the tth moment end Each grid information at a moment and The far-field terahertz signal power information at each moment is normalized, and then the obstacle information at each moment is associated with the normalized maximum value. At the same time, the near-field terahertz signal power information at each moment is associated with the normalized minimum value. The far-field terahertz signal power information at each moment is linearly mapped to the preset normalized minimum and maximum value intervals. Based on the above information, a coarse-grained terahertz spectrum environment real map at each moment in the virtual communication scenario is constructed.

[0044] In an embodiment, first, the terahertz signal power information for each grid in the set C(t) at each moment is divided into near-field and far-field regions. Specifically, the Rayleigh distance can be calculated based on the antenna array settings of the terahertz base station (such as aperture size) in the virtual communication scenario and the considered terahertz signal wavelength. Then, the terahertz signal power information within the circular region centered at the terahertz base station with the Rayleigh distance as the radius is the near-field terahertz signal power information. Where C1(t) is the set of grids corresponding to the terahertz communication nodes within the near-field circular region. Therefore, the terahertz signal power information outside this near-field circular region is the far-field terahertz signal power information. Where C2(t) is the set of grids corresponding to the terahertz communication nodes within the far-field circular region.

[0045] Next, the maximum-minimum normalization process is performed on the terahertz signal power information for each grid in the set C2(t) at each moment. Specifically, first, sort the terahertz signal power information Ψ(x,t)| x∈C2(t) to obtain the maximum value Ψ max (t) of the terahertz signal power spectral density and the minimum value Ψ min (t) of the terahertz signal power spectral density at each moment. Thus, the normalized data of the terahertz signal power information Ψ(x,t)| x∈C2(t) can be calculated as Ψ n (x,t)| x∈C2(t) =(Ψ(x,t) - Ψ min (t)) / (Ψ max (t) - Ψ min (t)), and its value range is [0,1].

[0046] Finally, reassign the obstacle information Ψ(x,t)| x∈D(t) = 0 with the maximum value (i.e., the value 1) after normalization, then the obstacle information at each moment is Ψ(x,t)| x∈D(t) = 1. At the same time, reassign the near-field terahertz signal power information Ψ(x,t)| x∈C1(t) with the minimum value (i.e., the value 0) after normalization, then the near-field terahertz signal power information at each moment is Ψ(x,t)| x∈C1(t) = 0. In addition, based on the value range [0,1], preset a sub-small and sub-large value interval [s0, s1], where 0 < s0 < s1 < 1. Then, linearly map the far-field terahertz signal power information Ψ n (x,t)| x∈C2(t) with this sub-small and sub-large value interval, then the far-field terahertz signal power information at each moment is <00********>By integrating the obstacle information, near-field terahertz signal power information, and far-field terahertz signal power information at each moment after the above processing, a real map of the coarse-grained terahertz spectrum environment of one channel at each moment in the virtual communication scene can be constructed. Then the map is expanded in three dimensions to obtain the three-channel coarse-grained terahertz spectrum environment real map Ψ(t)=[Ψ 1d (t),Ψ 1d (t),Ψ 1d (t)]. Therefore, from the first moment to the tth moment end The coarse-grained terahertz spectrum environment real map at the moment can be expressed as

[0048] Sub-step S114: The obtained real map of the coarse-grained terahertz spectrum environment at each moment Perform sparse sampling one by one to obtain a coarse-grained terahertz spectrum environment real sampling map at each moment

[0049] In an embodiment, with a sampling rate α s For t∈{1,…,t end The one-channel coarse-grained terahertz spectrum environment real map Ψ at time Ψ(t) 1d THz signal power information in (t) Sampling is performed to obtain the sampled terahertz signal power information Ψ′ of the sampling grid set C3(t) at each moment 1d (x,t)| x∈C3(t) and its three-dimensional expanded three-channel information Ψ′(x,t)| x∈C3(t) =[Ψ′ 1d (x,t),Ψ′ 1d (x,t),Ψ′ 1d (x, t)]. Thus, a preset vector v s =[v 1s ,v 2s ,v 3s ] to construct a coarse-grained terahertz spectrum environment real sampling map at time t in a virtual communication scenario Therefore, from the first moment to the tth moment end The coarse-grained terahertz spectrum environment real map at the moment can be expressed as

[0050] Sub-step S115: Based on the obtained and Training generative adversarial networks for terahertz passive sensing.

[0051] In an embodiment, the terahertz passive sensing problem dedicated to obtaining coarse-grained obstacle information and terahertz signal power information at each moment in a virtual communication scenario can be expressed as:

[0052]

[0053] in, The coarse-grained terahertz spectrum environment estimation map obtained based on Ψ′(t) contains coarse-grained obstacle estimation information and terahertz signal power estimation information. To solve this problem, the embodiment uses a generative adversarial network for terahertz passive sensing for training. The fully trained generative adversarial network can output a coarse-grained terahertz spectrum environment estimation map that is almost identical to Ψ(t) based on the input Ψ′(t). In an embodiment, the generative adversarial network for terahertz passive sensing may be designed and trained based on the generative adversarial network proposed by Goodfellow and its corresponding variants.

[0054] Specifically, let N e =1 and N b = 1, the coarse-grained terahertz spectrum environment real sampling map at several random moments Input to the generator network G of the generative adversarial network for terahertz passive perception, so that the generator network G outputs a coarse-grained terahertz spectrum environment estimation map corresponding to several moments Right now Among them, θ g is the neural network parameter of the generator network G, N e ∈{1,2,…,N em} and N em are the training cycle index and its maximum value, N b ∈{1,2,…,N bm} and N bm are the index of batch training times and their maximum values ​​relative to a certain training cycle, N B is the number of coarse-grained terahertz spectrum environment real sampling maps used in batch training, and

[0055] Further, Input to the discriminator network D of the generative adversarial network for terahertz passive sensing and compare it with The corresponding coarse-grained terahertz spectrum environment real map at several moments or terahertz spectrum environment estimation map The same input is sent to the discriminator network D, so that the discriminator network D outputs true or false judgment information Right now or Among them, θ d is the neural network parameter of the discriminator network D, true and false judgment information Used to represent the input of the discriminator network D or Whether it comes from the true or false judgment of the generator network G, and in the embodiment, the true or false judgment information It can be represented as a real number with a value range in [0,1].

[0056] Combating losses for terahertz passive sensing and self-weighted reconstruction loss The weighted sum of is used to train the generative adversarial network, that is, the training objective of the generative adversarial network is: This realizes the parameter θ g and θ d Of which, represents the expectation, λ is the value of V re The weight factor of (G), p is the penalty factor for adversarial loss; further, ∈ is uniformly distributed Random number; Denotes the relationship between D(Ψ′(t),Ψ′ p (t);θ d ) Find the value of Ψ′ p The gradient of (t).

[0057] Let N b =N b +1, and repeat the above steps until N b =N bm .

[0058] Let N e =N e +1, and repeat the above steps until N e =N em The generative adversarial network designed for terahertz passive perception is fully trained.

[0059] In an embodiment, the generator network of the generative adversarial network for terahertz passive perception, under the action of initialized or updated neural network parameters (i.e., neuron weights and biases, convolution kernel parameters, etc.), converts the coarse-grained terahertz spectrum environment real sampling map into a coarse-grained terahertz spectrum environment estimation map with the same specifications and similar distribution as the coarse-grained terahertz spectrum environment real map (obeying the distribution obtained by implicitly modeling the distribution of the real map by the generator network).

[0060] The generative adversarial network for terahertz passive sensing adopts a training strategy in which the generator network and the discriminator network compete with each other. Specifically, the discriminator network expects to maximize the probability of its correct identification, while the generator network expects to maximize the probability of the discriminator network's identification error. Therefore, under the feedback of the discriminator network's different identification results on the real map or the estimated map, the discriminator network and / or the generator network can continuously improve its network identification and / or generation capabilities. If the discriminator network and the generator network have sufficient network capacity (i.e., enough neural network parameters or enough training times N), the discriminator network and / or the generator network can continuously improve their network identification and / or generation capabilities. em N bm ), the two can achieve a dynamic balance. At this time, the distribution obtained by the implicit modeling of the generator network converges to the distribution of the real map, while the discriminator network cannot correctly distinguish between real samples and simulated samples.

[0061] The embodiment implements the training of a generative adversarial network for passive perception based on a real map of the terahertz spectrum environment in a virtual communication scenario, so as to facilitate the rapid implementation of the passive perception method in the actual communication scenario at a low cost and in a short time. In view of the similarity between the virtual communication scenario and the actual communication scenario, the coarse-grained obstacle information and terahertz signal power information initially obtained based on the trained generative adversarial network are highly accurate and can meet the on-demand implementation of the active perception method in the actual communication scenario. Therefore, the present application ensures the effectiveness and efficiency of the generative adversarial network for hybrid active and passive perception in the actual communication scenario by using simulation data for passive perception in the virtual communication scenario, further saving the cost and time in constructing the terahertz spectrum environment map.

[0062] Step S120: Periodically perform passive sensing in a field communication scenario based on the well-trained generative adversarial network obtained in step S110 to estimate t u = coarse-grained terahertz signal power information and obstacle information at time t0+uT, where obstacles in the field communication scene may change in position, shape, or number at any time, and there are several terahertz communication nodes such as terahertz users in the field communication scene, t u is the passive sensing moment after u passive sensing cycles in the field communication scenario, t0 and T are the initial moment and passive sensing cycle in the field communication scenario, respectively.

[0063] In the embodiment, the field communication scenario is a scenario in real life, which can be either an indoor scenario or an outdoor scenario.

[0064] In an exemplary embodiment, step S120 includes sub-steps S121 to S126 .

[0065] Sub-step S121: Let u = 0, divide the actual communication scene into several grids, and the actual communication scene tu = the set of terahertz communication nodes existing at time t0+uT Passively receive the downlink communication signal from the terahertz base station, thereby obtaining the terahertz signal power information of the corresponding grid, where t u The total number of terahertz communication nodes at time , and The locations of the terahertz communication nodes are randomly distributed.

[0066] In the embodiment, the field communication scene is also divided into N A =N L *N W *N H grids, where the information about the location, shape, number of obstacles and the location of the base station is unknown, and the length of each side of each grid is also equal to S LWH The perception resolution determined by the grid side length in the field communication scenario is much lower than the perception resolution of the communication perception integration technology based on active perception in the terahertz frequency band. Therefore, the relevant information obtained by the generative adversarial network for passive perception used in the field communication scenario is all coarse-grained information. Without loss of generality, the corresponding grid set of the three-dimensional area occupied by the scene is also set to where x i is the center position of the ith grid in the field communication scene. Set t u The grid set of terahertz communication nodes that exist at all times is in t u The center position of the i-th grid in the area occupied by the terahertz communication node at time t, t u The total number of terahertz communication nodes at time The locations of the terahertz communication nodes are randomly distributed.

[0067] In field communication scenarios A terahertz communication node needs to be u The downlink communication signal from the terahertz base station is passively received at all times, thereby obtaining the terahertz signal power information of the corresponding grid, and without loss of generality, it is marked as

[0068] Sub-step S122: t obtained in sub-step S121 u The far-field terahertz signal power information of each grid at the time is normalized and linearly mapped with the preset maximum and minimum values ​​and the normalized next minimum and next maximum value intervals. Based on the above information, the field communication scenario t u The coarse-grained terahertz spectrum environment sampling map Ψ′(t u ).

[0069] In the embodiment, first, t u Time about the set R(t u ) to divide the terahertz signal power information of each grid into the far and near fields. Specifically, the Rayleigh distance can be calculated based on the antenna array setting of the terahertz base station in the field communication scenario and the wavelength of the terahertz signal, and then we can get Far-field terahertz signal power information corresponding to each terahertz communication node and the grid set R2(t u ).

[0070] Then, according to the preset maximum value of the terahertz signal power information Ψ′ in the field communication scenario max and the minimum value Ψ′ min For the above far-field terahertz signal power information Perform normalization processing, that is Then, the normalized information is normalized according to the preset minimum and maximum value interval [s0, s1] Performing linear mapping, we can get And the three-channel information after three-dimensional expansion Therefore, by presetting the vector v s =[v 1s ,v 2s ,v 3s ] to construct the t in the field communication scene u Coarse-grained terahertz spectrum environment sampling map at each moment

[0071] Sub-step S123: Sub-step S122 obtained Ψ '(t u ) is input to the trained generative adversarial network obtained in step S110, and the generator network G of the generative adversarial network outputs t u Coarse-grained terahertz spectrum environment estimation map at each moment

[0072] Sub-step S124: Determine the value obtained in step S123 Whether each element is in the second smallest and second largest value interval or the second smallest and minimum value interval, if so, the coarse-grained terahertz signal power information of the corresponding grid is estimated by inverse processing That is to achieve t u Coarse-grained passive perception of terahertz signal power information at all times.

[0073] In the embodiment, first extract the The one-channel coarse-grained terahertz spectrum environment estimation map is denoted as like An element in the grid x In the interval of the second smallest and second largest value [s0, s1], that is, it meets the Then after inverse mapping, we can get Then the element value after inverse mapping After inverse normalization, we can get like Comply with The corresponding grid set of the elements is recorded as C2′(t u ), the far-field coarse-grained terahertz signal power information estimated in the actual communication scenario can be obtained

[0074] On the other hand, if An element in the grid x In the second smallest minimum interval [0,s0], that is, it meets the Then after assigning it a zero value, we can get If Comply with The corresponding grid set of the elements is recorded as C1′(t u ), the near-field coarse-grained terahertz signal power information estimated in the field communication scenario can be obtained It should be noted that, since the embodiment only focuses on the far-field coarse-grained terahertz signal power information, the near-field coarse-grained terahertz signal power information is set to zero. u )∪C2′(t u ) is used to estimate the coarse-grained terahertz signal power information

[0075] Sub-step S125: Determine the value obtained in step S123 Whether each element in is in the second largest maximum interval, if so, the coarse-grained obstacle information is estimated by considering the corresponding grid as an obstacle That is to achieve t u Coarse-grained passive perception of obstacle information at all times.

[0076] like An element in the grid x In the interval of the second largest maximum value [s1,1], that is, it meets the Then after assigning a value to it, we can get Will Comply with The corresponding grid set of the elements is recorded as D′(t u), the coarse-grained obstacle information estimated in the actual communication scene can be obtained It should be noted that, in this embodiment, the grid corresponding to the element that meets the above conditions is directly regarded as an obstacle, so the element value is set to a value to distinguish it from the terahertz signal power information, and D′(t u )∪C1′(t u )∪C2′(t u )=A(t). Thus, we can get the grid D′(t u ) is estimated to obtain coarse-grained obstacle information

[0077] Sub-step S126: let u=u+1, and execute sub-steps S121-S125 again until the construction and estimation of the terahertz spectrum environment map are stopped in the actual communication scenario.

[0078] Step S130: at the initial time t0 of the current field communication scene or at the time t0 when the coarse-grained obstacle information estimated in step S120 is updated u The terahertz base station actively senses at all times, estimates fine-grained obstacle information, and then updates the coarse-grained obstacle information.

[0079] In an exemplary embodiment, step S130 includes sub-steps S131 to S139 .

[0080] Sub-step S131: Determine the current time t u Is it the initial time t0 of the field communication scene? If so, execute sub-steps S133-S135; otherwise, execute sub-step S132.

[0081] Sub-step S132: Determine t u The coarse-grained obstacle information estimated at step S120 at the moment With the most recent u’ =t0+u'T, whether the difference of the coarse-grained obstacle information saved at the time exceeds a preset threshold. If so, execute sub-steps S136-S139, otherwise do not execute.

[0082] In an embodiment, when t u =t0+T=t1, t u’ = t0, and the corresponding coarse-grained obstacle information has been obtained and saved through sub-step S135 at time t0 When t u =t0+uT| u≥2 When t1≤t u’ ≤t u , and t u’ At this moment, the corresponding coarse-grained obstacle information has been obtained and saved through sub-step S139 like Figure 2 and Figure 3 As shown, Figure 2 It shows the actual communication scenario at time t0 or t u’ The three-dimensional diagram of obstacle information at the moment Figure 3 It shows the actual communication scenario at time t0 or t u’ The top view of obstacle information at the moment. Figure 4 and Figure 5 As shown, Figure 4 Shows the actual communication scenario u The three-dimensional diagram of obstacle information at the moment Figure 5 Shows the actual communication scenario u Obstacle information top view at the moment. Specifically, Figure 2 and Figure 3 Shows the time t0 or t u’ The two obstacles in the scene at the moment are used as examples; Figure 4 and Figure 5 Shows t u Take the three obstacles in the scene at this moment as an example. u The obstacle information at time t0 or t u’ At any moment, there are three situations: (the upper obstacle) moves, (the right obstacle) remains unchanged, and (the left obstacle) is added. The movement of (the upper obstacle) can be determined by Figure 5 The dotted part reveals.

[0083] and The difference between the two To represent, if the number of elements of the difference set Exceeds the preset threshold T com , that is, if Then execute sub-steps S136-S139, otherwise do not execute. or The elements in are all key-value pairs, that is, they contain information about the grid position x and the corresponding element value 1. Figure 6 As shown, Figure 6 It shows the actual communication scenario provided by Example 1. u Time relative to t u‘ Top view of the difference obstacle information at time t, where the dotted line reveals u Time and t u‘ The movement and sharing of obstacles above at all times.

[0084] Sub-step S133: Based on the coarse-grained obstacle information estimated in step S120 at time t0 The terahertz base station actively sends out several terahertz signal beams in the direction of the obstacle, and the difference between the beam directions of adjacent beams is the half-power beam width.

[0085] In the embodiment, the terahertz communication and perception integrated system is set to realize active perception based on frequency modulated continuous wave signals. Therefore, the terahertz base station actively sends a number of frequency modulated continuous wave signals in the direction of the obstacle, and uses beamforming technology to improve the signal gain and the angular resolution of active perception. Specifically, the signal s about the beam direction θ is sent x (t′0) can be expressed as: Where P is the signal power, N T is the number of transmitting antenna elements of the terahertz base station, a(θ,N T ) is the transmit beamforming vector based on the array response vector, is the carrier frequency f c The initial frequency modulated continuous wave signal at a given chirp rate f' is t'0, and t'0 is the time independent variable relative to the initial time t0. In the embodiment, a uniform circular array is used as the transmitting antenna array equipped with the terahertz base station, and its half-power beamwidth remains almost unchanged. In addition, the half-power beamwidth W θ It can be calculated by |θ1-θ2|, where θ1 and θ2 satisfy |a T (θ,N T )a(θ i＝1,2 ,N T ) / N T | 2 =0.5.

[0086] Set the estimated coarse-grained obstacle information The number of obstacles associated with The direction of each obstacle can be revealed by the maximum and minimum values ​​of the angles of its vertex position in polar coordinate form. Figure 2 and Figure 3 For example, at this time there is Setting coarse-grained obstacle information Middle There are a total of vertices, and the maximum and minimum values ​​of the angles in the polar coordinate form of the vertex positions are θ dmax (t0) and θ dmin (t0), and the grid set of the area occupied by the d-th obstacle is recorded as D′ d (t0). Therefore, the terahertz base station needs to actively send (θ dmax (t0)-θ dmin (t0)) / W θ THz signal beams, and the i-thd The beam direction of the beam is That is, the difference in beam directions between adjacent beams is the half-power beamwidth W θ .like Figure 3 As shown, Figure 3 The figure also shows a schematic diagram of the beam direction corresponding to the obstacle information at time t0 in an actual communication scenario. Figure 3 The 6 and 8 terahertz signal beams and their directions corresponding to the upper obstacle and the right obstacle are shown as examples. d The beam direction, the transmitted FMCW signal is marked as

[0087] Sub-step S134: The terahertz base station receives the echo signal from the direction of the obstacle, estimates and saves all fine-grained obstacle information of the field communication scene at time t0 That is, fine-grained active perception of obstacle information at time t0 is achieved.

[0088] In the embodiment, since the terahertz communication and perception integrated system realizes active perception based on frequency modulated continuous wave signals, it is not necessary to obtain coarse-grained obstacle information. For the d-th obstacle, the terahertz base station needs to realize the i-th d Specifically, under the action of the terahertz sensing channel and the receiving beamforming vector based on the array response vector, the received signal can be expressed as: in Contains the path loss, molecular absorption, and obstacle reflection coefficient of the corresponding terahertz sensing channel, N R The number of receiving antenna elements equipped for the terahertz base station, is the signal about the ith obstacle on the dth d The one-way propagation delay of a point target.

[0089] To achieve the d-th obstacle at the i-th d The embodiment will receive the signal With the transmission signal Multiply and filter between them, and then get This signal Carrier frequency It can be measured based on a carefully designed sampling rate and number of Fourier transform points and recorded as Therefore, the ith obstacle on the dth obstacle d The distance between a point target and the base station can be estimated as And the i d The direction of the point target and the base station can be estimated as Therefore, when i d Gradually increase from 0 to (θ dmax (t0)-θ dmin (t0)) / W θ When the beam direction changes from θ dmin (t0) with beam width W θ Gradually increase to θ dmax (t0), the terahertz base station can realize the fine-grained active perception of the d-th obstacle at time t0. Therefore, when d increases from 1 to When the fine-grained active perception of obstacle information at time t0 is achieved, all fine-grained obstacle information of the actual communication scene can be estimated and saved. in

[0090] Sub-step S135: Estimating the grid size of the coarse-grained terahertz spectrum environment map, determining whether there is fine-grained obstacle information in each grid, and then updating and saving the original coarse-grained obstacle information at time t0.

[0091] In an embodiment, fine-grained obstacle information is first calculated The rectangular coordinates of each element in the Cartesian coordinate system are obtained in Then, based on the grid set A(t0) of the actual communication scene, determine whether each rectangular coordinate falls into a grid of A(t0). i Taking a grid as an example, if one or more rectangular coordinates fall within the grid, the grid reflects the coarse-grained obstacle information. In the embodiment, it is assumed that in the coarse-grained obstacle information updated at time t0, the grid set that meets the above conditions is D′ n (t0). Therefore, we have x i ∈D′ n (t0) and corresponding In this way, the original coarse-grained obstacle information can be Update, that is, obtain and save the updated coarse-grained obstacle information

[0092] Sub-step S136: According to t u The coarse-grained obstacle information difference obtained in sub-step S132 is used at each moment. The terahertz base station actively sends out several terahertz signal beams in the directions where the obstacle information is different, and the difference in beam directions between adjacent beams is the half-power beam width.

[0093] According to the set of information differences representing coarse-grained obstacles The corresponding grid set D′ of the elements can be obtained com ,Right now Set the collection The number of associated different obstacles is Among them The maximum and minimum values ​​of the vertex angles of different obstacles in polar coordinate form are and And the grid set of the occupied area is recorded as D′ d (t u ).by Figure 6 For example, at this time there is Similarly, the communication and perception integrated system used by the terahertz base station is based on frequency modulated continuous wave signals to achieve active perception, and uses a uniform circular array as the transmitting antenna array equipped with the terahertz base station. Therefore, it is necessary to actively send a signal in the direction of the dth difference obstacle. THz signal beams, where the i-th d The beam direction of the beam is That is, the difference in beam directions between adjacent beams is the half-power beamwidth W θ .like Figure 6 As shown, Figure 6 It also shows that the actual communication scenario takes t into account u Time relative to t u‘ Schematic diagram of the beam direction when there is obstacle information at different times. Specifically, Figure 6 The 5 and 6 terahertz signal beams and their directions corresponding to the upper obstacle and the left obstacle are shown as examples. d The beam direction, the transmitted FMCW signal is marked as where t ′ u is relative to the initial time t u The time variable.

[0094] Sub-step S137: The terahertz base station receives the echo signal from the direction where the obstacle information is different, and estimates the direction where the obstacle information is different at t u Fine-grained obstacle information at all times

[0095] In the embodiment, since the terahertz communication and sensing integrated system realizes active sensing based on the frequency modulated continuous wave signal, the collection For the dth difference obstacle, the terahertz base station needs to realize the i-th d The FMCW echo signal of each beam Similarly, the d-th difference obstacle is received at the i-thd The fine-grained active sensing in each beam direction is also performed by the received signal With the transmission signal The distance between the corresponding point target and the base station is estimated by multiplying and filtering the corresponding carrier frequency. and direction Therefore, it can be estimated that the direction where the obstacle information is different is t u Fine-grained obstacle information at all times in

[0096] Sub-step S138: Update the most recently saved fine-grained obstacle information in the direction where the obstacle information differs, thereby obtaining and saving the actual communication scene in t u All fine-grained obstacle information at the moment That is to achieve t u Actively perceive the fine-grained obstacle information at all times.

[0097] In an embodiment, when t u = t0 + T = t1, the corresponding time of the most recent storage of fine-grained obstacle information is t u’ = t0, and at time t0, the corresponding fine-grained obstacle information has been obtained and saved respectively through sub-steps S134 and S135 and coarse-grained obstacle information When t u =t0+uT| u≥2 When the corresponding time of the most recent storage of fine-grained obstacle information satisfies t1≤t u’ ≤t u , and t u’ At this moment, the corresponding fine-grained obstacle information has been obtained and saved through sub-steps S138 and S139. and coarse-grained obstacle information

[0098] Calculate t u The coarse-grained obstacle information obtained at step S120 at the moment and t u’ Coarse-grained obstacle information at all times The union between The grid position set D′(t u )∩D′ n (t u’ )Retain t u’ Time-granular obstacle information Specifically, first calculate The rectangular coordinates of each element in in in θ dmax (t u’ ) and θ dmin (t u’ )for The number of associated obstacles and the maximum and minimum values ​​of the vertex angle of the d-th obstacle. Figure 2 and Figure 3 For example, at this time there is Then according to the grid set D′(t u )∩D′ n (t u’ ), determine whether each rectangular coordinate falls on D′(t u )∩D′ n (t u’ ) in a grid. i Take a grid as an example, if one or more rectangular coordinates fall within the grid, it can be considered that the coarse-grained and fine-grained obstacle information reflected by the grid is the same at t u’ Time to t u Therefore, in the embodiment, the rectangular coordinate set that meets the above conditions is set as D′ uu’ (t u’ ), thus retaining t u’ Time about the rectangular coordinate set D′ uu’ (t u’ )’s fine-grained obstacle information, i.e. and Thus, the representation of the fine-grained obstacle information in polar coordinate form can be obtained, that is, and

[0099] Therefore, the field communication scenario is u All fine-grained obstacle information at the moment Including about t u Coarse-grained obstacle information at all times and t u’ Coarse-grained obstacle information at all times Fine-grained obstacle information of common information and t u Fine-grained obstacle information with different directions at all times Right now in in θ dmax (t u ) and θ dmin (t u )for The number of associated obstacles and the maximum and minimum values ​​of the vertex angle of the d-th obstacle. Figure 4 and Figure 5 For example, there are

[0100] Sub-step S139: Estimating the grid size of the coarse-grained terahertz spectrum environment map, determining whether there is fine-grained obstacle information in each grid in the direction where the obstacle information is different, and then updating and saving the original t u Coarse-grained obstacle information at each moment.

[0101] In an embodiment, fine-grained obstacle information is first calculated The rectangular coordinates of each element in the Cartesian coordinate system are obtained in Then, according to the grid set A(t u ), determine whether each rectangular coordinate falls on A(t u ) in a grid. i Taking a grid as an example, if one or more rectangular coordinates fall within the grid, the grid reflects the coarse-grained obstacle information in the direction where the obstacle information is different. u In the coarse-grained obstacle information updated at any moment, the grid set that meets the above conditions is D′ u1 (t u ). Therefore, we have x i ∈D′ u1 (t u ) and corresponding In this way, the original coarse-grained obstacle information can be Update, that is, obtain and save the updated coarse-grained obstacle information where D′ n (t u )=D′ u1 (t u )∪(D′(t u )∩D′ n (t u’ )), where t u’ is the corresponding moment when the fine-grained and coarse-grained obstacle information is most recently saved.

[0102] Step S140: Based on the fine-grained obstacle information estimated in step S130, a digital twin of the current field communication scene is constructed, the fine-grained terahertz signal power information is estimated, and then the coarse-grained terahertz signal power information is updated, and all the information is combined to construct a terahertz spectrum environment map of different granularities.

[0103] In an exemplary embodiment, step S140 includes sub-steps S141 to S144 .

[0104] Sub-step S141: Based on the perception resolution of the terahertz communication perception integrated system based on active perception, the actual communication scene is divided into several small grids, a digital twin of the obstacle is created, and a virtual communication scene for the digital twin is obtained.

[0105] In an embodiment, the sensing resolution of the terahertz communication sensing integrated system based on active sensing includes distance resolution and angle resolution, wherein the distance resolution ΔR=c / 2(f max -f min ), and the angular resolution Δθ=0.886λ / D A Equivalent to the half-power beamwidth, where c is the speed of light, λ and D A are the wavelength of the terahertz signal used for active sensing and the aperture of the base station antenna, respectively. Therefore, the length, width, and height of the small grid set in the embodiment are ΔRsinΔθcosΔθ< LWH , ΔRsin 2 Δθ< LWH , ΔRcosΔθ< LWH , that is, the side length is the transformation of the distance resolution and angle resolution in polar coordinates to the rectangular coordinate system. Therefore, the field communication scene can be divided into several small grids according to the side length, and the corresponding small grid set is where e i is the center position of the ith small grid in the field communication scene, and also refers to the ith small grid, and N E is the total number of small grids in the actual communication scene.

[0106] According to t u The fine-grained obstacle information estimated at step S130 at the moment in Assign values ​​to the small grids at the corresponding positions of the actual communication scene. Specifically, determine Whether the rectangular coordinates of each element in falls within the set E(t u ), that is, if a rectangular coordinate falls on the e i grid, then let Thus, the fine-grained obstacle information corresponding to the small grid in the actual communication scene can be obtained. Where E′(t u ) is t u The grid set that meets the above conditions at all times. i Also refers to i small grids, so the embodiment here directly converts the set E′(t​​​u ) is considered as an obstacle, that is, the corresponding grid is completely covered by (a part of) the obstacle. The elements in are all key-value pairs, that is, they contain information about the grid position e and the corresponding element value 1. Therefore, the digital twin of the obstacle can be constructed in the form of a matrix representation.

[0107] Then in the grid set E(t u )-E′(t u ) is placed virtually at the center of each grid and is recorded as Thus, through the collection And the location of the terahertz base station to obtain the virtual communication scenario for digital twins.

[0108] Sub-step S142: Calculate the terahertz signal power information of each small grid in the virtual communication scenario for digital twins, that is, realize the estimation of fine-grained terahertz signal power information.

[0109] In an embodiment, obtaining the terahertz signal power information of a virtual communication scene for a digital twin based on a ray tracing method may include: controlling the terahertz base station in the virtual communication scene to emit a ray simulating a terahertz signal, and controlling the grid set E(t u )-E′(t u ) receives the ray omnidirectionally, and records the corresponding terahertz signal power spectrum density as Ψ(e,t u ). Therefore, t u The fine-grained terahertz signal power information estimated at the moment is

[0110] Sub-step S143: averaging a plurality of fine-grained terahertz signal power information within each grid in the coarse-grained terahertz spectrum environment estimation map, and then updating the original coarse-grained terahertz signal power information with the averaged data.

[0111] In an embodiment, a grid set A(t u ), determine the fine-grained terahertz signal power information Whether each element in falls into a grid of A(t0). i For example, if there are k grids i Elements If it falls within the grid, the coarse-grained terahertz signal power information corresponding to the grid will be updated by averaging the above element values, that is, Therefore, the updated coarse-grained terahertz signal power information is recorded as where C′ n (t u ) is a grid set of coarse-grained terahertz spectrum environment estimation maps that meets the above conditions.

[0112] Sub-step S144: Integrate the obtained terahertz signal power information and obstacle information of different granularities to construct a fine-grained terahertz spectrum environment map and a coarse-grained terahertz spectrum environment map respectively.

[0113] In this embodiment, the fine-grained obstacle information obtained in sub-step S141 is integrated and the fine-grained terahertz signal power information obtained in sub-step S142 Get the actual communication scenario at t u The latest fine-grained terahertz spectrum environment map at all times

[0114] Reintegrate the coarse-grained obstacle information obtained in sub-step S139 and the coarse-grained terahertz signal power information obtained in sub-step S143 Get the actual communication scenario at t u The latest coarse-grained terahertz spectrum environment map at all times And C′ n (t u )∪D′ n (t u )=A(t u ).

[0115] Therefore, by periodically utilizing the terahertz signals passively received by terahertz communication nodes for coarse-grained passive sensing and on-demand utilizing the terahertz signals actively emitted by terahertz base stations for fine-grained active sensing, the latest coarse-grained and fine-grained terahertz spectrum environment maps for the current scenario can be simultaneously obtained, without the need for dense deployment of a large number of terahertz radio monitoring nodes and the periodic traversal method that consumes a large amount of communication resources. This effectively solves the problems of traditional terahertz spectrum environment map construction methods, such as the severe consumption of communication and hardware resources, the ineffective utilization of terahertz communication nodes, and the long-term accumulation of recursive errors.

[0116] The saved communication resources can enable terahertz communication links to meet users' high-speed data transmission needs with lower latency, and the saved hardware resources can also enable mobile operators to deploy terahertz mobile communication networks at a lower cost. The coarse-grained and fine-grained terahertz spectrum environment maps constructed can help the implementation of different terahertz signal coverage technologies through obstacle information of different granularities and terahertz signal power spectrum density information.

[0117] Example 2

[0118] This embodiment also provides a method for constructing a terahertz spectrum environment map based on hybrid active and passive sensing. Based on embodiment 1, embodiment 2 provides another method for calculating terahertz signal power, specifically including:

[0119] In the embodiment, a segmented signal propagation model is used instead of a ray tracing method to implement the sub-step S112 of the embodiment 1 regarding the time from the first moment to the tth moment in the virtual communication scene. end Specifically, the segmented signal propagation model may include: according to the obstacle information in the virtual communication scene, clearly identifying the obstacle at the grid position x a The type of communication link between the terahertz base station (t) and each terahertz communication node in the area occupied by the terahertz communication node after considering the direction and width of the terahertz beam, which includes but is not limited to direct path propagation, non-direct path propagation, and blocked direct path propagation. Then, the corresponding path loss coefficient ρ and shadow fading coefficient are obtained according to the type of communication link. Then, according to the segmented signal propagation model Calculate the received power at each terahertz communication node, and then calculate the signal power spectrum density. Therefore, according to the segmented signal propagation model, the virtual communication scenario from the first moment to the tth moment can also be obtained. end The terahertz signal power information of each grid in the time set C(t) is also recorded as

[0120] In the embodiment, a segmented signal propagation model is used instead of a ray tracing method to implement the calculation of the terahertz signal power information of the virtual communication scene for the digital twin in sub-step S142 of embodiment 1. Specifically, the segmented signal propagation model may include: clarifying the distance between the terahertz base station and the grid set E(t u )-E′(t u ) between each terahertz communication node after considering the terahertz beam direction and width, and then obtain the corresponding path loss coefficient ρ and shadow fading coefficient according to the type of communication link Then, according to the segmented signal propagation model Calculate the received power at each terahertz communication node and then calculate the signal power spectrum density Ψ(e,t u ). Therefore, t u The fine-grained terahertz signal power information estimated at the moment is

[0121] Example 3

[0122] This embodiment also provides a method for constructing a terahertz spectrum environment map based on hybrid active and passive sensing. Based on embodiment 1, embodiment 3 provides a specific implementation of active sensing for another type of modulation form of terahertz signals:

[0123] In the embodiment, the terahertz communication and perception integrated system is set to use orthogonal frequency division multiplexing signals instead of frequency modulated continuous wave signals to implement the active perception of the terahertz base station actively sending out several terahertz signal beams in the direction of the obstacle in sub-step S133 of embodiment 1, and to implement the active perception of the terahertz base station actively sending out several terahertz signal beams in the direction where the obstacle information is different in sub-step S136 of embodiment 1. Specifically, the signal s about the beam direction θ sent out is x (t′0) can be expressed as: Where P is the signal power, N T is the number of transmitting antenna elements of the terahertz base station, a(θ,N T ) is the transmit beamforming vector based on the array response vector, f c is the signal carrier frequency, t′0 is the time variable relative to the initial time t0, and The number of symbols is N Sym , the number of subcarriers is N Sub The complex symbol transmitted by the μth symbol and the nth subcarrier is S(μN Sub +n), the symbol duration is T S , the cyclic prefix duration is T CP , baseband orthogonal frequency division multiplexing signal with subcarrier spacing of Δf and given rectangular pulse RECT(·).

[0124] Specifically, in the embodiment, a uniform circular array is used as the transmitting antenna array of the terahertz base station, and its half-power beam width remains almost unchanged. Therefore, when implementing sub-step S133, the emitted signal for the i-th d The OFDM signals in the beam directions are marked as When implementing sub-step S136, the embodiment sends the signal for the i-th dThe OFDM signals in the beam directions are marked as where t′ u is relative to the initial time t u The time variable.

[0125] In the embodiment, the terahertz communication and sensing integrated system is set to realize the obstacle information at time t0 in sub-step S134 of embodiment 1 based on the received orthogonal frequency division multiplexing echo signal from the direction of the obstacle, rather than the frequency modulated continuous wave echo signal. Specifically, in the embodiment, since the terahertz communication and perception integrated system realizes active perception based on orthogonal frequency division multiplexing signals, the coarse-grained obstacle information at time t0 is For the d-th obstacle, the terahertz base station needs to realize the i-th d Therefore, under the action of the terahertz sensing channel and the receiving beamforming vector based on the array response vector, the echo signal at time t0 is can be expressed as: in Contains the path loss, molecular absorption, and obstacle reflection coefficient of the corresponding terahertz sensing channel, N R The number of receiving antenna elements equipped for the terahertz base station, is the signal about the ith obstacle on the dth d The one-way propagation delay of a point target. To achieve the d-th obstacle at the i-th time d The embodiment first receives the signal Convert to baseband and remove the cyclic prefix part, then combine it with the transmitted signal Divide the frequency domain form of From this sequence Parameters in It can also be measured based on a carefully designed sampling rate and number of Fourier transform points and recorded as Therefore, the ith obstacle on the dth obstacle d The distance between a point target and the base station can be estimated as And the i d The direction of the point target and the base station can be estimated as

[0126] In the embodiment, the terahertz communication and perception integrated system is set to implement the sub-step S137 of the embodiment 1 based on the received orthogonal frequency division multiplexing echo signal from the direction where the obstacle information exists, rather than the frequency modulated continuous wave echo signal. uFine-grained obstacle information with different directions at all times Specifically, in the embodiment, since the terahertz communication and perception integrated system realizes active perception based on orthogonal frequency division multiplexing signals, for the collection For the dth difference obstacle, the terahertz base station needs to realize the i-th d OFDM echo signals of multiple beams Similarly, the d-th difference obstacle is received at the i-th d Fine-grained active sensing of beam directions is performed by receiving signals After baseband conversion, cyclic prefix removal, frequency domain conversion and other processing, the transmitted signal This is achieved by element-by-element division and corresponding measurement after frequency domain conversion, thereby estimating the distance between the corresponding point target and the base station and direction

[0127] The method provided in the embodiment not only simplifies the complex calculations of the ray tracing method through a segmented signal propagation model, thereby reducing the computing resource overhead of mobile operators when implementing terahertz spectrum environment map construction; but also avoids the additional hardware resource overhead caused by the terahertz base station when implementing communication and perception integration based on frequency modulated continuous wave by multiplexing the orthogonal frequency division multiplexing signals that have been applied to modern communication systems when implementing active perception, thereby further reducing the deployment cost of mobile operators.

Claims

1. A method for constructing a terahertz spectrum environment map based on hybrid active and passive sensing, characterized in that: The following steps are involved: Passive sensing step: The terahertz base station receives the distributed terahertz signals uploaded by the terahertz communication nodes and periodically inputs the received terahertz signals into a well-trained generative adversarial network to estimate the coarse-grained obstacle information and terahertz signal power information; Active sensing step: This step is triggered at the initial moment of the current field communication scene or when the coarse-grained obstacle information estimated in the passive sensing step is updated. The terahertz base station determines the obstacle position based on the obtained coarse-grained obstacle information, first sends a terahertz signal to the obstacle, then receives the corresponding terahertz echo signal, and estimates fine-grained obstacle information based on the terahertz echo signal, thereby updating the coarse-grained obstacle information. Afterwards, the terahertz base station constructs a digital twin of the current field communication scene based on the estimated fine-grained obstacle information, then estimates the fine-grained terahertz signal power information, thereby updating the coarse-grained terahertz signal power information. Map construction steps: The terahertz base station combines the latest coarse-grained obstacle information and coarse-grained terahertz signal power information to construct a coarse-grained terahertz spectrum environment map, and combines the latest fine-grained obstacle information and fine-grained terahertz signal power information to construct a fine-grained terahertz spectrum environment map.

2. The method according to claim 1, wherein: The training process of the generative adversarial network is: The terahertz base station constructs a virtual communication scenario including obstacles, terahertz communication nodes and terahertz base stations, wherein the obstacles undergo random changes in position, shape and / or number at random moments in the virtual communication scenario. The terahertz base station obtains and processes a coarse-grained terahertz spectrum environment real map of the virtual communication scenario, and uses the coarse-grained terahertz spectrum environment real map to train a generative adversarial network for terahertz passive perception. The generative adversarial network is used to receive input terahertz signals and output an estimated coarse-grained terahertz spectrum environment estimation map, thereby obtaining coarse-grained obstacle information and terahertz signal power information.

3. The method according to claim 2, wherein: The specific steps for generating a training set for a generative adversarial network are: The virtual communication scene is divided into several grids. At each moment, a terahertz base station is randomly placed in a certain grid, several obstacles are randomly placed in several grids, and the remaining grids are all filled with terahertz communication nodes. Calculate the time from the 1st moment to the tth moment in the virtual communication scenario end The terahertz signal power information of each grid at the moment, and the obstacle information is counted, where t end For the preset time; According to the obtained time from the 1st moment to the tth moment end The grid information of each moment is normalized for the far-field terahertz signal power information at each moment, and then the obstacle information at each moment is associated with the normalized maximum value, and the near-field terahertz signal power information at each moment is associated with the normalized minimum value. The far-field terahertz signal power information at each moment is linearly mapped to the preset normalized minimum and maximum value intervals. Based on the above information, a coarse-grained terahertz spectrum environment real map of each moment in the virtual communication scenario is constructed. Ψ(t) represents the real map of the coarse-grained terahertz spectrum environment at time t; Real map of the coarse-grained terahertz spectrum environment at each moment Perform sparse sampling one by one to obtain a coarse-grained terahertz spectrum environment real sampling map at each moment Will and The training set is used as input to train a generative adversarial network for terahertz passive sensing.

4. The method according to claim 3, wherein: The specific steps of inputting the training set into the generative adversarial network for terahertz passive sensing are as follows: Set the maximum number of training cycles N em And the maximum number of batch training times N in the training cycle bm , set the training cycle index to N e , the batch training index is N b , record the Nth e In the Nth training cycle b The corresponding time set of the real sampling map of the coarse-grained terahertz spectrum environment used in the next batch training is Let N e =1 and N b =1; Start executing the iteration steps; The coarse-grained terahertz spectrum environment real sampling map at several random moments Input to the generator network G of the generative adversarial network for terahertz passive perception, so that the generator network G outputs a coarse-grained terahertz spectrum environment estimation map corresponding to several moments Right now Among them, θ g is the neural network parameter of the generator network G; Will Input into the discriminator network D of the generative adversarial network for terahertz passive sensing, and compare it with The corresponding coarse-grained terahertz spectrum environment real map at several moments Terahertz spectrum environment estimation map The same input is sent to the discriminator network D, so that the discriminator network D outputs true or false judgment information Right now or Among them, θ d is the neural network parameter of the discriminator network D, the true and false judgment information It is used to represent the discriminator network D for the input or Whether it comes from the true or false judgment of the generator network G; Combating losses for terahertz passive sensing and self-weighted reconstruction loss The weighted sum of is used to train the generative adversarial network, that is, the training objective of the generative adversarial network is: Where λ is the value of V re The weight factor of (G), p is the penalty factor for adversarial loss, and ∈ is uniformly distributed A random number, Express expectations, Denotes the relationship between D(Ψ′(t),Ψ′ p (t);θ d ) Find the value of Ψ′ p the gradient of (t); Let N b =N b +1, and perform the above iterative steps again until N b =N bm ; Let N e =N e +1, and perform the above iterative steps again until N e =N em The generative adversarial network designed for terahertz passive perception is fully trained.

5. The method according to claim 1, wherein: The specific implementation of the passive perception step is: Set t u is the passive sensing moment after u passive sensing cycles in the field communication scenario, t0 is the initial moment in the field communication scenario, and T is the passive sensing cycle in the field communication scenario; Let u = 0 and start the iterative steps: The actual communication scene is divided into several grids. u = the set of terahertz communication nodes existing at time t0+uT Passively receive the downlink communication signal from the terahertz base station, thereby obtaining the terahertz signal power information of the corresponding grid, where i is the terahertz communication node number, t u The total number of terahertz communication nodes at time The positions of the terahertz communication nodes are randomly distributed; t u The far-field terahertz signal power information of each grid at the moment is normalized and linearly mapped with the preset maximum and minimum values ​​and the normalized next minimum and next maximum value intervals, and then the t u The coarse-grained terahertz spectrum environment sampling map Ψ′(t u ); Ψ′(t u ) is input to the well-trained generative adversarial network, and the generator network G of the generative adversarial network outputs t u Coarse-grained terahertz spectrum environment estimation map at each moment judge Whether each element is in the second smallest and second largest value interval or the second smallest and minimum value interval, if so, the coarse-grained terahertz signal power information of the corresponding grid is estimated by inverse processing That is to achieve t u Coarse-grained passive perception of terahertz signal power information at all times; judge Whether each element in is in the second largest maximum interval, if so, the coarse-grained obstacle information is estimated by considering the corresponding grid as an obstacle That is to achieve t u Coarse-grained passive perception of obstacle information at all times; Update u=u+1 and perform the iterative steps again until the construction and estimation of the terahertz spectrum environment map are stopped in the actual communication scenario.

6. The method according to claim 1, wherein: The specific implementation of the active perception step is: S131 determines whether the current moment is the initial moment t0 of the field communication scenario. If so, execute steps S133 to S135; otherwise, execute step S132; S132 Determine the passive sensing time t u Coarse-grained obstacle information estimated at every moment With the most recent u’ =t0+u'T Whether the difference in the coarse-grained obstacle information stored at the time exceeds the preset threshold, if so, execute steps S136-S139, otherwise not; t' u is relative to time t u The time variable, u' is t' u The corresponding u'th passive sensing period, t0 is the initial moment in the field communication scenario, and T is the passive sensing period in the field communication scenario; S133 estimates the coarse-grained obstacle information obtained at time t0 The terahertz base station actively sends out several terahertz signal beams in the direction of the obstacle, and the difference in the beam directions of adjacent beams is the half-power beamwidth; The S134 terahertz base station receives the echo signal from the direction of the obstacle, estimates and saves all fine-grained obstacle information of the actual communication scene at time t0 That is, to achieve fine-grained active perception of obstacle information at time t0; S135 estimates the grid size of the map based on the coarse-grained terahertz spectrum environment, determines whether there is fine-grained obstacle information in each grid, and then updates and saves the original coarse-grained obstacle information at time t0; S136 According to t u The coarse-grained obstacle information difference obtained in step S132 is used at each moment. The terahertz base station actively sends a number of terahertz signal beams in the directions where the obstacle information differs, and the difference in beam directions between adjacent beams is the half-power beamwidth. The S137 terahertz base station receives the echo signal from the direction where the obstacle information is different, and estimates the direction where the obstacle information is different at t u Fine-grained obstacle information at all times; S138 vs. u’ The fine-grained obstacle information saved at all times is updated in the direction where the obstacle information differs, thereby obtaining and saving the actual communication scene at t u All fine-grained obstacle information at the moment That is to achieve t u Actively perceive obstacle information at a fine-grained level at all times; S139 estimates the grid size of the map based on the coarse-grained terahertz spectrum environment, determines whether there is fine-grained obstacle information in each grid in the direction where the obstacle information is different, and then updates and saves the original t u Coarse-grained obstacle information at each moment.

7. The method according to claim 1, wherein: The map construction steps specifically include: Based on the perception resolution of the terahertz communication and perception integrated system based on active sensing, the actual communication scene is divided into several small grids, and digital twins of obstacles are created to obtain a virtual communication scene for digital twins. Calculate the terahertz signal power information of each small grid in the virtual communication scenario for digital twins, that is, realize the estimation of fine-grained terahertz signal power information; Averaging several fine-grained terahertz signal power information within each grid in the coarse-grained terahertz spectrum environment estimation map, and then updating the original coarse-grained terahertz signal power information with the averaged data; The terahertz signal power information and obstacle information of different granularities are integrated to construct fine-grained terahertz spectrum environment maps and coarse-grained terahertz spectrum environment maps respectively.

Images