Downlink channel knowledge map construction method and channel estimation method based on diffusion model

Abstract

The invention discloses a downlink channel knowledge map (CKM) construction method based on a diffusion model and a channel estimation method, and belongs to the technical field of wireless communication. According to the method, a base station service area is divided into grids, an off-line training data set is constructed by combining pilot frequency data and position labels of receiving equipment, and channel statistical information of different positions is learned by using a diffusion model to generate the CKM. In the channel estimation stage, based on the CKM and a small amount of pilot frequency data, noise is gradually removed through the reverse process of the diffusion model, and efficient channel estimation is achieved. Compared with a traditional method, the method has the advantages that the requirement for the number of pilots is remarkably reduced (only 1-2 pilots are needed), the estimation precision is improved (the NMSE is reduced by more than 40%), and the method is adaptive to a high-density MIMO system and is suitable for a 6G environment sensing communication scene.

Description

Technical Field

[0001] The present invention relates to the technical field of channel estimation, and in particular to a downlink channel knowledge map construction method and a channel estimation method based on a diffusion model. Background Art

[0002] In recent years, various emerging applications, such as autonomous driving, networked robots, and virtual reality, have placed increasing demands on the latency and throughput of wireless communication systems. The fifth generation of mobile communication technology (5G) can no longer meet all the requirements, and the application of millimeter waves and massive MIMO has made it possible to solve these problems. Due to the increasing density of access devices and the size of antennas, the energy consumption and cost of RF modules are also issues that cannot be ignored.

[0003] The above new challenges and opportunities bring about a promising paradigm shift from traditional context-unaware communication to new context-aware communication, which fully exploits the prior knowledge of the actual local environment to design and optimize the communication network. Context-aware communication can be seen as a further development of traditional location-aware communication, as it not only exploits location information but also the local actual environment information where the communication takes place. On the one hand, context-aware communication is a promising technology to address the new challenges facing the sixth generation of mobile communication technology (6G), such as the difficulty in obtaining real-time CSI (Channel State Information) as network density and channel size increase. On the other hand, context-aware communication becomes more feasible than ever due to the accessibility of high-quality location information, rich sources of location-specific data, and more powerful data mining capabilities to understand the environment.

[0004] CKM (Channel Knowledge Map) in context-aware communications has recently attracted widespread attention. Different from the physical environment map, CKM aims to directly reflect the wireless channel properties inherent to the local environment. Specifically, CKM is a place-indexed database marked with the locations of transmitters and / or receivers, which provides location-specific channel knowledge to help improve situational awareness and facilitate or even avoid complex real-time CSI acquisition. The abstract term "knowledge" is deliberately chosen to emphasize that any information related to the local wireless channel can be the expected query output of CKM, including but not limited to channel modeling parameters for a specific area, such as path loss, delay, Doppler, and angular spread; as well as location-specific channel knowledge, such as channel gain, delay, Doppler, angle of arrival (AoAs) and angle of departure (AoDs) of multipath, and even the impulse response of the entire channel.

[0005] Within a certain range, CKM only needs to be calculated offline based on the channel information obtained in advance. Once the CKM is built, the communication between the user and the base station only requires a few pilot signals to obtain relatively accurate CSI. In fact, with the continuous increase in computing resources, both users and base stations can calculate local CKM to assist in uplink or downlink communications. Traditional CKM construction usually uses a channel estimation algorithm to obtain certain channel parameters at the current location and stores the channel information of each path locally, which consumes a lot of storage space. Secondly, different channel estimation methods require different information when using CKM, and different CKM construction and usage methods may cause incompatibility.

[0006] With the development of artificial intelligence (AI), network architectures such as convolutional neural networks, adversarial neural networks, and transformers have had a significant impact in different fields. In recent years, the diffusion model has continued to develop in the field of image generation and has become one of the mainstream models in the field of image generation. In addition, the diffusion model can also solve the problem of solving linear equations with noise in a statistical sense, so it can be applied to channel estimation problems. In the process of training the diffusion model, the channel position information is added as a condition to control the sampling process of the diffusion model. In the inference stage, only a small number of pilots are needed to recover the channel coefficients that meet the requirements of channel estimation. Summary of the invention

[0007] The present invention aims to provide a method for constructing a downlink channel knowledge map and a channel estimation method based on a diffusion model. By adding location information as a label during the model training process, the model can learn channel statistics at different locations and use it as a CKM to help complete channel estimation. After the CKM is constructed, the diffusion model can generate location-based channel prior information in the constructed CKM and complete channel estimation with the help of a few pilots.

[0008] In order to solve the above technical problems, the specific technical solutions of the present invention are as follows:

[0009] A method for constructing a downlink channel knowledge map based on a diffusion model comprises the following steps:

[0010] Step S1: Divide the base station service area into n×n grids, and the receiving device moves within the grid and collects pilot data to generate an original data set with location tags;

[0011] Step S2: Based on the data set in step S1, an offline channel estimation algorithm is used to obtain channel parameters at each location to construct training data;

[0012] Step S3: Build a diffusion model. The forward process gradually adds noise according to the Markov chain, and the reverse process predicts the noise through the U-Net network. The loss function is:

[0013]

[0014] in β t is the noise parameter set in advance, x0 is the original input data, ∈ is the standard Gaussian process random sampling, ∈ θ is the output function of the U-Net network;

[0015] Step S4: After training is completed, the diffusion model generates location-based CKMs and stores channel statistics.

[0016] Preferably, in step S1, the grid division density satisfies n≥50, and the pilot data includes multipath delay, angle of arrival (AoA) and complex channel coefficient.

[0017] Preferably, the U-Net network input is a two-channel matrix of the real part and the imaginary part of the complex channel data, and is embedded with position coding and diffusion steps.

[0018] The present invention also provides a channel estimation method based on the CKM, comprising:

[0019] Step S5: The user receives pilot data Y=HX+N and extracts the current position p, where H is the channel matrix, N is the additive noise, X is the pilot matrix, and Y is the received signal;

[0020] Step S6: input p into CKM, obtain a priori channel information, and initialize the estimated channel;

[0021] Step S7: In the diffusion reverse process, iteratively execute:

[0022]

[0023] Among them, H t is the estimated channel obtained at the tth iteration, ∈ θ represents the noise prediction network trained in advance, α t , β t represents the preset noise parameters, is the cumulative noise attenuation coefficient, α t =1-β t ;

[0024] Step S8: Combine likelihood gradients The estimation is corrected and the estimated channel is finally output.

[0025] Preferably, the likelihood gradient is calculated as:

[0026]

[0027] σ 2 is the noise variance, and I is the identity matrix.

[0028] Preferably, in step S7, the number of iterations T≤100, and the final NMSE is lower than -20 dB.

[0029] Preferably, the method supports a large-scale MIMO system with a base station antenna number N. t ≥32, number of users K ≥16.

[0030] Preferably, the diffusion model is trained using the Adam optimizer with a learning rate of lr=10 -4 , batch size batch=64.

[0031] Preferably, the CKM adapts to the millimeter wave frequency band (28 GHz and above) and supports a dynamic update mechanism to periodically integrate newly collected channel data.

[0032] Preferably, the method is integrated into 6G base station equipment.

[0033] Beneficial effects:

[0034] 1. Apply the diffusion model to the construction of the downlink channel knowledge map (CKM). By dividing the base station service area into grids, an offline training data set is constructed based on the pilot data and location labels collected by the receiving device, so that the diffusion model can learn the channel statistics at different locations to generate CKM.

[0035] Compared with the traditional CKM construction method that consumes a lot of storage space, this method uses the characteristics of data distribution learned by the diffusion model to more efficiently store and characterize channel statistical information. At the same time, the introduction of the diffusion model makes the construction and use of CKM more compatible, avoiding incompatibility issues between different methods.

[0036] 2. In the channel estimation stage, based on the constructed CKM and a small amount of pilot data (only 1-2 pilots are required), the channel estimation is achieved by gradually removing noise with the help of the reverse process of the diffusion model.

[0037] This method significantly reduces the dependence on the number of pilot signals, reduces pilot data transmission, improves the data transmission rate, and alleviates the problem of data transmission rate being affected by the increase in pilot signals. It is especially suitable for 6G communication scenarios with extremely high requirements for data transmission rate.

[0038] 3. Through the collaborative work of the diffusion model and CKM, the normalized mean square error (NMSE) of channel estimation is reduced by more than 40%. In complex communication environments, such as 6G macro cell scenarios, more accurate channel estimation can be achieved, effectively improving communication quality and reliability, ensuring stable signal transmission, and reducing data transmission errors and losses.

[0039] 4. This method supports large-scale MIMO systems, with the number of base station antennas Nt ≥ 32 and the number of users K ≥ 16. It meets the communication needs with the increasing density of access devices and antenna size, adapts to the development trend of future communication networks, and provides an effective channel estimation solution for the application of high-density MIMO systems in 6G and subsequent communication technologies.

[0040] 5. CKM is compatible with millimeter wave frequency bands (28GHz and above) and supports a dynamic update mechanism that can periodically integrate newly collected channel data. This meets the application requirements of 6G for millimeter wave frequency bands. At the same time, the dynamic update mechanism enables CKM to adapt to changes in the communication environment, maintain accurate representation of channel information, and further improve the accuracy and adaptability of channel estimation.

[0041] 6. The channel estimation method is integrated into the 6G base station equipment, which is convenient for direct application in the actual 6G communication system, accelerating the transformation of the technology from theoretical research to practical deployment, and promoting the development and application of 6G communication technology. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 Schematic diagram of the area served by the base station.

[0043] Figure 2 Schematic diagram of U-Net structure.

[0044] Figure 3 Schematic diagram of the channel estimation process.

[0045] Figure 4 This is a graph showing the variation trend of NMSE as the number of pilot signals increases.

[0046] Figure 5 This is the trend diagram of NMSE changing with SNR. DETAILED DESCRIPTION

[0047] In the 6G macro cell scenario, there are a lot of NLoS paths due to the presence of scatterers such as buildings and trees. Real-time and accurate channel estimation is a challenging problem when there are many antennas. Although the base station to user channel changes in real time, the scatterers on the signal propagation path usually have relatively stable geometric dimensions and relatively little change in position, resulting in relatively stable channel statistics. Therefore, a channel knowledge map can be established in advance, and relatively stable statistical information can be used to assist in channel estimation.

[0048] like Figure 1 As shown, it is assumed that G areas are divided in the area served by the base station, and each area is represented by Furthermore, the base station side can know that the served user is located at a certain location, thus enabling a more accurate estimation.

[0049] The present invention discloses a downlink channel knowledge map (CKM) construction method and a channel estimation method based on a diffusion model, comprising the following steps:

[0050] 1. Offline CKM Construction

[0051] Step 1: Divide the base station service range into several grid points. In order to obtain the information required to construct CKM, use the receiving device to randomly move within the base station service range in advance, receive the pilot data sent by the base station at different locations and add location tags.

[0052] Step 2: Use the pilot data, received signal and location information in step 1 to construct an offline training data set. Since the offline computing power is sufficient, a more complex channel estimation method can be used to obtain the estimated channel data.

[0053] Step 3: Build a diffusion model and train it with the constructed data set. During the training process, randomly select the t-th step forward process to add noise, and use the network to predict the noise in the backward process. The process of network convergence means that the construction of CKM is completed.

[0054] 2. CKM Real-time Application

[0055] Step 1: A user receives pilot data sent by a base station at any location within the service range of the base station. The user side knows the pilot data and the current location.

[0056] Step 2: In each step of the backward process, the estimated channel, step t and current position obtained in the previous step are input into the network, and the network will output the predicted noise of the current step.

[0057] Step 3: Use the predicted noise output by the network combined with the gradient calculated from the received data and the pilot to obtain the estimated channel of the current step, until t reaches the preset number of steps to obtain the final estimated channel.

[0058] The present invention will be further explained below in conjunction with the embodiments.

[0059] Example 1: Offline CKM Construction

[0060] The diffusion model includes a forward process and a backward process. The forward process can be regarded as a Markov process with continuous noise, which is expressed as follows:

[0061]

[0062] where 0<β1<β2<…<β T <1 indicates the preset noise parameter, the purpose of which is to add noise to the original data step by step so that Therefore, according to the properties of Gaussian distribution and Markov process, the distribution can be obtained in one step:

[0063]

[0064] in α t =1-β t , so in the forward process x t The distribution of satisfies:

[0065]

[0066] in

[0067] If the reverse process is defined as a parameterized variational Markov chain, we can obtain:

[0068]

[0069] where ∈ θ represents a pre-trained noise prediction network that can fit the noise at each step according to the input. According to the following equation:

[0070]

[0071] And Jensen inequality:

[0072]

[0073] So the minimum evidence lower bound (ELBO) is:

[0074]

[0075] The right side of the above equation can be simplified to:

[0076]

[0077] The first term has nothing to do with the network, and the third term is only related to the one-step noise process, so we focus on the second term.

[0078] According to the Bayesian formula,

[0079]

[0080] So we can get

[0081]

[0082] Its mean and variance are

[0083] If we assume And the following formula holds:

[0084]

[0085] Then the ELBO lower bound can be further simplified as:

[0086]

[0087] To further simplify, Use x t The expression is:

[0088]

[0089] Therefore μ θ (x t ,t) must have a similar form, assuming

[0090]

[0091] Substituting both means into the previously derived ELBO yields:

[0092]

[0093] At this point, the loss function for training the diffusion model is derived. At the same time, it can be seen that the diffusion model is learning the distribution of data x0. Therefore, for the channel estimation problem, the channel information of multiple states can be used as a data set to learn the statistical information of the current channel. Usually, the network used to predict noise in the diffusion model is a U-Net structure. Compared with the traditional autoencoder structure, U-Net can better preserve data information. The U-Net structure adopted in this experiment is as follows Figure 2 As shown, the real and imaginary parts of the complex channel data are combined as a two-channel image. After convolution, downsampling and self-attention modules, the data dimensions are as shown in the figure, where the position information g and the diffusion model step t are combined and merged into the network data.

[0094] To construct CKM, it is necessary to collect channel information from different locations and create a dataset with location information for training. g For model training, the diffusion model will learn the statistical information p of different data sets based on the label g θ (H g ), thus completing the construction of CKM.

[0095] Example 2: Method of using CKM

[0096] In the mobile communication downlink transmission system, as the user moves, the environment around the user is constantly changing, resulting in changes in the scatterers between the communication links, which has a great impact on the channel model parameters such as multipath delay and Doppler shift. Therefore, the base station is usually required to send pilot data to estimate the channel data in real time to ensure the normal communication process. However, the increase in pilot data will significantly affect the data transmission rate, so a channel estimation method based on CKM of the diffusion model is proposed.

[0097] Consider a single-cell downlink FDD system with N t The uniform linear array of antennas, so the frequency domain response between the base station and each user is:

[0098]

[0099] where ξ l represents the standard complex Gaussian coefficient, ρ l represents the complex gain of the lth subpath, φ l represents the angle of arrival, a(φ l ) represents the frequency domain array response vector having the following form:

[0100]

[0101] When there is N in a certain area r Users receive N T pilot, the transmitted data can be expressed as The received signal can be expressed as

[0102] Y=HX+N

[0103] Where N represents additive complex Gaussian noise with zero mean and variance σ 2 I, the channel matrix can be expressed as Assume that the pilot signal is sent with a DFT matrix that satisfies tr(X H X) = N t N T .

[0104] According to the above model, the channel estimation problem can be expressed as estimating the channel matrix at the current moment based on the received signal. According to the maximum a posteriori criterion, the following problems need to be solved:

[0105] max H p(H|Y)∝p(H)p(Y|H)

[0106] The steps of the reverse process of the diffusion model are to continuously subtract the predicted noise from a Gaussian noise sample, so the noise at each step can be regarded as a gradient, that is:

[0107]

[0108] In the problem with posterior information, the above gradient should be changed to According to the Bayesian formula, we can get:

[0109]

[0110] The distribution can be learned through the diffusion model Therefore, the calculation of the above formula still lacks the likelihood probability of the second term, because p θ (Y|H t )=∫p θ (Y|H0)p θ (H0|H t )dH0, where p θ (H0|H t ) cannot be calculated accurately, so it is assumed So it satisfies:

[0111]

[0112] According to the previously constructed CKM, the statistical channel information of each position has been obtained, that is, p(H) in the above formula, so the following formula holds:

[0113]

[0114] in represents the standard complex Gaussian distribution, and then substituting the above formula into the formula of the received signal, we get

[0115]

[0116] Therefore, the likelihood probability can be calculated as

[0117]

[0118] So the gradient of the likelihood is

[0119]

[0120] According to the assumptions made during the CKM construction process:

[0121]

[0122] In the above formula Using the gradient of the posterior probability Substituting:

[0123]

[0124] This results in a complete channel estimation process, such as Figure 3 shown.

[0125] Simulation results:

[0126] In this downlink channel estimation, the minimum mean square error (MMSE), sparse Bayesian learning (SBL), and orthogonal matching pursuit algorithm (OMP) are selected as comparison methods. The base station is equipped with a uniform linear array of 32 antennas, serving 32 users in an area at the same time. The number of multipaths is 21. In a 50m*50m square area, the horizontal and vertical lengths are divided into d segments, with a total of The transmitted pilot is the column of the DFT matrix, and the standard mean square error (NMSE) is used as the measurement mark. The change trend of NMSE with the increase of the number of pilots is as follows: Figure 4 As shown. First, we can see that as the area is divided, the density As the number of pilots increases, the NMSE becomes smaller and smaller. Compared with the method without channel prior information, the method based on the diffusion model only needs 1 to 2 pilots to achieve good performance. Secondly, from the trend of the curve, the channel estimation assisted by the diffusion model CKM is close to convergence when the number of pilots is 4, and the performance exceeds that of SBL with 32 pilots. In addition Figure 5 The NMSE changes with SNR. It can also be seen that with the increase of the partition density As increases, NMSE becomes smaller, which means the estimation performance is getting better. Compared with the estimation method without CKM assistance, the estimation accuracy is significantly higher.

Claims

1. A method for constructing a downlink channel knowledge map based on a diffusion model, characterized in that: The steps include: Step S1: Divide the base station service area into n×n grids, and the receiving device moves within the grid and collects pilot data to generate an original data set with location tags; Step S2: Based on the data set in step S1, an offline channel estimation algorithm is used to obtain channel parameters at each location to construct training data; Step S3: Build a diffusion model. The forward process gradually adds noise according to the Markov chain, and the reverse process predicts the noise through the U-Net network. The loss function is: in β t is the noise parameter set in advance, x0 is the original input data, ∈ is the standard Gaussian process random sampling, ∈ θ is the output function of the U-Net network; Step S4: After training is completed, the diffusion model generates location-based CKMs and stores channel statistics.

2. A method for constructing a downlink channel knowledge map based on a diffusion model according to claim 1, characterized in that: In the step S1, the grid division density satisfies n≥50, and the pilot data includes multipath delay, angle of arrival (AoA) and complex channel coefficient.

3. The method for constructing a downlink channel knowledge map based on a diffusion model according to claim 1, characterized in that: The U-Net network input is a two-channel matrix of the real and imaginary parts of the complex channel data, and is embedded with position encoding and diffusion steps.

4. A channel estimation method based on the CKM according to claims 1-3, characterized in that: include: Step S5: The user receives pilot data Y=HX+N and extracts the current position p, where H is the channel matrix, N is the additive noise, X is the pilot matrix, and Y is the received signal; Step S6: input p into CKM, obtain a priori channel information, and initialize the estimated channel; Step S7: In the diffusion reverse process, iteratively execute: Among them, H t is the estimated channel obtained at the tth iteration, ∈ θ represents the noise prediction network trained in advance, α t , β t represents the preset noise parameters, is the cumulative noise attenuation coefficient, α t =1-β t ; Step S8: Combine likelihood gradients The estimation is corrected and the estimated channel is finally output.

5. The method according to claim 4, characterized in that The likelihood gradient is calculated as: σ 2 is the noise variance, and I is the identity matrix.

6. The method according to claim 4, characterized in that The number of iterations T in step S7 is ≤ 100, and the final NMSE is lower than -20 dB.

7. The method according to claim 4, characterized in that The method supports large-scale MIMO systems with N base station antennas. t ≥32, number of users K ≥16.

8. The method according to claim 1, characterized in that: The diffusion model is trained using the Adam optimizer with a learning rate of lr=10 -4 , batch size batch=64.

9. The method according to claim 1, characterized in that: The CKM is adapted to the millimeter wave frequency band (28 GHz and above) and supports a dynamic update mechanism to periodically integrate newly collected channel data.

10. The method according to claim 4, characterized in that The method is integrated into 6G base station equipment.

Images