An adaptive channel sensing and tracking method based on channel knowledge base

Abstract

The application discloses a kind of based on adaptive channel sensing and tracking method of channel knowledge base, comprising the following steps: defining wireless channel in MIMO system, channel knowledge base and channel observation model;Based on channel knowledge base and channel sensing sequence, define the joint probability distribution based on channel measurement sequence, channel sequence, user position sequence;Adopt the switching Kalman filter architecture of fusion channel knowledge base, realize channel sensing matrix adaptive design, and carry out user positioning and channel tracking.The application is based on historical low-dimensional (relative to high-dimensional channel) channel observation to predict high-dimensional channel, according to channel prediction error to design channel sensing matrix to be used for real-time channel measurement, to realize the differential entropy minimization of channel estimation;And based on the low-dimensional channel measurement observed by adaptive channel sensing matrix, utilize the channel space-time statistical distribution (channel covariance matrix) provided in channel knowledge base and switching Kalman filtering technology to realize channel tracking.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wireless channel estimation and tracking, and in particular, to an adaptive channel sensing and tracking method based on channel knowledge base. BACKGROUND

[0002] It is a key challenge to estimate and track wireless channel quickly and accurately to improve the quality of service of wireless communication network. For example, in a large-scale multiple-input multiple-output (MIMO) system, multi-antenna technology enables beamforming and efficient signal transmission, but beam alignment requires more accurate estimation of the channel. In some stationary scenarios, although a beam dictionary can be used to traverse all possible beams, so as to select the best beam according to channel feedback, the complexity of such beam scanning process is extremely high, and its effect is strongly related to the quality of the beam dictionary.

[0003] Current wireless channel estimation and tracking techniques can be classified into three categories: methods based on known channel statistical distribution, methods based on compressed sensing, and machine learning methods based on historical channel data. Assuming that the statistical distribution of the channel is known, existing techniques can use the least mean square error or Bayesian inference to obtain the estimation of the channel. If the statistical distribution of the channel is unknown, but the channel is known to be sparse, compressed sensing algorithms can be used to recover the channel from sparse observations. If the historical channel is known, Kalman filtering can be used to implement channel tracking, or deep learning methods such as long short-term memory networks, attention networks, and generative networks can be used to implement data-driven channel prediction.

[0004] When using Kalman filtering technology to implement channel tracking, Kalman filtering is based on a stable and invariant channel spatial statistical distribution, while the channel spatial statistical distribution in the actual scenario will change due to the obstruction of obstacles. Therefore, Kalman filtering technology cannot timely capture the spatial statistical distribution of the channel, and the accuracy of the actual channel tracking will be low, and the accuracy of the channel-assisted beam alignment will be lower.

[0005] The cost of channel prediction based on deep learning is high, and this technology requires a large number of channel vectors. In a large-scale multiple-input multiple-output system, the cost of collecting these channel vectors is extremely high. Channel prediction based on deep learning requires accurate channel vectors for model training, and cannot directly perform channel prediction based on low-dimensional noisy channel observations.

[0006] In short, the limitations of existing technologies mainly include

[0007] 1) Most of the prior art assumes that the spatial statistical properties of the channel are constant. This assumption is often not satisfied when the user moves or the environment changes, and when the communication link between the user and the base station changes from unobstructed to obstructed by buildings, vehicles, etc. The spatial statistical properties of the channel are significantly changed.

[0008] 2) Most of the prior art assumes that the user's position is known. This assumption can be used to obtain the obstruction state of the link between the user and the base station, and to infer the spatial statistical properties of the channel, for example, in the presence of a three-dimensional spatial obstacle map, user position, base station position, the user position and base station position can be connected, it is checked whether there is an obstacle obstruction on the line, and different channel empirical models are adopted according to whether there is an obstacle obstruction.

[0009] 3) Most of the prior art requires a large amount of accurate channel data for model training, and in a multi-antenna system, due to the extremely high dimension of the channel, the cost of obtaining such channel data is extremely high. SUMMARY

[0010] The purpose of the present application is to overcome the shortcomings of the prior art and provide a self-adaptive channel perception and tracking method based on a channel knowledge base.

[0011] The purpose of the present application is achieved by the following technical solution: a self-adaptive channel perception and tracking method based on a channel knowledge base, comprising the following steps:

[0012] Defining a wireless channel in a MIMO system, a channel knowledge base and a channel observation model;

[0013] Based on the channel knowledge base and the channel perception sequence, a joint probability distribution based on the channel measurement sequence, the channel sequence and the user position sequence is defined;

[0014] A switching Kalman filter architecture that fuses the channel knowledge base is used to realize adaptive design of the channel perception matrix, and user positioning and channel tracking are performed.

[0015] The beneficial effects of the present application are:

[0016] 1) The technology proposed in the present application does not need to assume the stationary properties of the channel. During user movement, the channel may be suddenly obstructed by buildings, and the assumption of the stationary properties of the channel in traditional technology is not established, and the present application can cope with the scene of channel mutation.

[0017] 2) The estimation and tracking of the wireless channel in the present application do not require user position information, and the present application can output the estimated user position information.

[0018] 3) The implementation cost of the present application is extremely low, and the amount of channel observation data required is extremely small. In actual use, only a small amount, for example 1, of pilot signals is required at each discrete time for channel sensing, and channel estimation and tracking, as well as user positioning, can be performed through the technology proposed by the present application. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 a schematic diagram of the principle of the present application;

[0020] Figure 2 a schematic diagram of the channel tracking performance based on a wireless spectrum map and an adaptive channel sensing matrix. DETAILED DESCRIPTION

[0021] The technical solutions of the present application will be described in further detail below with reference to the accompanying drawings, but the scope of protection of the present application is not limited to the following description.

[0022] The present application investigates adaptive channel sensing and tracking based on a channel knowledge base in a wireless network. The channel knowledge base is a knowledge base that characterizes the spatio-temporal statistical characteristics of a channel, which can map any spatial position coordinate to the spatio-temporal statistical characteristics of a channel. These spatio-temporal statistical characteristics include, but are not limited to, a channel covariance matrix, a signal departure angle, a signal arrival angle, a received signal strength, a signal-to-noise ratio, etc. Specifically, a high-dimensional channel is predicted based on low-dimensional (relative to a high-dimensional channel) channel observations, and a channel sensing matrix is designed based on channel prediction errors for channel measurement to achieve differential entropy minimization of channel estimation. Based on the channel measurements observed by the adaptive channel sensing matrix, channel tracking is achieved using the channel spatio-temporal statistical distribution (channel covariance matrix) provided in the channel knowledge base and a switching Kalman filter technology.

[0023] The present application is applicable to adaptive sensing matrix design and wireless channel measurement and tracking in scenarios including, but not limited to, single antenna, multiple antennas, etc., and the technical solutions proposed by the present application are not affected by environmental topology, the number of base stations, and the number of users. Specifically, for a multi-base station, multi-user scenario, only one channel knowledge base needs to be constructed for each base station, and adaptive channel sensing measurement, positioning, and channel tracking for a specific user can be performed using the channel knowledge base.

[0024] Below, the present application takes a wireless channel in a Multiple-Input Multiple-Output (MIMO) network as an example to illustrate the technical solutions of the present application. The channel knowledge base in this example is embodied as a mapping from a spatial position to a channel covariance matrix, but the channel spatial characteristics carried by the channel knowledge base are not limited to a channel covariance matrix, but can also be a signal departure angle, a signal arrival angle, a received signal strength, a signal-to-noise ratio, etc. such variables related to the spatial position of a signal transmitting and receiving device and capable of characterizing a certain aspect of a channel.

[0025] As shown in Figure 1 , a method for adaptive channel sensing and tracking based on channel knowledge base, comprising the following steps:

[0026] Defining the wireless channel in MIMO system, channel knowledge base and channel observation model;

[0027] A MIMO system is investigated, which contains one base station and one mobile user, the base station is equipped with large-scale MIMO antennas, which has N t antenna units, and the mobile user uses a single antenna for communication.

[0028] First, define the wireless channel as h. Since the base station has N t antenna units, the channel h is an N t dimensional column vector. Since the channel is associated with time and space, the channel h t at time t can be modeled as a first-order autoregressive model:

[0029] h t = γh t-1 + (1-γ 2 )u t (1)

[0030] Where γ is the first-order autoregressive coefficient, h t-1 is the channel at time t-1, and u t is a vector related to the spatial statistical characteristics of the channel. Define the user's position at time t as p t , and assume that u t ~ CN(0, C(p t )), i.e. u t obeys a complex Gaussian distribution with mean 0 and variance C(p t ) of the channel covariance matrix at the position p t , where C(p t ) is provided by the channel knowledge base. According to the above channel model (1), it can be deduced that h t ~ CN(0, C(p t )).

[0031] Define a channel knowledge base as

[0032]

[0033] Where x = (x1, x2, x3) T is a three-dimensional position vector, x is the set of all spatial grid center positions after discretizing the target area into equidistant grids, and C(x) is the average channel covariance matrix at the position of the grid with center position x, which is defined as here Represents the covariance hh of all channels h in the grid H Take the expectation. According to the definition, C(x) is an N t ×N t Dimensional matrix. Since the user position p t Falling in the grid of the target area, if the center of the grid is x, then u in the channel model t The covariance C(p t ) can be replaced by C(x).

[0034] Given M pilot signals, we can observe y at time t t Create the following model:

[0035] y t =A t h t +n t (3)

[0036] Here, A t is the channel sensing matrix at the base station at time t, each row of which corresponds to the observation vector of a pilot signal on the receiving end MIMO antenna. Therefore, A t There are M rows and N t Column, h t is an N t The column vector of the row is the real channel at time t, n t is the measurement noise at time t, which is a column vector with M rows. Assume That is n t Each element of is subject to the mean of O and the variance of Gaussian distribution, where I is a column vector with M rows. According to the above channel measurement model (3), it can be deduced that in It's A t The conjugate transpose of .

[0037] definition For the observation sequence from t = 1 to t = T, define For the user trajectory from t=1 to t=T, define For the channel sequence from t=1 to t=T, define is the channel sensing matrix sequence from t=1 to t=T.

[0038] Based on the channel knowledge base and the channel sensing sequence, a joint probability distribution based on the channel measurement sequence, the channel sequence, and the user position sequence is defined;

[0039] Based on the above model, we define Channel measurement sequence channel sequence user location sequence joint probability distribution based on channel knowledge base and channel-aware sequence In the process of constructing the channel knowledge base, the present technology only assumes that the channel measurement sequence and the channel knowledge base are known, while the real channel sequence and the user location sequence are unknown, and the channel-aware sequence needs to be designed, and the unknown quantities need to be concatenated with the known quantities by using the channel spatial features in the channel knowledge base , i.e., the user location indicates the channel spatial features (through the channel covariance matrix in the channel knowledge base ), which reflect the spatial distribution characteristics of the real channel, and part of the features of the real channel are captured by the channel-aware matrix and saved in the channel observation.

[0040] channel knowledge base can provide the channel spatial features of each location, such as the channel covariance matrix C(x). According to the channel model (1), the channel h t ~ CN(0, C(p t )) can be obtained, therefore, the complex Gaussian distribution of the channel h t is known, and in combination with the relationship between the current channel and the historical channel, channel tracking can be performed.

[0041] The switching Kalman filter architecture that fuses the channel knowledge base is adopted to realize adaptive design of the channel-aware matrix, and user positioning and channel tracking are performed.

[0042] 1) Input: channel knowledge base Output: adaptive channel-aware matrix A t , user location estimate channel estimate

[0043] 2) Initialization (including but not limited to the following methods): transmit M pilots to obtain the sparse observation y1 at t = 1. Obtain or estimate an initial location Initialize the channel estimate at the current time is obtained from the channel knowledge base. Initialize a channel estimation error matrix A1 is an adaptive channel-aware matrix at t = 1 randomly generated in the initialization stage;

[0044] 3) Channel estimation and tracking (t ≥ 2):​​

[0045] a) Predict the current channel based on historical information (tracking):

[0046] Channel Prediction: in, represents the channel predicted based on the information at time t-1, represents the channel estimate at time t-1.

[0047] Channel prediction error matrix update: in It is from the position to p t The transition probability can be estimated from the user's historical trajectory data, assuming that this information is known.

[0048] b) Adaptive perception matrix design:

[0049] The channel prediction error matrix Q t|t-1 Perform eigenvalue decomposition, that is in, Is a diagonal element Q t|t-1 The diagonal matrix of the eigenvalues ​​of is the value corresponding to each eigenvalue λ n The eigenvector w n The matrix composed of .

[0050] In order to minimize the differential entropy of channel estimation (detailed derivation see later), the adaptive sensing matrix is ​​designed as: A t =[w1, w2, ..., w M ] H

[0051] c) Transmitting pilot signals for channel observation, based on the adaptive sensing matrix A t Get channel observation y t .

[0052] d) Channel prediction error matrix correction:

[0053] Among them, π t (p t ) is a function that integrates the spatial statistical characteristics of the channel and is defined as

[0054] e) Channel estimation based on current observations:

[0055] Kalman filter optimal coefficient calculation:

[0056] Channel estimation:

[0057] Channel estimation error matrix update: Q t = (I - K t A t )Q t|t-1 .

[0058] f) User location estimation:

[0059] User location is estimated by maximizing the conditional probability distribution of user location where the conditional probability can be obtained from the channel model (1) since h t |h t-1 ~ CN (γh t-1 , (1 - γ 2 )C (p t )).

[0060] g) Output channel estimation and user location estimation

[0061] The simulation performance of the present application in one embodiment is shown below. The embodiment is deployed in a 740m x 710m urban environment, and the target area contains 7 base stations, each of which is equipped with a MIMO antenna consisting of N t = 64 antenna elements, and the base stations are randomly deployed on the top of some buildings.

[0062] A user is considered to move on the road with a speed of 10m / s, and the interval of the pilot signal transmitted by the user is 10ms. Figure 2 The performance of the proposed technique under different signal-to-noise ratios (SNRs) is shown. In the figure, the horizontal axis represents the value of the signal-to-noise ratio in dB, and the vertical axis represents the channel capacity efficiency ratio, which is the ratio of the maximum channel capacity achieved by the current channel estimation scheme to the maximum channel capacity achieved by the perfect channel. It can be seen from the figure that when the signal-to-noise ratio SNR is greater than or equal to 20dB, the present technique can achieve more than 97% of the perfect channel effect, while the performance of other comparative technical solutions is below 83%.

[0063] The above is the preferred embodiment of the present application, and it should be understood that the present application is not limited to the form disclosed herein, and should not be considered as excluding other embodiments, but can be used in other combinations, modifications and environments, and can be modified within the scope of the concepts described herein by the above teachings or related art or knowledge. Any modification and change made by those skilled in the art without departing from the spirit and scope of the present application shall be within the protection scope of the claims of the present application.

Claims

1. A method for adaptive channel sensing and tracking based on channel knowledge base, characterized in that: The method comprises the following steps: defining a wireless channel, a channel knowledge base and a channel observation model in a MIMO system; defining a joint probability distribution based on a channel measurement sequence, a channel sequence and a user position sequence based on the channel knowledge base and a channel sensing sequence; adopting a switching Kalman filter architecture integrated with the channel knowledge base to realize adaptive design of a channel sensing matrix and user positioning and channel tracking.

2. The method of claim 1, wherein the method further comprises: The definition of the wireless channel, the channel knowledge base and the channel observation model in the MIMO system comprises: In a MIMO system, including 1 base station and 1 mobile user, the base station is configured with a large-scale MIMO antenna, which has N t antenna units, and the mobile user uses a single antenna for communication. defining the wireless channel as h: Since the base station has N t antenna units, the channel h is an Nt-dimensional column vector. Since the channel is associated with both time and space, the channel h t at time t is modeled as a first-order autoregressive model: h t = γh t-1 + (1 - γ 2 )u t (1) Among them, γ is the first-order autoregressive coefficient, h t-1 is the channel at time t-1, u t It is a vector related to the channel spatial statistical characteristics; the position of the user at time t is defined as p t , assuming u t ~CN(0,C(p t )), that is u t Subject to mean 0 and variance p t The channel covariance matrix C(p t ) complex Gaussian distribution, where C(p t ) is provided by the channel knowledge base; According to the above channel model (1), it is obtained that: h t ~CN(0,C(p t )); defining a channel knowledge base as: Where x = (x1, x2, x3) T is a three-dimensional position vector, It is the set of all spatial grid center positions after discretizing the target area into equidistant grids. C(x) is the average channel covariance matrix of the positions in the grid with the center position x, which is defined as here Represents the covariance hh of all channels h in the grid H Take the expectation, C(x) is an N t ×N t Dimensional matrix; due to the user position p t Falling in the grid of the target area, if the center of the grid is x, then u in the channel model t The covariance C(p t ) is replaced by C(x); Given M pilot signals, the channel observation y at time t t The following model is established: y t = A t h t + n t (3) where A t is the channel sensing matrix at the base station end at time t, each row of which corresponds to the observation vector of a pilot signal on the MIMO antenna at the receiving end, A t has M rows and N t columns, h t is an N t row column vector, which is the real channel at time t, n t is the measurement noise at time t, which is an M row column vector; it is assumed that each element of n t obeys a Gaussian distribution with mean 0 and variance , and I is an M row column vector, and according to the above channel measurement model (3), it is derived that where A is the conjugate transpose of A t ; Definitions Let be the sequence of observations from t = 1 to t = T, define the sequence of user positions as Let be the trajectory of the user from t = 1 to t = T, define Let be the sequence of channels from t = 1 to t = T, define Let be the sequence of channel sensing matrices from t = 1 to t = T.

3. The method of claim 1, wherein: The definition of the joint probability distribution based on the channel measurement sequence, the channel sequence and the user position sequence based on the channel knowledge base and the channel sensing sequence comprises: Definitions channel measurement sequence channel sequence user location sequence joint probability distribution, which is based on the channel knowledge base and channel sensing sequence In the channel knowledge base construction process, the channel measurement sequence and the channel knowledge base are known, while the real channel sequence and the user location sequence are unknown, and the channel sensing sequence needs to be designed, which needs to be concatenated between the unknown and known quantities by using the channel spatial features in the channel knowledge base , which refers to the channel covariance matrix in the channel knowledge base , the channel spatial features reflect the spatial distribution characteristics of the real channel, and part of the features of the real channel are captured by the channel sensing matrix and saved in the channel observation.

4. The adaptive channel sensing and tracking method based on a channel knowledge base according to claim 2, characterized in that: The channel knowledge base The channel space characteristics of each position are provided, including channel covariance matrix C(x), which is derived from channel h t ~ CN(0, C(p t )), thus, the complex Gaussian distribution of channel h t is known, and in combination with the relationship between the current channel and the historical channel, user positioning and channel tracking are realized.

5. The method of claim 2, wherein: The switching Kalman filter architecture integrated with the channel knowledge base to realize adaptive design of the channel sensing matrix, user positioning and channel tracking comprises: A1, input: channel knowledge base output: adaptive channel-aware matrix A t user position estimate channel estimate A2, initialization: transmit M pilots, obtain sparse observation y1 at t=1, obtain or estimate an initial position by prior knowledge Initialize the channel estimation at current time wherein, Obtained from the channel knowledge base; initialize a channel estimation error matrix wherein, A1is a t = 1 time instance adaptive channel aware matrix randomly generated in an initialization phase; A3, when t≥2, user positioning and channel tracking are performed.

6. The method of claim 5, wherein: The step A3 comprises: A301, predicting the current time channel based on historical information to realize current time channel prediction: Channel prediction: wherein, represents a channel predicted based on information at time t-1, represents a channel estimation at time t-1; Channel prediction error matrix update: where is the transition probability from position to p t , which is estimated from the user's historical trajectory data, which is known; A302, adaptive sensing matrix design: Q is the channel prediction error matrix t|t-1 Eigenvalue decomposition is performed, i.e. where λ is the diagonal matrix of eigenvalues of Q t|t-1 is the matrix of eigenvectors corresponding to each eigenvalue λ n w n ;​​ To minimize the differential entropy of the channel estimate, the adaptive sensing matrix is designed as: A t = [w1, w2,..., w M ] H ; A303, transmitting pilot signals for channel observation, based on adaptive sensing matrix A t Obtaining channel observation y t ; A304, channel prediction error matrix correction: where π t (p t ) is a function that incorporates the channel spatial statistical characteristics, which is defined as: A305, channel estimation based on current observation: Kalman filter optimal coefficient calculation: Channel estimation: Channel estimate error matrix update: Q t = (I - K t A t )Q t|t-1 ; A306, user positioning: using the maximum user position conditional probability distribution to estimate the user position ， where the conditional probability From the channel model (1), since h t |h t-1 ~ CN(γh t-1 , (1 - γ 2 )C(p t )) A307, output channel estimates and user position estimates

Images