CN107086970A - Channel Estimation Method Based on Bayesian Algorithm - Google Patents

Abstract

The invention belongs to wireless communication technology field, a kind of channel estimation methods based on bayesian algorithm are particularly related to.The inventive method calculates the shared variance parameter of common sparse position with united Bayesian Estimation algorithm, compared with common Bayesian Estimation variance parameter, the accuracy of estimate variance is greatly improved, at the same replaced with GAMP algorithms bayesian algorithm avoid to posterior probability direct matrix in verse process.The beneficial effects of the invention are as follows：Compared with conventional method, this invention simplifies operand, arithmetic speed and operational precision are improved, the accuracy of channel estimation is improved.

Description

Channel Estimation Method Based on Bayesian Algorithm

technical field

The invention belongs to the technical field of wireless communication, and in particular relates to a channel estimation method based on a Bayesian algorithm.

Background technique

Massive MIMO (Multiple Input Multiple Output) system is one of the key technologies of the fifth-generation mobile communication system. Its main advantages are: system capacity increases with the number of antennas; transmission signal power is reduced; simple The linear precoder and detector can achieve the optimal performance; the channels tend to be orthogonal, so the co-channel interference in the cell is eliminated. The prerequisite for realizing these advantages is that the base station (BS) knows the channel state information (CSIT). In a Time Division Duplex (TDD) system, channel estimation is performed at the user end (MS) by utilizing the reciprocity of the uplink and downlink channels. For the FDD massive MIMO system, the channel estimation process is as follows: the base station broadcasts pilot signals to each user, and the mobile users use the received signal to estimate CSIT and then feed it back to the base station. In this case, the number of pilot signals is proportional to the number of base station antennas. Due to the huge number of antennas in massive MIMO systems, conventional channel estimation methods (such as the least squares method) will face huge training overhead, making the training time become shorter. Long, or even more than the desired dry time of the channel, making channel estimation meaningless.

Contents of the invention

The purpose of the present invention is to propose a channel estimation method based on Bayesian algorithm. The invention mainly utilizes the principle of compressed sensing and the common sparse position of the multi-task channel, and proposes an improved Bayesian algorithm on the basis of the Bayesian algorithm, so as to improve the accuracy of channel estimation.

In order to make it easier for those skilled in the art to understand the technical solution of the present invention, the principle of compressed sensing and the Bayesian algorithm used in the present invention will be described first.

Massive MIMO multi-task channel estimation system model:

Assume that the channel to be estimated is flat-block fading, that is, the channel state does not change within a certain period of time.

The system has a base station BS, and each BS is a large-scale antenna array configured with M antennas. The mathematical model of FDD massive MIMO multi-task channel estimation can be expressed as R _p =Φ _p H _p +W _p , where P represents the total number of tasks of the multi-task channel, R _p represents the received signal matrix of the pth task, and H _p represents the channel matrix of the p-th task between the BS and the MS, Φ _p is the measurement matrix of the P-th task, and W _p is the received noise signal matrix of the p-th task.

Standard compressed sensing mathematical model:

y=Αx+n, where Α is a perception matrix with a size of n×m, y is an n×1-dimensional compressed signal, x is an m×1-dimensional sparse signal, and its sparsity is k, that is, only k<<m elements are non-zero, and all other elements are 0. n is n×1-dimensional system noise and its elements obey the Gaussian distribution with mean value 0 and variance σ ² .

The improved Bayesian algorithm obtains the update formulas of channel parameters and noise parameters by maximizing the marginal likelihood function, uses the obtained parameter values to update the mean and variance estimates of the channel, and iteratively updates the channel parameters and noise parameters and the mean of the channel and variance information until the channel estimation requirements are met.

pass Get channel parameter α and noise β, namely G represents the length of the training overhead and is also the dimension of the received signal R _p

In a large-scale antenna array, the channel The column vectors of have the same sparse support set, that is, the sparse positions of different tasks of the channel are exactly the same, and the same sparse positions obey the same complex Gaussian distribution. The common sparse support set of P task channels H _p is expressed as Ω=Ω _p , p=1,...P.

Technical scheme of the present invention is:

Channel estimation method based on Bayesian algorithm, including:

The transmitting end:

The base station uses G time slots to broadcast G multitasking pilot signals to the UE: H=[H ₁ ,H ₂ ,...,H _p ]∈C ^M×P ; where P represents the total task of the multitasking channel number, M is the number of antennas, H _p =[h ₁ ,h ₂ ,...,h _M ] ^T , and the multitasking channels H _p have the same sparse characteristics;

Receiving end:

The receiving end obtains the received signal R according to the pilot signal and the task channel, so that the received signal matrix for P tasks is expressed as: R=[R ₁ ,R ₂ ,…R _p ]∈C ^G×P ; where R _p represents The received signal matrix of the pth task, p=1,2,...,P; it is characterized in that, the method for estimating the channel at the receiving end comprises the following steps:

S1. Set the iteration control variable ε and the maximum number of iterations N supported by P task signal sparseness;

S2. Given the initial value:

The element H _m of the mth row of the channel H has a mean of 0 and a variance of The same complex Gaussian distribution of ; is the variance distribution that the M row elements of the channel H obey; noise W=[W ₁ ,W ₂ ,…W _P ], W _p is the received noise signal matrix of the pth task, all elements obey the mean value of 0, and the variance is complex Gaussian distribution;

S3. Using the following formula 1 and formula 2 to obtain the posterior probability variance Σ _p and mean value u _p of the channel H respectively:

in, Represents the conjugate transposition of Φ _p , A=diag(α ₁ ,α ₂ ,…α _M ) represents the diagonalization of the variance α; since matrix inversion requires high computational complexity, the actual estimation uses the GAMP algorithm (Generalized ApproximateMessage Passing) to reduce the complexity of inversion;

S4. Update the channel parameter α _m and the noise parameter β using the following formula 3 and formula 4:

Among them, u _p (m) represents the mean value of the m-th position element of the p-th vector H _p of the sparse channel H, and Σ _p (m) represents the variance of the m-th position element of the p-th vector H _p of the sparse channel H; represent a sparse signal the mean value of represent a sparse signal Variance;

S5. Iterate step S3 and step S4 until the iteration control variable ε or the maximum number of iterations N is satisfied, then end the loop and enter step S6;

S6. Output the estimated mean value and variance of each task channel, and the obtained estimated mean value u=[u ₁ , u ₂ , . . . u _P ] is the final estimated result of the multi-task channel.

The method of the present invention uses the joint Bayesian estimation algorithm to calculate the shared variance parameter α _m of the common sparse position, compared with the ordinary Bayesian estimated variance parameter α _m , the accuracy of the estimated variance is greatly improved, and the GAMP algorithm is used instead The Bayesian algorithm avoids the process of direct matrix inversion of the posterior probability of H.

The beneficial effect of the present invention is: compared with the traditional method, the present invention simplifies the calculation amount, improves the calculation speed and calculation precision, and improves the accuracy of channel estimation.

Description of drawings

Figure 1 is a schematic diagram of multi-user massive MIMO channel joint sparsity;

Fig. 2 is the algorithm flowchart of the present invention;

Fig. 3 is the performance comparison figure of algorithm of the present invention and common BCS, DSAMP, MT-BCS algorithm under different expenses G;

Fig. 4 is a performance comparison diagram of the algorithm of the present invention and common BCS, DSAMP, and MT-BCS algorithms under different sparsity K.

detailed description

Below in conjunction with specific accompanying drawing and embodiment, the present invention is described in further detail:

Figure 1 is a schematic diagram of a multi-task massive MIMO channel.

Assume that the number of multiple tasks is P=3, the number of base station antennas is M=50, and the signal-to-noise ratio SNR=20dB. In the performance comparison chart under different overhead G, G is 6, 8, 10, 12, 14, 16, and the sparsity K is 4. In the performance comparison chart under different sparsity K, G takes 8, and K takes 2, 3, 4, 5, and 6.

Example

Figure 2 is a flow chart of multi-user massive MIMO channel estimation. According to the flow chart, taking the above parameters as an example, this example specifically includes:

S1. Initialization, specifically:

S11. The BS uses G time slots to broadcast G multi-task pilot signals H=[H ₁ ,H ₂ ,H ₃ ]∈C ^50×3 to the UE, where H _p =[h ₁ ,h ₂ , ...,h ₅₀ ] ^T , the multitasking channels H _p have the same sparse property.

S12. The received signal matrix of the multi-task at the user end is R=[R ₁ , R ₂ , R ₃ ], where R _p represents the received signal matrix of the p-th task, and p=1,2,3.

S2. Multi-task sparse support joint estimation, that is, use the Bayesian algorithm to jointly estimate multi-task signals, and obtain the mean value u _p (m) and variance Σ _p (m) of each position element. The specific iterative estimation algorithm is as follows:

S21. Set the iteration control variable ε= ^10-3 and the maximum number of iterations N=20 supported by P task signal sparseness;

S22, given the initial value:

The element H _m of the mth row of the channel H has a mean of 0 and a variance of The same complex Gaussian score of ; the initial value α ₁ =100, α ₂ =100, ... α _M =100 is the variance distribution of the M row elements of the channel H; the noise W = [W ₁ , W ₂ , ... W _P ], All elements have a mean of 0 and a variance of The complex Gaussian distribution of , where Pn is the noise variance obtained from the signal-to-noise ratio SNR.

S33. By the improved Bayesian inference algorithm, the update formula of the posterior probability variance Σ _p and the mean value u _p of the channel H is obtained: in Represents the conjugate transposition of Φ _p , A=diag(α ₁ ,α ₂ ,…α _M ) represents the diagonalization of variance α. Since matrix inversion requires high computational complexity, the GAMP algorithm (Generalized Approximate Message Passing) is used in actual estimation to reduce the complexity of inversion.

S34. Update the channel parameter α _m and the noise parameter β by the expectation maximization algorithm, and the update formula is as follows:

Among them, u _p (m) represents the mean value of the m-th position element of the p-th vector H _p of the sparse channel H, and Σ _p (m) represents the variance of the m-th position element of the p-th vector H _p of the sparse channel H; represent a sparse signal the mean value of represent a sparse signal Variance.

S35. Steps S33 and S34 are iterated until the iteration control variable ε or the maximum number of iterations N is satisfied, and the loop ends.

S4. Output the estimated mean value and variance of each task channel. The estimated mean value u=[u ₁ , u ₂ , . . . u _P ] is the final estimated result of the multi-task channel.

Fig. 3 is a graph comparing the performance of the algorithm of the present invention when it is applied to multi-task massive MIMO channel estimation with that of other sparse signal recovery algorithms for different overhead G when it is applied to the same channel estimation. As can be seen from the figure, the algorithm of the present invention has reached optimal performance when the base station sends 10 pilot signals. Compared with the common BCS, DSAMP, and MT-BCS algorithms, the estimated performance of this algorithm is close to the ideal situation Estimated performance below. The rest of the algorithms need the base station to send more pilot signals to achieve optimal performance. By comparison, it is illustrated that the algorithm of the present invention has obvious advantages in reducing the overhead of multi-task massive MIMO channel estimation, making the realization of massive MIMO channel estimation possible in practice.

Fig. 4 is a graph comparing the performance of the algorithm of the present invention when it is applied to multi-user massive MIMO channel estimation and other sparse signal recovery algorithms for different sparsity K when it is applied to the same channel estimation. It shows that the performance of the present invention is consistent under the environment of different sparsity K. The same conclusion as in Figure 3 can be drawn under different sparsity K.

Claims (1)

1. Channel estimation method based on Bayesian algorithm, including: The transmitting end: The base station uses G time slots to broadcast G multitasking pilot signals to the UE: H=[H ₁ ,H ₂ ,...,H _p ]∈C ^M×P ; where P represents the total task of the multitasking channel number, M is the number of antennas, H _p =[h ₁ ,h ₂ ,...,h _M ] ^T , and the multitasking channels H _p have the same sparse characteristics; Receiving end: The receiving end obtains the received signal R according to the pilot signal and the task channel, so that the received signal matrix for P tasks is expressed as: R=[R ₁ ,R ₂ ,…R _p ]∈C ^G×P ; where R _p represents The received signal matrix of the pth task, p=1,2,...,P; it is characterized in that, the method for estimating the channel at the receiving end comprises the following steps: S1. Set the iteration control variable ε and the maximum number of iterations N supported by P task signal sparseness; S2. Given the initial value: The element H _m of the mth row of the channel H has a mean of 0 and a variance of The same complex Gaussian distribution of ; is the variance distribution that the M row elements of the channel H obey; noise W=[W ₁ ,W ₂ ,…W _P ], W _p is the received noise signal matrix of the pth task, all elements obey the mean value of 0, and the variance is complex Gaussian distribution; S3. Using the following formula 1 and formula 2 to obtain the posterior probability variance Σ _p and mean value u _p of the channel H respectively: in, Represents the conjugate transposition of Φ _p , A=diag(α ₁ ,α ₂ ,…α _M ) represents the diagonalization of variance α; S4. Update the channel parameter α _m and the noise parameter β using the following formula 3 and formula 4: Among them, u _p (m) represents the mean value of the m-th position element of the p-th vector H _p of the sparse channel H, and Σ _p (m) represents the variance of the m-th position element of the p-th vector H _p of the sparse channel H; represent a sparse signal the mean value of represent a sparse signal Variance; S5. Iterate step S3 and step S4 until the iteration control variable ε or the maximum number of iterations N is satisfied, then end the loop and enter step S6; S6. Output the estimated mean value and variance of each task channel, and the obtained estimated mean value u=[u ₁ , u ₂ , . . . u _P ] is the final estimated result of the multi-task channel.