Method and device for outputting channel assessment result and electronic equipment

Abstract

The invention discloses a method and device for outputting a channel evaluation result and electronic equipment, and the method comprises the steps: firstly obtaining a channel evaluation parameter, inputting the channel evaluation parameter to a target network model, and obtaining a target channel evaluation result. It needs to be explained that K layers of networks are included in the target network model, the channel assessment result output by the previous layer of network in the K layers of networks serves as the input of the next layer of network, and the channel assessment result is obtained through calculation of J sub-layers in each layer of network in the K layers of networks. Through the method, a newly constructed network can be used, so that the convergence of the model can be realized more quickly, and the complexity is lower.

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to a method, device and electronic device for outputting channel assessment results. Background Art

[0002] Currently, in a traditional OFDM (Orthogonal Frequency Division Multiplexing) system, in order to obtain accurate channel state information, blind, semi-blind and non-blind channel estimation methods are proposed.

[0003] In the existing channel estimation methods based on compressed sensing, the angular domain sparsity caused by the limited scattering characteristics of millimeter-wave signals is generally used by default, and the latitude of the subcarrier can be processed in two different ways, resulting in two types of estimators. The first type is the angular frequency domain estimator, in which the angular domain sparse structure common to each subchannel can be used to improve the estimation accuracy in the angular domain. The second type is the angular delay estimator, which further transforms the frequency domain subchannel into a sparse delay domain and utilizes the dual-channel sparsity in the AD domain. Although both types of estimators can effectively perform channel estimation, they have encountered challenges in practical applications, mainly because the sparse transformation from the frequency domain to the delay domain leads to sparse vectors and higher-dimensional measurement matrices, which increases the complexity of the algorithm. Summary of the invention

[0004] The present application provides a method, device and electronic device for outputting channel evaluation results, so as to reduce the computational complexity of the channel evaluation results.

[0005] In a first aspect, an embodiment of the present application provides a method for outputting a channel assessment result, the method comprising:

[0006] Acquire channel assessment parameters, wherein the channel assessment parameters at least include an initial approximation parameter, an iteration step size, and an attenuation coefficient;

[0007] The channel evaluation parameters are input into a target network model to obtain a target channel evaluation result, wherein the target network model includes a K-layer network, the channel evaluation result output by an upper-layer network in the K-layer network is used as the input of a lower-layer network, and the output result of the K-th-layer network is used as the target channel evaluation result, and each layer of the K-layer network obtains the channel evaluation result through J sub-layer calculations, and K and J are integers greater than or equal to 1.

[0008] In an optional embodiment, the acquiring a channel assessment parameter includes:

[0009] receiving a frequency domain signal;

[0010] According to the frequency domain signal, an observation matrix, a first vector and a second vector are obtained, wherein the first vector is obtained by performing a pseudo-inverse calculation on the observation matrix, and the second vector is obtained by calculating a basis matrix and the observation matrix;

[0011] The observation matrix, the first vector, the second vector, the initial approximation parameter, the iteration step size and the attenuation coefficient are used as the channel estimation parameters.

[0012] In an optional embodiment, inputting the channel parameters into a target network model to obtain a target channel evaluation result includes:

[0013] Bring the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence into the first layer network in the target network model to obtain a first layer output result;

[0014] Bringing the first layer output result, the observation matrix, the first vector, the second vector, the second initial approximation parameter in the initial approximation parameter sequence, and the second iteration step in the iteration step sequence into the second layer network in the target network model to obtain the second layer output result;

[0015] Execute each network layer in the target network model in sequence until the Kth network layer outputs the target channel evaluation result.

[0016] In an optional embodiment, the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence are brought into the first layer network in the target network model to obtain the first layer output result, including:

[0017] Bring the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence into the first type of sublayer in the first layer network to obtain an output result of the first sublayer, wherein the first sublayer includes an update module and a projection module;

[0018] Inputting the first output result into the second type of sublayer in the first layer network to obtain an output result of the second sublayer;

[0019] The output results of each type of sub-layer are obtained in turn, and the output result of the Jth sub-layer is used as the first layer output result of the first layer network.

[0020] In an optional embodiment, before inputting the channel parameters into the target network model to obtain the target channel evaluation result, the method further includes:

[0021] Obtain the loss value of each layer of the target network model;

[0022] A target loss value is determined from all loss values, and the network model is adjusted according to the target loss value to obtain the target network model.

[0023] In an optional embodiment, before inputting the channel parameters into the target network model to obtain the target channel evaluation result, the method further includes:

[0024] Obtaining a loss value of the target network model;

[0025] The weight parameters in the network model are adjusted according to the loss value to obtain the target network model.

[0026] In a second aspect, an embodiment of the present application provides a device for outputting a channel assessment result, the device comprising:

[0027] An acquisition module, used to acquire channel assessment parameters, wherein the channel assessment parameters at least include an initial approximation parameter, an iteration step size, and an attenuation coefficient;

[0028] A processing module is used to input the channel evaluation parameters into a target network model to obtain a target channel evaluation result, wherein the target network model includes a K-layer network, the channel evaluation result output by an upper-layer network in the K-layer network is used as the input of a lower-layer network, and the output result of the K-th layer network is used as the target channel evaluation result, and each layer of the K-layer network obtains the channel evaluation result through J sub-layer calculations, and K and J are integers greater than or equal to 1.

[0029] In an optional embodiment, the processing module is specifically used to receive a frequency domain signal;

[0030] According to the frequency domain signal, an observation matrix, a first vector and a second vector are obtained, wherein the first vector is obtained by performing a pseudo-inverse calculation on the observation matrix, and the second vector is obtained by calculating a basis matrix and the observation matrix;

[0031] The observation matrix, the first vector, the second vector, the initial approximation parameter, the iteration step size and the attenuation coefficient are used as the channel estimation parameters.

[0032] In a third aspect, an embodiment of the present application provides an electronic device, including:

[0033] Memory, used to store computer programs;

[0034] The processor is used to implement the above-mentioned method steps of outputting the channel assessment result when executing the computer program stored in the memory.

[0035] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps for outputting channel evaluation results are implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A flow chart of a method for outputting channel assessment results provided by the present application;

[0037] Figure 2 Schematic diagram of the model convergence speed provided for this application;

[0038] Figure 3 Schematic diagram of the iterative process of the network model algorithm provided for this application;

[0039] Figure 4 A schematic diagram of the network model architecture provided for this application;

[0040] Figure 5 A schematic diagram of the j-th layer structure of the nonlinear function provided in this application;

[0041] Figure 6 A schematic diagram showing the relationship between the number of network model layers and time consumption provided for this application;

[0042] Figure 7 A schematic diagram of the structure of a device for outputting channel assessment results provided by the present application;

[0043] Figure 8 A schematic diagram of the structure of an electronic device provided in this application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below in conjunction with the accompanying drawings. The specific operating methods in the method embodiments can also be applied to device embodiments or system embodiments. It should be noted that in the description of the present application, "multiple" is understood as "at least two". "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A is connected to B, which can represent: A is directly connected to B and A is connected to B through C. In addition, in the description of the present application, words such as "first" and "second" are only used to distinguish the purpose of description, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

[0045] The embodiments of the present application are described in detail below in conjunction with the accompanying drawings.

[0046] First of all, in order to obtain accurate channel state information in traditional OFDM systems, blind, semi-blind and non-blind channel estimation methods are proposed.

[0047] The blind channel estimation method does not rely on the pilot sequence, but estimates the relevant parameters of the channel by analyzing the statistical characteristics of the input signal source and the output sequence of the channel. This method abandons the operation of transmitting pilot signals, thereby greatly improving the transmission efficiency. Although blind channel estimation can improve spectral efficiency in theory, compared with other algorithms, the computational complexity of blind channel estimation is relatively high, and phase ambiguity will occur when the wireless channel changes rapidly, thus limiting its application in actual communication systems. Semi-blind channel estimation technology combines blind estimation and traditional pilot-based estimation methods. This method uses a small portion of pilot data to assist channel estimation, which can usually provide better performance than pure blind estimation.

[0048] Nevertheless, in semi-blind channel estimation, the length of the transmitted pilot sequence is often limited, which will also lead to phase ambiguity, slow convergence, increase propagation errors, or fall into local optimality. These factors limit the wide range of its practical applications. The third category is non-blind channel estimation. Non-blind channel estimation inserts a certain number of known pilot symbols into the transmission sequence so that the receiver can use these pilot symbols and its received signal to perform channel estimation. Non-blind channel estimation can be divided into non-sparse channel estimation and sparse channel estimation according to the sparsity of the channel.

[0049] In these algorithms, the high-dimensional channel is directly recovered from the compressed received pilot signal. In the existing channel estimation methods based on compressed sensing, the angular domain sparsity caused by the limited scattering characteristics of millimeter-wave signals is generally used by default, and the dimensionality of the subcarrier can be processed in two different ways, resulting in two types of estimators. The first type is the angular frequency domain estimator, in which the angular domain sparse structure common to each subchannel can be used to improve the estimation accuracy in the angular domain. The second type is the angular delay estimator, which further transforms the frequency domain subchannel into a sparse delay domain and utilizes the dual-channel sparsity in the AD domain. Although both types of estimators can effectively perform channel estimation, the sparse transformation from the frequency domain to the delay domain leads to sparse vectors and higher-dimensional measurement matrices, which increases the complexity of the algorithm.

[0050] In order to reduce the complexity of the algorithm in the channel estimation process, a method for outputting channel evaluation results is provided in an embodiment of the present application. In this method, channel evaluation parameters are first obtained, and the channel evaluation parameters are input into the target network model to obtain the target channel evaluation result. It should be noted here that the target network model includes a K-layer network, and the channel evaluation result output by the upper network layer in the K-layer network is used as the input of the next network layer, and the channel evaluation result is obtained by calculating J sub-layers in each layer of the network in the K-layer network. Through the above method, the newly constructed network can be used, so that the convergence of the model can be achieved faster and with lower complexity.

[0051] Reference Figure 1 A method for outputting a channel assessment result provided by an embodiment of the present application is shown, and the method includes:

[0052] S1, obtaining channel assessment parameters;

[0053] S2, input the channel evaluation parameters into the target network model to obtain the target channel evaluation result.

[0054] First of all, in a SISO-OFDM system with N subcarriers, the received frequency domain signal is:

[0055] y=XH+w=XF N h+w (5-1)

[0056] Where, X = diag(x1, x2, …, x N ) is a diagonal matrix with its parameters placed on the diagonal of an N×N matrix with 1s. The sparsity k is the length of the discrete event equivalent channel after sampling, and N>k. y and w are N×1 column vectors, and h is padded with Nk zero values ​​and is also an N×1 column vector. F N is the discrete Fourier transform matrix of length N. w is Gaussian white noise.

[0057] The transmitted signal X is a combination of the modulated pilot symbol and the data symbol. For one OFDM symbol, the subcarrier set C = a subset Q in {1, 2, 3…, N} P ∈C is reserved for |Q P | = P pilot symbols. Therefore, the received pilot signal can be described as:

[0058]

[0059] In the formula, F N , QP is a discrete Fourier transform matrix of length N indexed by pilot subcarriers Q P Select P rows, and the matrix I QPIt is also an N×N identity matrix indexed by the pilot subcarriers Q P Select row P.

[0060] According to the compressed sensing algorithm, the above formula can be converted into the following form:

[0061] y P =Th+w P (5-3)

[0062] Where T∈C P×N is the basis matrix X P With the observation matrix F N , QP The product of. Since the vector h∈C N×1 It contains only a small number of non-zero elements and has a sparsity of K, which is transformed into a sparse recovery problem as follows:

[0063]

[0064] In the above formula, ε represents noise, ‖●‖0, ‖●‖2 respectively indicate the calculation of L0 norm and L2 norm.

[0065] The smooth L0 norm algorithm uses a continuous Gaussian function g σ (x) = exp(-|x| 2 / 2σ 2 ) to approximate the L0 norm, and use the steepest descent method and gradient projection method to optimize the cost function of the smoothing function. In order to improve the sparse recovery effect of the algorithm, an improved negative exponential function is used as a smoothing function to approach the L0 norm, that is, the ISL0 algorithm. The difference between the two functions is that when the parameter σ approaches 0 from infinity, the Gaussian function approaches the L0 norm from the L2 norm, while the negative exponential function approaches the L0 norm from the L1 norm. Therefore, the negative exponential function has a higher recovery probability.

[0066] For equation (5-4), the following negative exponential function:

[0067] f σ (x) = exp(-|x| / σ) (5-5)

[0068] This function satisfies:

[0069]

[0070] Assume h n is the nth element in h, then the cost function can be defined as:

[0071]

[0072] Then equation (5-4) can be transformed into:

[0073]

[0074] In order to obtain the optimal solution of equation (5-4), the ISL0 algorithm uses a decreasing sequence of σ to solve it. The specific calculation is done through the following process:

[0075] initialization:

[0076] Calculate y by using the pseudo-inverse matrix of T P =The least squares solution of Th gives h (0) , and given the approximation parameter sequence {σ1,...,σ I} and the iterative step sequence {μ1,...,μ I};

[0077] Loop i=1,…,I times:

[0078] (1) Let σ = σ i , μ=μ i ;

[0079] (2) Use the J-times steepest descent method to calculate the function F σ (h) Approximately maximize over the feasible domain h = {h:||yP-Th||2≤ε}:

[0080] 1) Order

[0081] 2) Loop j = 1, ..., J times:

[0082] i) Calculation

[0083] ii) Calculation

[0084] iii) Projection onto the feasible domain:

[0085] (3) Order i←i+1;

[0086] Final result: h = h (I)

[0087] If the iteration step size μ in the steepest descent method is too large, the function may not converge, and if it is very small, the efficiency of the calculation will decrease. The approximation parameter σ determines the function F σ The smoothness of (h): The larger the σ, the higher the F σ (h) is smoother, but less close to the L0 norm; the smaller σ is, the smaller F σ The less smooth (h) is, the closer it is to the L0 norm, making equation (5-8) difficult to solve.

[0088] In general, the initial approximation sequence of the ISL0 algorithm {σ1,...,σ I} and the iterative step sequence {μ1,...,μ I} are given in advance. σ1=2max|h (0) |,σ i =cσ i-1 , c is the attenuation coefficient, usually set between 0.5 and 1. Iteration step μ1 = ... = μ I =μ>0 is usually set to 1.

[0089] However, a fixed initialization method does not always ensure that the ISL0 algorithm achieves the best results. Improper settings may affect its accuracy and reduce the convergence speed. Figure 2 As shown in Figure 1, when the initial values ​​of σ, c and μ are fixed and set to different parameters, the recovery performance also changes. In the figure, the received signal and the sensing matrix T are generated according to the channel model and formula (5-3). The simulation parameter settings are shown in Table 1, and the frequency domain values ​​of the pilot are all 1.

[0090]

[0091] Table 1

[0092] From the above content, we can see that these different parameter settings are very unstable when the SNR changes. If c and μ are set smaller, the recovery performance is poor. If c and μ are set larger, the recovery performance is improved in high SNR conditions, but worse in low SNR conditions. If and are set too large, the recovery performance is improved in low SNR conditions, but still worse in high SNR conditions. Research has found that the size of c and μ determines whether the ISL0 algorithm can search for the optimal solution and the number of iterations required to converge. If c and μ are set too small, the convergence speed will become very slow, and the performance will be very poor at the same number of iterations. If c and μ are set too large, ISL0 may skip the optimal solution or fall into a local optimum, and the performance will also deteriorate. Therefore, it is limited in practical applications.

[0093] Therefore, in order to determine the optimal c and μ, the iterative process of the ISL0 algorithm is expanded into a directed flow graph of data between different sub-nodes, referring to Figure 3As shown, the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence are brought into the first layer network in the target network model to obtain the first layer output result; the first layer output result, the observation matrix, the first vector, the second vector, the second initial approximation parameter in the initial approximation parameter sequence, and the second iteration step in the iteration step sequence are brought into the second layer network in the target network model to obtain the second layer output result; each layer network in the target network model is executed in sequence until the Kth layer network outputs the target channel evaluation result.

[0094] Specifically, refer to Figure 3 As shown, in Figure 3 The i-th layer in the algorithm is represented as the i-th external loop. The i-th layer also contains 2j sub-layers. These 2j sub-layers can be regarded as two modules, namely the update module and projection module After the input received signal is iterated I×2J times, an estimated channel frequency domain response is obtained.

[0095] According to the above algorithm steps and Figure 3 The data iteration process diagram shows that the ISL0 algorithm can be regarded as Figure 4 The first step is to receive the frequency domain signal and obtain the observation matrix (T), the first vector (y P ), the second vector (h0 (i) ), the first vector is obtained by pseudo-inverse calculation of the observation matrix, and the second vector is obtained by calculating the matrix and the observation matrix.

[0096] Specifically, the input of ISL0-UNet is T, y P and h (0) , the training parameters are {σ1,...,σ I} sequence and {μ1,...,μ I} sequence. The nonlinear activation function is P J (●), represents the inner loop of the ISL0 algorithm, and the output is h (I) The i-th layer network output of ISL0-UNet can be expressed as:

[0097] h (i) =P J (y P ,T,h (i-1) ,σ i ,μ i ) (5-9)

[0098] Nonlinear activation function P J(●) represents the inner loop of the ISL0 algorithm, which can be equivalent to a J-layer sub-network in the network. The input of this sub-network is, P ,T,h0 (i) , σ i and μ i , the output is

[0099] In the embodiment of the present application, the nonlinear activation function P J The structure of the jth (j=1, 2, ..., J) layer in (●) is as follows Figure 5 As shown. Figure 5 The nonlinear activation function includes an update module and a projection module. The specific definitions of these two modules are as follows:

[0100] First, update the module Volume: The output vector of the Jth steepest descent data in the i-1 layer is projected. is the initial input of the i-th layer Then the update module can be expressed as:

[0101]

[0102] In the above formula, σ i , μ i are the approximation parameters and iteration step parameters of the i-th layer in ISL0-UNet, and the output vector Input to the projection module introduced next.

[0103] Projection module As the input of this module, the projection module can be expressed as:

[0104]

[0105] Where T + =T H (TT H ) -1 , when j≤J-1 and i≤I, the output vector h j (i) for Input; when j = J and i ≤ I-1, the output vector h J (i) is the i+1th layer in the network When j = J and i = I, the output vector h J (i) The final output h (I) , the h (I) It is the channel assessment parameter.

[0106] Based on the architecture of the above nonlinear activation function, in the process of training the network model, the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence are brought into the first type of sublayer in the first layer of the network to obtain the output result of the first sublayer. The first output result is input into the second type of sublayer in the first layer of the network to obtain the output result of the second sublayer; the output result of each type of sublayer is obtained in turn, and the output result of the Jth sublayer is used as the first layer output result of the first layer of the network.

[0107] The network model ISL0-UNet is constructed in the above manner, and then the training data is obtained by processing the subcarriers. Specifically, in the embodiment of the present application, the number of subcarriers can be set to N=256 and the number of pilots can be set to 32. The modulation mode adopts 4QAM. Different pilot insertion methods are used, and different methods determine the different observation matrices T. The specific pilot insertion methods are shown in Table 2:

[0108]

[0109] Table 2

[0110] And the frequency domain data value of the pilot is 1. According to (5-2), generate As training data labels, generate As the test label set. The number of multipaths L P is 5, according to the observation matrix T, Generate the corresponding training data set and test set in, The noise w is additive white Gaussian noise, and the signal-to-noise ratio ranges from 5 to 25 dB.

[0111] The network parameters of the constructed ISL0-UNet network model include the approximation sequence {σ1,...,σ I} and the iterative step sequence {μ1,...,μ I In the embodiment of the present application, two different parameter learning methods are provided to learn the modularity of the ISL0-UNet network. These two different parameters will train two different network models, such as ISL0-UNet1 and ISL0-UNet2.

[0112] Among them, the training parameter of ISL0-UNet1 is σ i =cσ i-1 ,μ1=…=μ I,c, c is the attenuation coefficient, so the ISL0-UNet1 network model only needs to train σ1, μ1 and c to obtain a better {σ1,...,σ I} and {μ1,...,μ I This training method only requires training three parameters and can reach convergence more quickly, reducing the situation where the final result falls into the local optimal solution.

[0113] For another training method, the training parameters of ISL0-UNet2 are no longer limited to σ i =cσ i-1 and μ1=…=μ I , but σ in each layer of the network model i , μ i . , the number of parameters required for training the network model increases to 2I. Through this method of training, the linear fitting ability of the network model can be made stronger.

[0114] For the initialization of ISL0-UNet, in order to ensure that the network model can adapt to various distributed data and sparse vectors, rather than the validity of a single data. Therefore, in the embodiment of the present application, the network model is input into h (0) = 0. Two different initialization methods are shown in Table 3:

[0115]

[0116] Table 3

[0117] In the above formula,

[0118] After initializing the network with the minimum L2 norm solution, the next value of σ (i.e., σ1) can be selected as the maximum absolute value of the acquired data (max(|h| (i) )).

[0119] Given a network model, iteratively optimize the number of layers I and the training data set The normalized mean square error (NMSE) algorithm is used as the loss function of the network model:

[0120]

[0121] In the formula, represents the output of the iterative optimization of the first layer in the network model, Φ is the training parameter, T,h (0) , The two training methods in the embodiment of the present application are as follows:

[0122] For ISL0-UNet1:

[0123] In this training method, the loss value of each layer of the target network model is obtained, the target loss value is determined among all the loss values, and the network model is adjusted according to the target loss value to obtain the target network model. Specifically, the optimal value of the network parameters of ISL0-UNet1 is It can be obtained by minimizing the network loss function Ε(Φ) through the back propagation algorithm.

[0124] For ISL0-UNet2:

[0125] In this training method, the loss value of the target network model is obtained, and the weight parameters in the network model are adjusted according to the loss value to obtain the target network model. In other words, the network parameters are trained using a layer-by-layer training method. Under this premise, the parameters of the first i+1 layers of the network are trained in two steps. First, keep The parameter Φ of the i+1th layer of the network is unchanged and is learned through formula (5-13) i+1 :

[0126]

[0127] in, Represents the output of the i+1th layer of the network, and then the parameters Set it as the initial value and update the parameters through formula (5-14):

[0128]

[0129] Use the layer-by-layer training method to obtain the optimal parameters from the first layer to the i-th layer After network parameter optimization, the channel estimation results can be obtained through a new test data set. We get formula (5-15):

[0130]

[0131] In the embodiment of the present application, the computational complexity of the algorithm is measured by using two methods: algorithm time consumption and real number multiplication. Figure 6 The time consumption comparison of the three methods is shown in Figure 1. It can be seen that as the number of network layers I increases, the computation time required by the ISL0 algorithm is higher than that of ISL0-UNet1 and ISL0-UNet2. For ISL0-UNet1, the number of parameters that need to be trained for each network layer is 2, while for ISL0-UNet2, the number is 2I. However, it should be emphasized that in practical applications, when the number of layers is the same, the time complexity of these two different networks is O(I).

[0132] The computational complexity of each network model is explained in detail below.

[0133] Taking the ISL0 algorithm as an example, since the initialization of the ISL0 algorithm requires solving the least squares norm solution first, which involves finding the pseudo-inverse matrix of the observation matrix T, the number of real number multiplications for solving the pseudo-inverse is first analyzed.

[0134] Assume that m pilots and n subcarriers are used to generate an m×n measurement matrix T. The pseudo matrix T of the measurement matrix T is + =T H (TT H ) -1 The number of real number multiplications required is O(4m 2 n+4m 3 / 3+4m 2 n), that is: Ο(8m 2 n+4m 3 / 3).

[0135] After obtaining the least squares norm solution, according to the steps of the ISL0 algorithm, in the J inner loops, calculate It takes 6n real number multiplications to calculate It takes 6n times to calculate the gradient projection Therefore, after initialization calculation, J inner loops, and 1 outer loop, the total number of real number multiplications required by the ISL0 algorithm is O((8mn+12n)IJ+8m 2 n+4m 3 / 3).

[0136] According to the complexity calculation method of the ISL0 algorithm, the computational complexity of OMP, mpNet and ISL0-UNet can also be obtained. The complexity of the three recovery algorithms is shown in Table 4, where A represents the number of atoms in the dictionary matrix constructed in mpNet, and D represents the number of network layers determined by the input signal-to-noise ratio.

[0137]

[0138]

[0139] Table 4

[0140] In order to more intuitively show the complexity relationship between various methods, Table 5 calculates the complexity of each algorithm according to different configurations, where the subcarrier is 256, the number of iterations of ISL0 and ISL0-UNet is the same, the number of network layers is 10, and the number of subnetwork layers is 5. Since OMP needs to provide sparsity K, mpNet needs to provide the number of atoms A and the number of network layers D, and ISL0 does not require additional prior information, only the computational complexity of ISL0-UNet and ISL0 is compared.

[0141]

[0142] Table 5

[0143] When other conditions are the same, when 16 pilots are configured, the complexity of ISL0-UNet is reduced by about 22.82% compared with ISL0, when 32 pilots are configured, the complexity of ISL0-UNet is reduced by about 38.43%, and when 64 pilots are configured, the complexity of ISL0-UNet is reduced by about 56.58%. It should be pointed out that in the case of the same number of layers, the computational complexity of ISL0-UNet1 and ISL0-UNet2 after offline training is consistent in the online application stage.

[0144] The expanded network structure of ISL0-UNet proposed in the embodiment of the present application is consistent with the calculation steps of the ISL0 algorithm, except that the input h of the network is initialized. (0) = 0, avoiding the calculation of y p =The least squares norm solution of Th. Therefore, when the number of iterations is the same, the computational complexity of ISL0-UNet is O((8mn+12n)IJ), which is one less complexity of calculating the pseudo-inverse matrix than ISL0. 2 n+4m 3 / 3).

[0145] Obviously, through the method provided in the embodiment of the present application, a channel estimation model combining a sparse recovery algorithm and a deep neural network architecture is constructed, and then the model-driven and data-driven methods are used. Therefore, the ISL0-UNet proposed in the present application has a higher precision estimation performance and lower computational complexity compared with ISL0, mpNet, etc. And compared with other sparse recovery algorithms such as LS, OMP, etc., ISL0-UNet can still have better recovery performance in an environment with large SNR changes, and can quickly obtain accurate estimation results when the sparsity is unknown.

[0146] In addition, when parameters such as the number of subcarriers are the same, as the number of iterative optimization network layers increases, even if ISL0-UNet uses fewer pilot resources, it can still achieve good recovery effects.

[0147] Based on the same inventive concept, an apparatus for outputting a channel evaluation result is also provided in an embodiment of the present application, referring to Figure 7 As shown, the device comprises:

[0148] An acquisition module 701 is used to acquire channel assessment parameters, wherein the channel assessment parameters at least include an initial approximation parameter, an iteration step size, and an attenuation coefficient;

[0149] Processing module 702 is used to input the channel evaluation parameters into the target network model to obtain a target channel evaluation result, wherein the target network model includes a K-layer network, the channel evaluation result output by the upper network layer in the K-layer network is used as the input of the lower network layer, and the output result of the K-th network layer is used as the target channel evaluation result, and each network layer in the K-layer network obtains the channel evaluation result through J sub-layer calculations, and K and J are integers greater than or equal to 1.

[0150] In an optional embodiment, the processing module 702 is specifically configured to receive a frequency domain signal;

[0151] According to the frequency domain signal, an observation matrix, a first vector and a second vector are obtained, wherein the first vector is obtained by performing a pseudo-inverse calculation on the observation matrix, and the second vector is obtained by calculating a basis matrix and the observation matrix;

[0152] The observation matrix, the first vector, the second vector, the initial approximation parameter, the iteration step size and the attenuation coefficient are used as the channel estimation parameters.

[0153] Based on the same inventive concept, an electronic device is also provided in an embodiment of the present application, and the electronic device can realize the function of the aforementioned device for outputting channel evaluation results, referring to Figure 8 , the electronic device comprises:

[0154] At least one processor 801, and a memory 802 connected to the at least one processor 801. The specific connection medium between the processor 801 and the memory 802 is not limited in the embodiment of the present application. Figure 8 In the example, the processor 801 and the memory 802 are connected via a bus 800. Figure 8 The connections between other components are shown in bold lines, and are not intended to be limiting. The bus 800 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, Figure 8 Only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. Alternatively, the processor 801 can also be called a controller, and there is no limitation on the name.

[0155] In the embodiment of the present application, the memory 802 stores instructions that can be executed by at least one processor 801. The at least one processor 801 can execute a method for outputting a channel assessment result discussed above by executing the instructions stored in the memory 802. The processor 801 can implement Figure 7 The functions of each module in the device shown.

[0156] Among them, the processor 801 is the control center of the device, and can use various interfaces and lines to connect the various parts of the entire control device. By running or executing instructions stored in the memory 802 and calling the data stored in the memory 802, the various functions of the device and process data, the device can be monitored as a whole.

[0157] In one possible design, the processor 801 may include one or more processing units, and the processor 801 may integrate an application processor and a modem processor, wherein the application processor mainly processes an operating system, a user interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the modem processor may not be integrated into the processor 801. In some embodiments, the processor 801 and the memory 802 may be implemented on the same chip, and in some embodiments, they may also be implemented separately on separate chips.

[0158] Processor 801 can be a general-purpose processor, such as a central processing unit (CPU), a digital signal processor, an application-specific integrated circuit, a field programmable gate array or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, and can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. A general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of a method for outputting a channel assessment result disclosed in the embodiments of the present application can be directly embodied as a hardware processor to be executed, or a combination of hardware and software modules in the processor can be executed.

[0159] The memory 802 is a non-volatile computer-readable storage medium that can be used to store non-volatile software programs, non-volatile computer executable programs and modules. The memory 802 may include at least one type of storage medium, such as a flash memory, a hard disk, a multimedia card, a card-type memory, a random access memory (Random Access Memory, RAM), a static random access memory (Static Random Access Memory, SRAM), a programmable read-only memory (Programmable Read Only Memory, PROM), a read-only memory (Read Only Memory, ROM), an electrically erasable programmable read-only memory (Electrically Erasable Programmable Read-Only Memory, EEPROM), a magnetic memory, a disk, an optical disk, etc. The memory 802 is any other medium that can be used to carry or store a desired program code in the form of an instruction or data structure and can be accessed by a computer, but is not limited thereto. The memory 802 in the embodiment of the present application can also be a circuit or any other device that can realize a storage function, for storing program instructions and / or data.

[0160] By programming the processor 801, the code corresponding to the method for outputting the channel assessment result described in the above embodiment can be fixed into the chip, so that the chip can execute the code when running. Figure 1 The steps of a method for outputting a channel estimation result in the embodiment shown are as follows: How to design and program the processor 801 is a technology well known to those skilled in the art and will not be described in detail here.

[0161] Based on the same inventive concept, an embodiment of the present application also provides a storage medium, which stores computer instructions. When the computer instructions are executed on a computer, the computer executes a method for outputting channel assessment results discussed above.

[0162] In some possible implementations, various aspects of a method for outputting channel assessment results provided by the present application may also be implemented in the form of a program product, which includes a program code. When the program product is run on an apparatus, the program code is used to enable the control device to execute the steps of a method for outputting channel assessment results according to various exemplary implementations of the present application described above in this specification.

[0163] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0164] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0165] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0166] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0167] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

1. A method for outputting a channel assessment result, characterized in that: The method comprises: Acquire channel assessment parameters, wherein the channel assessment parameters include at least an initial approximation parameter, an iteration step size, and an attenuation coefficient; The channel evaluation parameters are input into a target network model to obtain a target channel evaluation result, wherein the target network model includes a K-layer network, the channel evaluation result output by an upper-layer network in the K-layer network is used as the input of a lower-layer network, and the output result of the K-th-layer network is used as the target channel evaluation result, and each layer of the K-layer network obtains the channel evaluation result through J sub-layer calculations, and K and J are integers greater than or equal to 1.

2. The method according to claim 1, characterized in that The obtaining of channel assessment parameters comprises: receiving a frequency domain signal; According to the frequency domain signal, an observation matrix, a first vector and a second vector are obtained, wherein the first vector is obtained by performing a pseudo-inverse calculation on the observation matrix, and the second vector is obtained by calculating a basis matrix and the observation matrix; The observation matrix, the first vector, the second vector, the initial approximation parameter, the iteration step size and the attenuation coefficient are used as the channel estimation parameters.

3. The method according to claim 1, characterized in that Inputting the channel parameters into the target network model to obtain a target channel evaluation result includes: Bring the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence into the first layer network in the target network model to obtain a first layer output result; Bringing the first layer output result, the observation matrix, the first vector, the second vector, the second initial approximation parameter in the initial approximation parameter sequence, and the second iteration step in the iteration step sequence into the second layer network in the target network model to obtain the second layer output result; Execute each network layer in the target network model in sequence until the Kth network layer outputs the target channel evaluation result.

4. The method according to claim 3, characterized in that: Bringing the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence into the first layer network in the target network model to obtain a first layer output result, including: Bring the observation matrix, the first vector, the second vector, the first initial approximation parameter in the initial approximation parameter sequence, and the first iteration step in the iteration step sequence into the first type of sublayer in the first layer network to obtain an output result of the first sublayer, wherein the first sublayer includes an update module and a projection module; Inputting the first output result into the second type of sublayer in the first layer network to obtain an output result of the second sublayer; The output results of each type of sub-layer are obtained in turn, and the output result of the Jth sub-layer is used as the first layer output result of the first layer network.

5. The method according to claim 1, characterized in that Before inputting the channel parameters into the target network model to obtain the target channel evaluation result, the method further includes: Obtain the loss value of each layer of the target network model; A target loss value is determined from all loss values, and the network model is adjusted according to the target loss value to obtain the target network model.

6. The method according to claim 1, characterized in that Before inputting the channel parameters into the target network model to obtain the target channel evaluation result, the method further includes: Obtaining a loss value of the target network model; The weight parameters in the network model are adjusted according to the loss value to obtain the target network model.

7. A device for outputting a channel assessment result, characterized in that: The device comprises: An acquisition module, used to acquire channel assessment parameters, wherein the channel assessment parameters at least include an initial approximation parameter, an iteration step size, and an attenuation coefficient; A processing module is used to input the channel evaluation parameters into a target network model to obtain a target channel evaluation result, wherein the target network model includes a K-layer network, the channel evaluation result output by an upper-layer network in the K-layer network is used as the input of a lower-layer network, and the output result of the K-th layer network is used as the target channel evaluation result, and each layer of the K-layer network obtains the channel evaluation result through J sub-layer calculations, and K and J are integers greater than or equal to 1.

8. The device according to claim 7, characterized in that The processing module is specifically used to receive a frequency domain signal; According to the frequency domain signal, an observation matrix, a first vector and a second vector are obtained, wherein the first vector is obtained by performing a pseudo-inverse calculation on the observation matrix, and the second vector is obtained by calculating a basis matrix and the observation matrix; The observation matrix, the first vector, the second vector, the initial approximation parameter, the iteration step size and the attenuation coefficient are used as the channel estimation parameters.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to implement the method steps of any one of claims 1 to 6 when executing the computer program stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 6 are implemented.

Images