Joint uplink and downlink dictionary learning and channel estimation method and device

Abstract

The present invention provides a method and device for joint uplink and downlink dictionary learning and channel estimation applied to a large-scale antenna system, including: based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on uplink and downlink received signals, determining sparse coding vectors, and updating uplink and downlink dictionaries; performing downlink channel estimation based on sparse coding vectors and uplink and downlink dictionaries of the current time period; wherein the first objective function is based on the expected form of logarithms, the first objective function is related to uplink and downlink received signals, uplink and downlink dictionaries, and sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors, and the constraint condition of the first objective function satisfies less than or equal to the model matching error. The present invention is used to capture channel characteristic changes in communication scenarios through online dictionary updates, and improve the accuracy of downlink channel estimation in dynamic scenarios by using limited pilot overhead.

Description

Technical Field

[0001] The present invention relates to the field of wireless communication technology, and in particular to a method and device for joint uplink and downlink dictionary learning and channel estimation. Background Art

[0002] In the application of massive multi-input multi-output (MIMO) technology, the base station needs to accurately obtain the channel state information (CSI) of the downlink. Most current cellular systems, especially those with symmetric traffic and delay-sensitive applications, operate in frequency division duplexing (FDD) mode. However, compared with time division duplexing (TDD), it is more challenging to obtain downlink CSI in FDD systems. First, the base station must actively transmit pilot symbols to the user and request downlink CSI feedback. In this case, the traditional downlink channel estimation method requires training and feedback overhead proportional to the number of antennas on the base station side, which leads to serious pilot overhead problems. Second, in actual mobile communication scenarios, even if the pilot sequence length is equal to the number of base station antennas, the traditional downlink channel estimation method will fail due to the limited channel coherence time.

[0003] To overcome the above challenges, compressed sensing (CS) technology has been used to utilize the sparsity of the channel in the angular domain and has been applied to many existing downlink channel estimation works. Among them, the method based on discrete Fourier transform (DFT) is only applicable to uniform linear array (ULA) and has modeling errors caused by directional mismatch. The sparse channel model based on dictionary learning is data-driven, so it can adapt to any antenna array. In addition, by utilizing the angular reciprocity between the uplink and downlink channels, uplink-assisted downlink channel estimation can be achieved. Considering the increasingly mobile dynamic communication scenarios, the above-mentioned CS-based and uplink-assisted downlink channel estimation methods cannot capture the changes in channel characteristics in dynamic communication environments well, and therefore cannot be well applied to mobile scenarios.

[0004] Therefore, there is an urgent need for a downlink channel estimation method for dynamic scenarios. Summary of the invention

[0005] The present invention provides a joint uplink and downlink dictionary learning and channel estimation method and device, which are used to solve the defects in the prior art that it is difficult to perform downlink channel estimation in dynamic scenarios and the accuracy of downlink channel estimation in dynamic scenarios is low, and the channel characteristic changes in communication scenarios are captured by online dictionary updating, the accuracy of downlink channel estimation in dynamic scenarios is improved, and the pilot overhead of the system is reduced.

[0006] The present invention provides a joint uplink and downlink dictionary learning and channel estimation method, which is applied to a large-scale antenna system, comprising:

[0007] Collect uplink and downlink receiving signals in the same time slot;

[0008] Based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating an uplink and downlink dictionary;

[0009] Perform downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionaries of the current time period;

[0010] The first objective function is based on the expected form of logarithms, and is related to the uplink and downlink received signals, the uplink and downlink dictionaries, and the sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors. When the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error;

[0011] The uplink and downlink dictionaries are used to characterize changes in channel characteristics. The downlink channel estimation is performed in each time slot. The uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0012] According to a joint uplink and downlink dictionary learning and channel estimation method provided by the present invention, based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, the uplink and downlink received signals are subjected to joint uplink and downlink dictionary learning and channel estimation, a sparse coding vector is determined, and an uplink and downlink dictionary is updated, including:

[0013] The non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation subproblems and a long-term dictionary learning subproblem according to the relaxed dual-time-scale primal-dual decomposition theory, and correspondingly, the first objective function includes a second objective function corresponding to the short-term channel estimation subproblem and a third objective function corresponding to the long-term dictionary learning subproblem;

[0014] Based on the second objective function, determining a sparse coding vector;

[0015] Based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period.

[0016] According to a joint uplink and downlink dictionary learning and channel estimation method provided by the present invention, based on the second objective function, a sparse coding vector is determined, including:

[0017] determining a non-convex part and a convex part of the second objective function;

[0018] Determine a proxy function of the non-convex part based on a first-order Taylor expansion of the non-convex part of the second objective function;

[0019] Determine a target proxy function corresponding to the second objective function based on the convex part of the second objective function and the proxy function of the non-convex part;

[0020] Based on the uplink and downlink dictionaries and the target proxy function, J iterations are performed to obtain a stable solution to the short-term channel estimation subproblem, and the sparse coding vector is obtained.

[0021] According to a joint uplink and downlink dictionary learning and channel estimation method provided by the present invention, based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries of the next time period, including:

[0022] Determine a convex proxy function corresponding to the third objective function;

[0023] Based on the convex proxy function, updating the uplink and downlink dictionaries to determine the uplink and downlink dictionaries for the next time period;

[0024] Wherein, the convex proxy function includes a target proxy function corresponding to the third objective function and a constraint proxy function corresponding to the constraint function of the third objective function;

[0025] The target proxy function includes an approximation of the third target function, a first-order Taylor expansion term of the third target function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the third target function with respect to Lagrange multipliers, and a quadratic term of the third target function with respect to the uplink and downlink dictionaries and the Lagrange multipliers;

[0026] The constraint proxy function includes an approximation of the constraint function, a first-order Taylor expansion term of the constraint function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the constraint function with respect to Lagrange multipliers, and a quadratic term of the constraint function with respect to the uplink and downlink dictionaries and the Lagrange multipliers.

[0027] According to a joint uplink and downlink dictionary learning and channel estimation method provided by the present invention, based on the convex proxy function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries of the next time period, including:

[0028] Obtaining a current iteration parameter group, wherein the iteration parameter group includes an uplink and downlink dictionary and a Lagrange multiplier;

[0029] Based on the iteration parameter group and the convex proxy function, iteratively update the uplink and downlink dictionaries to determine a set of optimal solutions corresponding to the convex proxy function;

[0030] Based on the iterative parameter group and the optimal solution, an uplink and downlink dictionary for the next time period is determined.

[0031] According to a joint uplink and downlink dictionary learning and channel estimation method provided by the present invention, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem according to the relaxed dual-time-scale primal-dual decomposition theory, including:

[0032] According to the relaxed dual-time-scale primal-dual decomposition theory, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem at the t-th channel statistical coherence time;

[0033] The coherence time of the channel statistical time scale includes several frames, and each frame includes multiple time slots.

[0034] The present invention also provides a combined uplink and downlink dictionary learning and channel estimation device, comprising:

[0035] A signal collection module is used to collect uplink and downlink received signals in the same time slot;

[0036] A parameter determination module, configured to perform joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, determine a sparse coding vector, and update an uplink and downlink dictionary;

[0037] A downlink channel estimation module, used to perform downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period;

[0038] Wherein, the first objective function is an expected form based on logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionaries and the sparse coding vectors, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error;

[0039] The uplink and downlink dictionaries are used to characterize changes in channel characteristics. The downlink channel estimation is performed in each time slot. The uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0040] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the above-mentioned joint uplink and downlink dictionary learning and channel estimation methods is implemented.

[0041] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the combined uplink and downlink dictionary learning and channel estimation method as described in any one of the above is implemented.

[0042] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned joint uplink and downlink dictionary learning and channel estimation methods.

[0043] The joint uplink and downlink dictionary learning and channel estimation method and device provided by the present invention utilize the time scale separation of dictionary and sparse channel representation parameters, propose a non-convex dual-time-scale stochastic optimization problem for joint uplink and downlink dictionary learning and channel estimation, and achieve capturing channel characteristic changes in communication scenarios through online dictionary update, thereby solving the defects in the prior art that it is difficult to perform downlink channel estimation for dynamic scenarios and the accuracy of downlink channel estimation in dynamic scenarios is low, thereby improving the accuracy of downlink channel estimation in dynamic scenarios and reducing the pilot overhead of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 This is one of the flow charts of the joint uplink and downlink dictionary learning and channel estimation method provided by the present invention;

[0046] Figure 2 It is a schematic diagram of the time axis of the dual-time-scale joint uplink and downlink dictionary learning and channel estimation provided by the present invention;

[0047] Figure 3 This is the second flow chart of the joint uplink and downlink dictionary learning and channel estimation method provided by the present invention;

[0048] Figure 4 This is the third flow chart of the joint uplink and downlink dictionary learning and channel estimation method provided by the present invention;

[0049] Figure 5 It is a network structure diagram of the short-term channel estimation sub-algorithm provided by the present invention;

[0050] Figure 6 This is a fourth flow chart of the joint uplink and downlink dictionary learning and channel estimation method provided by the present invention;

[0051] Figure 7 This is the fifth flow chart of the joint uplink and downlink dictionary learning and channel estimation method provided by the present invention;

[0052] Figure 8 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0054] Combine the following Figure 1-Figure 7 The joint uplink and downlink dictionary learning and channel estimation method of the present invention is described.

[0055] Please refer to Figure 1 The joint uplink and downlink dictionary learning and channel estimation method proposed in the present invention is applied to a large-scale antenna system, comprising:

[0056] Step 10, collecting uplink and downlink received signals in the same time slot;

[0057] Step 20, based on the first objective function corresponding to the non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating an uplink and downlink dictionary;

[0058] Step 30, performing downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period;

[0059] The first objective function is based on the expected form of logarithms, and is related to the uplink and downlink received signals, the uplink and downlink dictionaries, and the sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors. When the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error;

[0060] The uplink and downlink dictionaries are used to characterize changes in channel characteristics. The downlink channel estimation is performed in each time slot. The uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0061] The model matching error is composed of channel noise and channel estimation error.

[0062] Assume that in a massive MIMO cell communication system, the base station is equipped with N antennas and each mobile user has only a single antenna. For uplink-assisted downlink channel estimation in FDD systems, the base station and the user terminal simultaneously send training pilots and transmit them through a narrowband block flat fading channel. Considering that the overcomplete dictionary can capture the characteristics of the data set in a more flexible way, it is assumed that the dictionary learned from the downlink channel measurement is Satisfy h d =D d β d ,in is the sparse coding vector, ||β d ||0=s<<N, the received signal of the downlink is expressed as:

[0063]

[0064] Given y d , d , P and D d , if we find β d , then the downlink channel estimation Similarly, the received signal of the uplink is expressed as:

[0065]

[0066] Uplink channel estimation in, Respectively represent the training pilots sent in the uplink and downlink, T u , T d represents the uplink and downlink training period, ρ u , d Indicates the received power of uplink and downlink, represents the received signal at the base station, where and Indicates the downlink single-antenna user receiving signal, Represent the receiving noise of uplink and downlink respectively.

[0067] In the FDD communication system, considering the angle reciprocity between the uplink and downlink channels, the same angle of arrival and departure (AoA / AoD) is converted into an uplink sparse coding vector β uand the downlink sparse coding vector β d The same position of the non-zero elements in u )=supp(β d ). Therefore, the common support information of uplink and downlink channels can be used to achieve better downlink channel estimation performance without increasing training overhead. Due to the movement of users and the presence of dynamic scatterers such as crowds and arbitrary vehicles, the channel environment is constantly changing, so the offline trained dictionary will become inaccurate and may cause severe performance degradation. Compared with traditional sparse channel estimation, the dictionary adapted to channel statistics can be updated on a long-term time scale, which can capture the dynamic spatial characteristics of the channel without additional overhead. Therefore, considering Figure 2 The dual time scale transmission protocol is shown.

[0068] like Figure 2 As shown in the figure, the coherence time of the channel statistics time scale is decomposed into several frames, each of which contains B time slots. It is assumed that the relevant parameters of the channel statistics remain unchanged during the entire coherence time of the channel statistics. Then, channel estimation is performed in each time slot. The short-term sparse channel vector acquired through each channel coherence time is used to update the uplink and downlink dictionaries in each frame. This transmission mechanism not only learns an effective sparse channel representation through the angular reciprocity between the uplink and downlink channels, but also ensures that the updated dictionary can capture dynamic channel characteristics in real time.

[0069] First, the uplink and downlink received signals in the same time slot are collected. Each mobile user feeds back the received downlink received signal to the base station. The base station uses the uplink and downlink received signals collected in the same time slot to perform joint uplink and downlink dictionary learning and channel estimation. The uplink and downlink received signals include the downlink received signal fed back by the mobile user and the uplink received signal collected by the base station in the same time slot as the downlink received signal. In the instantaneous channel estimation process, each pair of received signals {y u ,y d} Collect at the same user's location to enforce the support constraint supp(β u )=supp(β d ), that is, the downlink channel estimation based on compressed sensing is performed in each time slot, and the uplink and downlink dictionaries are updated on a long-term time scale, wherein the long-term time scale includes multiple time slots.

[0070] Then the non-convex dual-time-scale stochastic optimization problem can be expressed as

[0071]

[0072]

[0073] The above formula (3) is the first objective function of the non-convex dual-time-scale stochastic optimization problem, is the constraint function corresponding to the first objective function.

[0074] The first objective function is based on the expected form of the logarithm, involving a sparse smooth approximation of the mixed l2 / l1 norm, and μ>0 is a smooth parameter that controls the approximation error. Let in, is the group core, Represents a standard basis in which the element at the mth position is 1 and the elements at other positions are 0. is the set of short-term variables corresponding to all possible states of the system (i.e., observed values ​​y), where In the probability space The random state defined in , Ω is the sample space is the σ-algebra generated by the subsets of Ω, is The probability measure defined above.

[0075] When the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to the preset threshold, the constraint condition of the first objective function satisfies that the model matching error is less than or equal to the model matching error. Satisfy the conditions And the range of assumptions Let the combined pilot matrix Give a combined sparse representation within the model matching error ε From this, we can extract β d (y d ) and β u (y u ), and finally the sparse coding vector corresponding to the uplink and downlink channels is obtained. Sparse coding vector β d (y d ) is used for downlink channel estimation.

[0076] The joint uplink and downlink dictionary learning and channel estimation method provided by the present invention utilizes the time scale separation of dictionary and sparse channel representation parameters, proposes a non-convex dual-time-scale stochastic optimization problem for joint uplink and downlink dictionary learning and channel estimation, and realizes capturing the channel characteristic changes in the communication scenario through online dictionary update, thereby solving the defects in the prior art that it is difficult to perform downlink channel estimation for dynamic scenarios and the accuracy of downlink channel estimation in dynamic scenarios is low, thereby improving the accuracy of downlink channel estimation in dynamic scenarios and reducing the pilot overhead of the system.

[0077] In one embodiment, step 20, based on the first objective function corresponding to the non-convex dual time scale stochastic optimization problem, performs joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determines the sparse coding vector, and updates the uplink and downlink dictionary, please refer to Figure 3 ,include:

[0078] Step 21, dividing the non-convex dual-time-scale stochastic optimization problem into a series of short-term channel estimation subproblems and a long-term dictionary learning subproblem according to the relaxed dual-time-scale primal-dual decomposition theory, and correspondingly, the first objective function includes a second objective function corresponding to the short-term channel estimation subproblem and a third objective function corresponding to the long-term dictionary learning subproblem;

[0079] Step 22, determining a sparse coding vector based on the second objective function;

[0080] Step 23: Based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period.

[0081] The above non-convex dual-time-scale stochastic optimization problem There are the following challenges. First, the sparse smooth approximation of the group norm in the first objective function is non-convex. Second, through the error constraints corresponding to the first objective function, short-term variables (the sparse channel vector is a short-term variable) and long-term variables (the uplink and downlink dictionaries are long-term variables) are complexly coupled in a complex way. For this non-convex dual-time scale stochastic optimization problem, there is usually no effective method to solve it. To this end, this embodiment proposes a solution to the problem An effective algorithm is proposed, namely the dual-time-scale joint uplink / downlink dictionary learning and channel estimation (TTS-JDLCE) algorithm.

[0082] According to a new relaxed dual-time-scale primal-dual decomposition theory, the long-term variables (the combined dictionary D of the joint uplink and downlink) and the short-term variables (the sparse channel vector β) are separated from the problem The non-convex dual-time-scale stochastic optimization problem is decoupled from the original problem, thus dividing it into a long-term dictionary learning subproblem and a series of short-term channel estimation subproblems. Then, the sparse coding vector is determined and the uplink and downlink dictionaries are updated through the second objective function corresponding to the short-term channel estimation subproblem and the third objective function corresponding to the long-term dictionary learning subproblem, thereby effectively solving the original problem. Specifically, through the second objective function corresponding to the short-term channel estimation subproblem, a stable sparse coding vector is iteratively determined; and through the third objective function corresponding to the long-term dictionary learning subproblem, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period. Finally, based on the sparse coding vector obtained by solving the short-term channel estimation subproblem and the uplink and downlink dictionaries for the current time period, downlink channel estimation is performed to obtain the downlink channel estimation value for each time slot.

[0083] In this embodiment, a method is proposed to effectively solve the non-convex dual-time scale stochastic optimization problem, which includes the long-term variable (combined dictionary D of the joint uplink and downlink) and the short-term variable (sparse channel vector β) from the non-convex dual-time scale stochastic optimization problem. It decouples the problem from the original code and effectively solves the dictionary learning and channel estimation under the non-convex dual-time-scale stochastic optimization problem, obtains the optimal sparse coding vector, and further improves the accuracy of channel estimation.

[0084] In one embodiment, step 22, based on the second objective function, determines a sparse coding vector, see Figure 4 ,include:

[0085] Step 221, determining a non-convex part and a convex part of the second objective function;

[0086] Step 222, determining a proxy function of the non-convex part based on a first-order Taylor expansion of the non-convex part of the second objective function;

[0087] Step 223, determining a target proxy function corresponding to the second objective function based on the convex part of the second objective function and the proxy function of the non-convex part;

[0088] Step 224: Based on the uplink and downlink dictionaries and the target proxy function, perform J iterations to obtain a stable solution to the short-term channel estimation subproblem, and obtain the sparse coding vector.

[0089] When the long-term variables, i.e., the uplink and downlink dictionaries D and the Lagrange multiplier λ ≥ 0 are fixed, the second objective function corresponding to the short-term channel estimation subproblem is It is expressed as follows:

[0090]

[0091] The above short-term channel estimation subproblem is an unconstrained non-convex optimization problem, which means that under given long-term variables (uplink and downlink dictionaries D and Lagrange multipliers λ), the short-term sparse channel vector (also called sparse coding vector) is optimized with the goal of minimizing the sparsity of the joint sparse channel vector and the channel model error.

[0092] Define the short-term sparse channel vector β J(D,λ,y) represents The stationary solution of J (D,λ,y) satisfies The Karush-Kuhn-Tucker (KKT) condition (with a certain tolerable error) is:

[0093]

[0094] (5) where e J (D,λ) represents the error when the short-term sub-algorithm is run for a finite number of J iterations. When J→∞, J iterations are performed to obtain a stable solution to the short-term channel estimation sub-problem. The obtained β J Converge to A stationary point.

[0095] However, since the short-term channel estimation subproblem is an unconstrained non-convex optimization problem, the embodiment of the present application constructs a strong convex function, that is, the following formula (6) is used as The proxy function is used to obtain the optimal solution based on the Majorization-Minimization (MM) method. In the tth external iteration, the long-term variable (dictionary D t , Lagrange multiplier λ t ), in the (j-1)th internal iteration, the short-term variable (short-term sparse channel vector ), as the intermediate sparse coding vector.

[0096] First, for the non-convex part of the second objective function Derivation of the non-convex part of formula (4) The first-order Taylor expansion of As a proxy function for the non-convex part. Since omitting the constant term does not affect the optimality of the following optimization problem, in the tth external iteration, The proxy function without the constant term is used as the target proxy function. The target proxy function can be written as:

[0097]

[0098] Therefore, the j-th inner iteration of the short-term channel estimation subproblem can be expressed as:

[0099]

[0100] Let W = diag([w,w]) represent a diagonal weighted matrix, In order to express formula (7) more vividly, let f1(λ t ,GD t ,W)=(λt (GD t ) H GD t +WW) -1 , Through the short-term channel estimation sub-algorithm, based on the uplink and downlink dictionaries and the target proxy function, J iterations are performed to obtain a stable solution to the short-term channel estimation sub-problem, and the sparse coding vector is obtained. The network structure diagram of the short-term channel estimation sub-algorithm is shown in Figure 5 shown.

[0101] Specifically, for the initial outer iteration, let the initial long-term Lagrange multiplier λ be 0 =1, and the initial long-term dictionary D 0 By D d and D u Both are overcomplete DFT matrices. Figure 5 The optimization process of the sparse channel vector for the bth time slot at the tth external iteration is shown. The network structure diagram consists of a J-layer network, representing the J iterations of the short-term channel estimation sub-algorithm. The input of the network structure is the received signal of the bth time slot at the tth external iteration. and the initial point The initial point is set to a vector with all elements of the same size and applies to all external iterations. For the different layers of the network, the input of each layer is the output of the previous layer and the received signal and Represents the output of the j-th layer network. Long-term variable D t , t And receive signal It remains unchanged for different network layers. After J layers of network, the stable solution of the short-term channel estimation subproblem in the bth time slot can be obtained at the tth external iteration, that is, the sparse channel vector

[0102] These stable solutions have two functions: one is to estimate the downlink channel based on compressed sensing, and the other is to obtain the information about D based on these stable solutions. t and λ y The gradient information is used to construct the long-term proxy problem.

[0103] In this embodiment, since the short-term channel estimation subproblem is non-convex, it is difficult to solve the second objective function corresponding to the short-term channel estimation subproblem. In order to obtain the optimal solution for the short-term channel estimation subproblem, the MM method is adopted to replace the non-convex part of the second objective function with the proxy function of the non-convex part, so as to transform the second objective function corresponding to the short-term channel estimation subproblem into an easy-to-solve convex function, namely, the target proxy function, and then solve the short-term channel estimation subproblem, which improves the time of solving the short-term channel estimation subproblem and improves the computational efficiency of channel estimation.

[0104] In one embodiment, step 23, based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period. Figure 6 ,include:

[0105] Step 231, determining a convex proxy function corresponding to the third objective function;

[0106] Step 232, based on the convex proxy function, updating the uplink and downlink dictionaries to determine the uplink and downlink dictionaries for the next time period;

[0107] Wherein, the convex proxy function includes a target proxy function corresponding to the third objective function and a constraint proxy function corresponding to the constraint function of the third objective function;

[0108] The target proxy function includes an approximation of the third target function, a first-order Taylor expansion term of the third target function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the third target function with respect to Lagrange multipliers, and a quadratic term of the third target function with respect to the uplink and downlink dictionaries and the Lagrange multipliers;

[0109] The constraint proxy function includes an approximation of the constraint function, a first-order Taylor expansion term of the constraint function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the constraint function with respect to Lagrange multipliers, and a quadratic term of the constraint function with respect to the uplink and downlink dictionaries and the Lagrange multipliers.

[0110] The long-term dictionary learning subproblem corresponds to the third objective function As shown in formula (8):

[0111]

[0112]

[0113] in, is the constraint function corresponding to the third objective function.

[0114] The long-term dictionary learning subproblem is expressed in terms of obtaining β J(D, λ, y) and then optimize the long-term variables (D and λ) so that the sparsest sparse channel representation can be achieved by designing the long-term variables (D and λ) under the constraint of the channel modeling error, thereby obtaining the most suitable dictionary matrix D. From formula (8), it can be observed that the long-term dictionary learning subproblem involves the expectation of random states y, and it is difficult to obtain a closed-form solution. Select an effective optimization scheme and solve the long-term dictionary learning sub-problem through the long-term dictionary learning sub-algorithm The specific long-term dictionary learning sub-algorithm is described as follows.

[0115] First, determine the convex proxy function corresponding to the third objective function. At the tth external iteration, the third objective function corresponding to the long-term dictionary learning subproblem can be constructed using the Constrained Stochastic Successive Convex Approximation (CSSCA) framework. Recursive convex approximation of .

[0116] According to the preset proxy function Construct the convex proxy function corresponding to the third objective function, Represents the third objective function The convex proxy function includes a target proxy function corresponding to the third objective function and a constraint proxy function corresponding to the constraint function of the third objective function. is the target proxy function corresponding to the third objective function, that is, the proxy of the objective function, is the constraint proxy function corresponding to the constraint function of the third objective function, that is, the proxy of the constraint function. Specifically, As shown in the following formula (9):

[0117]

[0118] in, The first item and Respectively and The two are iteratively updated according to the following formula:

[0119]

[0120]

[0121] Among them, the step size {ρ t ∈0,1} satisfies ρ t →0,∑ tρ t =∞ and∑ t (ρ t ) 2 <∞. Assume for Abbreviation of .

[0122] The second and third terms of are the first-order Taylor expansion terms of the long-term variables D and λ, where the relevant gradient information is used. and The approximate functions of these gradients are iteratively updated according to the following formula:

[0123]

[0124]

[0125]

[0126]

[0127]

[0128] in,

[0129] at last The last term of is a quadratic term about the long-term variables D and λ used as a proximal regularization, the purpose is to relax the convergence condition of the resulting algorithm or to increase the convergence speed. x >0 can be any constant to ensure Uniformly strongly convex with respect to D and λ.

[0130] The optimal solution to the convex proxy problem is obtained by solving the following convex proxy function (17) or (18):

[0131]

[0132]

[0133] This is the third objective function corresponding to the long-term subproblem A convex approximation of . When problem (17) is infeasible, the optimal solution is The following questions are derived:

[0134]

[0135]

[0136] Both of the above problems are convex problems with simple quadratic objectives and constraints, and can be solved efficiently with off-the-shelf solvers such as CVX.

[0137] In this embodiment, the long-term dictionary learning sub-problem Contains random status The closed-form solution of the expected operation of b=1,2,...,λ is difficult to solve. It is proposed to establish a corresponding convex proxy function. The established convex proxy function can be successfully solved by existing solvers such as CVX, which reduces the difficulty of solving the long-term dictionary learning subproblem and improves the computational efficiency of channel estimation.

[0138] In one embodiment, step 232, based on the convex proxy function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period. Figure 7 ,include:

[0139] Step 2321, obtaining a current iteration parameter group, wherein the iteration parameter group includes an uplink and downlink dictionary and a Lagrange multiplier;

[0140] Step 2322, iteratively updating the uplink and downlink dictionaries based on the iteration parameter group and the convex proxy function, and determining a set of optimal solutions corresponding to the convex proxy function;

[0141] Step 2323: Determine the uplink and downlink dictionary for the next time period based on the iterative parameter group and the optimal solution.

[0142] The embodiment of the present application proposes a simple method for generating a (feasible) descending direction to solve the uplink and downlink dictionary at the latest time, namely the conditional gradient method, to improve the accuracy and computational efficiency of channel estimation. Given the current iteration parameter (D t ,λ y ) and the optimal solution of a set of convex surrogate functions Then the next feasible iteration parameter (D t+1 ,λ t+1 ) is obtained by recursive updating according to formulas (19) and (20). In this way, we obtain the latest long-term dictionary and Lagrange multiplier (D) of the next channel statistical coherence time. t+1 ,λ t+1 ). In addition, it is necessary to choose a suitable step size γ t Ensure that the algorithm converges and satisfies γ t →0,∑ t γ t =∞,∑ t (γ t ) 2 <∞ and

[0143] Update D and λ according to the following formula:

[0144]

[0145]

[0146] In this embodiment, a simple method for solving the latest uplink and downlink dictionaries is proposed to improve the accuracy and calculation efficiency of channel estimation.

[0147] In one embodiment, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem according to the relaxed dual-time-scale primal-dual decomposition theory, including:

[0148] According to the relaxed dual-time-scale primal-dual decomposition theory, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem at the t-th channel statistical coherence time;

[0149] The coherence time of the channel statistical time scale includes several frames, and each frame includes multiple time slots.

[0150] like Figure 2 As shown in the figure, the coherence time of the channel statistics time scale is decomposed into several frames, each of which contains B time slots. It is assumed that the relevant parameters of the channel statistics remain unchanged during the entire coherence time of the channel statistics. Then, channel estimation is performed in each time slot. The short-term sparse channel vector acquired through each channel coherence time is used to update the uplink and downlink dictionaries in each frame. This transmission mechanism not only learns an effective sparse channel representation through the angular reciprocity between the uplink and downlink channels, but also ensures that the updated dictionary can capture dynamic channel characteristics in real time.

[0151] The joint uplink and downlink dictionary learning and channel estimation device provided by the present invention is described below. The joint uplink and downlink dictionary learning and channel estimation device described below and the joint uplink and downlink dictionary learning and channel estimation method described above can refer to each other.

[0152] The joint uplink and downlink dictionary learning and channel estimation device proposed in this application includes:

[0153] A signal collection module is used to collect uplink and downlink received signals in the same time slot;

[0154] A parameter determination module, configured to perform joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, determine a sparse coding vector, and update an uplink and downlink dictionary;

[0155] A downlink channel estimation module, used to perform downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period;

[0156] The first objective function is based on the expected form of logarithms, and is related to the uplink and downlink received signals, the uplink and downlink dictionaries, and the sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors. When the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error;

[0157] The uplink and downlink dictionaries are used to characterize changes in channel characteristics. The downlink channel estimation is performed in each time slot. The uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0158] Furthermore, the parameter determination module is also used for:

[0159] The non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation subproblems and a long-term dictionary learning subproblem according to the relaxed dual-time-scale primal-dual decomposition theory, and correspondingly, the first objective function includes a second objective function corresponding to the short-term channel estimation subproblem and a third objective function corresponding to the long-term dictionary learning subproblem;

[0160] Based on the second objective function, determining a sparse coding vector;

[0161] Based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period.

[0162] Furthermore, the parameter determination module is also used to:

[0163] determining a non-convex part and a convex part of the second objective function;

[0164] Determine a proxy function of the non-convex part based on a first-order Taylor expansion of the non-convex part of the second objective function;

[0165] Determine a target proxy function corresponding to the second objective function based on the convex part of the second objective function and the proxy function of the non-convex part;

[0166] Based on the uplink and downlink dictionaries and the target proxy function, J iterations are performed to obtain a stable solution to the short-term channel estimation subproblem, and the sparse coding vector is obtained.

[0167] Furthermore, the parameter determination module is also used for:

[0168] Determine a convex proxy function corresponding to the third objective function;

[0169] Based on the convex proxy function, updating the uplink and downlink dictionaries to determine the uplink and downlink dictionaries for the next time period;

[0170] Wherein, the convex proxy function includes a target proxy function corresponding to the third objective function and a constraint proxy function corresponding to the constraint function of the third objective function;

[0171] The target proxy function includes an approximation of the third target function, a first-order Taylor expansion term of the third target function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the third target function with respect to Lagrange multipliers, and a quadratic term of the third target function with respect to the uplink and downlink dictionaries and the Lagrange multipliers;

[0172] The constraint proxy function includes an approximation of the constraint function, a first-order Taylor expansion term of the constraint function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the constraint function with respect to Lagrange multipliers, and a quadratic term of the constraint function with respect to the uplink and downlink dictionaries and the Lagrange multipliers.

[0173] Furthermore, the parameter determination module is also used for:

[0174] Obtaining a current iteration parameter group, wherein the iteration parameter group includes an uplink and downlink dictionary and a Lagrange multiplier;

[0175] Based on the iteration parameter group and the convex proxy function, iteratively update the uplink and downlink dictionaries to determine a set of optimal solutions corresponding to the convex proxy function;

[0176] Based on the iterative parameter group and the optimal solution, an uplink and downlink dictionary for the next time period is determined.

[0177] Furthermore, the parameter determination module is also used for:

[0178] The non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem at the t-th channel statistical coherence time;

[0179] The coherence time of the channel statistical time scale includes several frames, and each frame includes multiple time slots. Figure 8 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 8As shown, the electronic device may include: a processor (processor) 810, a communication interface (Communications Interface) 820, a memory (memory) 830 and a communication bus 840, wherein the processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call the logic instructions in the memory 830 to execute a method based on joint uplink and downlink dictionary learning and channel estimation, which includes: collecting uplink and downlink received signals in the same time slot; based on the first objective function corresponding to the non-convex dual time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating the uplink and downlink dictionary; performing downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period; wherein the first objective function is based on the expected form of logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionary and the sparse coding vector, and its goal is to minimize the sparsity of the sparse coding vector, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error; the uplink and downlink dictionary is used to characterize changes in channel characteristics, the downlink channel estimation is performed in each time slot, and the uplink and downlink dictionary is updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0180] In addition, the logic instructions in the above-mentioned memory 830 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0181] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the joint uplink and downlink dictionary learning and channel estimation method provided by the above methods, which method includes: collecting uplink and downlink received signals in the same time slot; based on the first objective function corresponding to the non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating the uplink and downlink dictionary; based on the sparse coding vector and the uplink and downlink of the current time period row dictionary, downlink channel estimation; wherein, the first objective function is based on the expected form of logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionaries and the sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies the model matching error; the uplink and downlink dictionaries are used to characterize changes in channel characteristics, the downlink channel estimation is performed in each time slot, and the uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0182] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which is implemented when the processor executes the joint uplink and downlink dictionary learning and channel estimation method provided by the above methods, the method comprising: collecting uplink and downlink received signals in the same time slot; based on the first objective function corresponding to the non-convex dual time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating the uplink and downlink dictionary; performing downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period; wherein the first objective function is based on the expected form of logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionary and the sparse coding vector, and its goal is to minimize the sparsity of the sparse coding vector, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies the model matching error less than or equal to the model matching error; the uplink and downlink dictionary is used to characterize changes in channel characteristics, the downlink channel estimation is performed in each time slot, and the uplink and downlink dictionary is updated on a long-term time scale, and the long-term time scale includes multiple time slots.

[0183] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0184] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A joint uplink and downlink dictionary learning and channel estimation method, characterized in that: Applied to large-scale antenna systems, including: Collect uplink and downlink receiving signals in the same time slot; Based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating an uplink and downlink dictionary; Perform downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionaries of the current time period; Wherein, the first objective function is based on the expected form of logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionaries and the sparse coding vectors, and its goal is to minimize the sparsity of the sparse coding vectors, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error; The uplink and downlink dictionaries are used to characterize changes in channel characteristics, the downlink channel estimation is performed in each time slot, and the uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots; The expression of the first objective function is: ； ； in, is the constraint function corresponding to the first objective function; the first objective function involves mixed The sparse smooth approximation of the norm, is a smoothing parameter that controls the approximation error; represents the group core; is the set of short-term variables corresponding to all possible states of the massive antenna system, where In the probability space The random state defined in is the sample space, yes Subset Generation Algebra, is The probability measure defined above; is the combined dictionary matrix composed of the uplink and downlink dictionaries; express The conditions for satisfaction; is the model matching error The combined sparse representation given in ; is the combined pilot matrix.

2. The joint uplink and downlink dictionary learning and channel estimation method according to claim 1, characterized in that: Based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, performing joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals, determining a sparse coding vector, and updating an uplink and downlink dictionary, including: The non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation subproblems and a long-term dictionary learning subproblem according to the relaxed dual-time-scale primal-dual decomposition theory, and correspondingly, the first objective function includes a second objective function corresponding to the short-term channel estimation subproblem and a third objective function corresponding to the long-term dictionary learning subproblem; Based on the second objective function, determining a sparse coding vector; Based on the third objective function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period.

3. The joint uplink and downlink dictionary learning and channel estimation method according to claim 2, characterized in that: Determining a sparse coding vector based on the second objective function includes: determining a non-convex part and a convex part of the second objective function; Determine a proxy function of the non-convex part based on a first-order Taylor expansion of the non-convex part of the second objective function; Determine a target proxy function corresponding to the second objective function based on the convex part of the second objective function and the proxy function of the non-convex part; Based on the uplink and downlink dictionaries and the target proxy function, J iterations are performed to obtain a stable solution to the short-term channel estimation subproblem, and the sparse coding vector is obtained.

4. The joint uplink and downlink dictionary learning and channel estimation method according to claim 2, characterized in that: Based on the third objective function, updating the uplink and downlink dictionaries to determine the uplink and downlink dictionaries for the next time period includes: Determine a convex proxy function corresponding to the third objective function; Based on the convex proxy function, updating the uplink and downlink dictionaries to determine the uplink and downlink dictionaries for the next time period; Wherein, the convex proxy function includes a target proxy function corresponding to the third objective function and a constraint proxy function corresponding to the constraint function of the third objective function; The target proxy function includes an approximation of the third target function, a first-order Taylor expansion term of the third target function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the third target function with respect to Lagrange multipliers, and a quadratic term of the third target function with respect to the uplink and downlink dictionaries and the Lagrange multipliers; The constraint proxy function includes an approximation of the constraint function, a first-order Taylor expansion term of the constraint function with respect to the uplink and downlink dictionaries, a first-order Taylor expansion term of the constraint function with respect to Lagrange multipliers, and a quadratic term of the constraint function with respect to the uplink and downlink dictionaries and the Lagrange multipliers.

5. The method for joint uplink and downlink dictionary learning and channel estimation according to claim 4, characterized in that: Based on the convex proxy function, the uplink and downlink dictionaries are updated to determine the uplink and downlink dictionaries for the next time period, including: Obtaining a current iteration parameter group, wherein the iteration parameter group includes an uplink and downlink dictionary and a Lagrange multiplier; Based on the iteration parameter group and the convex proxy function, iteratively update the uplink and downlink dictionaries to determine a set of optimal solutions corresponding to the convex proxy function; Based on the iterative parameter group and the optimal solution, an uplink and downlink dictionary for the next time period is determined.

6. The joint uplink and downlink dictionary learning and channel estimation method according to claim 2, characterized in that: According to the relaxed dual-time-scale primal-dual decomposition theory, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem, including: According to the relaxed dual-time-scale primal-dual decomposition theory, the non-convex dual-time-scale stochastic optimization problem is divided into a series of short-term channel estimation sub-problems and a long-term dictionary learning sub-problem at the t-th channel statistical coherence time; The coherence time of the channel statistical time scale includes several frames, and each frame includes multiple time slots.

7. A joint uplink and downlink dictionary learning and channel estimation device, characterized in that: include: A signal collection module is used to collect uplink and downlink received signals in the same time slot; A parameter determination module, configured to perform joint uplink and downlink dictionary learning and channel estimation on the uplink and downlink received signals based on a first objective function corresponding to a non-convex dual-time-scale stochastic optimization problem, determine a sparse coding vector, and update an uplink and downlink dictionary; A downlink channel estimation module, used to perform downlink channel estimation based on the sparse coding vector and the uplink and downlink dictionary of the current time period; Wherein, the first objective function is an expected form based on logarithms, the first objective function is related to the uplink and downlink received signals, the uplink and downlink dictionaries and the sparse coding vectors, and when the combined dictionary matrix composed of the uplink and downlink dictionaries satisfies that the bi-norm of each column vector is less than or equal to a preset threshold, the constraint condition of the first objective function satisfies that it is less than or equal to the model matching error; The uplink and downlink dictionaries are used to characterize changes in channel characteristics, the downlink channel estimation is performed in each time slot, and the uplink and downlink dictionaries are updated on a long-term time scale, and the long-term time scale includes multiple time slots; The expression of the first objective function is: ； ； in, is the constraint function corresponding to the first objective function; the first objective function involves mixed The sparse smooth approximation of the norm, is a smoothing parameter that controls the approximation error; represents the group core; is the set of short-term variables corresponding to all possible states of the massive antenna system, where In the probability space The random state defined in is the sample space, yes Subset Generation Algebra, is The probability measure defined above; is the combined dictionary matrix composed of the uplink and downlink dictionaries; express The conditions for satisfaction; is the model matching error The combined sparse representation given in ; is the combined pilot matrix.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the joint uplink and downlink dictionary learning and channel estimation method as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for joint uplink and downlink dictionary learning and channel estimation as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for joint uplink and downlink dictionary learning and channel estimation as claimed in any one of claims 1 to 6 is implemented.

Images