Joint PAPR reduction and INI suppression method based on deep expansion ADMM network

Abstract

The invention relates to a joint PAPR reduction and INI suppression method and device based on a deep expansion ADMM network, a multi-carrier communication waveform optimization method and device, computer equipment and a storage medium. Based on a constructed inter-parameter-set interference model and a guard interval injection technology, a multi-carrier communication waveform optimization method and device with inter-parameter-set interference minimization as an optimization target is constructed; taking the condition that the peak-to-average power ratio reaches a threshold value as a constraint condition, taking a mixed signal as an optimization problem model of an unknown quantity to be solved, substituting the obtained data sub-band input signal of the mixed parameter set multi-carrier system into the optimization problem model, and solving the optimization problem model by adopting a parallel alternating direction multiplier method to obtain a multi-carrier system data sub-band input signal. And taking the mixed signal as an output signal of a mixed parameter set multi-carrier system, and by adopting the method, the multi-carrier communication waveform with PAPR reduction and INI suppression compatibility can be generated.

Description

Technical Field

[0001] The present application relates to the technical field of wireless communication networks, and in particular to a joint PAPR reduction and INI suppression method based on a deeply expanded ADMM network. Background Art

[0002] The future mobile communication network will establish a unified connection architecture to create a world where everything is connected. However, faced with different usage scenarios and diverse service requirements, it is difficult for traditional mobile communication air interfaces to provide service access at the same time. The hybrid parameter set multi-carrier system supports flexible scheduling of heterogeneous resource blocks under a unified air interface by multiplexing different waveforms in the frequency domain, which can adapt to different usage scenarios and service requirements, and further expand the flexibility and adaptability of the mobile air interface at the waveform design level. However, the inherent defects of multi-carrier signals affect system design in new ways. On the one hand, there is a serious peak-to-average power ratio (PAPR) problem, and on the other hand, out-of-band leakage creates a new form of interference to other users, namely inter-parameter set interference (INI).

[0003] There is a close relationship between PAPR reduction technology and INI suppression technology. Reducing the PAPR of the signal changes the signal envelope, which may weaken the effect of INI suppression technology. Suppressing INI in multi-subband situations will affect the effectiveness of signal PAPR reduction technology, causing the signal peak power to increase again. Excessive PAPR worsens the out-of-band leakage problem after the signal passes through the PA, resulting in more serious INI. Therefore, PAPR reduction technology must be compatible with INI suppression technology.

[0004] However, most of the existing PAPR reduction technologies and INI suppression technologies have problems such as high computational complexity, bandwidth efficiency loss, and nonlinearity of the power amplifier. Summary of the invention

[0005] Based on this, it is necessary to provide a joint PAPR reduction and INI suppression method based on a deeply expanded ADMM network that can make PAPR reduction and INI suppression compatible with each other in order to address the above technical problems.

[0006] A multi-carrier communication waveform optimization method, the method being implemented in a mixed parameter set multi-carrier system, comprising:

[0007] Construct interference models between parameter sets;

[0008] Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved;

[0009] Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0010] In one embodiment, the parameter set interference model is expressed as:

[0011]

[0012] In the above formula, H1 represents the frequency response of the channel, C1 corresponds to the FFT matrix and the CP removal matrix, respectively. F2 represents a block diagonal matrix, which is the modulation matrix of the data subband input signal. X2 represents the frequency domain form of the data subband input signal. INI,1 It indicates that subband 1 is subject to inter-parameter set interference of subband 2, and the subscripts 1 and 2 represent the related parameters of the two adjacent subbands, subband 1 and subband 2, respectively.

[0013] In one embodiment, before solving the optimization problem model by using the parallelized alternating direction multiplier method, some variables are replaced so that the constraints are relaxed to convex constraints, and auxiliary variables are added to the optimization problem model so that there are equality constraints in the model, which are expressed as:

[0014]

[0015] G 1,i =0,G 2,i =0,i∈G c

[0016] In the above formula, G c represents the complementary set of G, β1 and β2 are penalty factors corresponding to the injected signal power, The auxiliary variable introduced is the final transmitted signal, z represents the mixed signal to be solved, σ represents the input signal power of the corresponding subband, F g1 F g2 They represent the modulation matrix of the guard interval injection signal of the corresponding parameter set, G1 and G2 represent the frequency domain form of the guard interval injection signal of the corresponding parameter set, and L sys Represents the number of time domain sampling points of the mixed signal.

[0017] In one embodiment, when the parallelized alternating direction multiplier method is used to solve the optimization problem model, the optimization problem model is converted into three sub-problems which are solved alternately in each iteration process. The three sub-problems are respectively expressed as:

[0018]

[0019] In the above formula, G represents the injection signal on the guard interval, represents the variable associated with the mixed signal to be solved, and y represents the parameter associated with the Lagrange multiplier.

[0020] In one embodiment, when the parallelized alternating direction multiplier method is used to solve the optimization problem model, the parallelized alternating direction multiplier method is expanded into a deep expansion network using deep expansion network technology, and the optimization problem is solved using the deep expansion network.

[0021] In one embodiment, the deep unfolding network includes a plurality of iterative layers connected in sequence and having the same structure, and each iterative layer corresponds to an iterative step of the parallelized alternating direction multiplication method;

[0022] The iteration layer is divided into a guard band step, a mixed signal step and a Lagrange multiplier update step according to the solution part of the iteration step, and in each of the steps, the model parameters are updated as optimizable learning parameters.

[0023] In one embodiment, when training the deep expansion network, staged training is adopted, including a first training stage and a second training stage;

[0024] During the first training phase, the deep unfolded network is trained using a labeled loss function to obtain a trained deep unfolded network, wherein the optimization problem model is solved according to a data subband input signal using the parallelized alternating direction multiplier method, and the obtained output signal is used as a label;

[0025] When using the trained deep expansion network to solve the mixed signal according to the data subband input signal, the trained deep expansion network is subjected to second-stage training at a preset interval, and an indicator-type loss function without label value is used for online training during the second-stage training.

[0026] The present application also provides a joint PAPR reduction and INI suppression device based on a deep unfolding ADMM network, the device comprising:

[0027] A parameter set interference model building module, used to build a parameter set interference model;

[0028] An optimization problem model building module is used to build an optimization problem model with minimization of inter-parameter interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved based on the inter-parameter interference model and the guard interval injection technology;

[0029] The mixed signal obtaining module is used to obtain the data subband input signal of the mixed parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, use the parallelized alternating direction multiplier method to solve the optimization problem model, obtain the mixed signal after waveform optimization, and use the mixed signal as the transmission signal of the mixed parameter set multi-carrier system.

[0030] A computer device comprises a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the following steps are implemented:

[0031] Construct interference models between parameter sets;

[0032] Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved;

[0033] Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0034] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the following steps:

[0035] Construct interference models between parameter sets;

[0036] Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved;

[0037] Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0038] The above-mentioned joint PAPR reduction and INI suppression method based on the deeply expanded ADMM network, through the constructed inter-parameter set interference model and the guard interval injection technology, constructs an optimization problem model with minimization of inter-parameter set interference as the optimization goal, takes the peak-to-average power ratio reaching the threshold value as the constraint condition, and takes the mixed signal as the unknown quantity to be solved. The data subband input signal of the mixed parameter set multi-carrier system is brought into the optimization problem model, and the optimization problem model is solved by the parallelized alternating direction multiplier method to obtain the waveform-optimized mixed signal, and the mixed signal is used as the output signal of the mixed parameter set multi-carrier system. This method can generate a multi-carrier communication waveform compatible with PAPR reduction and INI suppression. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A flowchart of a joint PAPR reduction and INI suppression method based on a deeply expanded ADMM network in one embodiment;

[0040] Figure 2 A schematic diagram of injecting a guard interval into a subcarrier between subbands of a hybrid parameter set system in one embodiment;

[0041] Figure 3 A schematic diagram of the structure of an iterative layer in a deep expansion network in one embodiment;

[0042] Figure 4 A schematic diagram of a flow chart of mixed signal generation based on a deep unfolding network in one embodiment;

[0043] Figure 5 A schematic diagram of a theoretical analysis process of INI in one embodiment;

[0044] Figure 6 It is a structural block diagram of a joint PAPR reduction and INI suppression device based on a deep unfolding ADMM network in one embodiment;

[0045] Figure 7 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0047] like Figure 1 As shown, the present application provides a joint PAPR reduction and INI suppression method based on a deep unfolding ADMM network, which is implemented in a mixed parameter set multi-carrier system and includes the following steps:

[0048] Step S100: construct an interference model between parameter sets.

[0049] Step S110, based on the inter-parameter set interference model and the guard interval injection technology, construct an optimization problem model with minimizing the inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and the mixed signal as the unknown quantity to be solved.

[0050] Step S120, obtaining a data subband input signal of the hybrid parameter set multi-carrier system, bringing the data subband input signal into the optimization problem model, solving the optimization problem model using a parallelized alternating direction multiplier method, obtaining a waveform-optimized hybrid signal, and using the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0051] In this embodiment, the Guard Band Injection (GBI) technology is adopted, and it is considered to modulate two frequency-domain superimposed non-orthogonal symbols on the guard band. The waveform parameters used by the two symbols are the same as the waveform parameters of the adjacent sub-bands, so that the two symbols can perform their respective functions to eliminate the interference between parameter sets caused by their respective adjacent sub-bands, and the mixed signal after frequency-domain superposition can suppress PAPR.

[0052] like Figure 2 As shown in Figure 1, the Guard Band Injection Signal (GBIS) is modulated on the guard band with a small number of subcarriers, which does not occupy precious data subcarrier resources. At the same time, the power of GBIS is minimized during modeling to save power. The frequency domain vector of GBIS is represented as Among them, N g is the number of subcarriers occupied by the protection band, and the time domain vector is expressed as Where J is the oversampling factor.

[0053] On the one hand, in order to avoid introducing signal distortion, the data subcarrier needs to be orthogonal to the subcarrier injected into the guard interval. On the other hand, GBIS needs to be able to suppress the INI generated by the superposition of different parameter sets. Therefore, in this embodiment, the frequency domain superposition of two non-orthogonal symbols is selected as GBIS, which can not only suppress the INI caused by the superposition of other subbands within the frequency band of the current subband, but also reduce the peak-to-average ratio of the mixed signal without causing distortion.

[0054] Furthermore, in a hybrid numerology multi-carrier system, in subband i, the channel impulse response of the user is considered to be constant within a symbol period. This constant state is expressed in the form of a vector as follows:

[0055] h i =[hi (0),h i (1),…,h i (L ch,i -1)] (1)

[0056] In formula (1), L ch In order to focus on the analysis of INI, it is assumed that the ISI (inter-symbol interference) introduced by the channel can be completely eliminated by the cyclic prefix (CP), while ignoring the interference of the channel Gaussian noise that the received signal may be subjected to. Therefore, the received signal of the user in subband i can be expressed as:

[0057]

[0058] In formula (2), F g G g Represents the injection signal of the guard band. i Represents a Toeplitz matrix, which is equivalent to the convolution operation of the channel, and its first column is composed of The first line consists of The first item on the right represents the desired sub-band signal, while the second item represents the INI generated by other sub-band signals on the current sub-band signal, and the third item is the signal injected in this chapter to suppress INI and reduce PAPR.

[0059] In step S100, in order to analyze the INI between sub-signals, in this embodiment, it is assumed that the SCS used by user 2 is greater than the SCS used by user 1 in the case of adjacent sub-bands. By analyzing the interference caused by adjacent sub-bands on different sub-carriers, the interference power on each sub-carrier can be derived. When the SCS used by different users meets the generalized synchronization assumption, that is, f2=2 μ f1. During the decoding process, user 1’s symbol is superimposed with user 2’s b s (b s =2) symbols, such as Figure 5 As shown in Figure 1, because the parameter sets used by different subbands are non-orthogonal, the interference between parameter sets accumulates. Therefore, user 1 is affected by the INI of user 2, that is, the interference model between parameter sets is expressed as:

[0060]

[0061] In formula (3), H1 represents the frequency response of the channel, C1 corresponds to the FFT matrix and the CP removal matrix, F2 represents a block diagonal matrix, which is the modulation matrix of the input signal, X2 represents the frequency domain form of the data subband input signal, and y INI,1It indicates that subband 1 is interfered by the inter-parameter set of subband 2, and the subscripts 1 and 2 respectively represent the related parameters of the two adjacent subbands, subband 1 and subband 2. Among them, since the channel is assumed to be ideal in the INI analysis, H1 is a unit matrix.

[0062] Furthermore, The element in the mth row and nth column It is expressed as:

[0063]

[0064] When adjacent subbands have the same parameter set, that is, μ = 0, 2 μ = 1, the subcarriers of each user are orthogonal to each other, that is, y INI,1 = 0, then the out-of-band leakage of the sub-band will not cause interference between parameter sets. When the guard interval reserved by the system is large enough and the frequency band occupied by user 1 and user 2 is far enough, the interference to each user can be ignored.

[0065] Similarly, the expression of INI received by the mth subcarrier of each symbol of user 2 can be derived as:

[0066]

[0067] Similarly, we can write, The element expression of the mth row and nth column in the matrix is:

[0068]

[0069] Formula (6) not only reflects the interference caused by different subcarrier spacing, but also the different positions of sub-signals in a receiving window will also affect y INI,2 have an impact.

[0070] In step S110, based on the parameter set interference model obtained in step S100, a parameter set interference minimization is constructed with the optimization goal of minimizing P INI,1 and P INI,2 The convex optimization model of the power weighted value of , and at the same time, taking the peak-to-average power ratio reaching the threshold value as a constraint condition, the optimization problem model with the mixed signal as the unknown quantity to be solved can be obtained, which is expressed as:

[0071]

[0072] In formula (7), G c is the complementary set of G, β1 and β2 are the penalty factors corresponding to the injected signal power, and the optimization problem model is the sum of two non-convex functions. And the constraint is linear.

[0073] Furthermore, in step S120, a parallelized ADMM algorithm is used to efficiently solve the convex optimization problem model, namely, formula (7).

[0074] Considering that the two non-orthogonal symbols in GIBS can suppress the interference between parameter sets caused by their respective waveform parameter subbands when they work alone, the PAPR of the mixed signal can be reduced after frequency domain superposition. This working mode is highly consistent with the "decomposition-cooperation" ADMM algorithm. Before using the parallelized alternating direction multiplier method to solve the optimization problem model, auxiliary variables are added to the optimization problem model so that there is an equality constraint in the model, which is expressed as:

[0075]

[0076] In formula (8), the separability between the objective function and the constraint variables, and The constraints between G1 and G2 are linear, which makes it easier to solve using the ADMM framework. The significant advantage of the ADMM algorithm is that it avoids solving the original complex optimization problem, but instead finds the saddle point of the augmented Lagrangian function.

[0077]

[0078] Formula (9) is the augmented Lagrangian function of the optimization problem model formula (8). In formula (9), ρ is the penalty factor of the augmented Lagrangian function, which must satisfy ρ>0.

[0079] In order to enable the designed algorithm to run efficiently on the distributed multi-baseband processing unit of the hybrid parameter set system, and at the same time make the data flow method suitable for another solution method, that is, the solution method based on the deep expansion network, the parallelized ADMM algorithm is also used to solve the optimization problem model. In each iterative calculation, the optimization problem model is converted into three sub-problems for alternating solutions. The three sub-problems are expressed as follows:

[0080]

[0081] In formula (10), G represents the injected signal, represents the variable associated with the mixed signal to be solved, and y represents the parameter associated with the Lagrange multiplier.

[0082] Specifically, when solving the three sub-problems in formula (10), since it is assumed that INI only affects adjacent sub-bands, in actual deployment, as long as the frequency domain bandwidth occupied by the user is not too narrow, it can be considered that the frequency distance is far enough and INI will not affect non-adjacent sub-bands, so i = 1, 2. The iterative process involves solving three optimization problems, i.e., sub-problems. Although the form is complex, each stage has an exact closed-form solution.

[0083] In the right and When updating, the first subproblem in formula (10) can be transformed into the following equivalent form:

[0084]

[0085] The closed-form solution of can be obtained by taking the partial derivative of formula (11) and setting it equal to zero:

[0086]

[0087] in,

[0088]

[0089] Similarly, we can get:

[0090]

[0091] in,

[0092]

[0093] In the right When updating, first create the auxiliary variable u l+1 , expressed as:

[0094]

[0095] Then the second sub-problem in formula (10) can be reformulated as:

[0096]

[0097] Obviously, the problem expressed by formula (17) is to solve the problem of the closest vector u l+1 of therefore, It can be defined as:

[0098]

[0099] In formula (18), This process can be regarded as the injection process of GBIS, that is, two non-orthogonal symbols are superimposed in the frequency domain to suppress the mixed signal. PAPR.

[0100] In this embodiment, a pseudo code is also provided for solving the optimization problem model by using the parallelized alternating direction multiplier method, which is represented by Algorithm 1:

[0101]

[0102] In this embodiment, in addition to the above-mentioned direct use of the parallelized alternating direction multiplier method to solve the optimization problem model, the parallelized alternating direction multiplier method is expanded into a deep expansion network (GBI-ADMMnet) using deep expansion network technology, and the optimization problem model is solved using the deep expansion network to improve the solution efficiency.

[0103] Optimizing the quadratic penalty coefficient ρ in the ADMM framework will improve the convergence effect of the algorithm and make the algorithm performance less dependent on the initial selection of ρ. In order to improve the flexibility of the algorithm and maximize the convergence speed of the algorithm, in this embodiment, not only is ρ of each iteration flexible, but also each step ρ in each iteration is flexible, so that it can better adapt to the characteristics of specific processing objects in different steps. In the iterative structure of the ADMM algorithm, the clipping rate γ is a parameter that needs to be set in advance, which indicates the degree of adjustment of the mixed signal in each iteration. However, the random characteristics of the mixed signal are not only reflected before the iterative input, but also after each iteration. The mixed signal is also random. For some iterative steps, the signal distortion is small, and increasing the adjustment degree can reduce the number of iterations. Therefore, γ is preferably able to determine the degree of adjustment of the current mixed signal according to the characteristics of the current mixed signal. Therefore, five learning parameters that can be optimized are set in the deep expansion network, including:

[0104] In this embodiment, ADMMnet, which has almost the same computational complexity as the ADMM algorithm, is used for optimization. Each iteration process of solving the optimization problem model using the parallelized alternating direction multiplier method (GBI-ADMM) can be divided into three steps: operating the guard band, operating the mixed signal, and updating the Lagrange multiplier. These three steps are respectively expanded into the corresponding guard band step, mixed signal step, and Lagrange multiplier update step in the deep expansion network, which constitutes the following: Figure 3 A single layer structure of the deeply expanded network is shown.

[0105] Specifically, the deep unfolded network constructed based on the GBI-ADMM algorithm includes multiple iterative layers that are connected in sequence and have the same structure. Each iterative layer corresponds to an iterative step of the parallelized alternating direction multiplier method. The iterative layer is divided into a guard band step, a mixed signal step and a Lagrange multiplier update step according to the solution part of the iterative step. In each step, the model parameters are updated as optimizable learning parameters.

[0106] Furthermore, the training parameters belongs to the t-layer Lagrange multiplier update step, Belongs to the t-layer mixed signal step, Belongs to the t-layer guard band step.

[0107] It should be noted that the number of iterations must be fixed when performing deep expansion of GBI-ADMM, because the number of layers in the deep expansion network is fixed.

[0108] In this embodiment, the forward propagation formula of the GBI-ADMMnet network is expressed as:

[0109]

[0110] In formula (19) to formula (22), is nonlinear, In the deep expansion network, it plays the role of the activation function, y represents the activation threshold, when |u t (n)|≤T, when|u t (n)|>T,

[0111] In this embodiment, in the above-mentioned deep expansion network, the system parameters and the original mixed signal calculated offline in advance are input into each layer of the network. This data connection method is similar to the "jump connection" in the residual network, which can alleviate the gradient vanishing problem to a certain extent and is conducive to the convergence of the deep expansion network.

[0112] In this embodiment, when training the above-mentioned deep expansion network, staged training is adopted, including a first training stage and a second training stage, wherein, during the first training stage, the deep expansion network is trained using a labeled loss function to obtain a trained deep expansion network, wherein the optimization problem model is solved according to the input signal using a parallelized alternating direction multiplier method, and the obtained output signal is used as a label. In the process of using the trained deep expansion network to solve the mixed signal according to the input signal, the trained deep expansion network is trained in the second stage according to a preset interval time, and an indicator-type loss function without a label value is used for online training during the second stage training.

[0113] That is to say, the first training stage is actually to train a usable deep unfolding network, which has the ability to obtain a mixed signal with optimized waveform according to the input signal, and the second training stage is actually an online training stage. After the trained deep unfolding network is put into practical application, it is trained online at regular intervals to maintain good performance.

[0114] Specifically, in the first training stage, the root mean square error (RMSE) between the network output signal and the expected output signal (label value) is used as the loss function, and its formula is expressed as:

[0115]

[0116] In formula (28), Represents the output value of the output layer T of the deep expansion network, and M is the training data set The number of n corresponds to the dimension of the training set, that is, the number of time domain sampling points, ||·|| F Represents the F-norm of the matrix.

[0117] Furthermore, in the first training stage, the GBI-ADMM algorithm in this article is used to solve multiple input signals to obtain corresponding mixed signals, and these input signals are used as training samples. At the same time, the corresponding solved mixed signals are used as labels to train the deep expansion network.

[0118] Specifically, in the second training stage, in order to weaken the dependence of the loss function on the label value, an indicator loss function is proposed to evaluate the PAPR and INI performance of the output signal, and used in conjunction with the RMSE loss function to provide an optimization direction for parameter updating. Among them, the indicator loss function is expressed as:

[0119]

[0120] In formula (29), κ INI and κ PAPR They respectively represent the weights of the two indicators INI and PAPR in the loss function. In this embodiment, their values ​​are preliminarily determined by entropy weight, and then appropriately adjusted according to the training output results and actual needs.

[0121] Furthermore, in formula (29), ||·|| ∞ There is no gradient for the infinite norm, so the sub-gradient is used instead, expressed as:

[0122]

[0123] therefore Where i is a vector The maximum index in , so (0,…1 i ,0) T yes A sub-gradient of .

[0124] Furthermore, from the calculation of the sub-gradient, we can see that during each iteration of training, the vector Only the value at the maximum value i is passed back as the error. In the process of reducing PAPR, it is effective to focus only on the maximum value in each iteration, but in order to improve efficiency, this chapter improves the sub-gradient here according to the algorithm principle.

[0125] When the number of back-propagated values ​​m is larger, the residual volatility is larger. This is because the algorithm principle is deviated, so the convergence cannot be guaranteed, but the CCDF curve becomes steeper. In order to reduce the number of algorithm iterations as much as possible while ensuring the normal convergence of the algorithm, preferably, m = 3, that is, the three largest values ​​are paid attention to during back-propagation.

[0126] In this embodiment, due to the complex structure of the deep expansion network, the loss function and With learning parameters It is difficult to find an explicit expression for the functional relationship between them, and it is very difficult to directly parse the gradient expression. The back propagation (BP) algorithm can be used to quickly parse and solve the gradient.

[0127] Furthermore, since the algorithm in this article involves the processing of complex signals, all gradient operations need to be performed in the complex domain. Therefore, the derivative of the complex matrix is ​​first defined:

[0128]

[0129] In formula (31), represents an arbitrary complex matrix, Re{·} and Im{·} represent the real part matrix and the imaginary part matrix, respectively.

[0130] The derivation process of formula (31), that is, the derivation process of gradient direction propagation, is as follows:

[0131] In the output layer (t=T):

[0132] Lagrange multiplier update step: The Lagrange multiplier step of the output layer T works at the T+1 layer according to the algorithm principle, so the loss function is the same as Y T No matter,

[0133]

[0134] Mixing step: Before deriving the gradient of this layer, the step function needs to be defined as follows:

[0135]

[0136] This step has an explicit relationship with the loss function, so the gradient can be directly calculated as:

[0137]

[0138] The previous article gives the sub-gradient of the mixed signal output by the network for the "sub-gradient loss function of multi-value return".

[0139]

[0140] Two loss functions for γ T The gradients are the same:

[0141]

[0142] In formula (36), sum(·) represents the sum of all elements in the matrix, and the symbol ⊙ represents the element-by-element multiplication of the matrix.

[0143] Guardband step: The two loss functions mentioned above are and The gradient of the steps is also different. The loss function The gradient of Delivered from:

[0144]

[0145] Loss Function The gradient of It is passed from the INI part, and is calculated explicitly from the INI part.

[0146]

[0147] For this step, the training parameters The gradient of can be further calculated by formula (50).

[0148] In the input layer and the middle layer (t = T-1:-1:1):

[0149] Lagrange multiplier update step: Step Y t The learning parameters are included From the perspective of back propagation, the corresponding gradient matrix input to this step is like Figure 3 As shown, the input gradient matrix is ​​used to represent the current layer gradient function for:

[0150]

[0151] In solving the current layer gradient matrix Afterwards, the derivative of the loss function with respect to the parameters learned at this step can be calculated:

[0152]

[0153] Further, still from the perspective of back propagation, the output steps directly connected to this step are The gradient matrix of the current layer can be used to calculate the gradient of the next step according to the chain rule. Then, the gradient matrix of the Lagrange multiplier update step of the tth layer is passed to the t-1 layer.

[0154] Mixing steps: Steps The learning parameters are included From the perspective of back propagation, the corresponding input gradient matrix of this step is like Figure 3 As shown, the input gradient matrix is ​​used to represent the current layer gradient function for:

[0155]

[0156] In formula (43), the relationship between the input gradient and the current layer gradient is expressed based on the chain rule, and the gradient of the loss function with respect to the two learning parameters of this step can be obtained:

[0157]

[0158] in,

[0159]

[0160] Using the gradient calculation results of formula (44) and formula (47), we can achieve the parameter The output gradient matrix of this step is At this point, the back-propagation process of the mixing step of the tth layer has been completed, passing the gradient matrix to the t-1 layer.

[0161] Guard Band Steps: Steps The learning parameters are respectively and From the perspective of back propagation, the input gradient matrix of this step is therefore, The relationship with the input gradient matrix is:

[0162]

[0163] The relationship with the input gradient matrix is:

[0164]

[0165] Solve the gradient matrix for the current step and Afterwards, the gradient of the loss function with respect to the two learning parameters of this step can be obtained:

[0166]

[0167] in, and They are defined by formula (51) and formula (52) respectively. At this point, the gradient derivation of the loss function for the five learning parameters of each layer based on the chain rule is completed, and the error direction propagation process is completed. The pseudo code of the training process of the GBI-ADMMnet deep expansion network is summarized in Algorithm 2.

[0168]

[0169] In this embodiment, a pseudo code for training a deep expansion model is also provided, which is represented by Algorithm 2:

[0170]

[0171] In this embodiment, when using the trained deep expansion model to solve the mixed signal, it includes: receiving each sub-band bit stream through the mixed parameter set multi-carrier system, mapping it through the constellation, generating a corresponding sub-band signal, and then inputting the generated sub-band signal into the deeply expanded GBI-ADMMnet network, and generating a mixed signal through a forward propagation process.

[0172] Specifically, before actual deployment, the learnable parameters of the GBI-ADMMnet network need to be trained according to the two-stage training method of the deeply expanded ADMM network described above. In actual deployment, each sub-band bitstream can be directly generated into a sub-band signal, and then the sub-band signal can be input into the GBI-ADMMnet network trained in the first stage to generate a mixed signal. Alternatively, the second-stage training method without label values ​​can be adopted to perform online training on the GBI-ADMMnet network to further optimize the network performance.

[0173] like Figure 4 As shown, it is a flowchart of generating a hybrid parameter set system generation signal based on the joint PAPR reduction and INI suppression method of the deeply expanded GBI-ADMMnet network.

[0174] In the above-mentioned joint PAPR reduction and INI suppression method based on deeply expanded ADMM network, in a multi-subband mixed parameter set system, the PAPR and INI of the transmitted signal are jointly reduced by adopting the guard interval injection technology, and two non-orthogonal symbols superimposed in the frequency domain are modulated on the guard band to eliminate the interference caused by each subband. The mixed signal after frequency domain superposition can also achieve peak-to-average ratio reduction, and an inter-parameter set interference model is obtained. Based on the inter-parameter set interference model, an optimization problem model with minimization of inter-parameter set interference as the optimization goal is constructed, and the peak-to-average power ratio reaching the threshold value is used as a constraint condition, and the mixed signal is used as the unknown quantity to be solved. The obtained mixed parameter set multi-carrier system input signal is brought into the optimization problem model, and the parallelized alternating direction multiplier method is used to solve the optimization problem model to obtain a mixed signal after waveform optimization. The decomposition-cooperative injection signal generation mode is combined with the alternating vector multiplier method (ADMM) to achieve efficient solution, and the optimality of the algorithm is proved by using Slater conditions and KKT conditions. Aiming at the problem of optimal configuration of model parameters of the alternating vector multiplier method, the iterative structure of ADMM is expanded into a layer-by-layer structure of a neural network to obtain the GBI-ADMMnet network. Each layer of ADMM contains 5 training parameters to maximize the flexibility of the network. A sub-gradient loss function with multi-value returns is designed so that network training no longer depends on label values.

[0175] It should be understood that although Figure 1 The steps in the flowchart are shown in sequence as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified in this document, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, Figure 1 At least part of the steps may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least part of the sub-steps or stages of other steps.

[0176] In one embodiment, Figure 6 As shown, a joint PAPR reduction and INI suppression device based on a deep unfolding ADMM network is provided, comprising: an inter-parameter set interference model construction module 200, an optimization problem model construction module 210 and a mixed signal acquisition module 220, wherein:

[0177] The parameter set interference model construction module 200 is used to construct the parameter set interference model;

[0178] An optimization problem model building module 210 is used to build an optimization problem model with minimization of inter-parameter interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and a mixed signal as an unknown quantity to be solved based on the inter-parameter interference model and the guard interval injection technology;

[0179] The mixed signal obtaining module 220 is used to obtain the data subband input signal of the mixed parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, use the parallelized alternating direction multiplier method to solve the optimization problem model, obtain the mixed signal after waveform optimization, and use the mixed signal as the transmission signal of the mixed parameter set multi-carrier system.

[0180] For the specific definition of the multi-carrier communication waveform optimization device, please refer to the definition of the multi-carrier communication waveform optimization method above, which will not be repeated here. Each module in the above-mentioned multi-carrier communication waveform optimization device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0181] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 7 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a multi-carrier communication waveform optimization method is implemented. The display screen of the computer device can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covered on the display screen, or a key, trackball or touchpad set on the computer device housing, or an external keyboard, touchpad or mouse, etc.

[0182] Those skilled in the art will understand that Figure 7 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0183] In one embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the following steps are implemented:

[0184] Construct interference models between parameter sets;

[0185] Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved;

[0186] Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0187] In one embodiment, a computer readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0188] Construct interference models between parameter sets;

[0189] Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved;

[0190] Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

[0191] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0192] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0193] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A multi-carrier communication waveform optimization method, characterized in that: The method is implemented in a mixed parameter set multi-carrier system, comprising: Construct interference models between parameter sets; Based on the inter-parameter set interference model and the guard interval injection technology, an optimization problem model is constructed with minimization of inter-parameter set interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved; Acquire a data subband input signal of the hybrid parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, solve the optimization problem model using a parallelized alternating direction multiplier method, obtain a waveform-optimized hybrid signal, and use the hybrid signal as a transmission signal of the hybrid parameter set multi-carrier system.

2. The multi-carrier communication waveform optimization method according to claim 1, characterized in that: The parameter set interference model is expressed as: In the above formula, H1 represents the frequency response of the channel, C1 corresponds to the FFT matrix and the CP removal matrix, respectively. F2 represents a block diagonal matrix, which is the modulation matrix of the data subband input signal. X2 represents the frequency domain form of the data subband input signal. INI,1 It indicates that subband 1 is subject to inter-parameter set interference of subband 2, and the subscripts 1 and 2 represent the related parameters of the two adjacent subbands, subband 1 and subband 2, respectively.

3. The multi-carrier communication waveform optimization method according to claim 2, characterized in that: Before solving the optimization problem model by using the parallelized alternating direction multiplier method, some variables are replaced so that the constraints are relaxed to convex constraints, and auxiliary variables are added to the optimization problem model so that there are equality constraints in the model, which are expressed as: G 1,i =0,G 2,i =0,i∈G c In the above formula, G c represents the complementary set of G, β1 and β2 are penalty factors corresponding to the injected signal power, The auxiliary variable introduced is the final transmitted signal, z represents the mixed signal to be solved, σ represents the input signal power of the corresponding subband, F g1 F g2 They represent the modulation matrix of the guard interval injection signal of the corresponding parameter set, G1 and G2 represent the frequency domain form of the guard interval injection signal of the corresponding parameter set, and L sys Represents the number of time domain sampling points of the mixed signal.

4. The multi-carrier communication waveform optimization method according to claim 3, characterized in that: When the parallelized alternating direction multiplier method is used to solve the optimization problem model, the optimization problem model is converted into three sub-problems which are solved alternately in each iteration process. The three sub-problems are respectively expressed as: In the above formula, G represents the injection signal on the guard interval, represents the variable associated with the mixed signal to be solved, and y represents the parameter associated with the Lagrange multiplier.

5. The multi-carrier communication waveform optimization method according to any one of claims 1 to 3, characterized in that: When the parallelized alternating direction multiplier method is used to solve the optimization problem model, the parallelized alternating direction multiplier method is expanded into a deep expansion network using deep expansion network technology, and the optimization problem is solved using the deep expansion network.

6. The multi-carrier communication waveform optimization method according to claim 5, characterized in that: The deep unfolding network includes a plurality of iterative layers connected in sequence and having the same structure, and each iterative layer corresponds to an iterative step of the parallelized alternating direction multiplication method; The iteration layer is divided into a guard band step, a mixed signal step and a Lagrange multiplier update step according to the solution part of the iteration step, and in each of the steps, the model parameters are updated as optimizable learning parameters.

7. The multi-carrier communication waveform optimization method according to claim 6, characterized in that: When training the deep expansion network, phased training is adopted, including a first training phase and a second training phase; During the first training phase, the deep unfolded network is trained using a labeled loss function to obtain a trained deep unfolded network, wherein the optimization problem model is solved according to a data subband input signal using the parallelized alternating direction multiplier method, and the obtained output signal is used as a label; When using the trained deep expansion network to solve the mixed signal according to the data subband input signal, the trained deep expansion network is subjected to second-stage training at a preset interval, and an indicator-type loss function without label value is used for online training during the second-stage training.

8. A joint PAPR reduction and INI suppression device based on a deep unfolded ADMM network, characterized in that: The device comprises: A parameter set interference model building module, used to build a parameter set interference model; An optimization problem model building module is used to build an optimization problem model with minimization of inter-parameter interference as the optimization goal, peak-to-average power ratio reaching a threshold value as a constraint condition, and mixed signals as unknown quantities to be solved based on the inter-parameter interference model and the guard interval injection technology; The mixed signal obtaining module is used to obtain the data subband input signal of the mixed parameter set multi-carrier system, bring the data subband input signal into the optimization problem model, use the parallelized alternating direction multiplier method to solve the optimization problem model, obtain the mixed signal after waveform optimization, and use the mixed signal as the transmission signal of the mixed parameter set multi-carrier system.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images