Joint optimization method and device for beam forming, power control and frequency selection

Abstract

The invention discloses a joint optimization method and device for beam forming, power control and frequency selection, and belongs to the field of wireless communication networks, and the method comprises the steps: constructing a resource management optimization model of all links; wherein the resource management optimization model comprises resource management optimization variables containing beam forming, power control and frequency selection joint optimization, a network rate objective function and constraint conditions; parallel iteration is carried out based on the resource management optimization model, and beam forming, power control and frequency selection of all optimized links are obtained; wherein the parallel iteration is realized through a parallel processor; the parallel iteration algorithm parameters are optimized through a neural network. Therefore, by implementing the present invention, the network performance and computational efficiency of a wireless communication network can be improved by optimizing beam forming, power control and frequency selection in parallel.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wireless communication networks, in particular to a joint optimization method and device of beamforming, power control and frequency selection. BACKGROUND

[0002] The joint optimization of beamforming, power control and frequency selection in wireless communication networks is a large-scale non-convex optimization problem, and existing optimization methods and deep neural network (DNN) methods cannot achieve satisfactory communication network performance and computing time trade-off. Specifically, the joint optimization method of beamforming, power control and frequency selection based on weighted minimum mean square error (WMMSE) can only be applied to single-carrier narrowband systems, and can only perform partial parallel update in each iteration, which takes a long time to calculate; the joint optimization method of beamforming, power control and frequency selection based on block maximization minimization (BMM) can only be applied to single-carrier narrowband systems, and can only perform serial update in each iteration, which takes a long time to calculate; the power control method based on fractional programming (FP) for a given beamforming can only be applied to single-carrier narrowband systems, and can only perform serial update in each iteration, which takes a long time to calculate; the power control method based on successive convex approximation (SCA) for a given beamforming can only be applied to single-carrier narrowband systems.

[0003] The above optimization methods cannot achieve complete parallel closed-form update in each iteration, and need multiple iterations to achieve better communication network performance, which takes a long time to calculate. The existing deep learning-based joint optimization method of beamforming, power control and frequency selection can only be applied to single-carrier narrowband systems, and uses a parallel neural network architecture and a computational approximation to achieve parallel fast calculation, uses all transmit powers to produce strong interference, and the performance of the wireless communication network is limited, and lacks universality and interpretability. The existing deep unfolding neural network can only be applied to single-carrier narrowband systems, and can improve the performance of the communication network, the generalization ability and the interpretability while reducing the calculation time, but due to the lack of complete parallel closed-form update structure and good initial point in the underlying algorithm, it cannot further reduce the calculation time and improve the performance of the communication network. SUMMARY

[0004] The present application provides a joint optimization method and device of beamforming, power control and frequency selection, which can optimize beamforming, power control and frequency selection in parallel to improve the network performance and calculation efficiency of wireless communication networks.

[0005] The embodiment of the present application provides a joint optimization method of beamforming, power control and frequency selection, comprising:

[0006] constructing a resource management optimization model of all transmitters and receivers; wherein the resource management optimization model comprises resource management optimization variables, a network rate objective function and constraint conditions involving joint optimization of beamforming, power control and frequency selection;

[0007] performing parallel iteration based on the resource management optimization model to obtain optimized beamforming, power control and frequency selection of all links; wherein the parallel iteration is realized by a parallel processor; and algorithm parameters of the parallel iteration are optimized by a neural network.

[0008] The embodiment of the application can provide a model basis for subsequent implementation of joint optimization of beamforming, power control and frequency selection for network rate maximization by constructing a resource management optimization model of all transmitters and receivers; and can efficiently solve resource management optimization variables by performing parallel iteration based on the resource management optimization model, the resource management optimization variables being capable of realizing network rate maximization. Compared with the prior art which cannot trade off network performance and computing efficiency in a wireless communication network, the application can improve network performance and computing efficiency of the wireless communication network by optimizing beamforming, power control and frequency selection in parallel.

[0009] Further, the constructing of the resource management optimization model of all transmitters and receivers comprises:

[0010] constructing a joint optimization problem of beamforming, power control and frequency selection of all transmitters and receivers to maximize network rate under power constraints;

[0011] constructing a resource management optimization model based on the joint optimization problem; wherein the resource management optimization model is specifically:

[0012]

[0013] wherein R(V,W;H) is network rate, including sum rate, weighted sum rate and worst rate; (V,W) is resource management optimization variable involving joint optimization of beamforming, power control and frequency selection; H is channel state information; for any and v i,f is a transmission beamforming vector corresponding to the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents power of the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents a beam direction of the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents a subcarrier used by the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents that the transmitter bs(i) does not transmit a signal to the receiver i on the subcarrier f (not using the subcarrier); for any and w i,f represents a receiving beamforming vector corresponding to the transmitter bs(i) transmitting a signal on the subcarrier f received by the receiver i; for any P b,max represents the maximum transmission power of the transmitter b; for any represents a set of receivers corresponding to the transmitter b; represents a set of all subcarriers; represents a set of all receivers; represents a set of all transmitters; and

[0014] The embodiment of the application can provide a model basis for subsequent joint optimization of beamforming, power control and frequency selection for maximizing network rate by constructing a resource management optimization model of all transmitters and receivers.

[0015] Further, the parallel iteration based on the resource management optimization model comprises:

[0016] At each iteration, a plurality of joint optimization approximate convex problem models are constructed based on the resource management optimization model, and parallel iteration is performed on all the approximate convex problem models until a convergence condition is met, and a resource management optimization variable of the current iteration is output; wherein, at each iteration, all the approximate convex problem models are updated according to the receiving beamforming vectors and the transmitting beamforming vectors of all the links of the last iteration; wherein, the approximate convex problem model comprises a first approximate convex problem model about all the transmitters and a second approximate convex problem model about all the receivers on all the subcarriers; any transmitter corresponds to a first approximate convex problem model, and any receiver on any subcarrier corresponds to a second approximate convex problem model.

[0017] The embodiment of the application can efficiently solve the resource management optimization variable by performing parallel iteration based on the resource management optimization model, and the resource management optimization variable can maximize the network rate.

[0018] Further, the plurality of joint optimization approximate convex problem models constructed based on the resource management optimization model comprise:

[0019] The plurality of joint optimization approximate convex problem models constructed based on the resource management optimization model comprise:

[0020] ​

[0021] wherein, and (V (t) ,W (t) ) is the value of the resource management optimization variable at the t-1th iteration; is a strong convex term coefficient, and is an algorithm parameter; g(V (t) ,W (t) ;H) is selected according to the type of R(V,W;H); is a gradient of g(V (t) ,W (t) ;H) with respect to v i,f .

[0022] Based on the resource management optimization model, a plurality of second approximate convex problem models of joint optimization are constructed; wherein, at the tth iteration, with respect to any receiver The second approximate convex problem model on any subcarrier is specifically as follows:

[0023]

[0024] wherein, is a strong convex term coefficient, and is an algorithm parameter; g(V (t) ,W (t) ;H) is selected according to the type of R(V,W;H), is a gradient of g(V (t) ,W (t) ;H) with respect to w i,f .

[0025] The embodiment of the application can provide a model basis for parallel iteration by constructing approximate convex problem models of all transmitters and receivers.

[0026] Further, the parallel iteration solving of all the approximate convex problem models comprises:

[0027] The closed-form optimal solution of all the first approximate convex problem models is calculated in parallel by analyzing KKT conditions; wherein, at the tth iteration, with respect to any transmitter The closed-form optimal solution of the first approximate convex problem model is specifically as follows:

[0028]

[0029] wherein,

[0030] The transmit beamforming vectors of all transmitters are updated in parallel; wherein, at the tth iteration, the transmit beamforming vector of any transmitter is specifically:

[0031]

[0032] wherein, γ (t) ∈(0, 1] is a step size, and is an algorithm parameter.

[0033] Embodiments of the present application can obtain the closed-form optimal solution of the joint optimization problem of all transmitters in parallel through analyzing the KKT condition.

[0034] Further, the parallel iteration for solving all the approximate convex problem models further comprises:

[0035] The closed-form optimal solution of all the second approximate convex problem models is calculated in parallel through analyzing the KKT condition; wherein, at the tth iteration, the closed-form optimal solution of the second approximate convex problem model of any receiver on any subcarrier is specifically:

[0036]

[0037] The receive beamforming vectors of all receivers on all subcarriers are updated in parallel; wherein, at the tth iteration, the receive beamforming vector of any receiver on any subcarrier is specifically:

[0038]

[0039] wherein, γ (t) ∈(0, 1] is a step size, and is an algorithm parameter.

[0040] Embodiments of the present application can obtain the closed-form optimal solution of the joint optimization problem of all receivers in parallel through analyzing the KKT condition.

[0041] Further, the algorithm parameter of the parallel iteration is optimized through a neural network, comprising:

[0042] The parallel iteration process is mapped to a neural network layer to construct a neural network; wherein, the first layer of the neural network layer adjusts the initial resource management optimization variable by introducing adjustable parameters; each of the remaining layers of the neural network layer realizes one iteration, and the algorithm parameter of each iteration is taken as the adjustable parameter of the corresponding neural network layer; each channel state information is taken as the input of the neural network, and the initial resource management optimization variable is taken as the hyperparameter of the neural network. ​​

[0043] a set of channel state samples containing S training data samples optimization of the adjustable parameters based on a loss function; wherein, ∈ and is an adjustable parameter; the loss function is specifically:

[0044]

[0045] The embodiment of the present application can adaptively optimize the algorithm parameters of parallel iteration by constructing a neural network.

[0046] Another embodiment of the present application further provides a joint optimization device of beamforming, power control and frequency selection, comprising a joint optimization module and a parallel iteration module.

[0047] The joint optimization module is configured to construct a resource management optimization model of all transmitters and receivers; wherein, the resource management optimization model comprises resource management optimization variables, a network rate target function and constraint conditions, which contain joint optimization of beamforming, power control and frequency selection.

[0048] The parallel iteration module is configured to perform parallel iteration based on the resource management optimization model to obtain optimized beamforming, power control and frequency selection of all links; wherein, the parallel iteration is realized by a parallel processor; and algorithm parameters of the parallel iteration are optimized by a neural network.

[0049] Another embodiment of the present application further provides a terminal device, characterized by comprising a processor, a memory and a computer program stored in the memory and configured to be executed by the processor; when the processor executes the computer program, a joint optimization method of beamforming, power control and frequency selection according to any one of claims 1-7 is realized.

[0050] Another embodiment of the present application further provides a computer readable storage medium item, comprising a stored computer program; wherein, when the computer program runs, the device where the computer readable storage medium is located is controlled to execute a joint optimization method of beamforming, power control and frequency selection according to any one of claims 1-7. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 A flowchart of an embodiment of the joint optimization method of beamforming, power control and frequency selection provided by the present application is shown in the figure;

[0052] Figure 2 A structure diagram of an embodiment of the neural network provided by the present application is shown in the figure;

[0053] Figure 3 Structure diagram of an embodiment of the joint optimization device of beamforming, power control and frequency selection provided by the present application. DETAILED DESCRIPTION

[0054] For the purpose of making the purpose, technical scheme and advantages of the present application more clear, the technical scheme in the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all the other embodiments obtained by those skilled in the art without making creative efforts fall within the scope of protection of the present application.

[0055] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs; the terminology used herein is for the purpose of describing the particular embodiments only and is not intended to be limiting of the present application; the terms "include" and "have" and any variations thereof used in the specification and the claims and the above description of drawings are intended to cover the non-exclusive inclusion.

[0056] In the description of the embodiments of the present application, the technical terms "first", "second" and the like are only used to distinguish different objects, and cannot be understood as indicating or implying relative importance or implicitly indicating the number, specific order or primary and secondary relationship of the indicated technical features. In the description of the embodiments of the present application, the meaning of "multiple" is more than two, unless otherwise explicitly and specifically limited.

[0057] Reference herein to "an embodiment" means that a particular feature, structure, or characteristic described in connection with the embodiment can be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily all refer to the same embodiment, nor is it necessarily independent or alternative embodiments to each other. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0058] In the description of the embodiments of the present application, the term "and / or" is only a description of the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which means that there are three cases of A alone, A and B together, and B alone. In addition, the character " / " in this paper generally represents a "or" relationship between the front and rear associated objects.

[0059] In the description of the embodiments of the present application, the term "multiple" refers to more than two (including two), and similarly, "multiple groups" refers to more than two groups (including two groups), and "multiple pieces" refers to more than two pieces (including two pieces).

[0060] See Figure 1 To solve the problem that the prior art cannot balance network performance and computing efficiency in a wireless communication network, an embodiment of the present application provides a joint optimization method for beamforming, power control and frequency selection, comprising steps S101 to S102:

[0061] In step S101, a resource management optimization model of all transmitters and receivers is constructed; wherein the resource management optimization model comprises resource management optimization variables, a network rate objective function and constraint conditions involving joint optimization of beamforming, power control and frequency selection.

[0062] In a specific embodiment, the wireless communication network contains B transmitters and K receivers, the set of transmitters is The set of receivers is Each transmitter transmits signals to K b receivers, and the corresponding set of receivers is bs(j) is the jth receiver corresponding transmitter; each transmitter b is equipped with N T transmit antennas, and each receiver is equipped with N R receive antennas, and the maximum transmit power is P b,max (unit: watt); all transmitters and receivers operate on a frequency band with a bandwidth of W, the number of subcarriers is F, and the set of subcarriers is Note that for all When K b >1, the wireless communication network can represent a multi-cell cellular network, when K b =1, the wireless communication network can represent a D2D network; for part K b >1, and the other K b =1, the wireless communication network can represent a hybrid network of cellular networks and D2D networks.

[0063] Large-scale fading and small-scale fading are used to model the channel of the wireless communication network, and block fading and frequency-selective fading models are used to model small-scale fading. Considering any time, for all and Let and represent the large-scale fading coefficient and the small-scale fading coefficient of the channel between receiver i and transmitter b on subcarrier f, respectively, and the total fading coefficient of the channel between transmitter b and receiver i on subcarrier f can be represented as The original channel state information is known.

[0064] Each transmitter and receiver employs linear beamforming to enhance the transmitted signal and reduce interference. For all Let denote the transmit beamforming vector corresponding to the signal transmitted by transmitter bs(i) to receiver i on subcarrier f, and let denote the receive beamforming vector corresponding to the signal transmitted by transmitter bs(i) to receiver i on subcarrier f, the corresponding beamforming constraints are as follows:

[0065]

[0066] where ||·||2is the l-2 norm. Note that for all and denote the power of the signal transmitted by transmitter bs(i) to receiver i on subcarrier f, denote the beam direction of the signal transmitted by transmitter bs(i) to receiver i on subcarrier f, denote the signal transmitted by transmitter bs(i) to receiver i on subcarrier f (using subcarrier f), denote the signal not transmitted by transmitter bs(i) to receiver i on subcarrier f (not using subcarrier f). Thus, the transmit beam vector v i,f can characterize the transmit beamforming, power control, and frequency selection.

[0067] for all and the signal transmitted by transmitter bs(i) to receiver i on subcarrier f is as follows:

[0068]

[0069] where s i,f ~ CN(0, 1) is the symbol transmitted by transmitter bs(i) to receiver i on subcarrier f. The signal received by receiver i on subcarrier f is as follows:

[0070]

[0071] where is the additive white Gaussian noise (AWGN) vector on subcarrier f for receiver i, is the noise power on subcarrier f for receiver i. Thus, the signal-to-interference-and-noise ratio (SINR) on subcarrier f for receiver i is:

[0072]

[0073] wherein, and

[0074] The achievable rate of receiver i on subcarrier f is denoted as is given by

[0075]

[0076] The performance indicator of the wireless communication network can be the sum rate, the weighted sum rate or the worst rate of all receivers, denoted as

[0077]

[0078] wherein, α i is the weight of the achievable rate of receiver i, and The above performance indicator of the wireless communication network is referred to as network rate.

[0079] In the above specific embodiments, by considering a joint optimization problem of beamforming, power control and frequency selection of all transmitters and receivers, the resource management optimization variables (V, W) involving joint optimization of beamforming, power control and frequency selection are optimized to maximize the network rate. Based on the joint optimization problem, a resource management optimization model is constructed; wherein the resource management optimization model is specifically:

[0080]

[0081] Since is equivalent to The resource management optimization model can be transformed into:

[0082]

[0083] wherein the network rate objective function is a non-convex function, and when , the network rate objective function is not differentiable due to the existence of operator. Therefore, it is necessary to transform the resource management optimization model into:

[0084]

[0085] wherein the value of g(V, W; H) can be divided into two cases according to different R(V, W; H):

[0086] The first case is when , g(V, W; H) = -R(V, W; H);

[0087] The second case is when When the network rate objective function is non-differentiable, a differentiable log-sum-exp function is used to approximate the non-differentiable max function max{x1,…x n} where c>0. The approximation error is upper bounded by , which decreases as c increases and can be arbitrarily small. Therefore, when ,

[0088]

[0089] Note that the approximation error upper bound between g(V,W;H) and decreases as c increases.

[0090] Step S102, performing parallel iteration based on the resource management optimization model to obtain beamforming, power control and frequency selection of all optimized links; wherein the parallel iteration is realized by a parallel processor; and algorithm parameters of the parallel iteration are optimized by a neural network.

[0091] In the above specific embodiments, the network rate objective function g(V,W;H) of the resource management optimization model is a differentiable non-convex function, and the constraint set is a convex set. Therefore, the resource management optimization model is a large-scale non-convex optimization model. The resource management optimization model can be solved by parallel iteration to realize complete parallelism and closed-form update of joint optimization of beamforming, power control and frequency selection of all links, and converge to a stationary point (which can be a local optimal point or a global optimal point). It is worth noting that the resource management optimization variable finally output by the parallel iteration is the optimal solution of the resource management optimization model, and since the resource management optimization model is constructed to maximize the network rate, the resource management optimization variable finally output by the parallel iteration can improve the performance of the wireless communication network.

[0092] The parallel iteration process is shown in Table 1:

[0093] Table 1-Parallel iteration process table

[0094]

[0095] Specifically, at the tth iteration, a strongly convex approximation function of g(V,W;H) at the point (V (t) ,W (t) ) with respect to V b , and w i,f , is selected, which is specifically:

[0096] ​​

[0097] Among them, (V) (t) W (t) ) is the resource management optimization variable obtained in the (t-1)th iteration; and These are the coefficients of strongly convex terms; and They are respectively Regarding v i,f and w i,f The gradient; and The value of g(V,W;H) can be divided into the following three cases:

[0098] In the first case, when hour,

[0099]

[0100] In the second scenario, when When, that is, in the first case, when α j =1;

[0101] The third scenario, when hour,

[0102]

[0103] Among them, e i,f (V (t) W (t) ;H) as follows:

[0104]

[0105] and

[0106]

[0107] Therefore, in the t-th iteration, based on the resource management optimization model, we can obtain the result for any transmitter. The first approximate convex problem model is as follows:

[0108]

[0109] In the t-th iteration, based on the resource management optimization model, we can obtain information about any receiver. In any subcarrier The second approximate convex problem model is as follows:

[0110]

[0111] The first and second approximate convex problem models are two quadratic constraint quadratic programming models, and through analyzing Karush-Kuhn-Tucker (KKT) conditions, closed-form optimal solutions of the approximate convex problem models on all subcarriers can be obtained in parallel and all receivers The closed-form optimal solutions of the approximate convex problem models on all subcarriers are denoted as and and Specifically,

[0112]

[0113] The transmit beamforming vectors and the receive beamforming vectors of all links are updated in parallel, and at the tth iteration, the transmit beamforming vectors and the receive beamforming vectors are updated as follows:

[0114]

[0115] wherein γ (t) ∈ (0, 1] is a step size, and the step size for iterating the transmit beamforming vectors and the step size for iterating the receive beamforming vectors are consistent. The step size at each iteration is and a strong convexity coefficient are algorithm parameters, wherein

[0116] The algorithm parameters can be obtained by constructing a neural network to expand the parallel iteration algorithm, and specifically, the parallel iteration process is mapped to a neural network layer, and a neural network is constructed; wherein a first layer of the neural network layer adjusts initial resource management optimization variables (V (0) ,W (0) ) by introducing adjustable parameters and and respectively, to obtain so as to obtain an improved initial point for the neural network Each of the remaining layers of the neural network layer implements one iteration, and algorithm parameters at each iteration are taken as adjustable parameters of the corresponding neural network layer; each channel state information H is input to the neural network for inference, and initial resource management optimization variables (V (0) ,W (0) ) are taken as hyperparameters of the neural network. For Let (V (t) ,W (t) ) represent an output of the (t+1)th layer of the neural network, corresponding to the tth iteration. The neural network finally outputs (V (T) ,W (T) ), which can also be written as (V (T) (H; ∩, γ, τ), W(T) (H; e, g, t) to reflect its relationship with the input H, and the relationship with the algorithm parameters g and t corresponding to the adjustable parameters of the neural network, and the specific architecture of the neural network is as shown in Figure 2 .

[0117] The training process of the neural network is as follows: let represent a set of channel state samples of S training data samples, wherein, The neural network is trained by using an end-to-end unsupervised training method and an Adam optimizer, and the loss function L(e, g, t) is as follows:

[0118]

[0119] In training the neural network, the neural network can be divided into multiple sub-networks according to layers, and new sub-networks are added after the trained sub-networks to continue training according to actual needs. Wherein, the neural network does not need to be retrained under the condition that the distance between the transmitter and the receiver and the change of the device transmission power are small.

[0120] It is worth noting that, in addition to being obtained by the deep expansion method and with the help of the neural network and the data sample optimization, the algorithm parameters can also be selected in a specific way; wherein, the specific selection refers to selecting specific algorithm parameters that satisfy the following conditions at a given arbitrary initial point:

[0121]

[0122] The embodiment of the application can provide a model basis for subsequent joint optimization of beamforming, power control and frequency selection for maximizing network rate by constructing a resource management optimization model of all transmitters and receivers; and can efficiently solve resource management optimization variables capable of maximizing network rate by parallel iteration based on the resource management optimization model. Compared with the prior art which cannot balance network performance and computing efficiency in a wireless communication network, the application can improve the network performance and computing efficiency of the wireless communication network by parallel optimization of beamforming, power control and frequency selection.

[0123] Optionally, in the embodiment of the application, the construction of the resource management optimization model of all transmitters and receivers comprises:

[0124] constructing a joint optimization problem of beamforming, power control and frequency selection of all transmitters and receivers to maximize the network rate under power constraints;

[0125] constructing a resource management optimization model based on the joint optimization problem; wherein, the resource management optimization model is specifically:

[0126]

[0127] wherein R(V,W;H) is the network rate, including sum rate, weighted sum rate and worst rate; (V,W) is the resource management optimization variable containing beamforming, power control and frequency selection joint optimization; H is the channel state information; for any and v i,f represents the transmit beamforming vector corresponding to the signal transmitted by the transmitter bs(i) to the receiver i on the subcarrier f; represents the power of the signal transmitted by the transmitter bs(i) to the receiver i on the subcarrier f; represents the beam direction of the signal transmitted by the transmitter bs(i) to the receiver i on the subcarrier f; represents the (use subcarrier) of the signal transmitted by the transmitter bs(i) to the receiver i on the subcarrier f; represents the (not use subcarrier) of the signal transmitted by the transmitter bs(i) to the receiver i on the subcarrier f; for any and w i,f represents the receive beamforming vector corresponding to the signal transmitted by the transmitter bs(i) and received by the receiver i on the subcarrier f; for any P b,max represents the maximum transmit power of the transmitter b; for any represents the receiver set corresponding to the transmitter b; represents the set of all subcarriers; represents the set of all receivers; represents the set of all transmitters; and

[0128] The embodiment of the application can provide a model basis for subsequent beamforming, power control and frequency selection joint optimization for maximizing the network rate by constructing the resource management optimization model of all transmitters and receivers.

[0129] Optionally, in the embodiment of the application, the parallel iteration based on the resource management optimization model comprises:

[0130] At each iteration, a plurality of jointly optimized approximate convex problem models are constructed based on the resource management optimization model, and parallel iteration solving is performed on all the approximate convex problem models until a convergence condition is met, and resource management optimization variables of the current iteration are output; wherein at each iteration, all the approximate convex problem models are updated according to the receive beamforming vectors and the transmit beamforming vectors of all the links of the last iteration; wherein the approximate convex problem models include: first approximate convex problem models about all the transmitters and second approximate convex problem models about all the receivers on all the subcarriers; any transmitter corresponds to a first approximate convex problem model, and any receiver on any subcarrier corresponds to a second approximate convex problem model.

[0131] The embodiment of the present application can efficiently solve the resource management optimization variables by parallel iteration based on the resource management optimization model, and the resource management optimization variables can maximize the network rate.

[0132] Optionally, in the embodiment of the present application, the plurality of jointly optimized approximate convex problem models are constructed based on the resource management optimization model, including:

[0133] The plurality of jointly optimized first approximate convex problem models are constructed based on the resource management optimization model; wherein at the tth iteration, the first approximate convex problem model about any transmitter is specifically:

[0134]

[0135] wherein, and is the value of the resource management optimization variable at the (t-1) th iteration; is a strong convex term coefficient, and is an algorithm parameter; g(V (t) ,W (t) is selected according to the type of R(V,W;H); is the gradient of g(V (t) ,W (t) ;H) with respect to v i,f .

[0136] The plurality of jointly optimized second approximate convex problem models are constructed based on the resource management optimization model; wherein at the tth iteration, the second approximate convex problem model about any receiver on any subcarrier is specifically:

[0137]

[0138] wherein, is a strong convexity coefficient, and is an algorithm parameter; g(V (t) ,W (t) ; H) is valued according to the type of R(V, W; H), is a gradient of g(V (t) ,W (t) ; H) with respect to w i,f .

[0139] The embodiment of the present application can provide a model basis for parallel iteration by constructing approximate convex problem models of all transmitters and receivers.

[0140] Optionally, in the embodiment of the present application, the parallel iteration solving of all the approximate convex problem models comprises:

[0141] The closed-form optimal solution of the first approximate convex problem model of any transmitter is obtained by analyzing KKT conditions and performing parallel calculation; wherein, the closed-form optimal solution of the first approximate convex problem model of any transmitter

[0142]

[0143] wherein,

[0144] The transmit beamforming vector of all transmitters is updated in parallel; wherein, the transmit beamforming vector of any transmitter is updated in the tth iteration; wherein, the transmit beamforming vector of any transmitter

[0145]

[0146] wherein, γ (t) ∈ (0, 1] is a step size, and is an algorithm parameter.

[0147] The embodiment of the present application can obtain the closed-form optimal solution of the joint optimization problem of all transmitters in parallel by analyzing KKT conditions.

[0148] Optionally, in the embodiment of the present application, the parallel iteration solving of all the approximate convex problem models further comprises:

[0149] The closed-form optimal solution of the second approximate convex problem model of any receiver on any subcarrier is obtained by analyzing KKT conditions and performing parallel calculation; wherein, the closed-form optimal solution of the second approximate convex problem model of any receiver

[0150]

[0151] The receive beamforming vectors of all receivers on all subcarriers are updated in parallel; wherein, at the tth iteration, the receive beamforming vector of any receiver The receive beamforming vector on any subcarrier is specifically:

[0152]

[0153] wherein, γ (t) ∈(0, 1] is a step size, and is an algorithm parameter.

[0154] The embodiment of the present application can obtain the closed optimal solution of the joint optimization problem of all receivers in parallel through analyzing the KKT condition.

[0155] Optionally, in the embodiment of the present application, the algorithm parameter of the parallel iteration is optimized through a neural network, including:

[0156] mapping the parallel iteration process into a neural network layer, and constructing a neural network; wherein, the first layer of the neural network layer adjusts the initial resource management optimization variable by introducing an adjustable parameter; each of the remaining layers of the neural network layer realizes one iteration, and takes the algorithm parameter of each iteration as the adjustable parameter of the corresponding neural network layer; taking each channel state information as the input of the neural network, and taking the initial resource management optimization variable as the hyperparameter of the neural network;

[0157] based on a channel state sample set containing S training data samples optimizing the adjustable parameter based on a loss function; wherein, ∈ and is the adjustable parameter; the loss function is specifically:

[0158]

[0159] The embodiment of the present application can adaptively optimize the algorithm parameter of the parallel iteration by constructing a neural network. As Figure 3 shown, on the basis of the above method item embodiment, the corresponding device item embodiment is provided;

[0160] An embodiment of the present application provides a joint optimization device of beamforming, power control and frequency selection, including a joint optimization module 301 and a parallel iteration module 302.

[0161] The joint optimization module 301 is used for constructing a resource management optimization model of all transmitters and receivers; wherein, the resource management optimization model includes resource management optimization variables, a network rate target function and constraint conditions which contain joint optimization of beamforming, power control and frequency selection.

[0162] The parallel iteration module 302 is configured to perform parallel iteration based on the resource management optimization model to obtain beamforming, power control and frequency selection of all links after optimization; wherein the parallel iteration is realized by a parallel processor; and algorithm parameters of the parallel iteration are optimized by a neural network.

[0163] Optionally, in the embodiment of the present application, the joint optimization module 301 comprises a joint optimization problem sub-module and a resource management optimization model sub-module.

[0164] The joint optimization problem sub-module is configured to construct a joint optimization problem of beamforming, power control and frequency selection of all transmitters and receivers to maximize network rate under power constraint.

[0165] The resource management optimization model sub-module is configured to construct a resource management optimization model based on the joint optimization problem; wherein the resource management optimization model is specifically as follows:

[0166]

[0167] wherein R(V,W;H) is network rate, including sum rate, weighted sum rate and worst rate; (V,W) is resource management optimization variable containing beamforming, power control and frequency selection joint optimization; H is channel state information; for any and v i,f is a transmission beamforming vector corresponding to the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents the power of the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents the beam direction of the transmitter bs(i) transmitting a signal to the receiver i on the subcarrier f; represents the transmitter bs(i) transmitting a signal (using the subcarrier) to the receiver i on the subcarrier f; represents the transmitter bs(i) not transmitting a signal (not using the subcarrier) to the receiver i on the subcarrier f; for any and w i,f is a receiving beamforming vector corresponding to the receiver i receiving the signal transmitted by the transmitter bs(i) on the subcarrier f; for any P b,max is the maximum transmission power of the transmitter b; for any is a receiver set corresponding to the transmitter b; is a set of all subcarriers; is a set of all receivers; for all transmitters; and

[0168] The embodiment of the present application can provide a model basis for subsequent joint optimization of beamforming, power control and frequency selection for maximizing network rate by constructing resource management optimization models of all transmitters and receivers.

[0169] Optionally, in the embodiment of the present application, the parallel iteration module 302 comprises a parallel iteration sub-module.

[0170] The parallel iteration sub-module is configured to, at each iteration, construct a plurality of joint optimization approximate convex problem models based on the resource management optimization model, and perform parallel iteration and solving on all the approximate convex problem models until a convergence condition is met, and output resource management optimization variables of the current iteration; wherein at each iteration, all the approximate convex problem models are updated according to the receive beamforming vectors and the transmit beamforming vectors of all the links of the last iteration; wherein the approximate convex problem models comprise first approximate convex problem models about all the transmitters and second approximate convex problem models about all the receivers on all the subcarriers; any transmitter corresponds to a first approximate convex problem model, and any receiver on any subcarrier corresponds to a second approximate convex problem model.

[0171] The embodiment of the present application can efficiently solve resource management optimization variables by parallel iteration based on the resource management optimization model, and the resource management optimization variables can maximize network rate.

[0172] Optionally, in the embodiment of the present application, the parallel iteration sub-module comprises a first approximate convex problem model unit and a second approximate convex problem model unit.

[0173] The first approximate convex problem model unit is configured to construct a plurality of joint optimization first approximate convex problem models based on the resource management optimization model; wherein at the tth iteration, the first approximate convex problem model about any transmitter is specifically:

[0174]

[0175] wherein, and (V (t) ,W (t) ) is the value of the resource management optimization variable at the (t-1)th iteration; is a strong convex term coefficient, which is an algorithm parameter; g(V (t) ,W (t) ; H) is selected according to the type of R(V, W; H); is the gradient of g(V (t) ,W (t) ; H) with respect to w i,f .

[0176] The second approximate convex problem model unit is configured to construct a plurality of jointly optimized second approximate convex problem models based on the resource management optimization model; wherein, at the tth iteration, the closed-form optimal solution of the second approximate convex problem model with respect to any receiver The second approximate convex problem model on any subcarrier is specifically as follows:

[0177]

[0178] wherein, is a strong convex term coefficient, and is an algorithm parameter; g(V (t) ,W (t) ; H) is valued according to the type of R(V,W;H), is the gradient of g(V (t) ,W (t) ; H) with respect to w i,f .

[0179] The embodiment of the present application can provide a model basis for parallel iteration by constructing approximate convex problem models of all transmitters and receivers.

[0180] Optionally, in the embodiment of the present application, the parallel iteration submodule further comprises a first optimal solution unit and a first parallel update unit.

[0181] The first optimal solution unit is configured to calculate the closed-form optimal solution of all the first approximate convex problem models in parallel by analyzing KKT conditions; wherein, at the tth iteration, the closed-form optimal solution of the first approximate convex problem model with respect to any transmitter is specifically as follows:

[0182]

[0183] wherein,

[0184] The first parallel update unit is configured to update the transmit beamforming vectors of all transmitters in parallel; wherein, at the tth iteration, the transmit beamforming vector of any transmitter is specifically as follows:

[0185]

[0186] wherein, γ (t) ∈(0,1] is a step size, and is an algorithm parameter.

[0187] The embodiment of the present application can obtain the closed-form optimal solution of the joint optimization problem in parallel through analyzing the KKT condition The closed-form optimal solution of the joint optimization problem.

[0188] Optionally, in the embodiment of the present application, the parallel iteration sub-module further comprises a second optimal solution unit and a second parallel update unit.

[0189] The second optimal solution unit is configured to calculate the closed-form optimal solution of the second approximate convex problem model in parallel through analyzing the KKT condition; wherein, at the tth iteration, the closed-form optimal solution of the second approximate convex problem model with respect to any receiver on any subcarrier is specifically as follows:

[0190]

[0191] The second parallel update unit is configured to update the receive beamforming vector of all receivers on all subcarriers in parallel; wherein, at the tth iteration, the receive beamforming vector of any receiver on any subcarrier is specifically as follows:

[0192]

[0193] wherein, γ (t) ∈(0, 1] is a step size, and is an algorithm parameter.

[0194] The embodiment of the present application can obtain the closed-form optimal solution of the joint optimization problem in parallel through analyzing the KKT condition The closed-form optimal solution of the joint optimization problem.

[0195] Optionally, in the embodiment of the present application, the parallel iteration module 302 further comprises a neural network sub-module and a parameter optimization sub-module.

[0196] The neural network sub-module is configured to map the parallel iteration process into a neural network layer, and construct a neural network; wherein, the first layer of the neural network layer adjusts the initial resource management optimization variable by introducing an adjustable parameter; each of the remaining layers of the neural network layer realizes one iteration, and takes the algorithm parameter of each iteration as the adjustable parameter of the corresponding neural network layer; takes each channel state information as the input of the neural network, and takes the initial resource management optimization variable as the hyperparameter of the neural network.

[0197] The parameter optimization sub-module is configured to optimize the adjustable parameter based on a channel state sample set containing S training data samples based on a loss function; wherein, ∈ and is an adjustable parameter; and the loss function is specifically:

[0198]

[0199] The embodiment of the application can adaptively optimize the algorithm parameters of parallel iteration by constructing a neural network.

[0200] It can be understood that the above device item embodiments correspond to the method item embodiments of the application, and can implement the joint optimization method of beamforming, power control and frequency selection provided by any one of the above method item embodiments of the application.

[0201] The embodiment of the application can provide a model basis for subsequent joint optimization of beamforming, power control and frequency selection for maximizing network rate by constructing resource management optimization models of all transmitters and receivers through the joint optimization module; and the resource management optimization variables for maximizing network rate can be efficiently solved through parallel iteration of the parallel iteration module. Compared with the prior art which cannot balance network performance and computing efficiency in a wireless communication network, the application can improve the network performance and computing efficiency of the wireless communication network by optimizing beamforming, power control and frequency selection in parallel.

[0202] It should be noted that the device embodiments described above are only schematic, and part or all of the modules thereof can be selected to achieve the purpose of the embodiment of the application. In addition, in the device embodiment provided by the application, the connection relationship between the modules indicates that there is a communication connection therebetween, which can be implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement it without creative labor.

[0203] On the basis of the above-mentioned embodiment of the joint optimization method based on beamforming, power control and frequency selection, another embodiment of the application provides a terminal device, which comprises a processor, a memory and a computer program stored in the memory and configured to be executed by the processor, and when the processor executes the computer program, a joint optimization method of beamforming, power control and frequency selection according to any one of the embodiments of the application is implemented.

[0204] For example, in this embodiment, the computer program can be divided into one or more modules, which are stored in the memory and executed by the processor to complete the application. The one or more modules can be a series of computer program instruction segments capable of completing a specific function, which are used to describe the execution process of the computer program in the terminal device.

[0205] The terminal device can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The terminal device can include, but is not limited to, a processor and a memory.

[0206] The processor can be a central processing unit (CPU), and can also be other general-purpose processors, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, and the like. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor and the like. The processor is a control center of the terminal device, and is connected to various parts of the terminal device through various interfaces and lines.

[0207] On the basis of the above-mentioned method embodiment, another embodiment of the present application provides a computer readable storage medium, including a stored computer program, wherein when the computer program runs, the device where the computer readable storage medium is located performs the joint optimization method of beamforming, power control and frequency selection provided in any one of the above-mentioned method embodiments of the present application.

[0208] The modules / units integrated in the device / terminal device, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on this understanding, all or part of the processes in the above-mentioned embodiments can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the processor executes the computer program, the steps of each method embodiment can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0209] The above is the preferred embodiment of the present application, it should be pointed out that, for those skilled in the art, without departing from the principles of the present application, can also make a number of improvements and refinements, these improvements and refinements are also considered to be within the scope of the present application.

Claims

1. A joint optimization method for beamforming, power control, and frequency selection, characterized in that, include: Construct resource management optimization models for all transmitters and receivers; wherein, the resource management optimization models include resource management optimization variables, network rate objective functions, and constraints that imply joint optimization of beamforming, power control, and frequency selection; Based on the resource management optimization model, parallel iteration is performed to obtain the optimized beamforming, power control, and frequency selection for all links; wherein, the parallel iteration is implemented through a parallel processor; and the algorithm parameters of the parallel iteration are optimized through a neural network.

2. The joint optimization method for beamforming, power control, and frequency selection as described in claim 1, characterized in that, The construction of a resource management optimization model for all transmitters and receivers includes: Construct a joint optimization problem for beamforming, power control, and frequency selection for all transmitters and receivers to maximize network rate under power constraints; Based on the aforementioned joint optimization problem, a resource management optimization model is constructed; specifically, the resource management optimization model is as follows: Where R(V,W;H) represents the network rate, including the sum rate, weighted sum rate, and worst-case rate; (V,W) represents the resource management optimization variables that imply joint optimization of beamforming, power control, and frequency selection; H represents the channel state information; for any and v i,f Let bs(i) be the transmit beamforming vector corresponding to the signal transmitted by transmitter bs(i) to receiver i on subcarrier f; This represents the power of the signal transmitted by transmitter bs(i) to receiver i on subcarrier f; This represents the beam direction in which transmitter bs(i) transmits a signal to receiver i on subcarrier f; This means that transmitter bs(i) transmits a signal to receiver i on subcarrier f (using a subcarrier); This means that transmitter bs(i) does not transmit a signal to receiver i on subcarrier f (does not use a subcarrier); for any and W i,f For receiver i, the receive beamforming vector corresponding to the signal transmitted by transmitter bs(i) on subcarrier f; for any P b,max Let be the maximum transmit power of transmitter b; for any Let be the set of receivers corresponding to transmitter b; For the set of all subcarriers; For the set of all receivers; For all transmitters; and 3. The joint optimization method for beamforming, power control, and frequency selection as described in claim 1, characterized in that, The parallel iteration based on the resource management optimization model includes: In each iteration, multiple joint optimization approximate convex problem models are constructed based on the resource management optimization model, and all approximate convex problem models are solved in parallel iteratively until the convergence condition is met, and the resource management optimization variables of the current iteration are output. In each iteration, all approximate convex problem models are updated according to the receive beamforming vectors and transmit beamforming vectors of all links in the previous iteration. The approximate convex problem models include: a first approximate convex problem model for all transmitters and a second approximate convex problem model for all receivers on all subcarriers; each transmitter corresponds to one first approximate convex problem model, and each receiver on any subcarrier corresponds to one second approximate convex problem model.

4. The joint optimization method for beamforming, power control, and frequency selection as described in claim 3, characterized in that, The construction of multiple joint optimization approximate convex problem models based on the resource management optimization model includes: Based on the resource management optimization model, a first approximate convex problem model for joint optimization is constructed; wherein, in the t-th iteration, for any transmitter... The first approximate convex problem model is as follows: in, and (V (t) W (t) Let ) be the value of the resource management optimization variable in the (t-1)th iteration; These are the coefficients of the strongly convex term, which are algorithm parameters; g(V) (t) W (t) ;H) is selected based on the type of R(V,W;H); For g(V) (t) W (t) ;H) Regarding v i,f The gradient; Based on the resource management optimization model, a second approximate convex problem model for joint optimization is constructed; wherein, in the t-th iteration, for any receiver... In any subcarrier The second approximate convex problem model is as follows: in, These are the coefficients of the strongly convex term, which are algorithm parameters; g(V) (t) W (t) The value of ;H) is determined according to the type of R(V,W;H). For g(V) (t) W (t) ;H) Regarding w i,f The gradient.

5. The joint optimization method for beamforming, power control, and frequency selection as described in claim 3, characterized in that, The parallel iterative solution of all the approximate convex problem models includes: By analyzing the KKT conditions, the closed-form optimal solutions of all the first approximate convex problem models are computed in parallel; where, in the t-th iteration, for any transmitter... The closed-form optimal solution of the first approximate convex problem model is as follows: in, The transmit beamforming vectors of all transmitters are updated in parallel; where, in the t-th iteration, any transmitter... The specific transmit beamforming vector is as follows: Where, γ (t) ∈(0,1] is the step size, which is an algorithm parameter.

6. The joint optimization method for beamforming, power control, and frequency selection as described in claim 3, characterized in that, The parallel iterative solution of all the approximate convex problem models also includes: By analyzing the KKT conditions, the closed-form optimal solutions of all second approximate convex problem models are computed in parallel; where, at the t-th iteration, with respect to any receiver... In any subcarrier The closed-form optimal solution of the second approximate convex problem model is as follows: The receive beamforming vectors of all receivers on all subcarriers are updated in parallel; where, in the t-th iteration, any receiver... In any subcarrier The specific receiving beamforming vector on the beam is: Where, γ (t) ∈(0,1] is the step size, which is an algorithm parameter.

7. The joint optimization method for beamforming, power control, and frequency selection as described in claim 1, characterized in that, The algorithm parameters for the parallel iteration are optimized using a neural network, including: The parallel iterative process is mapped to neural network layers to construct a neural network. The first layer of the neural network adjusts the initial resource management optimization variables by introducing adjustable parameters. Each of the remaining layers of the neural network performs one iteration, and the algorithm parameters of each iteration are used as the adjustable parameters of the corresponding neural network layer. The channel state information is used as the input of the neural network, and the initial resource management optimization variables are used as the hyperparameters of the neural network. Based on a channel state sample set containing S training data samples Optimization of adjustable parameters based on the loss function; where, ∈ and These are adjustable parameters; the loss function is specifically:

8. A joint optimization device for beamforming, power control, and frequency selection, characterized in that, include: Joint optimization module and parallel iterative module; The joint optimization module is used to construct resource management optimization models for all transmitters and receivers; wherein, the resource management optimization model includes resource management optimization variables, network rate objective function and constraints that imply joint optimization of beamforming, power control and frequency selection; The parallel iteration module is used to perform parallel iteration based on the resource management optimization model to obtain optimized beamforming, power control, and frequency selection for all links; wherein, the parallel iteration is implemented by a parallel processor; and the algorithm parameters of the parallel iteration are optimized by a neural network.

9. A terminal device, characterized in that, The system includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, it implements a joint optimization method for beamforming, power control, and frequency selection as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, include: A stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform a joint optimization method for beamforming, power control, and frequency selection as described in any one of claims 1-7.

Images