Communication system applying QDU technique for power allocation and precoding optimazation in non-orthogonal multiple access multiple input multiple output

Abstract

The present invention provides a communication system that applies the QDU technique for non-orthogonal multiple access - multiple input multiple output, which constructs a quantum neural network (QNN) using N qubits to obtain weighting factors, and divides the system into an upper-level layer and a lower-level layer based on the weighting factors to process power allocation of NOMA and precoder selection of MIMO, respectively.

Description

Technology Field

[0001] The present invention relates to a communication system that solves the power allocation and precoding optimization problems of NOMA (non-orthogonal multiple access) and MIMO (multiple input multiple output) by applying a technique called QDU (Quantum Deep unfolding). Background Technology

[0003] Neural networks have been regarded as an important method for future wireless communication. However, conventional neural networks generally use general functions (called perceptrons) at each layer.

[0004] Unlike conventional neural networks, deep unfolding uses analysis-based functions at each layer. Furthermore, existing research on deep unfolding-based optimization generally considers classical computation.

[0005] In particular, quantum neural networks (QNNs) can reduce computational complexity.

[0006] Meanwhile, existing NOMA (non-orthogonal multiple access) has issues with power allocation and precoding optimization, and various methods are being proposed to address these problems. Prior art literature

[0008] KR10-2034955B The problem to be solved

[0009] The present invention is proposed to solve the technical problems described above and provides a communication system that solves the problems of power allocation and precoding optimization of NOMA (non-orthogonal multiple access) and MIMO (multiple input multiple output) by applying a technique called QDU (Quantum Deep unfolding). means of solving the problem

[0011] According to one embodiment of the present invention for solving the above problem, a communication system applying a QDU technique for power allocation and precoding optimization of non-orthogonal multiple access - multiple input multiple output is provided, characterized by configuring a quantum neural network (QNN) using N qubits, deriving weighting factors through the QNN, dividing layers into an upper-level layer and a lower-level layer based on the weighting factors, obtaining an analysis-based power allocation of NOMA in the upper-level layer, and selecting an analysis-based precoder of MIMO using the power allocation in the lower-level layer. Effects of the invention

[0013] The present invention solves the problem of power allocation and precoding optimization of NOMA (non-orthogonal multiple access) and MIMO (multiple input multiple output) by applying a technique called QDU (Quantum Deep unfolding). Brief explanation of the drawing

[0015] Figure 1 is a diagram showing the architecture of the proposed QDU method. FIG. 2 is a diagram showing the initial result (average total communication capacity) using QDU in the system of the present invention. Specific details for implementing the invention

[0016] Hereinafter, in order to explain in detail enough for a person skilled in the art to easily implement the technical concept of the present invention, embodiments of the present invention will be described with reference to the attached drawings.

[0018] The concept of the present invention, motivated by the advantages of quantum computing, considers a quantum neural network (QNN) in a deep unfolding architecture called Quantum Deep Unfolding (QDU). Specifically, the present invention utilizes the Quantum Deep Unfolding (QDU) technique in NOMA-MIMO (non-orthogonal multiple access - multiple input multiple output) and is used for power allocation and precoding optimization. Furthermore, it facilitates practical deployment by employing unsupervised learning methods.

[0020] The present invention uses Quantum Deep Unfolding (QDU) – processed by the QDU processing unit – to optimize power allocation and precoding in NOMA-MIMO (non-orthogonal multiple access through multiple inputs and multiple outputs). The present invention considers analysis-based power allocation and precoding. Additionally, a Quantum Neural Network (QNN) generates value-altering vectors for power allocation and precoding. The QDU aims to maximize the achievable sum ratio.

[0022] In other words, the present invention proposes a communication system that solves the problems of NOMA (non-orthogonal multiple access) and MIMO (multiple input multiple output) by applying a technique called QDU (Quantum Deep unfolding) in the QDU processing unit.

[0023] The problems with the existing NOMA include power allocation and precorder selection, but these can be optimized using the QDU (Quantum Deep Unfolding) technique.

[0024] Based on the weighting factor obtained through the QNN (quantum neural network), the layers are divided into upper-level and low-level to simultaneously solve the power allocation of NOMA and the precoder selection problem of MIMO.

[0026] The present invention can be configured with an architecture as shown in FIG. 1, and a "weighting factor" can be obtained by constructing a quantum neural network (QNN) using N qubits. Based on the "weighting factor" obtained through the QNN, the layers are divided into "upper-level" and "low-level" to simultaneously solve the power allocation of NOMA and the precoder selection problem of MIMO.

[0028] Use a loss function that maximizes the total communication capacity as shown in Table 1.

[0029]

[0030]

[0032] We intend to find the user's power allocation coefficient in the NOMA power allocation problem using the formula shown in Table 2, and to find the optimization coefficient that maximizes the gain of the precorder in the MIMO precorder selection problem.

[0033] Table 2

[0034]

[0036] A detailed explanation of Tables 1 and 2 will be provided later.

[0038] Figure 2 shows the average total communication capacity calculated by applying QDU. The learning rate was set to 0.03, the number of unfolded layers to 2, the number of episodes to 1024, the number of users to 2, the distance between adjacent users to the cell to 0.2, and the distance between users outside the cell to 0.8.

[0040] - System Model

[0041] The present invention assumes the m-th base station (BS) located in the middle of a cell servicing a NOMA group. N Tx Let be the number of transmission arrays, and each user has a single receiving antenna. Users with strong channel gain values ​​and users with weak channel gain values ​​are u, respectively. m,str and u m,weak It is indicated as.

[0043] d m,str ≤ 1 and d m,weak ≤ 1 for the base station (BS) and the user (u) with a strong channel gain value, respectively. m,str ) and users with weak channel gain (u m,weak It is called the normalized distance between ).

[0044] ∥h m,str ∥ 2 and ∥h m,weak ∥ 2 and are called the channel gain values ​​for cell center users and cell edge users, respectively. Therefore, ∥h m,str ∥ 2 ≥ ∥h m,weak ∥ 2 Let's assume that.

[0045] u m,str and u m,weak The channel coefficients for are h, respectively. m,str,j ∼CN(0, d -κ m,str ) ∈ R and h m,weak,j ∼CN(0, d -κ m,weak ) ∈ R, where k is the channel coefficient.

[0047] Assuming channel reciprocity between the base station (BS) and each user,

[0048]

[0049]

[0050] u respectively m,str and um,weak It is based on the channel information obtained regarding.

[0052] The incompleteness of channel information is, respectively

[0053]

[0054] and

[0055]

[0057] It can be expressed as, where

[0058]

[0059] is a noise variable and σ 2 ch is noise dispersion.

[0060] P T and σ 2 noise These are called total transmission power and noise dispersion, respectively.

[0061]

[0062] This is called the signal-to-noise ratio of the transmitter.

[0066] - Proposed QDU (Quantum Deep Unfolding)

[0067] A. Objective Function and Loss Function

[0068] The objectives of the present invention can be presented as follows.

[0069] u m,str and u m,weak The achievable ratios of each

[0071]

[0072] and

[0073]

[0074] It can be expressed as.

[0076] The total proportion of the m-th NOMA group

[0077]

[0078] It can be expressed as.

[0080] The goal of optimization is

[0081]

[0082] It is the maximization of the average sum ratio that can be expressed as.

[0083] Then, the objective function can be expressed by the following equation.

[0085] <Equations 1a, 1b, 1c>

[0086]

[0088] Consequently, the loss function is

[0089]

[0090] It can be expressed as such. Due to the limitations of obtaining labels from the dataset, this task considers unsupervised learning.

[0092] B. QDU for NOMA-MIMO

[0093] For each m-th base station

[0094]

[0095]

[0096] Defines. Here

[0097]

[0098] These are the power allocation and precoding obtained from optimization, respectively.

[0100] The structure of the QDU proposed for NOMA-MIMO is as shown in Fig. 1. Each lth layer has two stepped sections (upper-level and lower-level) that can be described as follows.

[0102] 1) NOMA Power Allocation

[0103] Initially, analysis-based power allocation is

[0104]

[0105] It can be expressed as, where

[0106]

[0107] and

[0108]

[0109] are respectively u m,str and u m,weak It is the NOMA power allocation factor for.

[0111] As shown in Fig. 1 (obtained from QNN)

[0112] It is called the optimization vector for . Then

[0113] Calculate.

[0115] 2) Precoding

[0116] Analysis-based precoding is

[0117]

[0118] It can be expressed as,

[0119] Here am.

[0120] Therefore, analysis-based precoding Analysis-based power allocation It is selected using .

[0121] The equivalent channel matrix identical to the Frobenius norm of the actual MIMO channel matrix H is It is given as.

[0123] Considering the channel matrix, it can be expressed as follows.

[0124] <Equation 2>

[0125]

[0126] Here am.

[0127] Here, represents the number of antennas corresponding to the nth user at the mth base station.

[0128] Then

[0129]

[0130] Calculate.

[0131] Here (obtained from QNN) is an optimization vector for precoding (Fig. 1).

[0133] 3) Quantum Variational Circuits

[0134] The encoding operation of the quantum mutation circuit for the QDU can be expressed by Equation 3.

[0135] <Equation 3>

[0136]

[0137] Here is the encoding part.

[0139] Performance Analysis and Conclusion

[0140] 1) Numerical results: Figure 2 shows the initial results using QDU.

[0141] Simulation parameters can be represented as follows.

[0142] κ = 2, d m,str = 0.2, d m,weak = 0.8, also N shot = 1024 , N [n] layer = 2, η = 0.03.(Here, κ is the unfolded layer, d m,str Is Distance between adjacent cell users, d m,weak is the distance of the user outside the cell, N shot is episode, N [n] layer represents the user, and η represents the learning rate.

[0143] The optimization of QNN is considered as gradient descent using the parameter-shifting rule. QDU experiments were run on IBM Q using IBM Qiskit.

[0144] 2) Conclusion: In the present invention, QDU was used to optimize power allocation and precoding.

[0146] As such, those skilled in the art to which the present invention pertains will understand that the present invention may be implemented in other specific forms without altering its technical concept or essential features. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and their equivalents should be interpreted as being included within the scope of the present invention.

Claims

Claim 1 A communication system applying a QDU technique for power allocation and precoding optimization of non-orthogonal multiple access - multiple input multiple output, characterized by constructing a quantum neural network (QNN) using N qubits, deriving weighting factors through the QNN, dividing layers into an upper-level layer and a lower-level layer based on the weighting factors, obtaining an analysis-based power allocation of NOMA in the upper-level layer, and selecting an analysis-based precoding of MIMO using the power allocation in the lower-level layer. Claim 2 A quantum neural network (QNN) is constructed using N qubits, weighting factors are derived through the QNN, and layers are divided into an upper-level layer and a lower-level layer based on the weighting factors, wherein an analysis-based power allocation of NOMA is obtained in the upper-level layer and an analysis-based precoding selection of MIMO is processed using the power allocation in the lower-level layer, wherein the analysis-based power allocation as the power allocation of NOMA is It is expressed as, and here and are respectively u m,str and u m,weak As the power allocation coefficient of NOMA for, obtained from QNN cast It is called the optimization vector for, and Calculating - u m,str and u m,weak ∥ defines users with strong and weak channel gain values ​​at the m-th base station, respectively, and ∥h m,str ∥ 2 and ∥h m,weak ∥ 2 is defined as the channel gain value for cell center users and cell edge users at the m-th base station, respectively, and is the signal-to-noise ratio of the transmitter, and P T and σ 2 noise Defined as total transmission power and noise dispersion, respectively - Non-orthogonal multiple access - A communication system applying the QDU technique for power allocation and precoding optimization of multiple input multiple output. Claim 3 A quantum neural network (QNN) is constructed using N qubits, weighting factors are derived through the QNN, and layers are divided into an upper-level layer and a lower-level layer based on the weighting factors, an analysis-based power allocation of NOMA is obtained in the upper-level layer, and an analysis-based precoding selection of MIMO is processed using the power allocation in the lower-level layer, wherein, as the precoding selection of MIMO, the analysis-based precoding is, It is expressed as, and here And, the equivalent channel matrix identical to the Frobenius norm of the MIMO channel matrix H is Given as, and considering the channel matrix, it can be expressed as follows, and Here And, Calculating,- where, N Tx represents the number of transmission arrays, and represents the channel matrix of the m-th base station and the n-th user, and means analysis-based power allocation, and is defined as the number of antennas corresponding to the nth user at the mth base station. A communication system applying the QDU technique for power allocation and precoding optimization of non-orthogonal multiple access - multiple input multiple output, characterized by being an optimization vector for precoding obtained from QNN.

Images