KR20240029923A - Sub-Band Assignment and Power Control Method in IoT Cellular Networks - Google Patents

Abstract

In the method of the IoT cellular network system controlling frequency subband allocation and transmission power of IoT devices,
The system is,
base station;
An IoT device connected to the base station through uplink communication; and
It includes a control device that controls transmission power for IoT devices connected to the base station through communication,
The method by which the control device of the system controls frequency subband allocation and transmission power of IoT devices is,
(a) For all IoT devices connected to the base station, inferring a subband allocation set that maximizes the total transmission rate against interference between subbands due to spectrum leakage while setting the transmission power of each IoT device to the maximum. ; and
(b) determining a transmission power vector that maximizes the total data rate based on the subband allocation set obtained in step (a).

Description

Sub-Band Assignment and Power Control Method in IoT Cellular Networks}

The present invention relates to a method for subband allocation and transmission power control in an IoT cellular network.

In addition, it provides a subband allocation and transmission power control method in an IoT cellular network where spectrum leakage exists.

The Internet of Things (IoT) is receiving great attention in the 5th generation (5G) and future 6th generation (6G) communication systems. Specifically, the Internet of Things can support a variety of services such as smart cities, smart homes, transportation, surveillance, and disaster and disaster management through large-scale device connectivity. Additionally, research on the Internet of Medical Things (IoMT), which utilizes IoT for medical purposes, and technology that provides long-distance IoT communication through satellites is attracting attention.

Due to the rapid growth in demand for these new IoT-based services, the total number of connected IoT devices is expected to reach approximately 80 billion by 2030. Therefore, it is becoming increasingly important to develop a suitable IoT communication framework that can efficiently serve a large number of IoT users with limited resources.

Therefore, it is important to develop a communication framework that can fairly and efficiently support multiple IoT devices with limited resources.

Unlike OMA (Orthogonal Multiple Access), which requires multiple users to use different frequency resources, Non-orthogonal Multiple Access (NOMA) allows multiple users to simultaneously occupy the same frequency resource. Resources can be used more efficiently. It is a multiple access method that allows allocation of non-orthogonal resources by sharing frequency, time, and code resources for each user terminal, while allowing more terminals to be accommodated with designated resources.

This refers to a technology that improves spectral efficiency by simultaneously transmitting data from two or more terminals sharing the same resources (time slot, subcarrier, spreading code, etc.). The purpose is to increase resource efficiency by simultaneously transmitting data for two or more terminals on the same time, frequency, and space resources.

NOMA is a technology used to increase the capacity of 5G cells, breaking the orthogonality property of the existing orthogonal multiple access (OFDMA) method in terms of frequency resource allocation and simultaneously overlapping data from two or more terminals on the same frequency resource. Resource efficiency can be improved through allocation. For example, when a base station transmits overlapping data signals to a terminal in the cell center area with a large difference in channel quality and a terminal in the cell border area, small power is allocated to the terminal in the cell center area and high power is allocated to the terminal in the cell border area. do.

Terminals within a cell first decode terminals in the central area, which are cells with high signal strength, and then decode cell edge terminals, according to the sequential interference cancellation (SIC) method. A terminal in the center of a cell can decode its own signal, considering it as interference because the signal from the terminal at the cell border is relatively weak. Since the terminal at the center of the decoded cell is removed from the received signal before decoding the terminal at the cell border, the signal from the terminal at the cell border can be successfully decoded.

Because of these advantages, NOMA is considered one of the promising technologies to support large-scale connectivity in IoT networks.

To this end, much research is currently being conducted on resource and interference management of various wireless networks, including IoT networks.

Recent prior research has studied joint optimization of subband allocation and power control to maximize total transmission rate (see Non-Patent Document 1), where interference exists between IoT subbands and between IoT and LTE (Long Term Evolution) bands. This was pointed out as a problem.

Additionally, previous studies (Non-Patent Documents 2 and 3) proposed an energy-efficient iterative algorithm for resource allocation in IoT networks using Lagrange double decomposition (LDD) under the assumption of incomplete channel state information (CSI).

Additionally, in prior research (non-patent document 4), a stochastic optimization problem to minimize the total power consumption of a NOMA-based IoT downlink network was studied.

In addition, a prior study (non-patent document 5) proposed a two-stage optimization method that applied genetic algorithm (GA) to each stage to solve resource block allocation and power allocation problems in the NOMA network.

In addition, in prior research (non-patent document 6), a three-stage resource allocation method using convex relaxation and heuristic greedy algorithms for subcarrier allocation and transmission power control was used to maximize the total transmission rate of the multi-carrier NOMA (MC-NOMA) system. suggested. In addition, in prior research (non-patent document 7), efficient weighted total data rate optimization based on iterative joint user selection and variable power allocation (VPA) was conducted to solve subband allocation and power allocation problems for downlink multi-band NOMA. An algorithm was proposed.

Recently, deep learning using fully connected neural networks (FNNs), convolutional neural networks (CNNs), and recurrent neural networks (RNNs) has been widely adopted to solve various types of complex combinatorial optimization problems in wireless communications.

Previous research (non-patent document 8) proposed CNN-based transmission power control to maximize spectral efficiency and energy efficiency.

In a prior study (Non-Patent Document 9), an RNN-based energy-efficient resource allocation algorithm was proposed for the NOMA-IoT network under the assumption that imperfect continuous interference cancellation (SIC) is applied.

In a prior study (non-patent document 10), an FNN-based approach was proposed to solve the linear sum allocation problem according to integer programming constraints.

In prior research (non-patent document 11), a joint pilot and data power control method based on a deep neural network was developed to maximize total spectral efficiency in an uplink cellular massive multiple-input multiple-output (MIMO) system.

In a prior study (non-patent document 12), an FNN-based unsupervised learning approach was studied to optimize beamforming vectors using uplink-downlink duality in a multi-user, multiple-input, single-output (MU-MISO) downlink cellular system. It has been done.

In addition, in a prior study (Non-Patent Document 13), a DBN (Deep Belief Network)-based algorithm was proposed in an MC-NOMA system supporting downlink simultaneous wireless information and power transmission (SWIPT) to optimize subband allocation and power allocation problems, thereby optimizing the system. A method to minimize total transmission power while satisfying quality of service (QoS) was proposed.

In addition, a prior study (non-patent document 14) presented an energy efficiency algorithm for the joint resource allocation problem that applies LDD and semi-supervised learning based on the FNN structure to maximize energy efficiency in the NOMA mmWave heterogeneous network.

In these recent studies, deep learning-based approaches have been shown to provide significantly improved performance with lower computational complexity when solving various joint resource allocation problems in wireless communication systems compared to traditional convex relaxation algorithms.

An embodiment of the present invention presents a CNN-based joint optimization method for frequency subband allocation and transmission power control in an uplink IoT cellular network. In particular, unlike previous studies that considered similar subband allocation and power control problems, an embodiment of the present invention presented a control method that took into account spectrum leakage between subbands to reflect this.

[1] Republic of Korea Patent Publication No. 10-2196546 (drx and scheduling optimization method to reduce terminal power consumption in IoT environment)

[6] Y. Fu, L. Salan, C. W. Sung, and C. S. Chen, “Subcarrier and power allocation for the downlink of multicarrier NOMA systems”, IEEE Trans. Veh. Technol, vol. 67, no. 12, pp. 11833-11847, Dec. 2018. [7] D. Goswami and S. S. Das, “Iterative sub-band and power allocation in downlink multiband NOMA”, IEEE Syst. J., vol. 14, no. 4, pp. 5199-5209, Dec. 2020. [8] W. Lee, M. Kim, and D.-H. Cho, “Deep power control: Transmit power control scheme based on convolutional neural network”, IEEE Commun. Lett., vol. 22, no. 6, pp. 1276-1279, Jun. 2018. [9] M. Liu, T. Song, and G. Gui, “Deep cognitive perspective: Resource allocation for NOMA-based heterogeneous IoT with imperfect SIC”, IEEE Internet Things J., vol. 6, no. 2, pp. 2885-2894, Apr. 2019. [10] M. Lee, Y. Xiong, G. Yu, and G. Y. Li, “Deep neural networks for linear sum assignment problems”, IEEE Wireless Commun. Lett., vol. 7, no. 6, pp. 962-965, Dec. 2018. [11] T. V. Chien, T. N. Canh, E. Bjrnson, and E. G. Larsson, °Power control in cellular massive MIMO with varying user activity: A deep learning solution”, IEEE Trans. Wireless Commun, vol. 19, no. 9, pp. 5732-5748, 2020. [12] J. Kim, H. Lee, S.-E. Hong, and S.-H. Park, “Deep learning methods for universal MISO beamforming”, IEEE Wireless Commun. Lett., vol. 9, no. 11, pp. 1894-1898, Nov. 2020. [13] J. Luo, J. Tang, D. K. C. So, G. Chen, K. Cumanan, and J. A. Chambers, “A deep learning-based approach to power minimization in multi-carrier NOMA with SWIPT”, IEEE Access, vol. 7, pp. 17450-17460, 2019. [14] H. Zhang, H. Zhang, K. Long, and G. K. Karagiannidis, “Deep learning based radio resource management in NOMA networks: User association, subchannel and power allocation”, IEEE Trans. Netw. Sci. Eng., vol. 7, no. 4, pp. 2406-2415, Oct. 2020. [1] SH Chae, S.-W. Jeon, and C. Jeong, “Efcient resource allocation for IoT cellular networks in the presence of inter-band interference, IEEE Trans. Commun., vol. 67, no. 6, pp. 4299-4308, Jun. 2019. [2] J. A. Ansere, G. Han, L. Liu, Y. Peng, and M. Kamal, “Optimal resource allocation in energy-efficient Internet-of-Things networks with imperfect CSI.” IEEE Internet Things J., vol. 7, no. 6, pp. 5401-5411, Jun. 2020. [3] S. Boyd and L. Vandenberghe, “Convex Optimization”, Cambridge, UK: Cambridge Univ. Press, 2004. [4] D. Zhai, R. Zhang, L. Cai, B. Li, and Y. Jiang, “Energy-efficient user scheduling and power allocation for NOMA-based wireless networks with massive IoT devices,” IEEE Internet Things J, vol. . 5, no. 3, pp. 1857-1868, Jun. 2018. [4]

[6] Y. Fu, L. Salan, C.W. Sung, and C.S. Chen, “Subcarrier and power allocation for the downlink of multicarrier NOMA systems”, IEEE Trans. Veh. Technol, vol. 67, no. 12, pp. 11833-11847, Dec. 2018. [7] D. Goswami and SS Das, “Iterative sub-band and power allocation in downlink multiband NOMA”, IEEE Syst. J., vol. 14, no. 4, pp. 5199-5209, Dec. 2020. [8] W. Lee, M. Kim, and D.-H. Cho, “Deep power control: Transmit power control scheme based on convolutional neural network”, IEEE Commun. Lett., vol. 22, no. 6, pp. 1276-1279, Jun. 2018. [9] M. Liu, T. Song, and G. Gui, “Deep cognitive perspective: Resource allocation for NOMA-based heterogeneous IoT with imperfect SIC”, IEEE Internet Things J., vol. 6, no. 2, pp. 2885-2894, Apr. 2019. [10] M. Lee, Y. Xiong, G. Yu, and G. Y. Li, “Deep neural networks for linear sum assignment problems”, IEEE Wireless Commun. Lett., vol. 7, no. 6, pp. 962-965, Dec. 2018. [11] TV Chien, TN Canh, E. Bjrnson, and EG Larsson, °Power control in cellular massive MIMO with varying user activity: A deep learning solution”, IEEE Trans. Wireless Commun, vol. 19, no. 9, pp. 5732-5748, 2020. [12] J. Kim, H. Lee, S.-E. Hong, and S.-H. Park, “Deep learning methods for universal MISO beamforming”, IEEE Wireless Commun. Lett., vol. 9, no. 11, pp. 1894-1898, Nov. 2020. [13] J. Luo, J. Tang, DKC So, G. Chen, K. Cumanan, and J. A. Chambers, “A deep learning-based approach to power minimization in multi-carrier NOMA with SWIPT”, IEEE Access, vol. 7, pp. 17450-17460, 2019. [14] H. Zhang, H. Zhang, K. Long, and G.K. Karagiannidis, “Deep learning based radio resource management in NOMA networks: User association, subchannel and power allocation”, IEEE Trans. Netw. Sci. Eng., vol. 7, no. 4, pp. 2406-2415, Oct. 2020.

The purpose of the present invention is to provide a control method that maximizes the total transmission rate of the network by jointly optimizing subband allocation and transmission power control of IoT devices.

Another object of the present invention is to provide a method for subband allocation and transmission power control in an IoT cellular network system in which spectrum leakage exists.

The present invention is not limited to the purposes mentioned above, and other purposes not mentioned can be clearly understood from the description below.

According to one aspect of the present invention, in a method for an IoT cellular network system to control frequency subband allocation and transmission power of an IoT device, the system includes: a base station; And an IoT device connected to the base station through uplink communication; A control device that controls transmission power to the IoT device connected to the base station by communication, wherein the control device of the system allocates frequency subbands and transmits the IoT device. The method of controlling power is (a) for all IoT devices connected to the base station, setting the transmission power of each IoT device to the maximum and maximizing the total transmission rate against interference between subbands due to spectrum leakage. inferring a band allocation set; and (b) determining a transmission power vector that maximizes the total data rate based on the subband allocation set obtained in step (a).

In addition, step (a) includes inferring a subband allocation set (C) that maximizes P1 in Equation 5 below,

[Equation 5]

here ensures that all IoT devices have a maximum transmit power, e.g. , ), and C is the subband allocation matrix. )ego, represents the transmission rate of IoT devices considering interference between subbands due to spectrum leakage, and when there are K IoT devices, lim,

Step (b) includes determining the subband allocation set C obtained in (a) as C* , and then calculating the transmission power vector (P) that maximizes P2 in Equation 6 for the given C *. It is characterized by

[Equation 6]

In addition, step (a) is characterized by satisfying the condition that only one IoT device can be assigned to each subband, and each IoT device can be assigned to only one subband.

In addition, step (a) is characterized by learning a CNN model to infer a predicted value, and the learning method using the CNN model includes ① an IoT device matrix for random subband allocation and a random permutation ( ) is input as input data, and the subband allocation output of the CNN model ( ), select the candidate group of IoT devices in order of increasing probability to form N k -candidate sets. determining the order of IoT devices to be allocated to each subband in this manner; ② Selecting two subbands as input data for the learning model according to the above-described allocation order, and then assigning one IoT device from each candidate set - Here, if the same IoT device is selected for both subbands, it will be assigned to the previous subband Characterized by allocating the corresponding IoT device and not assigning the corresponding IoT device to the next subband; ③ In step ② above, calculating P1 and selecting an allocation set in which P1 is maximized, and assigning IoT devices to the corresponding subband accordingly - here, assuming that the transmission power of all IoT devices is maximum, and P1 Search for a subband allocation set that maximizes , and then include fixing to exclude assigned IoT devices from the remaining k-best set; ④ Fixing the IoT devices assigned in step ③, selecting the next subband, assigning IoT devices one by one from the corresponding k-best set, and finding the case where P1 maximizes; and ⑤ repeating steps ② - ④ up to the last subband to obtain a final subband allocation set; It is characterized by including.

In addition, if the same IoT device is selected for both subbands in step ②, the corresponding IoT device is assigned to the first subband, and the corresponding IoT device is not assigned to the second subband.

In addition, it is characterized by further including a process of excluding IoT devices assigned in step ③ from the remaining k-candidate set.

In addition, the CNN model includes a convolution unit, a flattening unit, and an activation unit. The convolution unit performs a convolution operation between high-dimensional input data and a convolution filter of the CNN model to extract spatial feature maps of the P1 and P2, The feature map extracted from the convolution unit is vectorized in the flattening unit to be input to the activation unit, and the output of the flattening unit is mapped and output using an activation function suitable for the optimization problem at the output end of the activation unit.

In addition, step (b) is characterized in that it is obtained by a CNN model-based learning method, and the CNN model-based learning method includes a standardized channel matrix from the subband allocation set obtained in step ⑤ (b-1). Constructing a learning data set into sets; and (b-2) a normalized power vector close to p* by learning from the CNN model to minimize the mean squared error of Equation 9 below A subband allocation and power control method comprising the step of learning to predict.

[Equation 9]

here here, represents the loss function of the CNN model, and are p* and the output vector, respectively It represents the ith element of .

In addition, in step (b-1) , p* is normalized to the range [0, 1], and then the output final activation function is set to the Sigmoid function.

Additionally, after step (b-1), the transmission power control vector is It is characterized by being determined as.

In addition, step (a) is characterized by inferring using an FNN model-based learning method, and the FNN model is characterized by learning the input and output relationship directly from a fully connected part, and the fully connected part (fully connected part) includes a plurality of hidden layers, each hidden layer is activated by the ReLU function, and the input data of the fully connected part is input by vectorizing a normalized channel matrix to obtain the P1. It is characterized by

According to an embodiment of the present invention, a control method for maximizing the total transmission rate of the network can be provided by jointly optimizing subband allocation and transmission power control of IoT devices in an IoT cellular network system where spectrum leakage exists.

The resource allocation method according to an embodiment of the present invention can reduce computational complexity by separating a non-convex integer programming problem that is difficult to solve directly into a two-step optimization problem.

Additionally, by applying the CNN model or FNN model in the proposed approach, performance can be greatly improved without increasing computational complexity.

Figure 1A shows an example of an IoT cellular network system according to an embodiment of the present invention.
Figure 1B shows an example of a control device according to an embodiment of the present invention.
Figure 2 shows the structure of a CNN model that finds the solution to optimal subband allocation and transmission power control according to an embodiment of the present invention.
Figure 3 shows a standardized channel matrix applied to an optimization problem according to an embodiment of the present invention.
Figure 4 shows an overall flowchart of a two-step optimization method for subband allocation and transmission power control according to an embodiment of the present invention.
Figure 5 shows the FNN structure according to an embodiment of the present invention.
Figure 6 shows the sum rates for K when N = 5 and ρ1 = -35dB.
Figure 7 shows the total transmission rate for N when K = 12 and ρ1 = -35dB.
Figure 8 shows the total transmission rate for ρ1 when N = 4 and K = 8.

The terms used in this application are only used to describe specific embodiments and are not intended to limit the invention. Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this application, when a part “includes” a certain component, this means that it may further include other components rather than excluding other components unless specifically stated to the contrary.

In addition, “…part”, “…gi”, “…” written in the specification. Terms such as “unit” and “…device” refer to a unit that processes at least one function or operation, which may be implemented as hardware, software, or a combination of hardware and software.

Additionally, when describing components of embodiments of the present invention, terms such as first and second may be used. These terms are only used to distinguish the component from other components, and the nature, order, or order of the component is not limited by the term. When a component is described as being 'connected', 'coupled' or 'connected' to another component, that component may be directly connected, coupled or connected to that other component, but that component and that other component It should be understood that another component may be 'connected', 'combined', or 'connected' between elements.

Hereinafter, a resource allocation method and system for maximizing the total transmission rate of an IoT cellular network according to the implementation of the present invention will be described in detail.

In this specification, bold lowercase letters (e.g. a, b, . . . ) are used to indicate vectors, and bold uppercase letters A, B, . . . Used to represent matrices. Calligraphy characters (e.g. X, Y , . . . ) are used to represent finite sets. The circularly symmetric complex Gaussian distribution with mean 0 and variance σ ² is It is defined as Also, [1:N] = {1, 2, . . . , N }.

One embodiment of the present invention provides a method for optimizing joint resource allocation in an uplink IoT cellular network where a base station uses multiple subbands to provide services to IoT users and there is interference between subbands due to spectrum leakage. .

For example, in a cellular IoT environment where spectrum leakage exists, a resource allocation method is presented to maximize the total transmission rate of devices in the network.

Since IoT devices are often composed of cheap circuits, the actual IoT transmission signal does not exist only in the frequency band (spectrum) that is the resource allocated to it, but also exists in some adjacent spectrum, causing interference. In the present invention, this phenomenon is defined as a spectral leakage phenomenon.

Figure 1A shows an example of an IoT cellular network system according to an embodiment of the present invention.

Referring to FIG. 1A, the IoT cellular network system according to an embodiment of the present invention includes a base station (BS) and IoT devices 10-1, 10-2, ... 10-K, each having one antenna, the same It includes an uplink IoT cellular network using a total of N subbands with bandwidth and a control device 900 that controls it.

The control device 900 according to an embodiment of the present invention is built into the base station (BS) or installed in a remote control room outside the base station (BS) and communicates with the base station (BS) and each IoT device (10-1) through a communication network. , 10-2, …10-K) communication resources and power are controlled.

According to an embodiment of the present invention, there are a total of K IoT devices and a total of N frequency subbands. In one embodiment of the present invention, it is assumed that IoT devices are transmitters and the base station is a receiver, performing uplink communication.

According to an embodiment of the present invention, IoT devices (10-1, 10-2, ...10-K) are allocated to one subband for communication with a base station (BS). (i.e., multiple users cannot be assigned to the same subband at the same time, i.e. OMA) IoT devices assigned to each subband transmit their data to the base station (BS).

The control device 900 according to an embodiment of the present invention may be implemented in a computer system, for example, as a computer-readable recording medium.

Figure 1b shows an example of a control device according to an embodiment of the present invention.

As shown in FIG. 1B, the control device 900 includes at least one of one or more processors 910, a memory 920, a storage unit 930, a user interface input unit 940, and a user interface output unit 950. elements, which can communicate with each other via bus 960. Additionally, the computer system 900 may also include a network interface 970 for accessing a network to which a base station (BS) and IoT devices 10-1, 10-2,...10-K are connected. The processor 910 may be a CPU or a semiconductor device that executes processing instructions stored in the memory 920 and/or the storage unit 930. The memory 920 and storage unit 930 may include various types of volatile/non-volatile storage media. For example, memory may include ROM 924 and RAM 925.

Accordingly, embodiments of the present invention may be implemented using a computer-implemented method or a non-volatile computer recording medium storing computer-executable instructions. When executed by a processor, the instructions may perform a method according to at least one embodiment of the present invention.

For example, when the instructions are executed by the processor 910, the subband allocation and transmission power calculation processing unit 980 calculates the optimal subband allocation according to a programmed algorithm, and the deep learning processing unit 990 In conjunction with , the optimal transmission power can be calculated. The deep learning processing unit 990 has a built-in deep learning program and learns according to a programmed algorithm to find the optimal solution for subband allocation method and transmission power control for each IoT device.

According to an embodiment of the present invention, OMA communication is considered and it is taken into account that interference between subbands may occur. In one embodiment of the present invention, it is assumed that the base station (BS) and all IoT devices each have one antenna. Additionally, it is assumed that each IoT device can transmit data by accessing only one of N subbands and that OMA (orthogonal multiple access), that is, multiple IoT devices cannot be assigned to the same frequency band.

In one embodiment of the present invention and It represents the set of all IoT devices and the set of all subbands, respectively.

also, is a user go It is defined as indicating whether it has been assigned to . in other words, means that IoT device k is assigned to subband n, otherwise am.

for your convenience Expressed as the following binary indicator, if subband m is assigned to user device i, , Otherwise , At this time and am.

Due to the above assumptions must satisfy the two constraints of Equation 1 below.

[Equation 1]

IoT devices a subband Assuming it is assigned to is a signal transmitted by the IoT device (i) at the indicated time (t), and can be expressed as [Equation 2] below.

[Equation 2]

here is an independent information symbol of IoT device i at time t, It indicates that and, Is It represents the transmission power related to the power constraint ( ) is satisfied.

In one embodiment of the present invention, it is assumed that spectral leakage between frequency subbands occurs due to out-of-band emissions from IoT devices. Channel coefficients from subband m to subband n of IoT device i It is denoted by and is assumed to remain fixed during communication and to be known to the base station (BS). This takes into account that in the case of IoT communication, the size of the data packet is small, so the time it takes to transmit the data packet is quite short.

Additionally, in one embodiment of the present invention, since an uplink communication scenario is considered, the base station (receiver side) can easily estimate channel state information (CSI) by receiving a pilot signal from an IoT device.

then Subband at time t denoted by The received signal from the base station is

is given as

here is the additive noise in subband n.

[Equation 3]

Here, Ri represents the transmission rate of the IoT device, and the equation on the right represents subband interference due to spectrum leakage.

Remark 1: For all n ≠m and About By setting , the channel model according to an embodiment of the present invention can recover the existing scenario assuming that there is no spectrum leakage between subbands.

Also constraint(1) By eliminating NOMA and allowing one IoT device to access multiple subbands, it can be expanded to more general scenarios.

The purpose of an embodiment of the present invention is to provide a method for maximizing the total transmission rate of a network by jointly optimizing subband allocation and transmission power control of IoT devices.

In one embodiment of the present invention, since a cellular IoT network is considered, the base station can centrally manage subband allocation and transmission power control for each IoT device using CSI.)

In the specification of the present invention, the subband allocation matrix The (m, i)th element of It is defined as

Additionally, the power control vector whose ith element is pi is It is defined as In one embodiment of the present invention, a control method for maximizing the total transmission rate through frequency subband allocation and transmission power control of IoT devices can be formulated as [Equation 4].

[Equation 4]

here is given by [Equation 3] for fixed C and p.

In Equation 4, (4a) is about maximizing the total transmission rate of IoT devices in the network. Additionally, (4b) controls so that only one IoT device can be assigned to each subband.

Additionally, (4c) controls so that each IoT device can be assigned to only one subband. Additionally, (4d) represents the power limitation conditions for each IoT device.

The optimization problem (Equation 4) is a non-convex integer programming problem due to the nature of subband allocation and the existence of spectral leakage between subbands. Therefore, it is difficult to solve easily with existing optimization methods using convex relaxation strategies, so finding a solution may take a considerable amount of time due to computational complexity.

In order to overcome this problem, an embodiment of the present invention proposes a new optimization method based on deep learning that can significantly improve the total transmission rate for finding a solution with low computational complexity.

[Deep learning-based algorithm proposal]

In one embodiment of the present invention, a two-step optimization method is proposed in which subband allocation and transmission power control are sequentially performed to reduce the computational complexity of solving (Equation 4).

Specifically, assuming that only on-off (on-ff) power control is applied, first solve the subband allocation problem P1 formulated as follows.

The first step of optimization according to an embodiment of the present invention is to assume the maximum transmission power of all IoT devices and infer a subband allocation set that maximizes P1.

[Equation 5]

here ensures that every IoT device has a maximum transmit power, e.g. , ) represents the power control vector when having . Then, by solving P1, determining C to be C* , and then solving the following optimization problem P2 for a given C *, the appropriate power control vector can be found.

[Equation 6]

In one embodiment of the present invention, supervised learning-based deep learning is applied to the optimization problems P1 and P2. In the following, we describe the detailed steps of the proposed approach.

[Structure of CNN]

According to an embodiment of the present invention, the spatial features of equations (5) and (6) can be effectively extracted using CNN.

Figure 2 shows the structure of a CNN model that finds the solution to optimal subband allocation and transmission power control according to an embodiment of the present invention.

Referring to FIG. 2, the structure of a CNN model according to an embodiment of the present invention includes three parts: a convolution part, a flattening part, and an activation part.

According to an embodiment of the present invention, the convolution unit extracts spatial feature maps of P1 and P2 by performing a convolution operation between high-dimensional input data and the convolution filter of the CNN model. Then, the feature map extracted from the convolution part is vectorized in the flattening part to be input to the activation part. Then, the output of the flattening part is mapped using an activation function appropriate for each optimization problem at the output of the activation part, allowing the model to effectively infer executable outputs.

Specifically, the convolution unit according to an embodiment of the present invention is composed of N _c connection layers. For each layer, the size of the filter is set to 3×3 and the depth of the filter is set to F _c . The output of the convolution part is flattened for final activation of the flattening part. Finally, the output of the Flattening section is supplied to the Activation section.

Unlike existing CNN-based operations (see [Non-patent Document 8] and [Non-patent Document 11]), the CNN learning method according to an embodiment of the present invention provides an additional fully connected layer to reduce complexity in the CNN structure. A technical feature is that the layers are not connected after the convolution part.

If fully connected layers exist, model complexity can be increased by increasing the number of model parameters. In one embodiment of the present invention, optimal performance can be obtained without an additional fully connected layer, thereby increasing the efficiency of deep learning learning.

According to one embodiment of the present invention, the CNN structures used to solve P1 and P2 are different only in the Activation part, and the remaining parts are the same.

Specifically, in the case of P1, the final output is characterized by being activated by the softmax function for multi-class classification. That is, when the input of the Activation unit is given as x, the ith element of the final output is It becomes.

Here, xi represents the ith element of x.

In the deep learning model used in P1, the model's output represents the probability value of IoT devices that can be assigned to each subband. Therefore, the softmax function is connected to the activation part to enable such multi-class classification. This is to ensure that the sum of all elements of the final output vector is 1 because the final predicted value in the solution according to an embodiment of the present invention is a probability value. According to an embodiment of the present invention, since it is assumed that only one IoT device is allocated to each subband, the problem can be solved by constructing N models that predict which IoT device will be allocated to each subband. →For example, replacing the original problem with N multiple classification problems.

Therefore, according to an embodiment of the present invention, the probability of being assigned to a frequency subband for each IoT device can be predicted by receiving the channel between the base station and the IoT device as input. Therefore, by utilizing the predicted values of the deep learning model, the integer constraints of the problem to be solved can be effectively handled.

Meanwhile, in the case of P2, the sigmoid function is used in the activation part to more easily satisfy the transmission power constraint (4d). That is, the ith element of the output vector is is set to .

More specifically, the deep learning model applied to P2 must predict the normalized optimal power control vector, so it is characterized by using the sigmoid function that maps the output value to the range [0, 1] as an activation function.

[Deep learning-based subband allocation method]

Figure 3 shows a standardized channel matrix for an optimization problem according to an embodiment of the present invention.

To generate a data set for training, we first search for each channel implementation to find the optimal subband allocation matrix C* that maximizes equation (5), and the power control vector is It is fixed with .

Then the normalized channel matrix in Figure 4(a) A data set is created by .

Here in Figure 3(a) is given by [Equation 7] below.

[Equation 7]

Here, this normalization method for input data according to an embodiment of the present invention ensures that the learning process converges quickly and is robust against the problem of falling to a local optimal solution when applying gradient descent to update model parameters. You can make it.

According to one embodiment of the present invention, due to the constraint equation (4b), the original problem is N classification problems by considering each column of C* as a one-hot target vector in which only one of the elements is 1 and all remaining elements are 0. can be decomposed into

Additionally, in one embodiment of the present invention, a step of adding a vector with all 0 columns to the left of C* is performed to apply one-hot encoding even when an IoT device is not assigned to a given subband. Therefore, the optimal subband allocation matrix is now It becomes. Then, the process of constructing N distributed models is performed to find an appropriate subband allocation matrix through C* -based learning.

Specifically, each model are individually learned to minimize the loss function [Equation 8], which is a categorical cross-entropy widely adopted in multi-class classification problems.

[Equation 8]

here is a one-hot target vector, represents the ith element of , where is the mth row vector of C* , is the output vector, Represents the jth element of . This is the model's prediction vector updated at the end of every training epoch in the minimization process of Equation 8.

Setting the loss function as in Equation (8) can effectively represent the probability that the output of the learned model is assigned to a specific integer class, and as a result, the integer constraint of (Equation 4) can be effectively considered in the optimization process.

However, there is no guarantee that the resulting subband allocation matrix obtained from the aforementioned decomposition approach satisfies equation (4c). This is because each model is trained independently, so the inference of each model may be the same in some cases, so one IoT device can be assigned to multiple subbands at the same time. In one embodiment of the present invention, to solve this problem, a sequential optimization algorithm based on the k-best candidate algorithm is applied to find the optimal k candidates in the search space and control computational complexity by adjusting k.

The operation sequence of the algorithm according to an embodiment of the present invention is as follows.

① IoT device matrix and random permutation for random subband allocation ( ) is input as input data, and the subband allocation output of the CNN model ( ) Based on this, a candidate group of IoT devices is selected in order of increasing probability to form a set of N k-best candidates. In this way, the order of IoT devices to be assigned to each subband is determined.

② As input data for the learning model, select two subbands according to the above-described allocation order, and then assign one IoT device from each candidate set. At this time, if the same IoT device is selected for both subbands, the corresponding IoT device is assigned to the previous subband, and the corresponding IoT device is not assigned to the next subband.

③ In step ② above, P1 is calculated to select an allocation set where P1 is maximized, and IoT devices are assigned to the corresponding subband accordingly (in this step, the transmission power of all IoT devices is assumed to be maximum, and P1 is maximized) (searches for a set of subband allocations). Afterwards, the assigned IoT devices are excluded from the remaining k-best candidate set.

④ Fix the IoT devices assigned in step ③, select the next subband, assign IoT devices one by one from the corresponding k-best candidate set, and find the case where P1 maximizes.

⑤ Steps ② - ④ are repeated up to the last ordered subband to obtain the final subband allocation set.

By performing the above algorithm for all possible orders (!), we obtain a set of subband allocations that maximize P1 among them.

Table 1 below shows the proposed sequential algorithm for subband allocation for a given s in code.

The operation sequence of the algorithm described above is specifically, the output vector Using the fact that each element of represents the probability of an IoT device being assigned to the corresponding subband m, the probabilities of k candidate IoT devices are selected in descending order, and the selected IoT devices are sequentially optimized to satisfy Equation (4c). Perform learning.

According to one embodiment of the present invention, the set The set of all possible permutations of It is displayed as also The nth element of It is expressed as given permutation and For a specific example, the k-best algorithm according to an embodiment of the present invention is performed in the following step method.

One) Is A modified version of represents.

2) k -best candidate IoT device matrix It is defined as is called the (i, j)th element of C. also and respectively and It is called the ith row vector of .

3) C all zero matrices, n = 1 and Initialize with here silver represents the set of positions of 1. for example, If = [1, 0, 1, 0, 1, 1] = {1, 3, 5, 6}.

4) For a given n, subband allocation is performed sequentially on subbands S _n and S _{n = 1} . The first column of indicates whether an IoT device is assigned to each subband.

in other words, means that the device is not assigned to subband s _n . in this case Set to .

Otherwise About By setting it to Select one of the candidate devices from and assign it to subband s _n .

5) Assign candidate IoT devices one by one. Similar to step 4 If Set . Additionally, if an IoT device is already assigned to subband s _n , the device is not assigned to subband s _n+1 . for example, If Set . here and am.

Also, for the next iteration, Add {1} to . Otherwise About Set . Then all and About, Calculate .

6) From the previous step Find the (i, j) pair that maximizes j and assign IoT devices to subband s _n and subband s _n+1 accordingly. We then exclude these assigned IoT devices from the remaining set of candidate IoT devices. This is because IoT devices already allocated in the previous step can no longer be allocated to other subbands to satisfy Equation (4c).

7) Next set n to n+1 and n = Repeat steps 5) - 6) above until this is achieved.

8) Returns the final result for the subband allocation matrix C.

Finally, Algorithm 1 can be performed for all possible s to obtain the optimal subband allocation matrix C that maximizes Equation (5).

Remark 3: The computational complexity of Algorithm 1 is determined by k-factor and N. given permutation Since the size of the set of candidate IoT devices is proportional to the k-factor, It has complexity. Also, the entire algorithm is Repeat for. As a result, the overall computational complexity of Algorithm 1 considering all possible permutations is provided by .

[Deep learning-based transmission power allocation method]

According to an embodiment of the present invention, after solving P1 by [deep learning-based subband allocation method], a method of solving the transmission power control problem P2 to maximize the total data rate for the determined subband allocation matrix is performed. do.

The transmission power allocation method according to an embodiment of the present invention is characterized by configuring a supervised learning-based CNN (Convolutional Neural Network) model and performing optimization.

Since the subband allocation matrix is determined in the previous step, it is a standardized channel matrix for transmission power control. is given as in Figure 3(b). here is the subband in the previous step Indicates the index of IoT devices assigned to .

To construct a training data set, subband assignment to randomly generated channels is performed once again to obtain a set of standardized channel matrices.

For the learning target, the optimal power vector that maximizes equation (6) is searched exhaustively in the power range (4d) with sufficiently small quantization levels, then the range It is normalized in .

As a result, the data set for training is a normalized channel matrix and the normalized optimal power vector. It consists of

After the training dataset is constructed, the power control model is trained to minimize the mean square error (MSE) of Equation 9, which is the loss function.

[Equation 9]

here represents the loss function of the CNN model. and are p* and the output vector, respectively Represents the ith element of .

The final output of the model for P2 optimization is activated by the sigmoid function, so that equation (9) is minimized to obtain the normalized power vector closest to p* A model can be trained to predict. The closest to the optimal transmission power vector p* according to the above learning can be predicted.

According to an embodiment of the present invention, in the CNN model training process, p* is normalized to the range [0, 1], and then the final activation function of the proposed model is set to the sigmoid function to obtain the maximum transmission power of the problem to be solved. The conditions can be satisfied effectively.

Finally, the transmission power control vector is calculated after the learning process. is decided.

The overall procedure of the above-described optimization method is summarized in Figure 4.

Figure 4 shows an overall flowchart of a two-step optimization method for subband allocation and transmission power control according to an embodiment of the present invention.

[FNN-based proposed algorithm]

What was explained above is a resource allocation method based on CNN.

In another embodiment of the present invention, it can also be implemented through FNN.

Unlike the CNN model, which learns by extracting local information between adjacent elements of the input data in the convolutional part, the FNN model is different in that it learns the input and output relationship directly from the fully connected part without an extraction process.

The FNN-based method according to an embodiment of the present invention can effectively analyze the input and output relationship with a simpler model structure than the CNN-based method.

To avoid duplication of explanation, we will focus on the modifications introduced by FNN.

Figure 5 shows the FNN structure according to an embodiment of the present invention.

Referring to Figure 5, the FNN model according to an embodiment of the present invention has Nc hidden layers like the CNN model, and each layer includes 64 neurons. Each hidden layer is activated by the ReLU function, and the activation part is the same as the CNN model.

Additionally, the FNN-based approach solves subband allocation and transmission power control problems sequentially and individually by applying the FNN model instead of the CNN model.

On the other hand, since data in the form of a matrix cannot be input as an input to the FNN structure, the normalized channel matrix of Figure 3 must be vectorized before entering the FNN structure as shown in Figure 5. The remaining parts are applied in the same way as the CNN-based method.

[Numerical evaluation]

The total transmission rate was numerically evaluated for the optimization method according to an embodiment of the present invention.

The bandwidth of each subband is set to 15kHz, σ ² = -168dBm/Hz, and P _max = 23dBm. Assume: Additionally, the radius of cell D is set to D[km] = 0.1, the base station is located in the center of the cell, and IoT devices are located randomly and uniformly within the cell.

The base station and IoT device i, denoted by di[km], are all About It's within range.

The path loss of IoT device i is provided by .

As shown in Equation 10: Assume the spectral leakage rate from subband m to subband n, denoted by .

[Equation 10]

where ρ1 is specified later, ρ2 = ρ1 - 20 dB, ρ3 = ρ1 - 30 dB, In the case of am. Then the channel coefficient go is assumed to follow.

Table 2 shows the relative total throughput performance for simulation time per sample (seconds) and k-factor of ( N, K ) = (5, 12).

Table 3 shows the learning time (seconds) and relative total transfer rate performance for the ( N, K ) = (4, 8) model structure.

In Table 2, the relative total transmission rate of the proposed algorithm is expressed as a percentage according to the k-factor. It was analyzed that Algorithm 1 provides a good balance between computational complexity and total throughput performance and can achieve close to optimal performance when k is sufficiently large (i.e., k ≥ 6).

Since the total transmission rate almost converges at k = 6, in one embodiment of the present invention, k-factor is set to 6 for the entire simulation.

Additionally, the FNN-based method and the CNN-based method are compared in Table 3. Referring to Table 3, it can be seen that the FNN model can be trained with shorter training time than the CNN model. However, because there is no spatial feature extraction during the learning process, the FNN-based method was analyzed to have lower performance than the CNN-based method.

Additionally, the CNN model was generally found to be more suitable for overfitting problems because it has significantly fewer model parameters than the FNN model.

The following benchmark schemes were considered for comparison:

- The global search method maximizes Equation (4) by considering all possible combinations of the subband allocation matrix C and the discretized power control vector p with a sufficiently small quantization level.

- The two-step global search method first exhaustively finds the best subband allocation matrix for problem P1. Then, given the subband allocation matrix, we exhaustively find the optimal p for problem P2.

- The WMMSE PC (Weighted Minimum Mean Square Error Power Control) method determines an appropriate subband allocation matrix in the same way as the proposed method. On the other hand, the power control vector is obtained by MMSE-based iterative calculation instead of a learning process for a given subband allocation matrix.

- Consider a no-power control (No PC) method in which each IoT device can only perform on-off power control. As in the WMMSE PC method, the subband allocation matrix is selected in the same manner as the proposed method.

- The simplified subband allocation method can simply assign the IoT device with the largest channel gain to each subband without a learning process and without considering inter-band interference due to spectrum leakage. To satisfy the conditions of Equation (4c), IoT devices are sequentially assigned to each subband using Algorithm 1. For a given subband allocation, the power control vector is obtained using the same method as the proposed method.

- Consider a convex relaxation method based on the existing Hungarian algorithm [see (non-patent document 1)].

The average total transmission rates of the considered schemes for K, N and ρ1 are shown in Figures 6, 7 and 8, respectively.

Figure 6 shows the sum rates for K when N = 5 and ρ1 = -35dB.

Figure 7 shows the total transmission rate for N when K = 12 and ρ 1 = -35 dB.

Figure 8 shows the total transmission rate for ρ1 when N = 4 and K = 8.

Overall, the results show that the proposed method using CNN provides performance close to the full search or two-stage full search method despite much lower complexity, and performs much better than the method of the prior art [see (Non-Patent Document 1)]. It shows that it provides.

Specifically, it can be observed that using the subband allocation algorithm proposed in Figure 6, the total data rate of the CNN-based method increases with K in contrast to the simplified subband allocation.

References due to spectral leakage were appropriately taken into account during the learning process.

Additionally, the proposed power control algorithm was shown to significantly improve the total transfer rate compared to the WMMSE PC or No PC methods. However, it was observed that the performance of the FNN-based method did not improve as K increased. This is analyzed to be because the depth of the input channel matrix increases in proportion to K.

More specifically, as the depth of the input channel matrix increases, there are more spatial features that are not reflected in the learning process of the FNN structure, leading to performance degradation. On the other hand, CNN-based methods can improve total throughput performance as K increases because more hidden spatial features of the input channel matrix can be learned in the learning stage even if K increases.

Additionally, it was confirmed in Figure 8 that the proposed two-stage optimization method can achieve performance close to full search, which is an exhaustive joint optimizer.

According to one embodiment of the present invention, the total transmission rate of all considered schemes increases with N as shown in Figure 7 and decreases as spectrum leakage becomes more severe as shown in Figure 8.

Additionally, the simulation times required for some selected cases are summarized in Table 4.

In Table 4 Indicates the simulation time (seconds) for 100 samples.

Simulations according to one embodiment of the present invention were performed in Python 3.8 with Tensorflow 2.5.0, Intel i7-10700F CPU, 32GB RAM, and NVIDIA GeForce RTX 3060 Ti GPU.

Referring to Table 4, it can be seen that the method proposed according to an embodiment of the present invention has much lower computational complexity than the two-stage full search and has a similar level of computational complexity to the previous convex relaxation method.

In addition, the simulation time difference between the FNN-based method and the CNN-based method is negligible, while the CNN-based method has much better performance than the FNN-based method, so it is analyzed that CNN, the proposed method that considers both performance and complexity, is a more efficient deep learning structure. do.

According to an embodiment of the present invention, a subband allocation and transmission power control method and a deep learning-based algorithm method for sequentially optimizing the same are presented to maximize the total transmission rate of an uplink IoT cellular network in the case of spectrum leakage.

According to one embodiment of the present invention, the proposed algorithm can be extended to NOMA and applied to analyze additional benefits.

In addition, it is also possible in the future to extend and apply the subband allocation and transmission power control method according to an embodiment of the present invention to the case where the base station has multiple antennas.

10-1, 10-2, … 10-K: IoT devices
900: Control device
BS: base station

Claims (11)

In the method of the IoT cellular network system controlling frequency subband allocation and transmission power of IoT devices,
The system is,
base station;
An IoT device connected to the base station through uplink communication; and
It includes a control device that controls transmission power for IoT devices connected to the base station through communication,
The method by which the control device of the system controls frequency subband allocation and transmission power of IoT devices is,
(a) For all IoT devices connected to the base station, inferring a subband allocation set that maximizes the total transmission rate against interference between subbands due to spectrum leakage while setting the transmission power of each IoT device to the maximum. ; and
(b) A subband allocation and transmission power control method comprising the step of obtaining a transmission power vector that maximizes the total transmission rate based on the subband allocation set obtained in step (a). According to paragraph 1,
The step (a) is
It includes the step of inferring a subband allocation set (C) that maximizes P1 in Equation 5 below,
[Equation 5]

here ensures that all IoT devices have a maximum transmit power, e.g. , ), and C is the subband allocation matrix. )ego, represents the transmission rate of IoT devices considering interference between subbands due to spectrum leakage, and when there are K IoT devices, lim,
Step (b) includes determining the subband allocation set C obtained in (a) as C* , and then calculating the transmission power vector (P) that maximizes P2 in Equation 6 for the given C *. A subband allocation and transmission power control method characterized by:
[Equation 6]
According to paragraph 2,
In step (a), only one IoT device can be assigned to each subband, and each IoT device satisfies the condition that only one subband can be assigned. According to paragraph 2,
Step (a) is characterized by learning a CNN model to infer a predicted value,
The learning method using the CNN model is,
① IoT device matrix and random permutation for random subband allocation ( ) is input as input data, and the subband allocation output of the CNN model ( ), determining the order of IoT devices to be assigned to each subband by selecting a candidate group of k IoT devices in order of increasing probability to form N k -candidate sets;
② Selecting two subbands as input data for the learning model according to the above-described allocation order, and then assigning one IoT device from each candidate set - Here, if the same IoT device is selected for both subbands, it will be assigned to the previous subband Characterized by allocating the corresponding IoT device and not assigning the corresponding IoT device to the next subband;
③ In step ② above, calculating P1 and selecting an allocation set in which P1 is maximized, and allocating IoT devices to the corresponding subband accordingly - here, assuming that the transmission power of all IoT devices is maximum, Search for a set of subband allocations that maximize P1, and then include fixing to exclude assigned IoT devices from the remaining k -candidate set;
④ Fixing the IoT devices allocated in step ③, selecting the next subband, assigning IoT devices one by one from the corresponding k -candidate set, and finding the case where P1 maximizes; and
⑤ Repeating steps ② - ④ above to the last subband in order to obtain a final subband allocation set; A subband allocation and transmission power control method comprising: According to paragraph 4,
In step ②, if the same IoT device is selected in both subbands, the corresponding IoT device is assigned to the previous subband, and the corresponding IoT device is not assigned to the next subband. Subband allocation and transmission power control method. According to paragraph 4,
A subband allocation and transmission power control method further comprising the step of excluding IoT devices allocated in step ③ from the remaining k-candidate set. According to paragraph 4,
The CNN model is,
It includes a convolution unit, a flattening unit, and an activation unit. The convolution unit performs a convolution operation between high-dimensional input data and the convolution filter of the CNN model to extract spatial feature maps of P1 and P2, and the feature maps extracted from the convolution unit are A method for subband allocation and transmission power control, characterized in that it is vectorized in the flattening unit for input to the activation unit, and the output of the flattening unit is mapped and output at the output end of the activation unit using an activation function suitable for the optimization problem. According to clause 4,
Step (b) is characterized in that it is obtained by a CNN model-based learning method,
The learning method is,
(b-1) constructing a learning data set with a standardized channel matrix set from the subband allocation set obtained in step ⑤; and
(b-2) Normalized power vector close to p* by learning from the CNN model to minimize the mean squared error of Equation 9 below A subband allocation and power control method comprising the step of learning to predict.
[Equation 9]

here here, represents the loss function of the CNN model, and are p* and the output vector, respectively It represents the ith element of .
According to clause 8,
A subband allocation and transmission power control method characterized by normalizing p* to the range [0, 1] in step (b-1) and then setting the output final activation function to the Sigmoid function. According to clause 8,
After step (b-1), the transmission power control vector is A subband allocation and transmission power control method, characterized in that determined. According to paragraph 2,
Step (a) is characterized by inference using an FNN model-based learning method,
The FNN model is characterized by learning the input and output relationship directly from a fully connected part,
The fully connected part includes multiple hidden layers, and each hidden layer is activated by the ReLU function,
A subband allocation and transmission power control method, characterized in that the input data of the fully connected part is input by vectorizing a standardized channel matrix to obtain the P1.