KR20230058400A - Federated learning method based on selective weight transmission and its terminal - Google Patents

Abstract

A federated learning method based on selective weight transmission and a terminal thereof are disclosed. A federated learning method according to an embodiment of the present specification includes the steps of changing a first parameter set associated with a local model to a second parameter set based on a local model in a federated learning method of a terminal in a wireless communication system; 1, selecting at least a part of the second parameter set based on a difference between the second parameter sets, and transmitting the selected at least part to a server. Accordingly, the terminal can save a frequency band allocated to parameter transmission for joint learning. The terminal and server of the present specification are artificial intelligence (Artificial Intelligence) modules, drones (Unmanned Aerial Vehicles, UAVs), robots, augmented reality (AR) devices, virtual reality (VR) devices, 5G / 6G services It can be associated with a device related to.

Description

Federated learning method based on selective weight transmission and its terminal

The present specification relates to a federated learning method based on selective weight transmission and a terminal thereof, and more specifically, to selective weight transmission capable of selectively transmitting some of the changed weights by comparing weights changed through a local model with pre-change weights. It relates to a federated learning method and its terminal.

A wireless communication system is widely deployed to provide various types of communication services such as voice and data. In general, a wireless communication system is a multiple access system capable of supporting communication with multiple users by sharing available system resources (bandwidth, transmission power, etc.). Examples of the multiple access system include a Code Division Multiple Access (CDMA) system, a Frequency Division Multiple Access (FDMA) system, a Time Division Multiple Access (TDMA) system, a Space Division Multiple Access (SDMA) system, and an Orthogonal Frequency Division Multiple Access (OFDMA) system. , SC-FDMA (Single Carrier Frequency Division Multiple Access) system, IDMA (Interleave Division Multiple Access) system, and the like.

This specification aims to address the aforementioned needs and/or problems.

In addition, an object of the present specification is to implement a joint learning method and terminal thereof based on selective weight transmission that saves an uplink frequency band by selectively transmitting some of the parameters for joint learning.

In addition, the present specification provides a federated learning method based on selective weight transmission that provides a non-centralized global model by receiving and accumulating decentralized data from two or more client devices and performing federated learning with the data, and a terminal thereof It aims to implement

A method for joint learning of a terminal in a wireless communication system according to an aspect of the present specification includes changing a first parameter set associated with a local model to a second parameter set based on a local model, and selecting at least some of the second parameter set based on the difference and transmitting the selected at least some to a server.

The method may further include obtaining one or more pieces of raw data and extracting one or more pieces of feature information based on the one or more pieces of raw data, wherein the changing the one or more pieces of feature information and one or more labels associated with the feature information can be performed based on

The method further includes allocating a label to at least some of the one or more pieces of characteristic information, wherein the allocating of the label comprises a command associated with the control operation in response to receiving an input related to the control operation of the terminal. may be assigned as a label for the at least part of the feature information.

The method further includes allocating a label to at least some of the one or more pieces of feature information, wherein the extracting is performed in response to receiving an input related to a control operation of the terminal, and allocating the label may allocate a command associated with the control operation as a label for the at least part of the feature information.

Also, in the selecting step, at least some parameters in which a difference between the first and second parameter sets exceeds a set reference value may be selected.

Also, the reference value may be 0.

Also, at least some of the parameters transmitted to the base station may be for learning a global model stored in the server.

A method for federated learning of a server in a wireless communication system according to another aspect of the present specification includes the steps of receiving first parameter sets for learning a global model from two or more terminals, the average of each element of the received first parameter sets (element -wise averaging) to change at least some of the parameter sets of the global model into a new second parameter set, and transmit the changed third parameter set to two or more terminals.

A joint learning method based on selective weight transmission according to an embodiment of the present specification and effects of the terminal are described as follows.

This specification can save uplink frequency bands by selectively transmitting some of the parameters for joint learning.

In addition, the present specification may provide a non-centralized global model by receiving and accumulating decentralized data from two or more client devices and performing federated learning with the data.

Effects obtainable in the present specification are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below. .

1 illustrates physical channels and typical signal transmission used in a 3GPP system.
2 is a diagram showing an example of a communication structure that can be provided in a 6G system.
3 illustrates the perceptron structure.
4 illustrates a multilayer perceptron structure.
5 illustrates a deep neural network structure.
6 illustrates a convolutional neural network structure.
7 illustrates a filter operation in a convolutional neural network.
8 illustrates a neural network structure in which a cyclic loop exists.
9 illustrates an operating structure of a recurrent neural network.
10 shows an example of an electromagnetic spectrum.
11 shows an example of a THz communication application.
12 shows an example of an electronic element-based THz wireless communication transceiver.
13 shows an example of a method of generating an optical element-based THz signal.
14 shows an example of an optical element-based THz wireless communication transceiver.
15 illustrates a photonic source based transmitter structure.
16 illustrates the structure of an optical modulator.
17 is an exemplary diagram of a wireless communication system to which various embodiments of the present specification are applied.
18 is a flowchart of a federated learning method according to the first embodiment of the present specification.
19 is a flowchart of a federated learning method according to a second embodiment of the present specification.
20 is a federated learning method according to the third embodiment of the present specification.
21 is a federated learning method according to the fourth embodiment of the present specification.
22 is an implementation according to various embodiments of the present specification.
23 and 24 are average calculation methods of reception parameter sets applied to various embodiments of the present specification.
25 illustrates a communication system applied to the present invention.
26 illustrates a wireless device applicable to the present invention.
27 illustrates a signal processing circuit for a transmission signal.
28 shows another example of a wireless device applied to the present invention.
29 illustrates a portable device applied to the present invention.
30 illustrates a vehicle or autonomous vehicle to which the present invention is applied.
31 illustrates a vehicle to which the present invention is applied.
32 illustrates an XR device applied to the present invention.
33 illustrates a robot applied to the present invention.
34 illustrates an AI device applied to the present invention.
The accompanying drawings, which are included as part of the detailed description to aid understanding of the present specification, provide examples of the present specification and describe technical features of the present specification together with the detailed description.

Hereinafter, the embodiments disclosed in this specification will be described in detail with reference to the accompanying drawings, but the same or similar components are given the same reference numerals regardless of reference numerals, and redundant description thereof will be omitted. The suffixes "module" and "unit" for components used in the following description are given or used together in consideration of ease of writing the specification, and do not have meanings or roles that are distinct from each other by themselves. In addition, in describing the embodiments disclosed in this specification, if it is determined that a detailed description of a related known technology may obscure the gist of the embodiment disclosed in this specification, the detailed description thereof will be omitted. In addition, the accompanying drawings are only for easy understanding of the embodiments disclosed in this specification, the technical idea disclosed in this specification is not limited by the accompanying drawings, and all changes included in the spirit and technical scope of this specification , it should be understood to include equivalents or substitutes.

Terms including ordinal numbers, such as first and second, may be used to describe various components, but the components are not limited by the terms. These terms are only used for the purpose of distinguishing one component from another.

It is understood that when an element is referred to as being "connected" or "connected" to another element, it may be directly connected or connected to the other element, but other elements may exist in the middle. It should be. On the other hand, when an element is referred to as “directly connected” or “directly connected” to another element, it should be understood that no other element exists in the middle.

Singular expressions include plural expressions unless the context clearly dictates otherwise.

In this application, terms such as "comprise" or "have" are intended to designate that there is a feature, number, step, operation, component, part, or combination thereof described in the specification, but one or more other features It should be understood that the presence or addition of numbers, steps, operations, components, parts, or combinations thereof is not precluded.

The following techniques may be used in various radio access systems such as CDMA, FDMA, TDMA, OFDMA, SC-FDMA, and the like. CDMA may be implemented with a radio technology such as Universal Terrestrial Radio Access (UTRA) or CDMA2000. TDMA may be implemented with a radio technology such as Global System for Mobile communications (GSM)/General Packet Radio Service (GPRS)/Enhanced Data Rates for GSM Evolution (EDGE). OFDMA may be implemented with radio technologies such as IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802-20, and Evolved UTRA (E-UTRA). UTRA is part of the Universal Mobile Telecommunications System (UMTS). 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) is a part of Evolved UMTS (E-UMTS) using E-UTRA, and LTE-A (Advanced) / LTE-A pro is an evolved version of 3GPP LTE. 3GPP NR (New Radio or New Radio Access Technology) is an evolved version of 3GPP LTE/LTE-A/LTE-A pro. 3GPP 6G may be an evolved version of 3GPP NR.

For clarity of explanation, the description is based on a 3GPP communication system (eg, LTE, NR, etc.), but the technical spirit of the present invention is not limited thereto. LTE refers to technology after 3GPP TS 36.xxx Release 8. In detail, LTE technology after 3GPP TS 36.xxx Release 10 is referred to as LTE-A, and LTE technology after 3GPP TS 36.xxx Release 13 is referred to as LTE-A pro. 3GPP NR refers to technology after TS 38.xxx Release 15. 3GPP 6G may mean technology after TS Release 17 and/or Release 18. "xxx" means standard document detail number. LTE/NR/6G may be collectively referred to as a 3GPP system. For background art, terms, abbreviations, etc. used in the description of the present invention, reference may be made to matters described in standard documents published prior to the present invention. For example, you can refer to the following document.

3GPP LTE

- 36.211: Physical channels and modulation

- 36.212: Multiplexing and channel coding

- 36.213: Physical layer procedures

- 36.300: Overall description

- 36.331: Radio Resource Control (RRC)

3GPP NRs

- 38.211: Physical channels and modulation

- 38.212: Multiplexing and channel coding

- 38.213: Physical layer procedures for control

- 38.214: Physical layer procedures for data

- 38.300: NR and NG-RAN Overall Description

- 38.331: Radio Resource Control (RRC) protocol specification

Physical Channels and Frame Structure

Physical channels and general signal transmission

1 illustrates physical channels and typical signal transmission used in a 3GPP system. In a wireless communication system, a terminal receives information from a base station through downlink (DL), and the terminal transmits information to the base station through uplink (UL). Information transmitted and received between the base station and the terminal includes data and various control information, and various physical channels exist according to the type/use of the information transmitted and received by the base station and the terminal.

When the terminal is turned on or newly enters a cell, the terminal performs an initial cell search operation such as synchronizing with the base station (S11). To this end, the terminal may receive a primary synchronization signal (PSS) and a secondary synchronization signal (SSS) from the base station to synchronize with the base station and obtain information such as a cell ID. After that, the terminal can acquire intra-cell broadcast information by receiving a physical broadcast channel (PBCH) from the base station. Meanwhile, the terminal may check the downlink channel state by receiving a downlink reference signal (DL RS) in the initial cell search step.

After completing the initial cell search, the UE acquires more detailed system information by receiving a Physical Downlink Control Channel (PDCCH) and a Physical Downlink Shared Channel (PDSCH) according to the information carried on the PDCCH. It can (S12).

Meanwhile, when accessing the base station for the first time or when there is no radio resource for signal transmission, the terminal may perform a random access procedure (RACH) for the base station (S13 to S16). To this end, the UE transmits a specific sequence as a preamble through a physical random access channel (PRACH) (S13 and S15), and responds to the preamble through a PDCCH and a corresponding PDSCH (RAR (Random Access Channel) Response message) may be received In the case of contention-based RACH, a contention resolution procedure may be additionally performed (S16).

After performing the above procedure, the UE receives PDCCH/PDSCH (S17) and physical uplink shared channel (PUSCH)/physical uplink control channel (PUSCH) as a general uplink/downlink signal transmission procedure. Control Channel; PUCCH) transmission (S18) may be performed. In particular, the terminal may receive downlink control information (DCI) through the PDCCH. Here, the DCI includes control information such as resource allocation information for the terminal, and different formats may be applied depending on the purpose of use.

On the other hand, the control information that the terminal transmits to the base station through the uplink or the terminal receives from the base station is a downlink / uplink ACK / NACK signal, CQI (Channel Quality Indicator), PMI (Precoding Matrix Index), RI (Rank Indicator) ) and the like. The UE may transmit control information such as the aforementioned CQI/PMI/RI through PUSCH and/or PUCCH.

Structure of uplink and downlink channels

Downlink channel structure

The base station transmits a related signal to the terminal through a downlink channel described later, and the terminal receives the related signal from the base station through a downlink channel described later.

(1) Physical Downlink Shared Channel (PDSCH)

PDSCH carries downlink data (eg, DL-shared channel transport block, DL-SCH TB), and modulation methods such as Quadrature Phase Shift Keying (QPSK), 16 Quadrature Amplitude Modulation (QAM), 64 QAM, and 256 QAM are used. Applied. A codeword is generated by encoding the TB. PDSCH can carry multiple codewords. Scrambling and modulation mapping are performed for each codeword, and modulation symbols generated from each codeword are mapped to one or more layers (Layer mapping). Each layer is mapped to a resource along with a demodulation reference signal (DMRS), generated as an OFDM symbol signal, and transmitted through a corresponding antenna port.

(2) Physical Downlink Control Channel (PDCCH)

The PDCCH carries downlink control information (DCI) and a QPSK modulation method or the like is applied. One PDCCH is composed of 1, 2, 4, 8, or 16 Control Channel Elements (CCEs) according to an Aggregation Level (AL). One CCE is composed of 6 REGs (Resource Element Groups). One REG is defined as one OFDM symbol and one (P)RB.

The UE obtains DCI transmitted through the PDCCH by performing decoding (aka, blind decoding) on a set of PDCCH candidates. A set of PDCCH candidates decoded by the UE is defined as a PDCCH search space set. The search space set may be a common search space or a UE-specific search space. The UE may obtain DCI by monitoring PDCCH candidates in one or more search space sets configured by MIB or higher layer signaling.

Uplink channel structure

The terminal transmits a related signal to the base station through an uplink channel described later, and the base station receives the related signal from the terminal through an uplink channel described later.

(1) Physical Uplink Shared Channel (PUSCH)

PUSCH carries uplink data (e.g., UL-shared channel transport block, UL-SCH TB) and/or uplink control information (UCI), and CP-OFDM (Cyclic Prefix - Orthogonal Frequency Division Multiplexing) waveform , Discrete Fourier Transform-spread-Orthogonal Frequency Division Multiplexing (DFT-s-OFDM) waveform. When the PUSCH is transmitted based on the DFT-s-OFDM waveform, the UE transmits the PUSCH by applying transform precoding. For example, when transform precoding is impossible (eg, transform precoding is disabled), the terminal transmits a PUSCH based on the CP-OFDM waveform, and when transform precoding is possible (eg, transform precoding is enabled), the terminal transmits the CP-OFDM The PUSCH may be transmitted based on a waveform or a DFT-s-OFDM waveform. PUSCH transmission is dynamically scheduled by the UL grant in DCI or semi-static based on higher layer (eg, RRC) signaling (and/or Layer 1 (L1) signaling (eg, PDCCH)) It can be scheduled (configured grant). PUSCH transmission may be performed on a codebook basis or a non-codebook basis.

(2) Physical Uplink Control Channel (PUCCH)

PUCCH carries uplink control information, HARQ-ACK and/or scheduling request (SR), and may be divided into multiple PUCCHs according to PUCCH transmission length.

6G system general

6G (radio communications) systems are characterized by (i) very high data rates per device, (ii) very large number of connected devices, (iii) global connectivity, (iv) very low latency, (v) battery- It aims to lower energy consumption of battery-free IoT devices, (vi) ultra-reliable connectivity, and (vii) connected intelligence with machine learning capabilities. The vision of the 6G system can be four aspects such as intelligent connectivity, deep connectivity, holographic connectivity, and ubiquitous connectivity, and the 6G system can satisfy the requirements shown in Table 1 below. That is, Table 1 is a table showing an example of requirements for a 6G system.

6G systems include Enhanced mobile broadband (eMBB), Ultra-reliable low latency communications (URLLC), massive machine-type communication (mMTC), AI integrated communication, Tactile internet, High throughput, High network capacity, High energy efficiency, Low backhaul and It can have key factors such as access network congestion and enhanced data security.

2 is a diagram showing an example of a communication structure that can be provided in a 6G system.

6G systems are expected to have 50 times higher simultaneous radiocommunication connectivity than 5G radiocommunication systems. URLLC, a key feature of 5G, will become even more important in 6G communications by providing end-to-end latency of less than 1 ms. The 6G system will have much better volume spectral efficiency as opposed to the frequently used area spectral efficiency. 6G systems can provide very long battery life and advanced battery technology for energy harvesting, so mobile devices will not need to be charged separately in 6G systems. New network characteristics in 6G may be as follows.

- Satellites integrated network: 6G is expected to be integrated with satellites to serve the global mobile population. Integration of terrestrial, satellite and public networks into one wireless communication system is critical for 6G.

- Connected intelligence: Unlike previous generations of wireless communications systems, 6G is revolutionary and will update the evolution of wireless from “connected things” to “connected intelligence.” AI can be applied at each step of the communication procedure (or each procedure of signal processing to be described later).

- Seamless integration wireless information and energy transfer: 6G wireless networks will transfer power to charge the batteries of devices such as smartphones and sensors. Therefore, wireless information and energy transfer (WIET) will be integrated.

- Ubiquitous super 3D connectivity: Access to networks and core network capabilities of drones and very low Earth orbit satellites will make super 3D connectivity in 6G ubiquitous.

In the new network characteristics of 6G as above, some general requirements can be as follows.

- Small cell networks: The idea of small cell networks has been introduced to improve received signal quality resulting in improved throughput, energy efficiency and spectral efficiency in cellular systems. As a result, small cell networks are an essential feature of 5G and Beyond 5G (5GB) and beyond communication systems. Therefore, the 6G communication system also adopts the characteristics of the small cell network.

- Ultra-dense heterogeneous network: Ultra-dense heterogeneous networks will be another important feature of 6G communication systems. Multi-tier networks composed of heterogeneous networks improve overall QoS and reduce costs.

- High-capacity backhaul: A backhaul connection is characterized by a high-capacity backhaul network to support high-capacity traffic. High-speed fiber and free space optical (FSO) systems may be possible solutions to this problem.

- Radar technology integrated with mobile technology: High-precision localization (or location-based service) through communication is one of the features of 6G wireless communication systems. Thus, radar systems will be integrated with 6G networks.

- Softwarization and virtualization: Softwarization and virtualization are two important features fundamental to the design process in 5GB networks to ensure flexibility, reconfigurability and programmability. In addition, billions of devices can be shared in a shared physical infrastructure.

Core implementation technology of 6G system

Artificial Intelligence

The most important and newly introduced technology for the 6G system is AI. AI was not involved in the 4G system. 5G systems will support partial or very limited AI. However, the 6G system will be AI-enabled for full automation. Advances in machine learning will create more intelligent networks for real-time communication in 6G. Introducing AI in communications can simplify and enhance real-time data transmission. AI can use a plethora of analytics to determine how complex target tasks are performed. In other words, AI can increase efficiency and reduce processing delays.

Time-consuming tasks such as handover, network selection, and resource scheduling can be performed instantly by using AI. AI can also play an important role in machine-to-machine, machine-to-human and human-to-machine communications. In addition, AI can be a rapid communication in BCI (Brain Computer Interface). AI-based communication systems can be supported by metamaterials, intelligent structures, intelligent networks, intelligent devices, intelligent cognitive radios, self-sustaining wireless networks, and machine learning.

Recently, there have been attempts to integrate AI with wireless communication systems, but these have been focused on the application layer and network layer, especially deep learning in the field of wireless resource management and allocation. However, these studies are gradually developing into the MAC layer and the physical layer, and in particular, attempts to combine deep learning with wireless transmission are appearing in the physical layer. AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in fundamental signal processing and communication mechanisms. For example, deep learning-based channel coding and decoding, deep learning-based signal estimation and detection, deep learning-based MIMO mechanism, AI-based resource scheduling and may include allocations, etc.

Machine learning may be used for channel estimation and channel tracking, and may be used for power allocation, interference cancellation, and the like in a downlink (DL) physical layer. Machine learning can also be used for antenna selection, power control, symbol detection, and the like in a MIMO system.

However, the application of DNN for transmission in the physical layer may have the following problems.

AI algorithms based on deep learning require a lot of training data to optimize training parameters. However, due to limitations in acquiring data in a specific channel environment as training data, a lot of training data is used offline. This is because static training on training data in a specific channel environment may cause a contradiction between dynamic characteristics and diversity of a radio channel.

In addition, current deep learning mainly targets real signals. However, the signals of the physical layer of wireless communication are complex signals. In order to match the characteristics of wireless communication signals, further research on a neural network for detecting complex domain signals is needed.

Hereinafter, machine learning will be described in more detail.

Machine learning refers to a set of actions that train a machine to create a machine that can do tasks that humans can or cannot do. Machine learning requires data and a running model. In machine learning, data learning methods can be largely classified into three types: supervised learning, unsupervised learning, and reinforcement learning.

Neural network training is aimed at minimizing errors in the output. Neural network learning repeatedly inputs training data to the neural network, calculates the output of the neural network for the training data and the error of the target, and backpropagates the error of the neural network from the output layer of the neural network to the input layer in a direction to reduce the error. ) to update the weight of each node in the neural network.

Supervised learning uses training data in which correct answers are labeled in the learning data, and unsupervised learning may not have correct answers labeled in the learning data. That is, for example, learning data in the case of supervised learning related to data classification may be data in which each learning data is labeled with a category. Labeled training data is input to the neural network, and an error may be calculated by comparing the output (category) of the neural network and the label of the training data. The calculated error is back-propagated in a reverse direction (ie, from the output layer to the input layer) in the neural network, and the connection weight of each node of each layer of the neural network may be updated according to the back-propagation. The amount of change in the connection weight of each updated node may be determined according to a learning rate. The neural network's computation of input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently according to the number of iterations of the learning cycle of the neural network. For example, a high learning rate is used in the early stages of neural network learning to increase efficiency by allowing the neural network to quickly achieve a certain level of performance, and a low learning rate can be used in the late stage to increase accuracy.

The learning method may vary depending on the characteristics of the data. For example, in a case where the purpose of the receiver is to accurately predict data transmitted by the transmitter in a communication system, it is preferable to perform learning using supervised learning rather than unsupervised learning or reinforcement learning.

The learning model corresponds to the human brain, and the most basic linear model can be considered. ) is called

The neural network cord used as a learning method is largely divided into deep neural networks (DNN), convolutional deep neural networks (CNN), and recurrent Boltzmann Machine (RNN). there is.

An artificial neural network is an example of connecting several perceptrons.

Referring to FIG. 3, when the input vector x=(x1,x2,...,xd) is input, each component is multiplied by a weight (W1,W2,...,Wd), and after summing up the results, The entire process of applying the activation function is called a perceptron. The huge artificial neural network structure may extend the simplified perceptron structure shown in FIG. 3 and apply input vectors to different multi-dimensional perceptrons. For convenience of description, an input value or an output value is referred to as a node.

Meanwhile, the perceptron structure shown in FIG. 3 can be described as being composed of a total of three layers based on input values and output values. An artificial neural network in which H number of (d + 1) dimensional perceptrons exist between the 1st layer and the 2nd layer and K number of (H + 1) dimensional perceptrons between the 2nd layer and the 3rd layer can be expressed as shown in FIG. 4 .

The layer where the input vector is located is called the input layer, the layer where the final output value is located is called the output layer, and all the layers located between the input layer and the output layer are called hidden layers. In the example of FIG. 4 , three layers are disclosed, but when counting the number of layers of an actual artificial neural network, since the count excludes the input layer, it can be regarded as a total of two layers. The artificial neural network is composed of two-dimensionally connected perceptrons of basic blocks.

The above-described input layer, hidden layer, and output layer can be jointly applied to various artificial neural network structures such as CNN and RNN, which will be described later, as well as multi-layer perceptrons. As the number of hidden layers increases, the artificial neural network becomes deeper, and a machine learning paradigm that uses a sufficiently deep artificial neural network as a learning model is called deep learning. In addition, the artificial neural network used for deep learning is called a deep neural network (DNN).

The deep neural network shown in FIG. 5 is a multi-layer perceptron composed of 8 hidden layers + 8 output layers. The multilayer perceptron structure is expressed as a fully-connected neural network. In a fully-connected neural network, there is no connection relationship between nodes located on the same layer, and a connection relationship exists only between nodes located on adjacent layers. DNN has a fully-connected neural network structure and is composed of a combination of multiple hidden layers and activation functions, so it can be usefully applied to identify the correlation characteristics between inputs and outputs. Here, the correlation characteristic may mean a joint probability of input and output.

‘On the other hand, depending on how a plurality of perceptrons are connected to each other, various artificial neural network structures different from the aforementioned DNN can be formed.

In DNN, nodes located inside one layer are arranged in a one-dimensional vertical direction. However, in FIG. 6, it can be assumed that the nodes are two-dimensionally arranged with w nodes horizontally and h nodes vertically (the structure of the convolutional neural network in FIG. 6). In this case, a weight is added for each connection in the connection process from one input node to the hidden layer, so a total of hХw weights must be considered. Since there are hХw nodes in the input layer, a total of h ² w ² weights are required between two adjacent layers.

The convolutional neural network of FIG. 6 has a problem that the number of weights increases exponentially according to the number of connections, so instead of considering all mode connections between adjacent layers, it is assumed that there is a filter with a small size. As shown in , weighted sum and activation function calculations are performed for overlapping filters.

One filter has weights corresponding to the number of filters, and learning of weights can be performed so that a specific feature on an image can be extracted as a factor and output. In FIG. 7 , a 3Х3 size filter is applied to the 3Х3 region at the top left of the input layer, and an output value obtained by performing a weighted sum and an activation function operation for a corresponding node is stored in z22.

While scanning the input layer, the filter performs weighted sum and activation function calculations while moving horizontally and vertically at regular intervals, and places the output value at the position of the current filter. This operation method is similar to the convolution operation for images in the field of computer vision, so the deep neural network of this structure is called a convolutional neural network (CNN), and the hidden layer generated as a result of the convolution operation is called a convolutional layer. Also, a neural network having a plurality of convolutional layers is referred to as a deep convolutional neural network (DCNN).

In the convolution layer, the number of weights can be reduced by calculating a weighted sum by including only nodes located in a region covered by the filter from the node where the current filter is located. This allows one filter to be used to focus on features for a local area. Accordingly, CNN can be effectively applied to image data processing in which a physical distance in a 2D area is an important criterion. Meanwhile, in the CNN, a plurality of filters may be applied immediately before the convolution layer, and a plurality of output results may be generated through a convolution operation of each filter.

Meanwhile, there may be data whose sequence characteristics are important according to data attributes. Considering the length variability and precedence relationship of these sequence data, input each element on the data sequence one by one at each time step, and input the output vector (hidden vector) of the hidden layer output at a specific time point together with the next element on the sequence A structure in which this method is applied to an artificial neural network is called a recurrent neural network structure.

Referring to FIG. 8, a recurrent neural network (RNN) inputs an element (x1(t), x2(t), ..., xd(t)) of a line t on a data sequence into a fully connected neural network In the process, the immediately preceding time point t-1 applies the weighted sum and activation function by inputting the hidden vector (z1 (t-1), z2 (t-1), ..., zH (t-1)) It is a structure that The reason why the hidden vector is transmitted to the next time point in this way is that information in the input vector at previous time points is regarded as being accumulated in the hidden vector of the current time point.

Referring to FIG. 8 , the recurrent neural network operates in a sequence of predetermined views with respect to an input data sequence.

The hidden vector (z1(1),z2(1),.. .,zH(1)) is input together with the input vector (x1(2),x2(2),...,xd(2)) of time 2, and the vector of the hidden layer (z1( 2),z2(2) ,...,zH(2)). This process is repeated until time point 2, time point 3, ,,, time point T.

Meanwhile, when a plurality of hidden layers are arranged in a recurrent neural network, it is referred to as a deep recurrent neural network (DRNN). Recurrent neural networks are designed to be usefully applied to sequence data (eg, natural language processing).

As a neural network core used as a learning method, in addition to DNN, CNN, and RNN, Restricted Boltzmann Machine (RBM), deep belief networks (DBN), and Deep Q-Network It includes various deep learning techniques such as computer vision, voice recognition, natural language processing, and voice/signal processing.

Recently, there have been attempts to integrate AI with wireless communication systems, but these have been focused on the application layer and network layer, especially deep learning in the field of wireless resource management and allocation. However, these studies are gradually developing into the MAC layer and the physical layer, and in particular, attempts to combine deep learning with wireless transmission are appearing in the physical layer. AI-based physical layer transmission means applying a signal processing and communication mechanism based on an AI driver rather than a traditional communication framework in fundamental signal processing and communication mechanisms. For example, deep learning-based channel coding and decoding, deep learning-based signal estimation and detection, deep learning-based MIMO mechanism, AI-based resource scheduling and may include allocations, etc.

Terahertz (THz) communication

The data rate can be increased by increasing the bandwidth. This can be done using sub-THz communication with wide bandwidth and applying advanced massive MIMO technology. THz waves, also known as submillimeter radiation, typically represent a frequency band between 0.1 THz and 10 THz with corresponding wavelengths in the range of 0.03 mm-3 mm. The 100 GHz-300 GHz band range (sub THz band) is considered a major part of the THz band for cellular communications. 6G cellular communication capacity increases when added to the sub-THz band mmWave band. Of the defined THz bands, 300 GHz-3 THz is in the far-infrared (IR) frequency band. The 300 GHz-3 THz band is part of the broad band, but is at the border of the wide band, just behind the RF band. Thus, this 300 GHz-3 THz band exhibits similarities to RF. 10 shows an example of an electromagnetic spectrum.

The main characteristics of THz communications include (i) widely available bandwidth to support very high data rates, and (ii) high path loss at high frequencies (highly directional antennas are indispensable). The narrow beamwidth produced by the highly directional antenna reduces interference. The small wavelength of the THz signal allows a much larger number of antenna elements to be incorporated into devices and BSs operating in this band. This enables advanced adaptive array technology to overcome range limitations.

Optical wireless technology

OWC technology is intended for 6G communications in addition to RF-based communications for all possible device-to-access networks. These networks access network-to-backhaul/fronthaul network connections. OWC technology is already in use after the 4G communication system, but will be more widely used to meet the needs of the 6G communication system. OWC technologies such as light fidelity, visible light communication, optical camera communication, and FSO communication based on a wide band are already well-known technologies. Communications based on optical wireless technology can provide very high data rates, low latency and secure communications. LiDAR can also be used for ultra-high resolution 3D mapping in 6G communication based on broadband.

FSO backhaul network

The transmitter and receiver characteristics of an FSO system are similar to those of a fiber optic network. Thus, data transmission in FSO systems is similar to fiber optic systems. Therefore, FSO can be a good technology to provide backhaul connectivity in 6G systems along with fiber optic networks. With FSO, very long-distance communication is possible even at a distance of 10,000 km or more. FSO supports high-capacity backhaul connectivity for remote and non-remote locations such as ocean, space, underwater and isolated islands. FSO also supports cellular BS connections.

Massive MIMO technology

One of the key technologies to improve spectral efficiency is to apply MIMO technology. As MIMO technology improves, so does the spectral efficiency. Therefore, massive MIMO technology will be important in 6G systems. Since MIMO technology uses multiple paths, multiplexing technology and beam generation and operation technology suitable for the THz band must be considered as important so that data signals can be transmitted through more than one path.

block chain

Blockchain will be an important technology for managing large amounts of data in future communication systems. Blockchain is a form of distributed ledger technology, where a distributed ledger is a database that is distributed across numerous nodes or computing devices. Each node replicates and stores an identical copy of the ledger. Blockchain is managed as a peer-to-peer network. It can exist without being managed by a centralized authority or server. Data on a blockchain is collected together and organized into blocks. Blocks are linked together and protected using cryptography. Blockchain is the perfect complement to the IoT at scale with inherently improved interoperability, security, privacy, reliability and scalability. Thus, blockchain technology provides multiple capabilities such as interoperability between devices, traceability of large amounts of data, autonomous interaction of other IoT systems, and large-scale connection reliability in 6G communication systems.

3D Networking

The 6G system integrates terrestrial and air networks to support vertical expansion of user communications. 3D BS will be provided via low-orbit satellites and UAVs. Adding a new dimension in terms of height and related degrees of freedom makes 3D connections quite different from traditional 2D networks.

quantum communication

In the context of 6G networks, unsupervised reinforcement learning of networks is promising. Supervised learning approaches cannot label the vast amount of data generated by 6G. Labeling is not required in unsupervised learning. Thus, this technique can be used to autonomously build representations of complex networks. Combining reinforcement learning and unsupervised learning allows networks to operate in a truly autonomous way.

drone

Unmanned Aerial Vehicles (UAVs) or drones will be an important element in 6G wireless communications. In most cases, high-speed data wireless connectivity is provided using UAV technology. BS entities are installed on UAVs to provide cellular connectivity. UAVs have certain features not found in fixed BS infrastructures, such as easy deployment, strong line-of-sight links, and degrees of freedom with controlled mobility. During emergencies, such as natural disasters, the deployment of terrestrial communications infrastructure is not economically feasible and cannot provide services in sometimes volatile environments. UAVs can easily handle this situation. UAVs will become a new paradigm in the field of wireless communication. This technology facilitates three basic requirements of a wireless network: eMBB, URLLC and mMTC. UAVs can also support multiple purposes, such as enhancing network connectivity, fire detection, disaster emergency services, security and surveillance, pollution monitoring, parking monitoring, accident monitoring, and more. Therefore, UAV technology is recognized as one of the most important technologies for 6G communication.

Cell-free Communication

The tight integration of multiple frequencies and heterogeneous communication technologies is critical for 6G systems. As a result, users can seamlessly move from one network to another without having to make any manual configuration on the device. The best network is automatically selected from available communication technologies. This will break the limitations of the cell concept in wireless communication. Currently, user movement from one cell to another causes too many handovers in high-density networks, resulting in handover failures, handover delays, data loss and ping-pong effects. 6G cell-free communication will overcome all of this and provide better QoS. Cell-free communication will be achieved through multi-connectivity and multi-tier hybrid technologies and different heterogeneous radios of devices.

Integration of wireless information and energy transmission

WIET uses the same fields and waves as wireless communication systems. In particular, sensors and smartphones will be charged using wireless power transfer during communication. WIET is a promising technology for extending the lifetime of battery charging wireless systems. Thus, battery-less devices will be supported in 6G communications.

Integration of sensing and communication

Autonomous radio networks are capable of continuously sensing dynamically changing environmental conditions and exchanging information between different nodes. In 6G, sensing will be tightly integrated with communications to support autonomous systems.

Integration of access backhaul networks

In 6G, the density of access networks will be enormous. Each access network is connected by fiber and backhaul connections such as FSO networks. To cope with the very large number of access networks, there will be tight integration between access and backhaul networks.

Holographic Beamforming

Beamforming is a signal processing procedure that adjusts an antenna array to transmit radio signals in a specific direction. A subset of smart antennas or advanced antenna systems. Beamforming technology has several advantages such as high call-to-noise ratio, interference avoidance and rejection, and high network efficiency. Hologram beamforming (HBF) is a new beamforming method that differs significantly from MIMO systems because it uses software-defined antennas. HBF will be a very effective approach for efficient and flexible transmission and reception of signals in multi-antenna communication devices in 6G.

big data analytics

Big data analysis is a complex process for analyzing various large data sets or big data. This process ensures complete data management by finding information such as hidden data, unknown correlations and customer preferences. Big data is collected from various sources such as videos, social networks, images and sensors. This technology is widely used to process massive data in 6G systems.

Large Intelligent Surface (LIS)

In the case of THz band signals, there may be many shadow areas due to obstructions due to strong linearity. By installing LIS near these shadow areas, LIS technology that expands the communication area, strengthens communication stability, and provides additional additional services becomes important. An LIS is an artificial surface made of electromagnetic materials and can change the propagation of incoming and outgoing radio waves. LIS can be seen as an extension of massive MIMO, but its array structure and operating mechanism are different from massive MIMO. LIS also has low power consumption in that it operates as a reconfigurable reflector with passive elements, i.e. it only passively reflects signals without using an active RF chain. There are advantages to having In addition, since each passive reflector of the LIS must independently adjust the phase shift of an incident signal, it may be advantageous for a wireless communication channel. By properly adjusting the phase shift through the LIS controller, the reflected signal can be collected at the target receiver to boost the received signal power.

Terahertz (THz) wireless communication general

THz wireless communication uses wireless communication using THz waves having a frequency of approximately 0.1 to 10 THz (1 THz = 1012 Hz), and may mean terahertz (THz) band wireless communication using a very high carrier frequency of 100 GHz or more. . THz waves are located between RF (Radio Frequency)/millimeter (mm) and infrared bands, and (i) transmit non-metal/non-polarizable materials better than visible light/infrared rays, and have a shorter wavelength than RF/millimeter waves and have high straightness. Beam focusing may be possible. In addition, since the photon energy of the THz wave is only a few meV, it is harmless to the human body. A frequency band expected to be used for THz wireless communication may be a D-band (110 GHz to 170 GHz) or H-band (220 GHz to 325 GHz) band with low propagation loss due to molecular absorption in the air. Standardization discussions on THz wireless communication are being discussed centering on the IEEE 802.15 THz working group in addition to 3GPP, and standard documents issued by the IEEE 802.15 Task Group (TG3d, TG3e) can specify or supplement the content described in this specification. there is. THz wireless communication may be applied to wireless cognition, sensing, imaging, wireless communication, THz navigation, and the like.

11 is a diagram showing an example of a THz communication application.

As shown in FIG. 11, THz wireless communication scenarios can be classified into macro networks, micro networks, and nanoscale networks. In macro networks, THz wireless communication can be applied to vehicle-to-vehicle connections and backhaul/fronthaul connections. In micro networks, THz wireless communication is applied to indoor small cells, fixed point-to-point or multi-point connections such as wireless connections in data centers, and near-field communication such as kiosk downloading. It can be.

Table 2 below is a table showing an example of a technique that can be used in THz waves.

THz wireless communication can be classified based on the method for generating and receiving THz. The THz generation method can be classified as an optical device or an electronic device based technology.

12 is a diagram illustrating an example of an electronic element-based THz wireless communication transceiver.

Methods of generating THz using electronic devices include a method using a semiconductor device such as a Resonant Tunneling Diode (RTD), a method using a local oscillator and a multiplier, and an integrated circuit based on a compound semiconductor HEMT (High Electron Mobility Transistor). There is a MMIC (Monolithic Microwave Integrated Circuits) method using , a method using a Si-CMOS-based integrated circuit, and the like. In the case of FIG. 12, a doubler, tripler, or multiplier is applied to increase the frequency, and the radiation is emitted by the antenna after passing through the subharmonic mixer. Since the THz band forms high frequencies, a multiplier is essential. Here, the multiplier is a circuit that makes the output frequency N times greater than the input, matches the desired harmonic frequency, and filters out all other frequencies. In addition, beamforming may be implemented by applying an array antenna or the like to the antenna of FIG. 12 . 12, IF denotes an intermediate frequency, tripler and multipler denote a multiplier, PA denotes a power amplifier, LNA denotes a low noise amplifier, and PLL denotes a phase-locked circuit (Phase -Locked Loop).

13 is a diagram illustrating an example of a method for generating a THz signal based on an optical device, and FIG. 14 is a diagram illustrating an example of a THz wireless communication transceiver based on an optical device.

Optical device-based THz wireless communication technology refers to a method of generating and modulating a THz signal using an optical device. An optical element-based THz signal generation technology is a technology that generates an ultra-high speed optical signal using a laser and an optical modulator and converts it into a THz signal using an ultra-high speed photodetector. Compared to a technique using only an electronic device, this technique can easily increase the frequency, generate a high-power signal, and obtain a flat response characteristic in a wide frequency band. As shown in FIG. 13, a laser diode, a broadband optical modulator, and a high-speed photodetector are required to generate a THz signal based on an optical device. In the case of FIG. 13 , a THz signal corresponding to a wavelength difference between the lasers is generated by multiplexing light signals of two lasers having different wavelengths. In FIG. 13, an optical coupler means a semiconductor device that transmits an electrical signal using light waves in order to provide electrical isolation and coupling between circuits or systems, and UTC-PD (Uni-Travelling Carrier Photo- Detector is one of the photodetectors, which uses electrons as active carriers and reduces the movement time of electrons through bandgap grading. UTC-PD is capable of photodetection above 150 GHz. In FIG. 14, EDFA (Erbium-Doped Fiber Amplifier) represents an erbium-added optical fiber amplifier, PD (Photo Detector) represents a semiconductor device capable of converting an optical signal into an electrical signal, and OSA represents various optical communication functions (photoelectric conversion, light-to-optical conversion, etc.) represents an optical module (Optical Sub Assembly) that is modularized into one component, and DSO represents a digital storage oscilloscope.

The structure of the photoelectric converter (or photoelectric converter) will be described with reference to FIGS. 15 and 16 . 15 illustrates a structure of a transmitter based on a photoinc source, and FIG. 16 illustrates a structure of an optical modulator.

In general, a phase or the like of a signal may be changed by passing an optical source of a laser through an optical wave guide. At this time, data is loaded by changing electrical characteristics through a microwave contact or the like. Accordingly, the optical modulator output is formed as a modulated waveform. A photoelectric modulator (O/E converter) is an optical rectification operation by a nonlinear crystal, an O/E conversion by a photoconductive antenna, and a bundle of electrons in light flux. THz pulses can be generated according to emission from relativistic electrons, etc. A THz pulse generated in the above manner may have a unit length of femto second to pico second. An O/E converter uses non-linearity of a device to perform down conversion.

Considering the THz spectrum usage, it is necessary to use several contiguous GHz bands for fixed or mobile service for terahertz systems. more likely to use According to outdoor scenario criteria, available bandwidth may be classified based on oxygen attenuation of 10^2 dB/km in a spectrum up to 1 THz. Accordingly, a framework in which the available bandwidth is composed of several band chunks may be considered. As an example of the framework, if the length of a THz pulse for one carrier is set to 50 ps, the bandwidth (BW) becomes about 20 GHz.

Effective down conversion from the IR band to the THz band depends on how to utilize the nonlinearity of the O/E converter. That is, in order to down-convert to the desired terahertz band (THz band), the photoelectric converter (O / E converter) having the most ideal non-linearity to move to the corresponding terahertz band (THz band) design is required. If an O/E converter that does not fit the target frequency band is used, there is a high possibility that an error will occur with respect to the amplitude and phase of the corresponding pulse.

In a single carrier system, a terahertz transmission/reception system may be implemented using one photoelectric converter. Although it depends on the channel environment, as many photoelectric converters as the number of carriers may be required in a multi-carrier system. In particular, in the case of a multi-carrier system using several broadbands according to a plan related to the above-mentioned spectrum use, the phenomenon will be conspicuous. In this regard, a frame structure for the multi-carrier system may be considered. A signal down-frequency converted based on the photoelectric converter may be transmitted in a specific resource region (eg, a specific frame). The frequency domain of the specific resource domain may include a plurality of chunks. Each chunk may consist of at least one component carrier (CC).

The above Salpin 6G communication technology can be applied in combination with the methods proposed in this specification, which will be described later, or can be supplemented to specify or clarify the technical characteristics of the methods proposed in this specification. Meanwhile, the communication service proposed in this specification may be applied in combination with a communication service based on 3G, 4G, and/or 5G communication technology as well as the 6G communication technology described above.

17 is an exemplary diagram of a wireless communication system to which various embodiments of the present specification are applied.

Referring to FIG. 17 , a wireless communication system may include a first terminal 1701, a second terminal 1702, and a base station.

The first terminal 1701 stores the first local model 1701m, and the second terminal 1702 stores the second local model 1702m. The first local model 1701m and the second local model 1702m may be models having the same neural network structure, but are not limited thereto. The first terminal 1701 and the second terminal 1702 are base stations for joint learning, and one or more parameters (eg, weights or gradients) calculated based on the first local model 1701m and the second local model 1702m. is transmitted to the base station.

The base station transmits information received from the first terminal 1701 and the second terminal 1702 to the cloud server.

As such, the communication system for federated learning saves frequency bands by transmitting a parameter or parameter set calculated based on a local model rather than raw data, and as personal information included in raw data is transmitted to the server, Personal information security issues can be solved.

18 is a flowchart of a federated learning method according to the first embodiment of the present specification.

One or more processors of the terminal change a first parameter set related to the local model to a second parameter set based on the local model (S1801).

The parameter set may be one or more parameters or a sequence of one or more parameters in another embodiment. More specifically, the parameter set refers to a cluster of parameters composed of two or more parameters, but the one or more parameters refer to a single parameter and the sequence is a cluster of parameters divided into time-series according to a time stamp means Also, a parameter set may be mixed with a parameter vector or a parameter matrix.

Meanwhile, in various embodiments of the present specification, the local model may be implemented as a machine learning model. The machine learning model is SVM (Support Vector Machine), MLP (Multi-Layer Perceptron), 1D-CNN (1-dimentional Convolutional Neural Network), 2D-CNN (2-dimentional Convolutional Neural Network), LSTM (Long-Short Term Memory) and GRU (Gated Recurrent Unit), but the technical idea according to various embodiments of the present specification is not limited thereto.

Parameter sets related to the local model refer to a set of weights. At this time, the weight is determined according to the connection relationship (ie, node connection) between the neural networks constituting the local model. These weights are for calculating accurate inference results in the machine learning model, and detailed explanations are omitted as they are obvious to those with ordinary technical knowledge in the field of artificial intelligence.

Meanwhile, the change of the parameter set is performed according to the learning operation of the machine learning model. In the learning operation, for example, when data of a preset batch size is accumulated in the terminal, the terminal performs learning of the local model. At this time, the terminal calculates a gradient based on the cost function or loss function for learning of the learning model, calculates the amount of change in weight based on the calculated gradient, and then calculates the learning rate (Learning Rate) multiplied by Accordingly, the first parameter set is changed to the second parameter set.

Meanwhile, during this learning process, the terminal calculates the difference between the first and second parameter sets. The first parameter set represents weights of the local model before learning, and the second parameter set represents weights of the local model after learning. That is, the terminal can check the weight change before and after learning. The terminal selects at least some of the second parameter set based on the weight change confirmed later, and a detailed description and effect of this operation will be described with reference to S1802.

One or more processors select at least some of the second parameter sets based on the difference between the first and second parameter sets (S1802).

Here, the second parameter set refers to parameters to be transmitted to the base station for federated learning. Conventional federated learning methods simply transmit weights or gradients generated based on a local model to a base station. However, there is a problem in that more UL frequency bands are required in proportion to the number of weight parameters when transmitted to the network.

Accordingly, the federated learning method according to various embodiments of the present specification selects at least some of the calculated second parameter sets based on the difference between the first and second parameter sets. In one embodiment, when the total number of weights of the local model is M, the parameter set has first through Mth weights. At this time, the terminal compares the first to Mth weights for each element to calculate a difference value.

If the calculated difference value is 0, the weight of the corresponding entry does not change, so even if the value is transmitted to the base station, it does not significantly affect the performance of federated learning. In addition, even if the calculated difference value is not necessarily 0, if it is less than a set specific value, the weight of the corresponding entry does not necessarily need to be transmitted to the base station because the degree of effect on the efficiency of federated learning is small.

That is, the federated learning method according to some embodiments of the present specification calculates a difference value between the first and second parameter sets, and selects at least some of the weight elements constituting the second parameter set according to the calculated difference value. and omitting the remainder, the UL frequency band can be saved.

One or more processors transmit the selected at least some parameter sets to the base station (S1803).

At least some of the transmitted parameter sets are received by the base station and transmitted to a server for federated learning. Then, the server learns a global AI model based on the at least some parameter sets. The learned global AI model or parameter sets (ie, changed weights) of the global AI model are transmitted to one or more terminals associated with the server.

Meanwhile, transmission to the server may be directly transmitted by a terminal connected to the base station, but is not limited thereto. For example, the terminal may transmit at least some parameter sets selected to a WIFI access point (AP). At this time, the WIFI AP may transmit at least some of the received parameter sets to the base station.

In addition, it has been described that, among one or more operations of the federated learning method according to the first embodiment of FIG. 18, changing parameters or parameter sets of a local model is performed in a terminal, but is not limited thereto. For example, learning of a local model may also be performed on a Mobile Edge Computing (MEC) server. Likewise, MEC servers allow traffic and service computing to move from a centralized cloud closer to the network edge. That is, analyzing, processing, and storing data at the edge of the network reduces latency and improves real-time performance of high-bandwidth applications, without necessarily sending all data to the cloud for processing.

19 is a flowchart of a federated learning method according to a second embodiment of the present specification.

One or more processors of the terminal perform pre-processing on data received or input to the terminal (S1901).

For example, a terminal according to some embodiments of the present specification may be implemented as a natural language processing device (NLP Device). If the terminal is an NLP device, the terminal may receive a voice signal or voice data through a microphone for a natural language process (NLP). Voice signals received through a microphone may be generally converted into a Melspectrogram, but are not limited thereto. As such, the Melsepctrogram can represent the characteristics of the voice signal, and as a result, the NLP device can check the reverberation effect according to the user's speech style or the size of the room.

In some cases, the NLP device needs personalized data for a specific user, but in some cases, it is necessary to collect decentralized learning data from factors such as voice data of foreigners or reverberation effects according to the size of a room.

Meanwhile, in federated learning methods according to some embodiments of the present specification, auto-labeling is performed on preprocessed data. In order to implement federated learning, a label for data to be collected or collected from a terminal applied to various embodiments of the present specification is required. Accordingly, one or more processors of the terminal allocate labels based on unsupervised learning to collected raw data or characteristic information (eg, a melspectrogram) converted from the raw data. In this case, the unsupervised learning includes, for example, K-means Clustering or DBSCAN Clustering, but is not limited thereto.

More specifically, the terminal may receive an input related to one or more physical motions through a sensor or one or more input devices (eg, a touch screen, a button, etc.). For example, when the terminal is an NLP device, the terminal includes one or more input devices for controlling the operation of the terminal when a voice command is not recognized as well as a microphone for receiving a voice command.

One or more labels for auto-labeling are determined based on inputs to the one or more input devices. As an example, a case of a wireless vacuum cleaner will be exemplarily described. When a wireless cleaner receives a voice command, it generally performs an operation corresponding to a command included in the voice command, but may not perform a corresponding operation. At this time, it is common for the wireless cleaner to receive the same control command (eg, start cleaning, start charging, turbo mode, etc.) as an operation corresponding to the command through one or more input devices (eg, physical buttons).

That is, the wireless cleaner may store a cached voice command or feature information converted from the voice command, and assign, provide, or determine a command related to or corresponding to a control command input through the one or more input devices as a label. . As such, the assigned, provided or determined labels are then used as labels for federated learning.

Meanwhile, allocation, provision, or determination of labels may be equally performed for two or more pieces of raw data or feature information. Specifically, before receiving a control command through the one or more input devices, the terminal may configure a cluster of two or more pieces of cached raw data or characteristic information. The cluster is constructed based on unsupervised learning. The terminal may allocate, provide, or determine a command associated with or corresponding to the control command as a label for the cluster.

Meanwhile, the auto-labeling technique can be applied even in an indoor environment where two or more terminals exist. When two or more terminals are disposed in different locations indoors, the two or more terminals can obtain different raw data or characteristic information for one target or object. For example, in the case of an image, images that can be obtained are different depending on where two or more terminals are located. As another example, voice commands may also be distinguished from each other according to the place where they are placed.

The two or more terminals may store raw data or feature information for one target or object or transmit them to other terminals. The other terminal may be any one of the two or more terminals. The transmission may be performed through a WIFI network, but is not limited thereto.

The two or more terminals or the other terminals may allocate or provide a label associated with or corresponding to a control command input through one or more input devices for stored or received raw data or feature information. In addition, even in an environment where two or more terminals exist, label assignment based on the aforementioned unsupervised learning may be performed.

In addition, in an indoor environment where two or more terminals exist, valid information of two or more acquired raw data or feature information may be extracted. The valid information refers to information from which noise unnecessary for learning a local model or a global model has been removed. For example, any one of the two or more terminals or the other terminal extracts common information by comparing two or more pieces of collected raw data or feature information, and calculates a difference between the original raw data or feature information and the common information. The difference between the feature information and the common information constitutes intermediate information.

The effective information is determined based on the sum of the intermediate information. Specifically, the information with the smallest sum of the intermediate information is classified as valid information, and the rest is classified as noise information. At this time, the valid information is stored in the cache, but the remaining information may be classified as noise and not stored.

One or more processors accumulate the preprocessed data as much as the batch size (S1902).

Before transmitting the decentralized data, the terminal caches the preprocessed data as much as the batch size.

One or more processors change a first parameter set related to the local model into a second parameter set based on the local model (S1903).

One or more processors perform training on the local model based on preprocessed data of the batch size. As a result of the execution, the parameter set of the local model is changed from the first parameter set to the second parameter set.

More specifically, the parameter set may be one or more parameters or a sequence composed of one or more parameters in another embodiment. More specifically, the parameter set refers to a cluster of parameters composed of two or more parameters, but the one or more parameters refer to a single parameter and the sequence is a cluster of parameters divided into time-series according to a time stamp means Also, a parameter set may be mixed with a parameter vector or a parameter matrix.

Meanwhile, in various embodiments of the present specification, the local model may be implemented as a machine learning model. The machine learning model is SVM (Support Vector Machine), MLP (Multi-Layer Perceptron), 1D-CNN (1-dimentional Convolutional Neural Network), 2D-CNN (2-dimentional Convolutional Neural Network), LSTM (Long-Short Term Memory) and GRU (Gated Recurrent Unit), but the technical idea according to various embodiments of the present specification is not limited thereto.

Parameter sets related to the local model refer to a set of weights. At this time, the weight is determined according to the connection relationship (ie, node connection) between the neural networks constituting the local model. These weights are for calculating accurate inference results in the machine learning model, and detailed explanations are omitted as they are obvious to those with ordinary technical knowledge in the field of artificial intelligence.

Meanwhile, the change of the parameter set is performed according to the learning operation of the machine learning model. In the learning operation, for example, when data of a preset batch size is accumulated in the terminal, the terminal performs learning of the local model. At this time, the terminal calculates a gradient based on a cost function or a loss function for learning of the learning model, calculates the amount of change in weight based on the calculated gradient, and then calculates the learning rate (Learning Rate) multiplied by Accordingly, the first parameter set is changed to the second parameter set.

Meanwhile, during this learning process, the terminal calculates the difference between the first and second parameter sets. The first parameter set represents weights of the local model before learning, and the second parameter set represents weights of the local model after learning. That is, the terminal can check the weight change before and after learning. The terminal selects at least some of the second parameter set based on the weight change confirmed later, and a detailed description and effect of this operation will be described with reference to S1904.

One or more processors select at least some of the second parameter sets based on the difference between the first and second parameter sets (S1904).

Here, the second parameter set refers to parameters to be transmitted to the base station for federated learning. Conventional federated learning methods simply transmit weights or gradients generated based on a local model to a base station. However, there is a problem in that more UL frequency bands are required in proportion to the number of weight parameters when transmitted to the network.

Accordingly, the federated learning method according to various embodiments of the present specification selects at least some of the calculated second parameter sets based on the difference between the first and second parameter sets. In one embodiment, when the total number of weights of the local model is M, the parameter set has first through Mth weights. At this time, the terminal compares the first to Mth weights for each element to calculate a difference value.

If the calculated difference value is 0, the weight of the corresponding entry does not change, so even if the value is transmitted to the base station, it does not significantly affect the performance of federated learning. In addition, even if the calculated difference value is not necessarily 0, if it is less than a set specific value, the weight of the corresponding entry does not necessarily need to be transmitted to the base station because the degree of effect on the efficiency of federated learning is small.

That is, the federated learning method according to some embodiments of the present specification calculates a difference value between the first and second parameter sets, and selects at least some of the weight elements constituting the second parameter set according to the calculated difference value. and omitting the remainder, the UL frequency band can be saved.

One or more processors transmit the selected at least some parameter sets to the base station (S1905).

At least some of the transmitted parameter sets are received by the base station and transmitted to a server for federated learning. Then, the server learns a global AI model based on the at least some parameter sets. The learned global AI model or parameter sets (ie, changed weights) of the global AI model are transmitted to one or more terminals associated with the server.

Meanwhile, transmission to the server may be directly transmitted by a terminal connected to the base station, but is not limited thereto. For example, the terminal may transmit at least some parameter sets selected to a WIFI access point (AP). At this time, the WIFI AP may transmit at least some of the received parameter sets to the base station.

In addition, it has been described that, among one or more operations of the federated learning method according to the second embodiment of FIG. 19 , the change of parameters or parameter sets of the local model is performed in the terminal, but is not limited thereto. For example, learning of a local model may also be performed in Mobile Edge Computing (MEC). As such, MEC allows traffic and service computing to be moved from a centralized cloud closer to the edge of the network. That is, analyzing, processing, and storing data at the edge of the network reduces latency and improves real-time performance of high-bandwidth applications, without necessarily sending all data to the cloud for processing.

20 is a federated learning method according to the third embodiment of the present specification.

Here, the third embodiment may be combined with the first or second embodiment, but is not limited thereto. When the third embodiment is combined with the first or second embodiment, the combined embodiments can be implemented by a communication system including a terminal and a server. On the other hand, the third embodiment refers to the learning operation by the server, but is not limited to the server, and may be replaced by another terminal or MEC server capable of performing the function of the server.

One or more processors of the server receive parameter sets for global model learning from two or more terminals (S2001).

The parameter set for learning the global model includes at least a part of the second parameter set changed by the local model. That is, the server can perform effective global model learning by receiving meaningful parameter sets from two or more terminals.

One or more processors change at least some of the parameter sets of the global model into a new parameter set calculated based on an average of each element of the received parameter sets (S2002).

In some embodiments, the one or more processors may calculate an average by summing the first to Nth entries of the parameter sets for each element and dividing the number of received parameter sets.

In some other embodiments, the one or more processors may sum the first through Nth entries for each element, and identify or determine the number of parameters based on the summation for each element. For example, assuming that two parameters are received in the first entry and three parameters are received in the second entry, one or more processors divide the sum of the first entry by 2, and the second entry For the value added by , the average for each element can be calculated by dividing by 3.

One or more processors transmit the parameter set of the global model to the two or more terminals (S2003).

Then, the local model is updated by retransmitting the parameter set of the global model to one or more terminals. As such, by repeating the above processes, the local model can be changed similarly to the global model.

21 is a federated learning method according to the fourth embodiment of the present specification.

Here, the fourth embodiment may be combined with the first or second embodiment, but is not limited thereto. When the fourth embodiment is combined with the first or second embodiment, the combined embodiments can be implemented by a communication system including a terminal and a server. On the other hand, the fourth embodiment refers to the learning operation by the server, but is not limited to the server, and may be replaced by another terminal or MEC server capable of performing the function of the server.

One or more processors of the server receive parameter sets for global model learning from two or more terminals (S2101).

The parameter set for learning the global model includes at least a part of the second parameter set changed by the local model. That is, the server can perform effective global model learning by receiving meaningful parameter sets from two or more terminals.

One or more processors change at least some of the parameter sets of the global model into a new parameter set calculated based on an average of each element of the received parameter sets (S2102).

In some embodiments, the one or more processors may calculate an average by summing the first to Nth entries of the parameter sets for each element and dividing the number of received parameter sets.

In some other embodiments, the one or more processors may sum the first through Nth entries for each element, and identify or determine the number of parameters based on the summation for each element. For example, assuming that two parameters are received in the first entry and three parameters are received in the second entry, one or more processors divide the sum of the first entry by 2, and the second entry For the value added by , the average for each element can be calculated by dividing by 3.

One or more processors transmit the parameter set of the global model to the two or more terminals (S2103).

Then, the local model is updated by retransmitting at least a part of the parameter set of the global model to one or more terminals. For example, the global model may have a parameter set composed of first to Kth parameters. In this case, in order to save a frequency band, at least a part of the entire parameter set may be transmitted instead of the entire parameter set. More specifically, the server may select at least some of the parameter sets of the global model changed as a result of learning. At this time, at least some of the selected values are determined as values that are changed beyond the set reference value. At least some of the changed parameters are retransmitted to one or more terminals to update the local model. As such, by repeating the above processes, the local model can be changed similarly to the global model.

22 is an implementation according to various embodiments of the present specification.

Referring to FIG. 22 , the communication system for federated learning includes a first terminal 2201, a second terminal 2202, a base station 2203, and a server 2204. Here, the first and second terminals 2201 and 2202 are exemplary, and the number of terminals is not limited to two in various embodiments of the present specification.

The first terminal 2201 stores the first local model 2201m in its memory, and the second model 2202 stores the second local model 2202m in its memory. Here, the neural network structures of the first and second models 2201m and 2202m may be the same. When the first and second models 2201m and 2202m have the same neural network structure, the first and second models 2201m and 2202m have the same first to Mth weights (W ₁ , ..., W _M ) , but the values may be different from each other.

The first and second terminals 2201 and 2202 change the first parameter set into the second parameter set based on the first and second models 2201m and 2202m, respectively. Thereafter, the first and second terminals 2201 and 2202 calculate a difference between the first parameter set and the second parameter set. As a result, the first terminal calculates a first set of difference values (2201s), and the second terminal calculates a second difference value set (2202s).

Referring again to FIG. 22 , it is confirmed that the weight value W ₂ located in the second entry among the difference value sets 2201s of the first terminal 2201 is 0. In this way, even if the weight without the difference value is transmitted to the server 2204, it does not need to be transmitted because it has little or no effect on learning. Accordingly, the first terminal 2201 transmits the first transmission parameter set 2201t excluding the weight of the second entry to the base station 2203.

Unlike this, the second terminal 2202 transmits the second transmission parameter set 2202t including all changed parameters to the base station 2203 since there is no entry in which the difference value is 0 among the difference value sets.

Thereafter, the server 2204 receives the first and second transmission parameter sets 2201t and 2202t and performs learning of the global model.

23 and 24 are average calculation methods of reception parameter sets applied to various embodiments of the present specification.

The server calculates the average of the K-th entries by adding up the K-th entries of the received parameter sets and dividing by the number of parameter sets having the K-th entries (S2301).

Referring to FIG. 24, when receiving the first and second transmission parameter sets 2401t and 2402t, the server divides 2 (Nk=2) for the first, third, and fourth entries, and in the second entry For each element, the average can be calculated by dividing by 1 (Nk = 1).

The server sets the calculated value as the corresponding entry of the new parameter set 2403t whenever the average is calculated for each entry (S2302).

The server changes the existing parameter set into a new parameter set based on the new parameter set consisting of the average of each element (S2303).

Devices used in wireless communication systems

Although not limited thereto, various proposals of the present invention described above may be applied to various fields requiring wireless communication/connection (eg, 5G) between devices.

Hereinafter, it will be exemplified in more detail with reference to the drawings. In the following drawings/description, the same reference numerals may represent the same or corresponding hardware blocks, software blocks or functional blocks unless otherwise specified.

25 illustrates a communication system applied to the present invention.

Referring to FIG. 25, a communication system 1 applied to the present invention includes a wireless device, a base station and a network. Here, the wireless device means a device that performs communication using a radio access technology (eg, 5G New RAT (NR), Long Term Evolution (LTE)), and may be referred to as a communication/wireless/5G device. Although not limited thereto, wireless devices include robots 100a, vehicles 100b-1 and 100b-2, XR (eXtended Reality) devices 100c, hand-held devices 100d, and home appliances 100e. ), an Internet of Thing (IoT) device 100f, and an AI device/server 400. For example, the vehicle may include a vehicle equipped with a wireless communication function, an autonomous vehicle, a vehicle capable of performing inter-vehicle communication, and the like. Here, the vehicle may include an Unmanned Aerial Vehicle (UAV) (eg, a drone). XR devices include Augmented Reality (AR)/Virtual Reality (VR)/Mixed Reality (MR) devices, Head-Mounted Devices (HMDs), Head-Up Displays (HUDs) installed in vehicles, televisions, smartphones, It may be implemented in the form of a computer, wearable device, home appliance, digital signage, vehicle, robot, and the like. A portable device may include a smart phone, a smart pad, a wearable device (eg, a smart watch, a smart glass), a computer (eg, a laptop computer, etc.), and the like. Home appliances may include a TV, a refrigerator, a washing machine, and the like. IoT devices may include sensors, smart meters, and the like. For example, a base station and a network may also be implemented as a wireless device, and a specific wireless device 200a may operate as a base station/network node to other wireless devices.

The wireless devices 100a to 100f may be connected to the network 300 through the base station 200 . AI (Artificial Intelligence) technology may be applied to the wireless devices 100a to 100f, and the wireless devices 100a to 100f may be connected to the AI server 400 through the network 300. The network 300 may be configured using a 3G network, a 4G (eg LTE) network, or a 5G (eg NR) network. The wireless devices 100a to 100f may communicate with each other through the base station 200/network 300, but may also communicate directly (eg, sidelink communication) without going through the base station/network. For example, the vehicles 100b-1 and 100b-2 may perform direct communication (eg, vehicle to vehicle (V2V)/vehicle to everything (V2X) communication). In addition, IoT devices (eg, sensors) may directly communicate with other IoT devices (eg, sensors) or other wireless devices 100a to 100f.

Wireless communication/connection 150a, 150b may be performed between the wireless devices 100a to 100f/base station 200-base station 200/wireless devices 100a to 100f. Here, wireless communication/connection may be performed through various radio access technologies (eg, 5G NR) for uplink/downlink communication 150a and sidelink communication 150b (or D2D communication). Through the wireless communication/connection 150a and 150b, the wireless device and the base station/wireless device may transmit/receive radio signals to each other. For example, the wireless communication/connection 150a and 150b may transmit/receive signals through various physical channels based on all/partial processes of FIG. To this end, based on the various proposals of the present invention, various configuration information setting processes for transmitting / receiving radio signals, various signal processing processes (eg, channel encoding / decoding, modulation / demodulation, resource mapping / demapping, etc.) At least a part of a resource allocation process may be performed.

26 illustrates a wireless device applicable to the present invention.

Referring to FIG. 26 , the first wireless device 100 and the second wireless device 200 may transmit and receive radio signals through various radio access technologies (eg, LTE, NR). Here, {the first wireless device 100, the second wireless device 200} is the {wireless device 100x, the base station 200} of FIG. 25 and/or the {wireless device 100x, the wireless device 100x. } can correspond.

The first wireless device 100 includes one or more processors 102 and one or more memories 104, and may additionally include one or more transceivers 106 and/or one or more antennas 108. The processor 102 controls the memory 104 and/or the transceiver 106 and may be configured to implement the functions, procedures and/or methods described/suggested above. For example, the processor 102 may process information in the memory 104 to generate first information/signal, and transmit a radio signal including the first information/signal through the transceiver 106. In addition, the processor 102 may receive a radio signal including the second information/signal through the transceiver 106, and then store information obtained from signal processing of the second information/signal in the memory 104. The memory 104 may be connected to the processor 102 and may store various information related to the operation of the processor 102 . For example, memory 104 may perform some or all of the processes controlled by processor 102, or store software code including instructions for performing procedures and/or methods described/suggested above. . Here, the processor 102 and memory 104 may be part of a communication modem/circuit/chip designed to implement a wireless communication technology (eg, LTE, NR). The transceiver 106 may be coupled to the processor 102 and may transmit and/or receive wireless signals via one or more antennas 108 . The transceiver 106 may include a transmitter and/or a receiver. The transceiver 106 may be used interchangeably with a radio frequency (RF) unit. In the present invention, a wireless device may mean a communication modem/circuit/chip.

The second wireless device 200 includes one or more processors 202, one or more memories 204, and may further include one or more transceivers 206 and/or one or more antennas 208. The processor 202 controls the memory 204 and/or the transceiver 206 and may be configured to implement the functions, procedures and/or methods described/suggested above. For example, the processor 202 may process information in the memory 204 to generate third information/signal, and transmit a radio signal including the third information/signal through the transceiver 206. In addition, the processor 202 may receive a radio signal including the fourth information/signal through the transceiver 206 and store information obtained from signal processing of the fourth information/signal in the memory 204 . The memory 204 may be connected to the processor 202 and may store various information related to the operation of the processor 202 . For example, memory 204 may perform some or all of the processes controlled by processor 202, or may store software code including instructions for performing procedures and/or methods described/suggested above. . Here, the processor 202 and memory 204 may be part of a communication modem/circuit/chip designed to implement a wireless communication technology (eg, LTE, NR). The transceiver 206 may be coupled to the processor 202 and may transmit and/or receive wireless signals via one or more antennas 208 . The transceiver 206 may include a transmitter and/or a receiver. The transceiver 206 may be used interchangeably with an RF unit. In the present invention, a wireless device may mean a communication modem/circuit/chip.

Hereinafter, hardware elements of the wireless devices 100 and 200 will be described in more detail. Although not limited to this, one or more protocol layers may be implemented by one or more processors 102, 202. For example, one or more processors 102, 202 may implement one or more layers (eg, functional layers such as PHY, MAC, RLC, PDCP, RRC, SDAP). One or more processors 102, 202 may generate one or more Protocol Data Units (PDUs) and/or one or more Service Data Units (SDUs) according to functions, procedures, proposals and/or methods disclosed herein. One or more processors 102, 202 may generate messages, control information, data or information according to functions, procedures, suggestions and/or methods disclosed herein. One or more processors 102, 202 generate PDUs, SDUs, messages, control information, data or signals (e.g., baseband signals) containing information according to the functions, procedures, proposals and/or methods disclosed herein , can be provided to one or more transceivers 106, 206. One or more processors 102, 202 may receive signals (eg, baseband signals) from one or more transceivers 106, 206 and generate PDUs, SDUs according to functions, procedures, proposals and/or methods disclosed herein. , messages, control information, data or information can be obtained.

One or more processors 102, 202 may be referred to as a controller, microcontroller, microprocessor or microcomputer. One or more processors 102, 202 may be implemented by hardware, firmware, software, or a combination thereof. For example, one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), one or more Digital Signal Processing Devices (DSPDs), one or more Programmable Logic Devices (PLDs), or one or more Field Programmable Gate Arrays (FPGAs). may be included in one or more processors 102 and 202. The functions, procedures, proposals and/or methods disclosed in this document may be implemented using firmware or software, and the firmware or software may be implemented to include modules, procedures, functions, and the like. Firmware or software configured to perform the functions, procedures, suggestions and/or methods disclosed herein may be included in one or more processors (102, 202) or stored in one or more memories (104, 204) and may be stored in one or more processors (102, 202). 202). The functions, procedures, suggestions and/or methods disclosed herein may be implemented using firmware or software in the form of codes, instructions and/or sets of instructions.

One or more memories 104, 204 may be coupled with one or more processors 102, 202 and may store various types of data, signals, messages, information, programs, codes, instructions and/or instructions. One or more memories 104, 204 may be comprised of ROM, RAM, EPROM, flash memory, hard drives, registers, cache memory, computer readable storage media, and/or combinations thereof. One or more memories 104, 204 may be located internally and/or external to one or more processors 102, 202. Additionally, one or more memories 104, 204 may be coupled to one or more processors 102, 202 through various technologies, such as wired or wireless connections.

One or more transceivers 106, 206 may transmit user data, control information, radio signals/channels, etc., as referred to in the methods and/or operational flow charts herein, to one or more other devices. One or more of the transceivers 106, 206 may receive user data, control information, radio signals/channels, etc. referred to in functions, procedures, proposals, methods and/or operational flow charts, etc. disclosed herein from one or more other devices. For example, one or more transceivers 106 and 206 may be connected to one or more processors 102 and 202 and transmit and receive wireless signals. For example, one or more processors 102, 202 may control one or more transceivers 106, 206 to transmit user data, control information, or radio signals to one or more other devices. Additionally, one or more processors 102, 202 may control one or more transceivers 106, 206 to receive user data, control information, or radio signals from one or more other devices. In addition, one or more transceivers 106, 206 may be coupled with one or more antennas 108, 208, and one or more transceivers 106, 206, via one or more antennas 108, 208 may perform functions, procedures disclosed herein. , It can be set to transmit and receive user data, control information, radio signals / channels, etc. mentioned in proposals, methods, and / or operational flowcharts. In this document, one or more antennas may be a plurality of physical antennas or a plurality of logical antennas (eg, antenna ports). One or more transceivers (106, 206) convert the received radio signals/channels from RF band signals in order to process the received user data, control information, radio signals/channels, etc. using one or more processors (102, 202). It can be converted into a baseband signal. One or more transceivers 106 and 206 may convert user data, control information, and radio signals/channels processed by one or more processors 102 and 202 from baseband signals to RF band signals. To this end, one or more of the transceivers 106, 206 may include (analog) oscillators and/or filters.

27 illustrates a signal processing circuit for a transmission signal.

Referring to FIG. 27 , the signal processing circuit 1000 may include a scrambler 1010, a modulator 1020, a layer mapper 1030, a precoder 1040, a resource mapper 1050, and a signal generator 1060. there is. Although not limited to this, the operations/functions of FIG. 27 may be performed by processors 102 and 202 and/or transceivers 106 and 206 of FIG. 26 . The hardware elements of FIG. 27 may be implemented in processors 102 and 202 and/or transceivers 106 and 206 of FIG. 26 . For example, blocks 1010-1060 may be implemented in the processors 102 and 202 of FIG. 26 . Also, blocks 1010 to 1050 may be implemented in the processors 102 and 202 of FIG. 26 , and block 1060 may be implemented in the transceivers 106 and 206 of FIG. 26 .

The codeword may be converted into a radio signal through the signal processing circuit 1000 of FIG. 27 . Here, a codeword is an encoded bit sequence of an information block. Information blocks may include transport blocks (eg, UL-SCH transport blocks, DL-SCH transport blocks). The radio signal may be transmitted through various physical channels (eg, PUSCH, PDSCH) of FIG.

Specifically, the codeword may be converted into a scrambled bit sequence by the scrambler 1010. A scramble sequence used for scrambling is generated based on an initialization value, and the initialization value may include ID information of a wireless device. The scrambled bit sequence may be modulated into a modulation symbol sequence by modulator 1020. The modulation scheme may include pi/2-Binary Phase Shift Keying (pi/2-BPSK), m-Phase Shift Keying (m-PSK), m-Quadrature Amplitude Modulation (m-QAM), and the like. The complex modulation symbol sequence may be mapped to one or more transport layers by the layer mapper 1030. Modulation symbols of each transport layer may be mapped to the corresponding antenna port(s) by the precoder 1040 (precoding). The output z of the precoder 1040 can be obtained by multiplying the output y of the layer mapper 1030 by the N*M precoding matrix W. Here, N is the number of antenna ports and M is the number of transport layers. Here, the precoder 1040 may perform precoding after performing transform precoding (eg, DFT transformation) on complex modulation symbols. Also, the precoder 1040 may perform precoding without performing transform precoding.

The resource mapper 1050 may map modulation symbols of each antenna port to time-frequency resources. The time-frequency resource may include a plurality of symbols (eg, CP-OFDMA symbols and DFT-s-OFDMA symbols) in the time domain and a plurality of subcarriers in the frequency domain. The signal generator 1060 generates a radio signal from the mapped modulation symbols, and the generated radio signal can be transmitted to other devices through each antenna. To this end, the signal generator 1060 may include an inverse fast Fourier transform (IFFT) module, a cyclic prefix (CP) inserter, a digital-to-analog converter (DAC), a frequency uplink converter, and the like. .

The signal processing process for the received signal in the wireless device may be configured in reverse to the signal processing process 1010 to 1060 of FIG. 27 . For example, a wireless device (eg, 100 or 200 of FIG. 26 ) may receive a wireless signal from the outside through an antenna port/transceiver. The received radio signal may be converted into a baseband signal through a signal restorer. To this end, the signal restorer may include a frequency downlink converter, an analog-to-digital converter (ADC), a CP remover, and a fast Fourier transform (FFT) module. Thereafter, the baseband signal may be restored to a codeword through a resource de-mapper process, a postcoding process, a demodulation process, and a de-scramble process. The codeword may be restored to an original information block through decoding. Accordingly, a signal processing circuit (not shown) for a received signal may include a signal restorer, a resource demapper, a postcoder, a demodulator, a descrambler, and a decoder.

28 shows another example of a wireless device applied to the present invention. A wireless device may be implemented in various forms according to use-examples/services (see FIGS. 25 and 29 to 34).

Referring to FIG. 28, wireless devices 100 and 200 correspond to the wireless devices 100 and 200 of FIG. 26, and include various elements, components, units/units, and/or modules. ) can be configured. For example, the wireless devices 100 and 200 may include a communication unit 110 , a control unit 120 , a memory unit 130 and an additional element 140 . The communication unit may include communication circuitry 112 and transceiver(s) 114 . For example, communication circuitry 112 may include one or more processors 102, 202 of FIG. 26 and/or one or more memories 104, 204. For example, transceiver(s) 114 may include one or more transceivers 106, 206 of FIG. 26 and/or one or more antennas 108, 208. The control unit 120 is electrically connected to the communication unit 110, the memory unit 130, and the additional element 140 and controls overall operations of the wireless device. For example, the control unit 120 may control electrical/mechanical operations of the wireless device based on programs/codes/commands/information stored in the memory unit 130. In addition, the control unit 120 transmits the information stored in the memory unit 130 to the outside (eg, another communication device) through the communication unit 110 through a wireless/wired interface, or transmits the information stored in the memory unit 130 to the outside (eg, another communication device) through the communication unit 110. Information received through a wireless/wired interface from other communication devices) may be stored in the memory unit 130 .

The additional element 140 may be configured in various ways according to the type of wireless device. For example, the additional element 140 may include at least one of a power unit/battery, an I/O unit, a driving unit, and a computing unit. Although not limited thereto, the wireless device may be a robot (Fig. 25, 100a), a vehicle (Fig. 25, 100b-1, 100b-2), an XR device (Fig. 25, 100c), a mobile device (Fig. 25, 100d), a home appliance. (FIG. 25, 100e), IoT device (FIG. 25, 100f), digital broadcast terminal, hologram device, public safety device, MTC device, medical device, fintech device (or financial device), security device, climate/environmental device, It may be implemented in the form of an AI server/device (Fig. 25, 400), a base station (Fig. 25, 200), a network node, and the like. Wireless devices can be mobile or used in a fixed location depending on the use-case/service.

In FIG. 28, various elements, components, units/units, and/or modules in the wireless devices 100 and 200 may be entirely interconnected through a wired interface or at least partially connected wirelessly through the communication unit 110. For example, in the wireless devices 100 and 200, the control unit 120 and the communication unit 110 are connected by wire, and the control unit 120 and the first units (eg, 130 and 140) are connected through the communication unit 110. Can be connected wirelessly. Additionally, each element, component, unit/unit, and/or module within the wireless device 100, 200 may further include one or more elements. For example, the control unit 120 may be composed of one or more processor sets. For example, the controller 120 may include a set of a communication control processor, an application processor, an electronic control unit (ECU), a graphic processing processor, a memory control processor, and the like. As another example, the memory unit 130 may include random access memory (RAM), dynamic RAM (DRAM), read only memory (ROM), flash memory, volatile memory, and non-volatile memory. volatile memory) and/or a combination thereof.

Hereinafter, an implementation example of FIG. 28 will be described in more detail with reference to drawings.

29 illustrates a portable device applied to the present invention. A portable device may include a smart phone, a smart pad, a wearable device (eg, a smart watch, a smart glass), and a portable computer (eg, a laptop computer). A mobile device may be referred to as a mobile station (MS), a user terminal (UT), a mobile subscriber station (MSS), a subscriber station (SS), an advanced mobile station (AMS), or a wireless terminal (WT).

Referring to FIG. 29, the portable device 100 includes an antenna unit 108, a communication unit 110, a control unit 120, a memory unit 130, a power supply unit 140a, an interface unit 140b, and an input/output unit 140c. ) may be included. The antenna unit 108 may be configured as part of the communication unit 110 . Blocks 110 to 130/140a to 140c respectively correspond to blocks 110 to 130/140 of FIG. 28 .

The communication unit 110 may transmit/receive signals (eg, data, control signals, etc.) with other wireless devices and base stations. The controller 120 may perform various operations by controlling components of the portable device 100 . The control unit 120 may include an application processor (AP). The memory unit 130 may store data/parameters/programs/codes/commands necessary for driving the portable device 100 . In addition, the memory unit 130 may store input/output data/information. The power supply unit 140a supplies power to the portable device 100 and may include a wired/wireless charging circuit, a battery, and the like. The interface unit 140b may support connection between the portable device 100 and other external devices. The interface unit 140b may include various ports (eg, audio input/output ports and video input/output ports) for connection with external devices. The input/output unit 140c may receive or output image information/signal, audio information/signal, data, and/or information input from a user. The input/output unit 140c may include a camera, a microphone, a user input unit, a display unit 140d, a speaker, and/or a haptic module.

For example, in the case of data communication, the input/output unit 140c acquires information/signals (eg, touch, text, voice, image, video) input from the user, and the acquired information/signals are stored in the memory unit 130. can be stored The communication unit 110 may convert the information/signal stored in the memory into a wireless signal, and directly transmit the converted wireless signal to another wireless device or to a base station. In addition, the communication unit 110 may receive a radio signal from another wireless device or base station and then restore the received radio signal to original information/signal. After the restored information/signal is stored in the memory unit 130, it may be output in various forms (eg, text, voice, image, video, haptic) through the input/output unit 140c.

30 illustrates a vehicle or autonomous vehicle to which the present invention is applied. Vehicles or autonomous vehicles may be implemented as mobile robots, vehicles, trains, manned/unmanned aerial vehicles (AVs), ships, and the like.

Referring to FIG. 30 , a vehicle or autonomous vehicle 100 includes an antenna unit 108, a communication unit 110, a control unit 120, a driving unit 140a, a power supply unit 140b, a sensor unit 140c, and an autonomous driving unit. A portion 140d may be included. The antenna unit 108 may be configured as part of the communication unit 110 . Blocks 110/130/140a to 140d respectively correspond to blocks 110/130/140 of FIG. 28 .

The communication unit 110 may transmit/receive signals (eg, data, control signals, etc.) with external devices such as other vehicles, base stations (e.g. base stations, roadside base stations, etc.), servers, and the like. The controller 120 may perform various operations by controlling elements of the vehicle or autonomous vehicle 100 . The controller 120 may include an Electronic Control Unit (ECU). The driving unit 140a may drive the vehicle or autonomous vehicle 100 on the ground. The driving unit 140a may include an engine, a motor, a power train, a wheel, a brake, a steering device, and the like. The power supply unit 140b supplies power to the vehicle or autonomous vehicle 100, and may include a wired/wireless charging circuit, a battery, and the like. The sensor unit 140c may obtain vehicle conditions, surrounding environment information, and user information. The sensor unit 140c includes an inertial measurement unit (IMU) sensor, a collision sensor, a wheel sensor, a speed sensor, an inclination sensor, a weight detection sensor, a heading sensor, a position module, and a vehicle forward. /Can include a reverse sensor, battery sensor, fuel sensor, tire sensor, steering sensor, temperature sensor, humidity sensor, ultrasonic sensor, illuminance sensor, pedal position sensor, and the like. The autonomous driving unit 140d includes a technology for maintaining a driving lane, a technology for automatically adjusting speed such as adaptive cruise control, a technology for automatically driving along a predetermined route, and a technology for automatically setting a route when a destination is set and driving. technology can be implemented.

For example, the communication unit 110 may receive map data, traffic information data, and the like from an external server. The autonomous driving unit 140d may generate an autonomous driving route and a driving plan based on the acquired data. The controller 120 may control the driving unit 140a so that the vehicle or autonomous vehicle 100 moves along the autonomous driving path according to the driving plan (eg, speed/direction adjustment). During autonomous driving, the communicator 110 may non-/periodically obtain the latest traffic information data from an external server and obtain surrounding traffic information data from surrounding vehicles. In addition, during autonomous driving, the sensor unit 140c may acquire vehicle state and surrounding environment information. The autonomous driving unit 140d may update an autonomous driving route and a driving plan based on newly acquired data/information. The communication unit 110 may transmit information about a vehicle location, an autonomous driving route, a driving plan, and the like to an external server. The external server may predict traffic information data in advance using AI technology based on information collected from the vehicle or self-driving vehicles, and may provide the predicted traffic information data to the vehicle or self-driving vehicles.

31 illustrates a vehicle to which the present invention is applied. A vehicle may be implemented as a means of transportation, a train, an air vehicle, a ship, and the like.

Referring to FIG. 31 , the vehicle 100 may include a communication unit 110, a control unit 120, a memory unit 130, an input/output unit 140a, and a position measurement unit 140b. Here, blocks 110 to 130/140a to 140b respectively correspond to blocks 110 to 130/140 of FIG. 28 .

The communication unit 110 may transmit/receive signals (eg, data, control signals, etc.) with other vehicles or external devices such as base stations. The controller 120 may perform various operations by controlling components of the vehicle 100 . The memory unit 130 may store data/parameters/programs/codes/commands supporting various functions of the vehicle 100 . The input/output unit 140a may output an AR/VR object based on information in the memory unit 130. The input/output unit 140a may include a HUD. The location measurement unit 140b may obtain location information of the vehicle 100 . The location information may include absolute location information of the vehicle 100, location information within a driving line, acceleration information, and location information with neighboring vehicles. The location measurement unit 140b may include GPS and various sensors.

For example, the communication unit 110 of the vehicle 100 may receive map information, traffic information, and the like from an external server and store them in the memory unit 130 . The location measurement unit 140b may acquire vehicle location information through GPS and various sensors and store it in the memory unit 130 . The controller 120 may generate a virtual object based on map information, traffic information, vehicle location information, etc., and the input/output unit 140a may display the created virtual object on a window in the vehicle (1410, 1420). In addition, the controller 120 may determine whether the vehicle 100 is normally operated within the driving line based on the vehicle location information. When the vehicle 100 abnormally deviate from the driving line, the controller 120 may display a warning on a window in the vehicle through the input/output unit 140a. In addition, the controller 120 may broadcast a warning message about driving abnormality to surrounding vehicles through the communication unit 110 . Depending on circumstances, the controller 120 may transmit vehicle location information and information on driving/vehicle abnormalities to related agencies through the communication unit 110 .

32 illustrates an XR device applied to the present invention. The XR device may be implemented as an HMD, a head-up display (HUD) provided in a vehicle, a television, a smartphone, a computer, a wearable device, a home appliance, a digital signage, a vehicle, a robot, and the like.

Referring to FIG. 32, the XR device 100a may include a communication unit 110, a control unit 120, a memory unit 130, an input/output unit 140a, a sensor unit 140b, and a power supply unit 140c. . Here, blocks 110 to 130/140a to 140c respectively correspond to blocks 110 to 130/140 of FIG. 28 .

The communication unit 110 may transmit/receive signals (eg, media data, control signals, etc.) with external devices such as other wireless devices, portable devices, or media servers. Media data may include video, image, sound, and the like. The controller 120 may perform various operations by controlling components of the XR device 100a. For example, the controller 120 may be configured to control and/or perform procedures such as video/image acquisition, (video/image) encoding, and metadata generation and processing. The memory unit 130 may store data/parameters/programs/codes/commands necessary for driving the XR device 100a/creating an XR object. The input/output unit 140a may obtain control information, data, etc. from the outside and output the created XR object. The input/output unit 140a may include a camera, a microphone, a user input unit, a display unit, a speaker, and/or a haptic module. The sensor unit 140b may obtain XR device status, surrounding environment information, user information, and the like. The sensor unit 140b may include a proximity sensor, an illuminance sensor, an acceleration sensor, a magnetic sensor, a gyro sensor, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, an optical sensor, a microphone, and/or a radar. there is. The power supply unit 140c supplies power to the XR device 100a and may include a wired/wireless charging circuit, a battery, and the like.

For example, the memory unit 130 of the XR device 100a may include information (eg, data, etc.) necessary for generating an XR object (eg, AR/VR/MR object). The input/output unit 140a may obtain a command to operate the XR device 100a from a user, and the control unit 120 may drive the XR device 100a according to the user's driving command. For example, when a user tries to watch a movie, news, etc. through the XR device 100a, the control unit 120 transmits content request information to another device (eg, the mobile device 100b) or through the communication unit 130. can be sent to the media server. The communication unit 130 may download/stream content such as movies and news from another device (eg, the portable device 100b) or a media server to the memory unit 130 . The control unit 120 controls and/or performs procedures such as video/image acquisition, (video/image) encoding, metadata generation/processing, etc. for content, and acquisition through the input/output unit 140a/sensor unit 140b. An XR object may be created/output based on information about a surrounding space or a real object.

In addition, the XR device 100a is wirelessly connected to the portable device 100b through the communication unit 110, and the operation of the XR device 100a may be controlled by the portable device 100b. For example, the mobile device 100b may operate as a controller for the XR device 100a. To this end, the XR device 100a may acquire 3D location information of the portable device 100b and then generate and output an XR object corresponding to the portable device 100b.

33 illustrates a robot applied to the present invention. Robots may be classified into industrial, medical, household, military, and the like depending on the purpose or field of use.

Referring to FIG. 33 , the robot 100 may include a communication unit 110, a control unit 120, a memory unit 130, an input/output unit 140a, a sensor unit 140b, and a driving unit 140c. Here, blocks 110 to 130/140a to 140c respectively correspond to blocks 110 to 130/140 of FIG. 28 .

The communication unit 110 may transmit/receive signals (eg, driving information, control signals, etc.) with external devices such as other wireless devices, other robots, or control servers. The controller 120 may perform various operations by controlling components of the robot 100 . The memory unit 130 may store data/parameters/programs/codes/commands supporting various functions of the robot 100. The input/output unit 140a may obtain information from the outside of the robot 100 and output the information to the outside of the robot 100 . The input/output unit 140a may include a camera, a microphone, a user input unit, a display unit, a speaker, and/or a haptic module. The sensor unit 140b may obtain internal information of the robot 100, surrounding environment information, user information, and the like. The sensor unit 140b may include a proximity sensor, an illuminance sensor, an acceleration sensor, a magnetic sensor, a gyro sensor, an inertial sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, an optical sensor, a microphone, a radar, and the like. The driving unit 140c may perform various physical operations such as moving a robot joint. In addition, the driving unit 140c may make the robot 100 drive on the ground or fly in the air. The driving unit 140c may include actuators, motors, wheels, brakes, propellers, and the like.

34 illustrates an AI device applied to the present invention. AI devices include fixed or mobile devices such as TVs, projectors, smartphones, PCs, laptops, digital broadcasting terminals, tablet PCs, wearable devices, set-top boxes (STBs), radios, washing machines, refrigerators, digital signage, robots, and vehicles. It can be implemented with possible devices and the like.

Referring to FIG. 34, the AI device 100 includes a communication unit 110, a control unit 120, a memory unit 130, an input/output unit 140a/140b, a running processor unit 140c, and a sensor unit 140d. can include Blocks 110 to 130/140a to 140d respectively correspond to blocks 110 to 130/140 of FIG. 28 .

The communication unit 110 transmits wired and wireless signals (eg, sensor information, user input, and learning) to external devices such as other AI devices (eg, FIG. models, control signals, etc.) can be transmitted and received. To this end, the communication unit 110 may transmit information in the memory unit 130 to an external device or transmit a signal received from the external device to the memory unit 130 .

The controller 120 may determine at least one feasible operation of the AI device 100 based on information determined or generated using a data analysis algorithm or a machine learning algorithm. In addition, the controller 120 may perform the determined operation by controlling components of the AI device 100 . For example, the controller 120 may request, retrieve, receive, or utilize data from the learning processor unit 140c or the memory unit 130, and may perform a predicted operation among at least one feasible operation or an operation determined to be desirable. Components of the AI device 100 may be controlled to execute an operation. In addition, the control unit 120 collects history information including user feedback on the operation contents or operation of the AI device 100 and stores it in the memory unit 130 or the running processor unit 140c, or the AI server ( 25, 400) can be transmitted to an external device. The collected history information can be used to update the learning model.

The memory unit 130 may store data supporting various functions of the AI device 100 . For example, the memory unit 130 may store data obtained from the input unit 140a, data obtained from the communication unit 110, output data from the learning processor unit 140c, and data obtained from the sensing unit 140. In addition, the memory unit 130 may store control information and/or software codes necessary for operation/execution of the control unit 120 .

The input unit 140a may obtain various types of data from the outside of the AI device 100. For example, the input unit 120 may obtain learning data for model learning and input data to which the learning model is to be applied. The input unit 140a may include a camera, a microphone, and/or a user input unit. The output unit 140b may generate an output related to sight, hearing, or touch. The output unit 140b may include a display unit, a speaker, and/or a haptic module. The sensing unit 140 may obtain at least one of internal information of the AI device 100, surrounding environment information of the AI device 100, and user information by using various sensors. The sensing unit 140 may include a proximity sensor, an illuminance sensor, an acceleration sensor, a magnetic sensor, a gyro sensor, an inertial sensor, an RGB sensor, an IR sensor, a fingerprint recognition sensor, an ultrasonic sensor, an optical sensor, a microphone, and/or a radar. there is.

The learning processor unit 140c may learn a model composed of an artificial neural network using learning data. The running processor unit 140c may perform AI processing together with the running processor unit of the AI server (400 in FIG. 25). The learning processor unit 140c may process information received from an external device through the communication unit 110 and/or information stored in the memory unit 130 . In addition, the output value of the learning processor unit 140c may be transmitted to an external device through the communication unit 110 and/or stored in the memory unit 130.

The above specification can be implemented as computer readable code on a medium on which a program is recorded. The computer-readable medium includes all types of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable media include Hard Disk Drive (HDD), Solid State Disk (SSD), Silicon Disk Drive (SDD), ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage device, etc. , and also includes those implemented in the form of a carrier wave (eg, transmission over the Internet). Accordingly, the above detailed description should not be construed as limiting in all respects and should be considered illustrative. The scope of this specification should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of this specification are included in the scope of this specification.

Claims (15)

In the associated learning method of a terminal in a wireless communication system,
changing a first parameter set associated with the local model into a second parameter set based on the local model;
selecting at least some of the second parameter sets based on a difference between the first and second parameter sets; and
Transmitting the selected at least a part to a server;
Including, federated learning method. According to claim 1,
obtaining one or more raw data;
extracting one or more feature information based on the one or more raw data;
including,
The changing step is
Performed based on the one or more characteristic information and one or more labels associated with the characteristic information, the federated learning method. According to claim 2,
allocating a label to at least some of the one or more pieces of characteristic information;
Including more,
The step of assigning the label,
In response to receiving an input related to the control operation of the terminal, allocating a command related to the control operation as a label for the at least some feature information. According to claim 2,
allocating a label to at least some of the one or more pieces of characteristic information;
Including more,
The extraction step is
Performed in response to receiving an input related to a control operation of the terminal,
The step of assigning the label,
Allocating a command associated with the control operation as a label for the at least some feature information. According to claim 1,
The selection step is
The federated learning method of selecting at least some parameters in which a difference between the first and second parameter sets exceeds a set reference value. According to claim 5,
The federated learning method, wherein the reference value is 0. According to claim 1,
At least some parameters transmitted to the base station,
For learning of a global model stored in the server. In the federated learning method of a server in a wireless communication system,
Receiving first parameter sets for global model learning from two or more terminals;
changing at least some of the parameter sets of the global model into a new second parameter set calculated based on element-wise averaging of the received first parameter sets; and
transmitting the changed third parameter set to the two or more terminals;
Including, federated learning method. According to claim 8,
The first parameter sets,
Parameter sets calculated based on local models stored in the two or more terminals, a federated learning method. According to claim 9,
The local models stored in the two or more terminals,
A federated learning method with the same neural network structure. According to claim 9,
The first parameter sets,
Parameters selected based on differences in parameter sets before and after training of the local model, a federated learning method. According to claim 11,
The changed parameters are
A federated learning method comprising at least some parameters in which the difference between the parameter sets before and after learning exceeds a set reference value. According to claim 8,
The global model is
A federated learning method having the same neural network structure as the local model. According to claim 8,
The sending step is
The federated learning method of transmitting only parameters whose values are changed based on the second parameter set among the changed third parameter sets. one or more transceivers;
one or more processors; and
one or more memories coupled to the one or more processors and storing instructions;
The instructions, when executed by the one or more processors, cause the one or more processors to support operations for federated inference, the operations comprising:
changing a first parameter set associated with the local model into a second parameter set based on the local model;
selecting at least some of the second parameter sets based on a difference between the first and second parameter sets; and
transmitting at least a portion of the selected portion to a server;
Including, terminal.

Images