KR20240001134A - Method and device for transmitting and receiving signals in a wireless communication system using an auto encoder - Google Patents

Abstract

This specification provides a method for a transmitter to transmit a signal in a wireless communication system using an auto encoder.
More specifically, the method performed by the transmitting end includes encoding input data based on a pre-trained transmitting end encoder neural network; And transmitting the signal to a receiving end based on the encoded input data, wherein the transmitting end encoder neural network includes (i) a first encoder neural network that receives the input data and outputs a first codeword. a network, (ii) an interleaver that receives the first codeword, interleaves it, and outputs the interleaved first codeword, and (iii) receives the interleaved first codeword and encodes the encoded codeword. It is characterized by serially including a second encoder neural network that outputs a second codeword constituting the input data.

Description

Method and device for transmitting and receiving signals in a wireless communication system using an auto encoder

This specification relates to transmitting and receiving signals using an auto encoder, and more specifically, to a method and device for transmitting and receiving signals in a wireless communication system using an auto encoder.

Wireless communication systems are being 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 that can support communication with multiple users by sharing available system resources (bandwidth, transmission power, etc.). Examples of multiple access systems include Code Division Multiple Access (CDMA) systems, Frequency Division Multiple Access (FDMA) systems, Time Division Multiple Access (TDMA) systems, Space Division Multiple Access (SDMA), and Orthogonal Frequency Division Multiple Access (OFDMA) systems. , SC-FDMA (Single Carrier Frequency Division Multiple Access) system, IDMA (Interleave Division Multiple Access) system, etc.

The purpose of this specification is to provide a method and device for correcting transmission errors occurring in communication using a neural network.

The purpose of this specification is to provide a method and device for reducing the complexity of a neural network for correcting transmission errors occurring in communication.

The technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

This specification provides a method and device for transmitting and receiving signals in a wireless communication system using an auto encoder.

More specifically, the present specification provides a method for a transmitter to transmit a signal in a wireless communication system using an auto encoder, including the steps of encoding input data based on a pre-trained transmitter encoder neural network; And transmitting the signal to a receiving end based on the encoded input data, wherein the transmitting end encoder neural network includes (i) a first encoder neural network that receives the input data and outputs a first codeword. a network, (ii) an interleaver that receives the first codeword, interleaves it, and outputs the interleaved first codeword, and (iii) receives the interleaved first codeword and encodes the encoded codeword. It is characterized by serially including a second encoder neural network that outputs a second codeword constituting the input data.

In addition, in this specification, the first encoder neural network consists of n1 first encoder component neural networks, and the second encoder neural network consists of n2 first encoder component neural networks and n3 second encoder component neural networks. It may be characterized in that n1, n2, and n3 are positive integers other than 0.

In addition, in this specification, based on the length of the input data being k, the code rate of the first encoder neural network is k/n1, and the code rate of the second encoder neural network is n1/( n2+n3), and the overall code rate of the transmitter encoder neural network may be k/(n2+n3).

Additionally, in this specification, the first encoder component neural network and the second encoder component neural network each consist of at least one layer, and the first encoder component neural network and the second encoder component neural network are It may be characterized in that it is configured based on a neural network unit including (i) a filtering function for filtering the input value and (ii) an activation function that receives the output of the filtering function. there is.

Additionally, in this specification, the first encoder component neural network and the second encoder component neural network may each be composed of two layers.

Additionally, the present specification may be characterized in that the values of n1 and n2 are the same, and n3 is 1.

In addition, in this specification, the first encoder component neural network includes a first neural network unit that receives one input value and a second neural network unit that receives an output of the first neural network unit, It may be characterized in that linear regression is applied to the output of the second neural network component, and an eLu (exponential linear unit) function is applied to the output of the second neural network component to which the linear regression is applied. there is.

In addition, in this specification, the second encoder component neural network includes a third neural network unit that receives the n1 input values and a fourth neural network unit that receives the output values of the third neural network unit. And, the linear regression may be applied to the output value of the fourth neural network structural unit, and the eLu function may be applied to the output value of the fourth neural network structural unit to which the linear regression has been applied.

In addition, in this specification, the first neural network structural unit and the third neural network structural unit constitute a first layer, which is one of the two layers, and the second neural network structural unit and the fourth neural network structural unit The unit may be characterized as configuring a second layer, which is the remaining one of the two layers.

In addition, the present specification may be characterized in that the number of output channels through which output values of each of the first to fourth neural network structural units are output is 100.

In addition, in this specification, the i+1th bit in the first codeword bit stream input to the interleaver is interleaved based on the equation below,

[Equation]

Here, L is the size of the interleaver, f0 is a constant offset, f1 is a prime number that satisfies the disjoint relationship with L, and f2 is the power of 2 among the divisors of L. It may be characterized as a number corresponding to a square number (power of 2).

Additionally, in this specification, f0 may be 0, and a parallel operation may be performed on the bit string of the first codeword equal to the value of f2.

In addition, the present specification provides that the specific neural network unit constituting the first layer among the at least one layer further includes a pooling layer for reducing the number of output channels of the filtering function, The pooling layer may be located between the filtering function and the activation function.

In addition, the present specification may be characterized in that each neural network structural unit constituting the layers located after the first layer further includes the pooling layer.

In addition, this specification provides a transmitter for transmitting and receiving signals in a wireless communication system using an auto encoder, including a transmitter for transmitting a wireless signal; A receiver for receiving wireless signals; at least one processor; and at least one computer memory operably connectable to the at least one processor and storing instructions that, when executed by the at least one processor, perform operations, the operations comprising: pre-learning Encoding input data based on a transmitter encoder neural network; And transmitting the signal to a receiving end based on the encoded input data, wherein the transmitting end encoder neural network includes (i) a first encoder neural network that receives the input data and outputs a first codeword. a network, (ii) an interleaver that receives the first codeword, interleaves it, and outputs the interleaved first codeword, and (iii) receives the interleaved first codeword and encodes the encoded codeword. It is characterized by serially including a second encoder neural network that outputs a second codeword constituting the input data.

In addition, this specification provides that a method for a receiving end to receive a signal in a wireless communication system using an auto encoder is to receive a signal containing data encoded based on a pre-learned transmitting end encoder neural network. Receiving from a transmitting end; Decoding the encoded data included in the signal based on a pre-trained receiving end decoder neural network, wherein the receiving end decoder neural network is at least one sequentially connected basic decoder neural network. Includes, wherein each of the at least one basic decoder neural network (i) performs decoding on the second codeword output from the second encoder neural network included in the transmitter encoder neural network, and outputs the second decoded codeword a second decoder neural network that outputs, (ii) a de-interleaver (de) that receives the second decoded codeword, de-interleaves it, and outputs the de-interleaved second decoded codeword; -interleaver) and (iii) performing decoding on the de-interleaved second decoded codeword and outputting a first decoded codeword based on decoding on the de-interleaved second decoded codeword. It is characterized in that it includes a first decoder neural network serially.

In addition, the present specification provides a transmitter for transmitting and receiving signals in a wireless communication system, including a transmitter for transmitting a wireless signal; A receiver for receiving wireless signals; at least one processor; and at least one computer memory operably connectable to the at least one processor and storing instructions that, when executed by the at least one processor, perform operations, the operations comprising: pre-learning Receiving a signal containing data encoded based on a neural network of a transmitter encoder from a transmitter; Decoding the encoded data included in the signal based on a pre-trained receiving end decoder neural network, wherein the receiving end decoder neural network is at least one sequentially connected basic decoder neural network. Includes, wherein each of the at least one basic decoder neural network (i) performs decoding on the second codeword output from the second encoder neural network included in the transmitter encoder neural network, and outputs the second decoded codeword a second decoder neural network that outputs, (ii) a de-interleaver (de) that receives the second decoded codeword, de-interleaves it, and outputs the de-interleaved second decoded codeword; -interleaver) and (iii) performing decoding on the de-interleaved second decoded codeword and outputting a first decoded codeword based on decoding on the de-interleaved second decoded codeword. It is characterized in that it includes a first decoder neural network serially.

In addition, the present specification provides that, in a non-transitory computer readable medium (CRM) storing one or more instructions, one or more instructions executable by one or more processors are transmitted through a pre-trained transmitter encoder neural network. The input data is encoded based on a neural network, and the signal is transmitted to the receiving end based on the encoded input data. The encoder neural network at the transmitting end (i) receives the input data and transmits the signal to the receiving end. 1 A first encoder neural network that outputs a codeword, (ii) an interleaver that receives the first codeword, interleaves it, and outputs the interleaved first codeword, and (iii) It is characterized by serially comprising a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data.

In addition, the present specification relates to a device including one or more memories and one or more processors functionally connected to the one or more memories, wherein the one or more processors are configured to use a pre-trained transmitter encoder neural network of the device. The device is controlled to encode input data based on a network (neural network), and the device is controlled to transmit the signal to a receiving end based on the encoded input data, wherein the transmitting end encoder neural network (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword, interleaves it, and outputs the first codeword serially comprising an interleaver that outputs an interleaver and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data. It is characterized by

This specification has the effect of effectively correcting transmission errors occurring in communication by utilizing a neural network.

This specification has the effect of reducing the complexity of a neural network for correcting transmission errors occurring in communication.

The effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned can be clearly understood by those skilled in the art from the description below. .

The accompanying drawings, which are included as part of the detailed description to aid understanding of the present invention, provide embodiments of the present invention, and together with the detailed description, explain technical features of the present invention.
Figure 1 illustrates physical channels and typical signal transmission used in a 3GPP system.
Figure 2 is a diagram showing an example of a communication structure that can be provided in a 6G system.
Figure 3 is a diagram showing an example of a perceptron structure.
Figure 4 is a diagram showing an example of a multi-layer perceptron structure.
Figure 5 is a diagram showing an example of a deep neural network.
Figure 6 is a diagram showing an example of a convolutional neural network.
Figure 7 is a diagram showing an example of a filter operation in a convolutional neural network.
Figure 8 shows an example of a neural network structure in which a circular loop exists.
Figure 9 shows an example of the operational structure of a recurrent neural network.
Figure 10 is a diagram showing an example of a wireless communication system using an auto encoder.
Figure 11 is a diagram showing an example of the conceptual structure of a communication chain using an auto encoder.
Figure 12 is a conceptual diagram showing an example of a communication system based on an auto encoder proposed in this specification.
Figures 13 and 14 show examples of turbo codes to help understand the method proposed in this specification.
Figure 15 is a diagram showing an example of the overall structure of the serial-connected neural network autoencoder proposed in this specification.
Figure 16 is a diagram showing an example of a transmitter encoder neural network to which the method proposed in this specification can be applied.
Figure 17 is a diagram showing an example of a receiving-end decoder neural network to which the method proposed in this specification can be applied.
Figure 18 is a diagram showing an example of a receiving-end decoder neural network to which the method proposed in this specification can be applied.
Figure 19 is a diagram showing an example of a basic neural network unit for configuring a neural network to which the method proposed in this specification can be applied.
Figures 20 and 21 are diagrams showing an example of a transmitter encoder neural network constructed based on the neural network structural unit proposed in this specification.
Figure 22 is a diagram showing an example of a receiving-end decoder neural network constructed based on the neural network structural unit proposed in this specification.
Figure 23 is a diagram showing examples of various activation functions that can be used in the neural network construction method proposed in this specification.
Figures 24 and 25 are diagrams showing an example of the interleaver design method proposed in this specification.
Figure 26 is a diagram to help understand the pooling layer configuration method proposed in this specification.
Figure 27 is a diagram showing changes in the learning ability of a neural network as the number of filters used in the neural network increases.
Figure 28 is a diagram showing an example of the pooling layer configuration method proposed in this specification.
Figure 29 is a diagram showing an example of a neural network unit (Conv1dCode) structure to which a pooling layer is applied.
Figure 30 is a diagram showing the results of a comparative analysis of complexity based on whether or not the pooling layer configuration method proposed in this specification is applied to the decoder neural network at the receiving end.
Figure 31 is a diagram showing an example of a rate-matching method for achieving a high code rate proposed in this specification.
Figure 32 is a diagram showing channel coding performance when applying the pooling layer configuration method proposed in this specification.
Figures 33 and 34 are diagrams showing channel coding performance when applying the interleaver design method proposed in this specification.
Figure 35 is a diagram showing channel coding performance according to the use of various activation functions proposed in this specification.
Figure 36 is a flowchart showing an example of a method for transmitting and receiving signals in a wireless communication system using an auto encoder proposed in this specification.
Figure 37 illustrates a communication system applied to the present invention.
Figure 38 illustrates a wireless device to which the present invention can be applied.
Figure 39 illustrates a signal processing circuit for a transmission signal.
Figure 40 shows another example of a wireless device applied to the present invention.
Figure 41 illustrates a portable device to which the present invention is applied.
Figure 42 illustrates a vehicle or autonomous vehicle to which the present invention is applied.
Figure 43 illustrates a vehicle to which the present invention is applied.
Figure 44 illustrates an XR device applied to the present invention.
Figure 45 illustrates a robot applied to the present invention.
Figure 46 illustrates an AI device applied to the present invention.

The following technologies can be used in various wireless access systems such as CDMA, FDMA, TDMA, OFDMA, SC-FDMA, etc. CDMA can be implemented with wireless technologies such as Universal Terrestrial Radio Access (UTRA) or CDMA2000. TDMA can be implemented with wireless technologies such as Global System for Mobile communications (GSM)/General Packet Radio Service (GPRS)/Enhanced Data Rates for GSM Evolution (EDGE). OFDMA can be implemented with wireless technologies such as IEEE 802.11 (Wi-Fi), IEEE 802.16 (WiMAX), IEEE 802-20, Evolved UTRA (E-UTRA), etc. UTRA is part of the Universal Mobile Telecommunications System (UMTS). 3rd Generation Partnership Project (3GPP) Long Term Evolution (LTE) is 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 idea 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 refer to technologies after TS Release 17 and/or Release 18. “xxx” refers to the standard document detail number. LTE/NR/6G can be collectively referred to as a 3GPP system. Regarding background technology, terms, abbreviations, etc. used in the description of the present invention, matters described in standard documents published before the present invention may be referred to. 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 NR

- 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 Channel and Frame Structure

Physical channels and typical signal transmission

Figure 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). The information transmitted and received between the base station and the terminal includes data and various control information, and various physical channels exist depending on the type/purpose of the information they transmit and receive.

When the terminal is turned on or enters a new cell, it performs an initial cell search task such as synchronizing with the base station (S11). To this end, the terminal can synchronize with the base station by receiving a primary synchronization signal (PSS) and a secondary synchronization signal (SSS) from the base station and obtain information such as a cell ID. Afterwards, the terminal can receive broadcast information within the cell by receiving a physical broadcast channel (PBCH) from the base station. Meanwhile, the terminal can check the downlink channel status by receiving a downlink reference signal (DL RS) in the initial cell search stage.

After completing the initial cell search, the terminal acquires more specific system information by receiving a physical downlink control channel (PDCCH) and a physical downlink shared channel (PDSCH) according to the information carried in the PDCCH. You can do it (S12).

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

The terminal that has performed the above-described procedure will then perform PDCCH/PDSCH reception (S17) and Physical Uplink Shared Channel (PUSCH)/Physical Uplink Control Channel (Physical Uplink) as a general uplink/downlink signal transmission procedure. Control Channel (PUCCH) transmission (S18) can be performed. In particular, the terminal can receive downlink control information (DCI) through PDCCH. Here, DCI includes control information such as resource allocation information for the terminal, and different formats may be applied depending on the purpose of use.

Meanwhile, the control information that the terminal transmits to the base station through uplink or that the terminal receives from the base station includes downlink/uplink ACK/NACK signals, CQI (Channel Quality Indicator), PMI (Precoding Matrix Index), and RI (Rank Indicator). ), etc. may be included. The terminal can transmit control information such as the above-described CQI/PMI/RI through PUSCH and/or PUCCH.

Structure of uplink and downlink channels

Downlink channel structure

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

(1) Physical downlink shared channel (PDSCH)

PDSCH carries downlink data (e.g., DL-shared channel transport block, DL-SCH TB), and modulation methods such as QPSK (Quadrature Phase Shift Keying), 16 QAM (Quadrature Amplitude Modulation), 64 QAM, and 256 QAM are used. Applies. A codeword is generated by encoding 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 resources along with DMRS (Demodulation Reference Signal), generated as an OFDM symbol signal, and transmitted through the corresponding antenna port.

(2) Physical downlink control channel (PDCCH)

PDCCH carries downlink control information (DCI) and QPSK modulation method is applied. One PDCCH consists of 1, 2, 4, 8, or 16 CCEs (Control Channel Elements) depending on the AL (Aggregation Level). One CCE consists of six REGs (Resource Element Group). One REG is defined by one OFDM symbol and one (P)RB.

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

Uplink channel structure

The terminal transmits related signals to the base station through an uplink channel, which will be described later, and the base station will receive the related signals from the terminal through an uplink channel, which will be 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. , DFT-s-OFDM (Discrete Fourier Transform - spread - Orthogonal Frequency Division Multiplexing) waveform, etc. are transmitted. When the PUSCH is transmitted based on the DFT-s-OFDM waveform, the terminal transmits the PUSCH by applying transform precoding. For example, if transform precoding is not possible (e.g., transform precoding is disabled), the terminal transmits PUSCH based on the CP-OFDM waveform, and if transform precoding is possible (e.g., transform precoding is enabled), the terminal transmits CP-OFDM. PUSCH can be transmitted based on the waveform or DFT-s-OFDM waveform. PUSCH transmission is scheduled dynamically by UL grant within DCI, or semi-statically based on upper layer (e.g., RRC) signaling (and/or Layer 1 (L1) signaling (e.g., PDCCH)). Can be scheduled (configured grant). PUSCH transmission can be performed based on codebook or non-codebook.

(2) Physical Uplink Control Channel (PUCCH)

PUCCH carries uplink control information, HARQ-ACK, and/or scheduling request (SR), and can be divided into multiple PUCCHs depending on the PUCCH transmission length.

6G system general

6G (wireless communications) systems require (i) very high data rates per device, (ii) very large number of connected devices, (iii) global connectivity, (iv) very low latency, (v) battery- The goal is to reduce the 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 the requirements of a 6G system.

The 6G system includes 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.

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

The 6G system is expected to have simultaneous wireless communication connectivity that is 50 times higher than that of the 5G wireless communication system. URLLC, a key feature of 5G, will become an even more important technology in 6G communications by providing end-to-end delay of less than 1ms. The 6G system will have much better volumetric spectral efficiency, unlike the frequently used area spectral efficiency. 6G systems can provide ultra-long battery life and advanced battery technologies for energy harvesting, so mobile devices in 6G systems will not need to be separately charged. New network characteristics in 6G may include:

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

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

- Seamless integration wireless information and energy transfer: 6G wireless networks will deliver 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: Connectivity of drones and very low Earth orbit satellites to networks and core network functions will create super 3D connectivity in 6G ubiquitous.

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

- Small cell networks: The idea of small cell networks was 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 for 5G and Beyond 5G (5GB) communications systems. Therefore, the 6G communication system also adopts the characteristics of a small cell network.

- Ultra-dense heterogeneous network: Ultra-dense heterogeneous networks will be another important characteristic of the 6G communication system. Multi-tier networks comprised of heterogeneous networks improve overall QoS and reduce costs.

- High-capacity backhaul: Backhaul connections are characterized by high-capacity backhaul networks to support high-capacity traffic. High-speed fiber and free-space optics (FSO) systems may be possible solutions to this problem.

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

- Softwarization and virtualization: Softwarization and virtualization are two important features that form the basis of the design process in 5GB networks to ensure flexibility, reconfigurability, and programmability. Additionally, billions of devices may be shared on a shared physical infrastructure.

Core implementation technology of 6G system

Artificial Intelligence

The most important and newly introduced technology in the 6G system is AI. AI was not involved in the 4G system. 5G systems will support partial or very limited AI. However, 6G systems will be AI-enabled for full automation. Advances in machine learning will create more intelligent networks for real-time communications in 6G. Introducing AI in communications can simplify and improve real-time data transmission. AI can use numerous 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 M2M, machine-to-human and human-to-machine communications. Additionally, AI can enable 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, attempts have been made 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, this research is gradually advancing to the MAC layer and Physical layer, and attempts are being made to combine deep learning with wireless transmission, especially in the Physical layer. AI-based physical layer transmission means applying signal processing and communication mechanisms based on AI drivers, rather than traditional communication frameworks, in terms of 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 allocation, etc.

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

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

Deep learning-based AI algorithms require a large amount of training data to optimize training parameters. However, due to limitations in acquiring data from a specific channel environment as training data, a lot of training data is used offline. This means that static training on training data in a specific channel environment may result in a contradiction between the dynamic characteristics and diversity of the wireless channel.

Additionally, current deep learning mainly targets real signals. However, signals of the physical layer of wireless communication are complex signals. More research is needed on neural networks that detect complex domain signals to match the characteristics of wireless communication signals.

Below, we will look at machine learning in more detail.

Machine learning refers to a series of operations that train machines to create machines that can perform tasks that are difficult or difficult for humans to perform. Machine learning requires data and a learning model. In machine learning, data learning methods can be broadly divided into three types: supervised learning, unsupervised learning, and reinforcement learning.

Neural network learning is intended to minimize errors in output. Neural network learning repeatedly inputs learning data into the neural network, calculates the output of the neural network and the error of the target for the learning data, and backpropagates the error of the neural network from the output layer of the neural network to the input layer to reduce the error. ) is the process of updating the weight of each node in the neural network.

Supervised learning uses training data in which the correct answer is labeled, while unsupervised learning may not have the correct answer labeled in the training data. That is, for example, in the case of supervised learning on data classification, the learning data may be data in which each training data is labeled with a category. Labeled learning data is input to a neural network, and error can be calculated by comparing the output (category) of the neural network with the label of the learning data. The calculated error is backpropagated in the reverse direction (i.e., from the output layer to the input layer) in the neural network, and the connection weight of each node in each layer of the neural network can be updated according to backpropagation. The amount of change in the connection weight of each updated node may be determined according to the learning rate. The neural network's calculation of input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate may be applied differently depending on the number of repetitions of the learning cycle of the neural network. For example, in the early stages of neural network training, a high learning rate can be used to ensure that the neural network quickly achieves a certain level of performance to increase efficiency, and in the later stages of training, a low learning rate can be used to increase accuracy.

Learning methods may vary depending on the characteristics of the data. For example, in a communication system, when the goal is to accurately predict data transmitted from a transmitter at a receiver, 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 can be considered the most basic linear model. However, deep learning is a machine learning paradigm that uses a highly complex neural network structure, such as artificial neural networks, as a learning model. ).

Neural network cores used as learning methods are broadly 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 multiple perceptrons.

Referring to Figure 3, when the input vector x=(x1,x2,...,xd) is input, each component is multiplied by the weights (W1,W2,...,Wd), and the results are added together, The entire process of applying the activation function σ() is called a perceptron. A large artificial neural network structure can extend the simplified perceptron structure shown in FIG. 3 and apply input vectors to different multi-dimensional perceptrons. For convenience of explanation, input or output values are referred to as nodes.

Meanwhile, the perceptron structure shown in FIG. 3 can be described as consisting of a total of three layers based on input and output values. An artificial neural network in which there are H (d+1)-dimensional perceptrons between the 1st layer and the 2nd layer, and K (H+1)-dimensional perceptrons between the 2nd layer and the 3rd layer can be expressed as shown in Figure 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 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 actual number of artificial neural network layers, the input layer is counted excluding the input layer, so it can be viewed as a total of two layers. An artificial neural network is constructed by two-dimensionally connecting perceptrons of basic blocks.

The above-described input layer, hidden layer, and output layer can be jointly applied not only to the multi-layer perceptron, but also to various artificial neural network structures such as CNN and RNN, which will be described later. As the number of hidden layers increases, the artificial neural network becomes deeper, and the machine learning paradigm that uses a sufficiently deep artificial neural network as a learning model is called deep learning. Additionally, the artificial neural network used for deep learning is called a deep neural network (DNN).

The deep neural network shown in Figure 5 is a multi-layer perceptron consisting of 8 hidden layers and 8 output layers. The multi-layer perceptron structure is expressed as a fully-connected neural network. In a fully connected neural network, no connection exists between nodes located on the same layer, and 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 correlation characteristics between input and output. Here, the correlation characteristic may mean the joint probability of input and output. Figure 5 is a diagram showing an example of a deep neural network.

Meanwhile, depending on how multiple perceptrons are connected to each other, various artificial neural network structures different from the above-described DNN can be formed.

In DNN, nodes located inside one layer are arranged in a one-dimensional vertical direction. However, in Figure 6, it can be assumed that the nodes are arranged two-dimensionally with w nodes horizontally and h nodes vertically (convolutional neural network structure in Figure 6). In this case, a weight is added to each connection during 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 h2w2 weights are needed between two adjacent layers.

Figure 6 is a diagram showing an example of a convolutional neural network.

The convolutional neural network in FIG. 6 has a problem in that the number of weights increases exponentially depending on the number of connections, so instead of considering all mode connections between adjacent layers, it is assumed that a small filter exists, and FIG. As shown, weighted sum and activation function calculations are performed on areas where filters overlap.

One filter has a weight corresponding to its size, and the weight can be learned so that a specific feature in the image can be extracted and output as a factor. In Figure 6, a filter of size 3Х3 is applied to the upper leftmost 3Х3 area of the input layer, and the output value as a result of performing the weighted sum and activation function calculation for the corresponding node is stored in z22.

The filter scans the input layer and moves at a certain distance horizontally and vertically, performs weighted sum and activation function calculations, and places the output value at the current filter position. This operation method is similar to the convolution operation on images in the field of computer vision, so a deep neural network with this structure is called a convolutional neural network (CNN), and the hidden layer generated as a result of the convolution operation is is called a convolutional layer. Additionally, a neural network with multiple convolutional layers is called a deep convolutional neural network (DCNN).

Figure 7 is a diagram showing an example of a filter operation in a convolutional neural network.

In the convolution layer, the number of weights can be reduced by calculating a weighted sum by including only nodes located in the area covered by the filter from the node where the current filter is located. Because of this, one filter can be used to focus on features for a local area. Accordingly, CNN can be effectively applied to image data processing where physical distance in a two-dimensional area is an important decision criterion. Meanwhile, CNN may have multiple filters applied immediately before the convolution layer, and may generate multiple output results through the convolution operation of each filter.

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

Referring to FIG. 8, a recurrent neural network (RNN) connects elements (x1(t), In the input process, the immediately previous point t-1 is a structure in which the hidden vectors (z1(t1), z2(t1),..., zH(t1)) are input together and a weighted sum and activation function are applied. The reason for passing the hidden vector to the next time point like this is because the information in the input vector from previous time points is considered to be accumulated in the hidden vector at the current time point.

Figure 8 shows an example of a neural network structure in which a circular loop exists.

Referring to FIG. 8, the recurrent neural network operates in a predetermined time point order with respect to the input data sequence.

Hidden vector (z1(1),z2(1),.. when the input vector (x1(t), .,zH(1)) is input together with the input vector (x1(2),x2(2),...,xd(2)) at viewpoint 2, and the vector (z1( 2) Determine z2(2) ,...,zH(2)). This process is performed repeatedly until time point 2, time point 3, ,,, time point T.

Figure 9 shows an example of the operational structure of a recurrent neural network.

Meanwhile, when multiple hidden layers are placed within a recurrent neural network, it is called a deep recurrent neural network (DRNN). Recurrent neural networks are designed to be usefully applied to sequence data (for example, natural language processing).

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

Recently, attempts have been made 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, this research is gradually advancing to the MAC layer and Physical layer, and attempts are being made to combine deep learning with wireless transmission, especially in the Physical layer. AI-based physical layer transmission means applying signal processing and communication mechanisms based on AI drivers, rather than traditional communication frameworks, in terms of 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 allocation, etc.

Term Definition

For convenience of explanation, the following symbols/abbreviations/terms may be used interchangeably in this specification.

- AE: Auto Encoder

- FC: Fully Connected

- IR: Incremental Redundancy

- LDPC: Low Density Parity Check

- SCNN-AE: Serially Concatenated Neural-Network AutoEncoder

- NN: Neural Network

- Tx: Transmitter

- Rx: Receiver

- RS: Reference Symbols

Auto encoder

Various attempts are being made to apply neural networks to communication systems. In particular, among various attempts to apply neural networks to communication systems, attempts to apply neural networks to the physical layer mainly focused on optimizing specific functions of the receiver. For example, the performance of the receiving end can be improved by configuring the channel decoder with a neural network. As another example, in a MIMO system with multiple transmit/receive antennas, performance improvement can be achieved by implementing a MIMO detector as a neural network.

Another approach to apply a neural network to a communication system is to use an autoencoder in the communication system. Here, an autoencoder is a type of artificial neural network that has the characteristic of outputting the same information as the information input to the autoencoder. Since the goal of a communication system is to ensure that the signal transmitted from the transmitting end is restored at the receiving end without distortion, the characteristics of the autoencoder can suit the goal of the communication system.

When applying the auto-encoder to a communication system, the transmitting and receiving ends of the communication system are each composed of a neural network, and through this, performance improvement can be achieved by performing optimization from an end-to-end perspective. You can.

Figure 10 is a diagram showing an example of a wireless communication system using an auto encoder. The transmitting end 101 is composed of a neural network expressed as f(s), and the receiving end 103 is composed of a neural network expressed as g(s). That is, the neural networks (f(s) and g(s)) are components of the transmitting end 101 and the receiving end 103, respectively. In other words, the transmitting end 101 and the receiving end 103 are each configured as a neural network (or based on a neural network).

Recently, deep learning technology has developed, and AI (artificial intelligence)/ML (machine learning) is being applied in many areas. In particular, when the AI/ML is applied to tasks that require repeatability/continuity, the AI/ML can be very efficient.

New channel coding technologies called low-density parity-check (LDPC) and polar coding were introduced in 5G. The channel coding performance of the LDPC and polar coding exceeds that of the Turbo code or Tail-biting convolutional codes (TBCC) used in 4G. However, channel coding using LDPC and polar coding is designed to be optimized for additive white Gaussian noise (AWGN) channels when the technology is developed and reflected in 5G standards.

In communications, in order to minimize overhead, an MCS (Modulation and code-rate set) is defined by combining modulation and code-rate, and the MCS is defined according to the characteristics of the channel. It is set and used for communication. Here, the communication is used to include both wired and wireless communication, and the same applies hereinafter.

In 3GPP communication, the value of FER (Frame Error Rate) (or BLER (Block Error Rate)), which represents the data error rate, satisfies 10^(-1) (or 10^(-2)). MCS is used. A FER value of 10^(-1) indicates that, on average, one transmission error occurs in 10 signal transmissions, and a BLER value of 10^(-2) indicates that, on average, one transmission error occurs in 100 signal transmissions. Indicates that a transmission error has occurred. To solve problems caused by transmission errors, a retransmission operation is defined in the 3GPP standard, and a hybrid automatic repeat request (HARQ) technique is used to increase the efficiency of the retransmission operation. At this time, if a coding technique that is not optimized for the channel is used in communication using a neural network, the following problems may occur.

- The probability of an error occurring during signal transmission increases, and retransmission (HARQ, ARQ) must be performed to correct this.

- For retransmission, the base station must store the data to be retransmitted. Additionally, the terminal must store the previously received data in order to combine the previously received data with the retransmitted data. Accordingly, the memory of the base station/terminal for storing data to be retransmitted/data received earlier is wasted.

- Data throughput is reduced due to retransmission of data with transmission errors.

- Resources for retransmitting data are wasted.

This specification proposes a method of configuring a coding scheme optimized for the channel between the transmitting end and the receiving end through a neural network using an auto-encoder architecture. More specifically, the method proposed in this specification relates to a coding technique for correcting errors occurring in a communication channel using an autoencoder neural network.

BLER and retransmission rate can be improved through the method proposed in this specification. Additionally, the data throughput of the system can be increased based on improved BLER and retransmission rate, and end-to-end delay can be improved by reducing turn-around delay. ) can be improved.

Figure 11 is a diagram showing an example of the conceptual structure of a communication chain using an auto encoder.

More specifically, in FIG. 11, based on the structural perspective of an autoencoder, the transmitting end 101 can be interpreted as an encoder f(.), which is one of the components constituting the autoencoder, and the receiving end 103 is It can be interpreted as a decoder g(.), one of the components constituting the auto encoder. In addition, a channel exists between the transmitting end 101, which is the encoder f(.) constituting the auto encoder, and the receiving end 103, which is the decoder g(.) constituting the auto encoder. According to the above interpretation, the transmitting end 101 may be called a 'transmitting end encoder', and the receiving end 103 may be called a 'receiving end decoder', and may be called variously within the scope of the same/similar interpretation. It can be.

The structure of the autoencoder can be divided into two structures: Under-complete AE and Over-complete AE. The Under-complete AE is a structure in which X, the output data of the transmitter encoder f(.), is expressed in a smaller amount than U, the input data to the transmitter encoder f(.). ) can be used in technologies such as

On the other hand, over-complete AE is a structure in which the output data, Redundancy is added to output data X. This can be used in technologies that add parity, such as channel coding. The methods proposed in this specification use an autoencoder based on an over-complete AE structure. At this time, the amount of redundancy added to the input data can be adjusted based on the code-rate (R).

The relationship between U, X and R is defined according to the equation below.

Again in FIG. 11, the transmitter encoder f(.)(101) encodes (over-completes) the input data U with an amount inversely proportional to the code-rate (R). The encoded input data is transmitted to the receiving end decoder g(.)(103) through a channel, and the receiving end decoder g(.)(103) decodes the encoded data and restores the input data before encoding.

How to configure a neural network for the transmitter encoder/receiver decoder

The method proposed in this specification relates to a (channel) coding technique for correcting errors occurring in a communication channel using an autoencoder neural network (AE-NN). More specifically, this specification proposes a specific AE-NN structure to ensure that a (channel) coding technique for correcting errors occurring in a communication channel can be efficiently performed. Constructing the structure of the specific AE-NN is based on (i) the structure of the transmitting end encoder neural network of the transmitting end constituting the communication system and (ii) the structure of the receiving end encoder neural network of the receiving end constituting the communication system in a specific manner. This can mean that it is composed.

Figure 12 is a conceptual diagram showing an example of a communication system based on an auto encoder proposed in this specification.

More specifically, in FIG. 12, u (1201) is information to be transmitted by the transmitter 101, and the information is coded through the transmitter encoder neural network (NN-encoder) of the transmitter 101. The information coded based on the transmitter encoder neural network is transmitted from the transmitter 101 to the receiver 103 through a channel between the transmitter 101 and the receiver 103. Afterwards, the receiving end 103 receives the coded information passing through the channel and restores the coded information based on the receiving end decoder neural network (NN-decoder). Here, the receiving end 103 restoring the coded information means that the receiving end 103 estimates, based on the receiving end decoder neural network, (i) a channel through which the coded information is transmitted, and (ii) the It can be interpreted as including operations for decoding the coded information based on the estimated channel, and the above interpretation can be equally applied hereinafter.

The transmitting end encoder neural network and the receiving end encoder neural network are pre-trained to be optimized for the channel situation, and the transmitting end 101 and the receiving end 103 are pre-trained based on the channel situation based on the transmitting end encoder neural network and the receiving end encoder neural network. Optimized communication can be performed.

The auto-encoder-based coding method proposed in this specification applies the Turbo code method, one of the error correction codes, to artificial intelligence.

Prior to a detailed description of the application of the turbo code method to artificial intelligence, Figures 13 and 13 and 14 show examples of turbo codes to help understand the method proposed in this specification.

)가 위치하는 형태로 구성이 된다.As shown in FIGS. 13 and 14, the turbo code basically includes two component codes and an interleaver (interleaver) between the two component codes.

) is located.

로 표시되는 인터리버(1303)가 위치하는 형태로 구성되는 것을 알 수 있다. 여기서, 상기 PCCC 터보 코드의 입력 u1은 x0, x1 및 x2로 구성되어 출력되는데, 상기 x0은 입력 u1과 동일하고, 상기 x1은 입력 u1이 제 1 구성요소 코드(1301)에 의하여 인코딩되어 출력된 것이다. 또한, 상기 x2는 입력 u1이 인터리버(1303)에 의해서 인터리빙된 u2가 제 2 구성요소 코드(1302)에 의하여 인코딩되어 출력된 것이다. 상기 PCCC 터보 코드는 하나의 입력 u1을 입력 받아 3개의 값(x0, x1 및 x2)을 출력하므로, 상기 PCCC 터보 코드의 코드 레이트의 값은 1/3이다.More specifically, Figure 13 relates to parallel concatenated convolutional code (PCCC) Turbo code. Referring to FIG. 13, the PCCC turbo code has a first component code 1301 indicated as ENC 1 and a second component code 1302 indicated as ENC 2 located in parallel, and the component codes 1301 and Between 1302)

It can be seen that the interleaver 1303 indicated by is configured in such a way that it is located. Here, the input u1 of the PCCC turbo code is composed of x0, x1, and x2 and output. The x0 is the same as the input u1, and the x1 is the input u1 encoded by the first component code 1301 and output. will be. In addition, the x2 is output by encoding u2, which is input u1 interleaved by the interleaver 1303, by the second component code 1302. Since the PCCC turbo code receives one input u1 and outputs three values (x0, x1, and x2), the code rate value of the PCCC turbo code is 1/3.

로 표시되는 인터리버(1403)가 직렬적으로 위치하는 형태로 구성되는 것을 알 수 있다. 여기서, 상기 SCCC 터보 코드의 입력 u1은 x0, x1 및 x2로 구성되어 출력된다. 보다 구체적으로, 입력 u1은 제 1 구성요소 코드(1401)에 의하여 인코딩되어 c1으로 출력된다. 이후, 상기 c1은 인터리버(1403)에 의해서 인터리빙 되고, 제 2 구성요소 코드(1402)는 상기 인터리빙된 c1을 입력 u2를 입력 받는다. 상기 u2는 제 2 구성요소 코드(1402)에 의하여 인코딩되고, 결과적으로 x0, x1 및 x2로 구성된 c2가 출력된다. 상기 SCCC 터보 코드는 하나의 입력 u1을 입력 받아 3개의 값(x0, x1 및 x2)을 출력하므로, 상기 SCCC 터보 코드의 코드 레이트의 값은 1/3이다.Additionally, Figure 14 relates to SCCC (Serial concatenated convolutional codes) turbo codes. Referring to FIG. 14, the SCCC Turbo code has a first component code 1401 indicated as ENC 1 and a second component code 1402 indicated as ENC 2, and the components of the component codes 1401 and 1402 are located. Between

It can be seen that the interleaver 1403, indicated by , is configured to be located in series. Here, the input u1 of the SCCC turbo code is composed of x0, x1, and x2 and is output. More specifically, input u1 is encoded by the first component code 1401 and output as c1. Afterwards, the c1 is interleaved by the interleaver 1403, and the second component code 1402 receives the interleaved c1 as input u2. The u2 is encoded by the second component code 1402, and as a result, c2 consisting of x0, x1, and x2 is output. Since the SCCC turbo code receives one input u1 and outputs three values (x0, x1, and x2), the code rate value of the SCCC turbo code is 1/3.

In general, SCCC turbo codes have somewhat lower waterfall performance than PCCC turbo codes, but can have better error floor performance than PCCC turbo codes.

Since the SCCC turbo code has better error floor performance than the PCCC turbo code, this specification proposes a method that applies the form of the SCCC-turbo code to artificial intelligence for stable data transmission. Here, the method in which the form of the SCCC-Turbo code is applied to artificial intelligence can be called a Serially Concatenated Neural-Network AutoEncoder (SCNN-AE).

Figure 15 is a diagram showing an example of the overall structure of the serial-connected neural network autoencoder proposed in this specification.

In Figure 15, the transmitter 101 is configured based on a transmitter encoder neural network having a specific structure. The transmitter 101 encodes input data u based on the transmitter encoder neural network, and the encoded input data (x0, x1, and x2) are included in the signal transmitted by the transmitting end and are transmitted to the receiving end 103 on the channel.

)(101-3)를 포함한다(로 구성된다). 여기서, f1(101-1)은 상기 송신단 인코더 뉴럴 네트워크의 입력 데이터를 입력 받고, 이를 인코딩하여 제 1 코드워드(codeword)를 출력한다. 인터리버(101-3)는 상기 제 1 코드워드를 f1(101-2)로부터 입력 받아 이를 인터리빙(interleaving)하고, 상기 인터리빙된 제 1 코드워드를 출력한다. 또한, f2(101-3)는 상기 인터리빙된 제 1 코드워드를 인터리버(101-3)로부터 입력 받아 제 2 코드워드(x0, x1 및 x2)를 출력한다. 상기 제 2 코드워드는 상기 송신단 인코더 뉴럴 네트워크에 포함된 전력 노멀라이저(Power Normalizer)(101-4)를 거쳐서 채널 상으로 전송된다. 이 때, 전력 노멀라이저는 송신단(101)의 출력 전력의 크기가 1이 되도록 조정하는 역할을 한다.More specifically, the transmitter encoder neural network of the transmitter 101 includes a neural network f1 (101-1), a neural network f2 (101-2), and an interleaver (

) includes (consists of) (101-3). Here, f1 (101-1) receives input data from the transmitter encoder neural network, encodes it, and outputs a first codeword. The interleaver 101-3 receives the first codeword from f1 (101-2), interleaves it, and outputs the interleaved first codeword. Additionally, f2 (101-3) receives the interleaved first codeword from the interleaver (101-3) and outputs second codewords (x0, x1, and x2). The second codeword is transmitted on the channel through a power normalizer 101-4 included in the transmitter encoder neural network. At this time, the power normalizer serves to adjust the output power of the transmitter 101 to be 1.

In summary, the transmitter encoder neural network is (i) a first encoder neural network that receives input data and outputs a first codeword, (ii) receives the first codeword and interleaves it. (iii) an interleaver that receives the interleaved first codeword and outputs the interleaved first codeword, and (iii) a second device that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data. It can be described as involving serially an encoder neural network. In addition, the transmitter encoder neural network may be described as further including a power normalizer that adjusts the output power of the signal transmitted by the transmitter 101 to be 1.

Returning to Figure 15, the receiving end 103 is composed of at least one basic receiving end decoder neural network having a specific structure connected sequentially. The signal transmitted by the transmitter 101 is received, and the encoded input data (x0, x1, and x2) of the transmitter 101 included in the signal is restored. In Figure 15, Imax represents the maximum value (i.e., maximum decoding iteration) of the basic receiving end decoder neural network that can be included (configurable) in the receiving end decoder neural network.

)(103-3)를 포함한다. 여기서, g2(103-2)는 상기 채널상으로 전송된 송신단(101)의 신호에 포함된 f2(101-2)와 관련된 코드워드(제 2 코드워드(x0, x1 및 x2))에 대한 디코딩(decoding)을 수행하고, 제 2 디코딩된 코드워드를 출력한다. 또한, 디-인터리버(103-3)는 g2(103-2)로부터 상기 제 2 디코딩된 코드워드를 입력 받아 이를 디-인터리빙(de-interleaving)하고, 상기 디-인터리빙된 제 2 디코딩된 코드워드를 출력한다. 이 때, 상기 디-인터리빙된 제 2 디코딩된 코드워드는 f2(101-2)와 관련된 코드워드에 디코딩과 디-인터리빙이 순차적으로 수행된 것이므로, 상기 디-인터리빙된 제 2 디코딩된 코드워드는 f1(101-1)과 관련된 코드워드(제 1 코드워드)일 수 있다. g1(103-1)은 상기 f1(101-1)과 관련된 코드워드를 디-인터리버(103-3)로부터 입력 받아 이를 디코딩하고, 제 1 디코딩된 코드워드를 출력한다. 상기 제 1 디코딩된 코드워드는 인터리빙되어 상기 가장 먼저 위치하는 기본 수신단 디코더 뉴럴 네트워크의 다음에 연결된 기본 수신단 디코더 뉴럴 네트워크로 입력된다. 이 때, 상기 제 1 디코딩된 코드워드의 인터리빙은 (i) 상기 가장 먼저 위치하는 기본 수신단 디코더 뉴럴 네트워크 및 (ii) 상기 가장 먼저 위치하는 기본 수신단 디코더 뉴럴 네트워크의 다음에 연결된 기본 수신단 디코더 뉴럴 네트워크 사이에 위치하는 인터리버에 의하여 수행된다. 상기 제 1 디코딩된 코드워드가 인터리빙되어 상기 가장 먼저 위치하는 기본 수신단 디코더 뉴럴 네트워크의 다음에 연결된 기본 수신단 디코더 뉴럴 네트워크로 입력될 때, 상기 가장 먼저 위치하는 기본 수신단 디코더 뉴럴 네트워크의 다음에 연결된 기본 수신단 디코더 뉴럴 네트워크는 채널 상으로 전송된 송신단(101)이 전송한 신호를 함께 입력 받는다. 상기 설명된 내용은 수신단 디코더 뉴럴 네트워크에 포함된 적어도 하나 이상의 기본 수신단 디코더 뉴럴 네트워크들에 각각 포함된 g2 및 g1에 대해서도 적용될 수 있음은 물론이다.For the convenience of a more detailed explanation, the basic receiving end decoder neural network located first among at least one basic receiving end decoder neural network included in the receiving end decoder neural network (constituting the receiving end decoder neural network) will be explained as an example. . The basic receiving end decoder neural network of the receiving end 103 includes neural network g2 (103-2), neural network g1 (103-1), and de-interleaver (

)(103-3). Here, g2 (103-2) decodes the codeword (second codeword (x0, x1, and x2)) related to f2 (101-2) included in the signal of the transmitter 101 transmitted on the channel. (decoding) is performed, and the second decoded codeword is output. In addition, the de-interleaver 103-3 receives the second decoded codeword from g2 (103-2), de-interleaves it, and deinterleaves the second decoded codeword. outputs. At this time, the de-interleaved second decoded codeword is obtained by sequentially performing decoding and de-interleaving on the codeword related to f2 (101-2), so the de-interleaved second decoded codeword is It may be a codeword (first codeword) related to f1 (101-1). g1 (103-1) receives the codeword related to f1 (101-1) from the de-interleaver (103-3), decodes it, and outputs the first decoded codeword. The first decoded codeword is interleaved and input to a basic receiving end decoder neural network connected next to the first basic receiving end decoder neural network. At this time, the interleaving of the first decoded codeword is between (i) the basic receiving end decoder neural network located first and (ii) the basic receiving end decoder neural network connected next to the basic receiving end decoder neural network located first. It is performed by an interleaver located in . When the first decoded codeword is interleaved and input to the basic receiving end decoder neural network connected next to the first-positioned basic receiving end decoder neural network, the basic receiving end connected next to the first-positioned basic receiving end decoder neural network The decoder neural network receives the signal transmitted by the transmitter 101 on the channel. Of course, the above description can also be applied to g2 and g1 respectively included in at least one basic receiving end decoder neural network included in the receiving end decoder neural network.

To summarize what has been described above, the receiving end decoder neural network includes at least one basic decoder neural network connected sequentially. Here, each of the at least one basic decoder neural network (i) performs decoding on the second codeword output from the second encoder neural network included in the transmitter encoder neural network and outputs the second decoded codeword. A second decoder neural network, (ii) a de-interleaver that receives the second decoded codeword, de-interleaves it, and outputs the deinterleaved second decoded codeword. ) and (iii) a first device that performs decoding on the de-interleaved second decoded codeword and outputs a first decoded codeword based on the decoding on the de-interleaved second decoded codeword. It can be described as including a decoder neural network serially.

In addition, among the at least one basic decoder neural network, the basic receiving end decoder neural network located first in the receiving end decoder neural network is a codeword (second codeword (x0) included in the transmitting end signal received on the channel by the receiving end 103. , only x1 and x2)) are input. On the other hand, the basic receiving end decoder neural networks other than the basic receiving end decoder neural network located first include (i) the first decoded codeword output from the basic decoder neural network located immediately before it and (ii) the receiving end 103 on the channel. All codewords (second codewords (x0, x1, and x2)) included in the transmitter signal received are input. At this time, this means that the basic receiving end decoder neural network located first receives only the codewords (second codewords (x0, x1, and x2)) included in the transmitting end signal received on the channel by the receiving end 103. It may be understood that the codeword received from the basic receiver decoder neural network located immediately before is 0.

Figure 16 is a diagram showing an example of a transmitter encoder neural network to which the method proposed in this specification can be applied.

More specifically, FIG. 16 is a diagram related to a specific implementation form of the first encoder neural network and the second encoder neural network included in (constituting) the encoder neural network at the transmitting end. As shown in FIG. 15, the transmitter encoder neural network in FIG. 16 includes a first encoder neural network 101-1, an interleaver 101-3, and a second encoder neural network 101-2 in series. It can be composed of:

In Figure 16, the first encoder neural network (f1) (101-1) is composed of two encoder component neural networks (f1 (1) (101-1a) and f1 (2) (101-1b), and 2 Encoder Neural Network (f2)(101-2) consists of 3 encoder component neural networks (f2(1)(101-2a), f2(2)(101-2b) and f2(3)(101-2c). It can be composed of: Here, f1(1)(101-1a), f1(2)(101-1b), f2(1)(101-2a), and f2(2)(101-2b) may be composed of the same type of neural network. and the same type of neural network constituting f1(1)(101-1a), f1(2)(101-1b), f2(1)(101-2a), and f2(2)(101-2b). The first encoder component may be called a neural network. Additionally, f2(3)(101-2c) may be composed of a different type of neural network from the first encoder component neural network, and may be referred to as a second encoder component neural network.

Additionally, in Figure 16, the first encoder neural network (f1(1,2)) 101-1 receives input (u) and generates a codeword (c1) for the received input (u). The codeword (c1) may be generated based on encoding of the input (u) received by the first encoder neural network (f1(1,2)) 101-1. Here, the first encoder neural network (f1(1,2)) 101-1 is configured such that the code-rate value of the codeword (c1) is 1/2. That is, the first encoder neural network (f1(1,2)) 101-1 receives an input (u) with a length of 1 and outputs a codeword (c1) with a length of 2. Additionally, the second encoder neural network (f2(1,2,3)) 101-2 receives the codeword (c1) and generates a codeword (c2) for the received codeword (c1). The codeword (c2) may be generated based on encoding of the codeword (c1) received by the second encoder neural network (f2(1,2,3)) 101-2. Here, the second encoder neural network (f2(1,2,3)) 101-2 is configured such that the code-rate value of the codeword (c2) is 2/3. That is, the second encoder neural network (f2(1,2,3)) 101-1 receives a codeword (c1) with a length of 2 and outputs a codeword (c2) with a length of 3.

In FIG. 16, for convenience of explanation, the number of encoder component neural networks constituting the first encoder neural network 101-1 is two, and the number of encoder component neural networks constituting the second encoder neural network 101-2 is 2. Although the number of has been described as an example of 3, it goes without saying that the first encoder neural network 101-1 and the second encoder neural network 101-2 can be configured in various forms. That is, the first encoder neural network 101-1 is composed of n1 first encoder component neural networks, and the second encoder neural network 101-2 is composed of n2 first encoder component neural networks and n3 second encoder component neural networks. The encoder component may be composed of a neural network. At this time, n1, n2, and n3 may be positive integers other than 0. At this time, the input (u) of the first encoder neural network (101-1) may be k, the code rate may be k/n1, and the The code rate may be n1/(n2+n3). Here, the values of n1 and n2 may be the same, and the value of n3 may be different from n1 and n2. In particular, the value of n3 may be 1. The total code-rate becomes k/(n2+n3).

Figure 17 is a diagram showing an example of a receiving-end decoder neural network to which the method proposed in this specification can be applied.

(103-3)는 도 15에서 설명한 디-인터리버를 나타내고,

는 도 15에서 설명한 인터리버를 나타낸다.More specifically, Figure 17 shows the specific operation of the basic receiving end decoder neural network that constitutes the receiving end decoder neural network. In FIG. 17, g2 represents the second decoder neural network 103-2 described in FIG. 15, and g1 represents the first decoder neural network 103-1 described in FIG. 15. also,

(103-3) represents the de-interleaver described in Figure 15,

represents the interleaver described in FIG. 15.

The second decoder neural network 103-2 performs decoding on the data (x0, x1, x2) transmitted from the transmitter 101 on the channel, and is the input of the second encoder neural network constituting the transmitter encoder neural network. Create A Posteriori u2 corresponding to . From the generated A Posteriori u2, u2 (A Priori) that has correlation with the input (x0, x1 and x2) is removed (+: operation) (103-5), and de-interleaver (π-1) ( After passing through 103-3), a codeword (c1) (Extrinsic information) corresponding to the output of the first encoder neural network constituting the transmitter encoder neural network is obtained.

Afterwards, the first decoder neural network (g1) 103-1 receives the codeword (c1) (A Priori) and generates A Posteriori c1 through decoding of the codeword (c1) (A Priori). do. Next, to generate extrinsic c1, A priori c1 is removed from A posteriori c1 generated by the first decoder neural network (103-1) (103-6) and then interleaved (103-4). Here, the interleaved data is delivered to the input of a basic receiving end decoder neural network connected immediately after the basic receiving end decoder neural network that performed the operations described above.

Figure 18 is a diagram showing an example of a receiving-end decoder neural network to which the method proposed in this specification can be applied.

More specifically, FIG. 18 shows an example of a receiving decoder neural network configured by successively connecting (n-iteration) n basic receiving decoder neural networks described in FIG. 17. In FIG. 18, the subtraction operations 103-5 and 103-6 for generating extrinsic information described in FIG. 17 are included in g2,i and g1,i. Among the continuously connected basic receiver decoder neural networks, except for the first basic receiver decoder neural network, the remaining basic receiver decoder neural networks are connected to (i) data transmitted on the channel and (ii) the basic receiver decoder neural network located right in front of it. Data generated based on the performed decoding is input together, and decoding is performed based on the data in (i) and (ii) above. By repeatedly performing decoding based on the data in (i) and (ii), there is a difference between (a) the data restored from the last basic receiving end decoder neural network and (b) the input data (u) transmitted by the transmitting end encoder. Error can be reduced.

Specific neural network structure - (Proposal 1)

In the following, (i) the first encoder neural network and the second encoder neural network constituting the transmitting end encoder neural network and (ii) the first decoder neural network and the second decoder neural network constituting the receiving end decoder neural network are described above. Let's look at the specific structure.

When applying AI to (channel) coding, the space of information that AI needs to learn becomes an important issue. More specifically, since information has random characteristics, for information of length k, AI must learn for 2k cases. To solve the problem that the size of the space that AI must learn increases exponentially as the length (size) of information increases, the transmitter encoder neural network/receiver decoder neural network can be configured to have specific structural restrictions. , a neural network can learn the specific structure.

In the case of SCNN-AE proposed in this specification, a convolutional neural network (CNN) is used to configure the transmitter encoder and the receiver decoder. This is because filtering using CNN's Kernel plays a similar role to a convolution code. From the perspective of the transmitter encoder stage, the more hidden layers there are in the neural network constituting the transmitter encoder, the better the input is separated. On the other hand, as the number of hidden layers increases, the complexity of the neural network increases and the degree of performance improvement becomes insignificant compared to the increase in the number, so it is important to configure an appropriate number of hidden layers.

Figure 19 is a diagram showing an example of a basic neural network unit for configuring a neural network to which the method proposed in this specification can be applied.

In this specification, (i) a first encoder neural network (f1) and a second encoder neural network (f2) constituting the transmitting end encoder neural network and (ii) a first decoder neural network (g1) constituting the receiving end decoder neural network. And the method of configuring the second decoder neural network (g2) is a method of configuring f1, f2, g1, and g2 based on the same structure, but applying only the parameters applied to each differently. That is, the same neural network structural units shown in FIG. 19 are used to construct f1, f2, g1, and g2, and the number of hidden layers (number of neural network structural units) of each of f1, f2, g1, and g2, Parameters such as the number of filters may be applied differently to each of f1, f2, g1, and g2.

In Figure 19, Cin represents the dimension of the input data, k is the kernel size (kernel-size) (filter-size), and Cout represents the dimension of the output data, where f1, f2, and g1 And the neural network unit that is the basis of the g2 configuration is a filtering function (Conv1d) (1910) for filtering the input data and an activation function (ACT) that receives the output of the filtering function and non-linearizes it. )(1920). Neural networks use a function with non-linearity as an activation function to maintain independence between layers. Additionally, to solve the vanishing gradient problem during back-propagation, which is essential in neural network learning, the ReLu (Rectified Linear Unit) function or a modified version of the ReLu function is used as an activation function. f1(1,2), f2(1,2,3)/g1, and g2 of the transmitter encoder/receiver decoder are one neural network unit (Conv1dCode) forming one hidden layer, and Cin/Cout has different parameters.

Figures 20 and 21 are diagrams showing an example of a transmitter encoder neural network constructed based on the neural network structural unit proposed in this specification.

First, FIG. 20 is a diagram of an example of a first encoder component neural network constituting the first encoder neural network and the second encoder neural network of FIG. 16. As shown in FIG. 20, the first encoder component neural network may be composed of two layers, and at the last stage, linear regression and eLu (exponential linear unit) functions are used as activation functions. Co represents the number of filters used for convolution of CNN in the transmitter encoder. More specifically, in the first encoder component neural network, the first layer of the two layers receives one input data and outputs output data with the number of output channels being Co. Additionally, the second layer receives output data with Co number of output channels output from the first layer, and outputs output data with Co number of output channels. Afterwards, linear regression may be applied to the output data output from the second layer, and the eLu function may be applied to the output data of the second layer to which the linear regression has been applied.

Next, FIG. 21 is a diagram of an example of a second encoder component neural network constituting the second encoder neural network of FIG. 16. As shown in FIG. 21, the second encoder component neural network may be composed of two layers, and linear regression and the eLu function are used as activation functions at the last stage. Co represents the number of filters used for convolution of CNN in the transmitter encoder. More specifically, in the second encoder component neural network, the first layer of the two layers receives input data with two input channels and outputs output data with Co number of output channels. Additionally, the second layer receives output data with Co number of output channels output from the first layer, and outputs output data with Co number of output channels. Thereafter, linear regression may be applied to the output data output from the second layer, and the eLu function may be applied to the output data of the second layer to which the linear regression has been applied and output.

20 and 21, if the value of Co is increased, the accuracy of the neural network increases, but the learning time and complexity also increase, so the method proposed in this specification can be preferably applied when Co = 100. .

Figure 22 is a diagram showing an example of a receiving-end decoder neural network constructed based on the neural network structural unit proposed in this specification.

First, FIG. 22 is a diagram of an example of configuring the first decoder neural network (g1) and the second decoder neural network (g2) of FIG. 17. As shown in FIG. 22, the first encoder component neural network may be composed of five layers, and linear regression and sigmoid functions are used as activation functions at the last stage. Ci represents the input dimension in the first decoder neural network (g1) and the second decoder neural network (g2). For the first decoder neural network (g1), Ci = 5, Co = 100, and FT = 5. For the second decoder neural network (g2), Ci=8, Co=100, and FT=5. Here, the reason why Ci = 8 in g2 is that the input (x0, x1, This is because =5). In the case of g1, since no input data is received from the channel, Ci = 5.

Here, Co and FT are values set before learning of the decoder neural network at the receiving end. As Co and FT become larger, the accuracy of the neural network increases, but learning time and complexity also increase.

How to determine activation functions - (Proposal 2)

This proposal concerns a method of selecting an activation function that can be applied to the neural network construction method proposed in this specification.

The following requirements exist regarding the activation function used in neural networks.

1. In order to build deep layers in a neural network, a non-linear function is required.

2. During back-propagation, a function that does not have a vanishing gradient problem must be used.

An activation function that satisfies the above requirements is the ReLu (Rectified Linear Unit) function. The ReLu function has the advantage of fast calculation and fast convergence, but has the disadvantage that when the input is negative, the gradient becomes '0' and learning is not possible. To compensate for the shortcomings of the above ReLu function, three modified ReLu functions can be used.

Figure 23 is a diagram showing examples of various activation functions that can be used in the neural network construction method proposed in this specification.

Referring to FIG. 23, three functions to compensate for the shortcomings of the ReLu function, namely the PReLu function 2320, the LeakyRelu function 2330, and the eRelu function 2340, are shown. It can be seen that the slopes of the activation functions 2320, 2330, and 2340 shown in FIG. 23 are not 0 even when the input value is negative.

The SCNN-AE proposed in this specification can be designed to determine an appropriate activation function based on experimental results, performance, and complexity of various ReLu-based functions. In this proposal, we propose to select an activation function to be used in the neural network based on experimental results, performance, and complexity based on ReLu (2310), PReLu (2320), and eReLu (2340) among the functions shown in FIG. 23.

Table 2 above is a table summarizing the mathematical equations of the ReLu(2310), PReLu(2320), and eReLu(2340) functions and their respective advantages/disadvantages.

The eReLu function uses an exponential function, which increases the complexity of the function, but its nonlinearity is stronger for negative values than the other two functions, so it can perform better than the other two functions for negative values. PReLu is a form that can provide benefits in terms of both performance and complexity, and can be expected to have performance intermediate to the remaining two functions. In particular, the value of the constant 'a' of PReLu can be determined through training of a neural network.

Interleaver Design Method - (Proposal 3)

This proposal concerns the design of an interleaver used in the transmitter encoder neural network and the receiver decoder neural network. The following benefits can be obtained through the use of the interleaver.

1) Using an interleaver in encoding has the effect of increasing the block length, which has the greatest impact on coding performance.

2) By using an interleaver/de-interleaver in decoding, iterative decoding is possible between the first decoder neural network (g1) and the second decoder neural network (g2) that constitute the decoder neural network at the receiving end. At the decoder stage, two decoders decode using uncorrelated information. In general, as the size of the interleaver increases, randomness increases and decoder performance can increase.

In this proposal, the interleaver is designed to satisfy the following two requirements.

1. Randomness must be guaranteed at a certain level.

2. Considering ease of implementation, it can be expressed as a formula rather than a table to create an interleaver pattern. This is because the complexity of implementation increases when tables are organized by information size.

3. If hardware (H/W) is supported, the structure must be capable of parallel processing.

A method of generating an interleaver that satisfies the above requirements may be a Quadratic Polynomial Permutation (QPP) interleaver. The QPP interleaver can be generated based on the equation below.

Here, L is the size of the interleaver, and f0 is a constant offset. In particular, in this proposal, f0 is defined as f0=0. f1 is a prime number that satisfies the disjoint relationship with L. In order for f1 to be disjointly related to L regardless of the value of L, the value of f1 may be determined to be a prime number. f2 is a number corresponding to the power of 2 among the factors of L. Setting f2 to a power of 2 is to simplify the formula. Here, the maximum parallel operation unit may be determined based on the value of f2. Possible parallel facts can consist of common divisors of f2. For example, if f2=16, the divisors of 16 are (1,2,4,8,16), and these divisors are possible parallel facts.

At this time, because it is complicated to define QPP parameters for all interleaver lengths L, the interleaver size L corresponding to the pivot is defined, and the information size K between the pivots is in the relationship such that L >= K. define. In other words, the smallest L among those larger than K is determined as the size of the interleaver. Interleaving can be performed according to the rules in Table 3 below.

Figures 24 and 25 are diagrams showing an example of the interleaver design method proposed in this specification.

More specifically, Figures 24 and 25 relate to the case where the length of the interleaver L = 16, f0 = 0, f1 = 3, f2 = 4 in Equation 2 above, and information size K = 16. . By applying L = 16, f0 = 0, f1 = 3, and f2 = 4 to Equation 2 above, interleaving can be performed based on the equation below.

More specifically, Figures 24 and 25 relate to the case where the length of the interleaver L = 16, f0 = 0, f1 = 3, f2 = 4 in Equation 2 above, and information size K = 16. . By applying L = 16, f0 = 0, f1 = 3, and f2 = 4 to Equation 2 above, interleaving can be performed based on the equation below.

Figure 24 is a diagram showing a case where parallel computation is not performed, that is, a single bank memory.

Figure 25 is a diagram showing a case in which two interleaving operations are performed in parallel, that is, a 2-bank memory. Referring to Figure 25, it can be seen that in the case of 2-bank memory, when the 1st bank (0~7) and the 2nd bank (8~15) are activated at the same time (i.e., memory conflict) does not occur. That is, if one value from 0 to 7 is activated in the 1st bank, only one value from 8 to 15 is activated in the 2nd bank. Conversely, if one value from 8 to 15 is activated in the 1st bank, only one value from 0 to 7 is activated in the 2nd bank.

Pooling layer configuration method - (Proposal 4)

In a deep learning structure in the form of CNN, the pooling technique is used for the following two reasons.

1) Overfitting can be prevented. Here, overfitting refers to a phenomenon in which the neural network overlearns the training data, thereby increasing errors in the actual data.

2) The complexity of the neural network can be reduced because the dimensions of the extracted features can be reduced while maintaining the features extracted from the input data.

In general, CNN-type neural networks are widely used in image processing (recognition, classification, etc.). At this time, using pooling in CNN of image processing means that the Cout (Number of out channels) in one image (input data) is not changed, and the features extracted from the one image in each channel are used. It can be understood as reducing the dimension of (output data).

Figure 26 is a diagram to help understand the pooling layer configuration method proposed in this specification.

More specifically, Figure 26 relates to an example of an existing method for reducing the dimension of output data in each channel by applying pooling in CNN. In CNN, the convolution layer performs filtering on input data to generate a feature map, and an activation function is applied to the feature map to generate an activation map. That is, an activation map can be created as the final output result of one convolution layer. At this time, the number of channels in the feature map/activation map has a value equal to the number of filters applied to the input data. Pooling is applied to the activation map. By applying pooling, the size (dimension) of the activation map can be reduced.

Referring to FIG. 26, pooling of a size of 2*2 is applied to the activation map 2610 of a size of 4*4, and as a result, the size of the activation map 2610 is reduced to generate output data of a size of 2*2. You can. In Figure 26, 2620 represents a max pooling method that generates output data of the pooling layer by selecting the largest value within the pooling window, and 2630 represents a method that takes the average of the values included in the pooling window. Indicates the average pooling method that generates the output data of the pooling layer.

Since the video image has a large correlation between the top, bottom, left, and right pixels, even if the dimension of the output data of the convolution layer is reduced by applying pooling according to the method described in Figure 26, the features of the output data are maintained. It can be easy.

On the other hand, unlike the case of video image processing, input data is random in coding to which the methods proposed in this specification are applied. Therefore, there is no correlation between the front and back (or left and right sides) of the input data. For this reason, it may be difficult to apply pooling in the manner described in FIG. 26 in coding to which the methods proposed in this specification are applied.

Figure 27 is a diagram showing changes in the learning ability of a neural network as the number of filters used in the neural network increases. More specifically, Figure 27 relates to a case where the encoder neural network at the transmitting end is composed of a 2-layer CNN, and the decoder neural network at the receiving end is composed of a 5-layer CNN.

As shown in Figure 27, the learning ability of the neural network can increase as a larger number of filters are used, but the number of filters is directly related to the complexity of the CNN.

This proposal applies a pooling technique to reduce the number of output channels, which can reduce the complexity of the neural network while minimizing performance loss. In other words, while the existing CNN pooling technique maintains the number of output channels of the convolution layer and reduces the size of the data in each channel, the pooling technique of this proposal maintains the size of the data in each channel and reduces the size of the data in the convolution layer. This method reduces the number of output channels.

In addition, this proposal additionally proposes configuring the location of the pooling layer differently from the location in the existing CNN.

Figure 28 is a diagram showing an example of the pooling layer configuration method proposed in this specification.

Referring to FIG. 28, 2810 shows a conventional CNN form in which a filtering function, an activation function, and a pooling layer for filtering input data are located sequentially.

On the other hand, in 2820, the locations of the activation function and the pooling layer are changed, so the CNN has a form in which the filtering function, the pooling layer, and the activation function for filtering input data are located sequentially. At 2820, the pooling layer may reduce the number of channels of output data of the filtering function. When pooling is applied to the output data of the filtering function by a pooling layer, the number of output channels of the output data of the filtering function is reduced, and thus the number of activation functions is also reduced by the reduced number of output channels.

Applying pooling in this proposal has the advantage of reducing the complexity of the neural network, but if pooling is applied excessively, a problem may arise where features of the output data disappear. Therefore, this proposal can be preferably applied when pooling is applied only to the first convolutional layer among at least one convolutional layer constituting the encoder neural network at the transmitting end and the decoder neural network at the receiving end.

Figure 29 is a diagram showing an example of a neural network unit (Conv1dCode) structure to which a pooling layer is applied.

More specifically, Figure 29 illustrates a case where the pooling of this proposal is applied to the neural network configuration unit (Conv1dCode) described in Figure 19.

In FIG. 29, 2910 represents the neural network structural unit described in FIG. 19. 2920 further includes a pooling layer in the neural network unit that uses a method of reducing the number of channels of the output data of this proposal, and the neural network unit includes a filtering function (Conv1d), activation function, and pooling for filtering the input data. It consists of a structure in which the layers are located sequentially. 2930 further includes a pooling layer in the neural network unit that uses a method of reducing the number of channels of the output data of this proposal, and the neural network unit includes a filtering function for filtering the input data, a pooling layer, and an activation function sequentially. It consists of a structure located at . By placing the pooling layer in front of the activation function, the number of channels of the filtering function output data is reduced, so the complexity of the CNN can be reduced. In Figure 29, Cin is the number of input channels of input data, k is the kernel size (kernel-size), and Cout is the number of output channels of output data of the filtering function. One of the maximum pooling and average pooling described above may be selected as the pooling method. To reduce noise effects, this proposal can be preferably applied when average pooling is used.

Figure 30 is a diagram showing the results of a comparative analysis of complexity based on whether or not the pooling layer configuration method proposed in this specification is applied to the decoder neural network at the receiving end. More specifically, Figure 30 relates to the case where Block length = 100, decoder feature size (Decoder feature size) = 5, kernel size (kernel_size) = 5, number of filters (# of filter) = 100, and pooling This relates to the results of a comparative analysis of the complexity of a receiving decoder neural network (3010) where pooling was not applied and a receiving decoder neural network (3020) to which pooling was applied only in the first layer.

Referring to the addition operation of the first layer in the receiving end decoder neural network 3020 in which pooling is applied only in the first layer, the pooling applied in the receiving end decoder neural network 3020 is applied to 2 output channels for a total of 100 output channels. Since average pooling, which takes the average by adding each piece, is used, 50 addition operations are additionally performed compared to the case of the receiving decoder neural network 3010 in which pooling is not applied. On the other hand, referring to the number of activation functions (ACT) of the first layer in the receiving decoder neural network 3020 in which pooling is applied only to the first layer, the number of output channels is reduced from 100 to 50 through pooling, so pooling Compared to the case of the receiving end decoder neural network 3010 where this is not applied, 50 fewer activation functions are used.

In Figure 30, looking at the comparative analysis results for the second to fifth layers and the final result, the number of multiplication (Mul), addition (Add), and activation functions (ACT) is approximately lower overall than in the case where pooling was not applied. It is reduced by about 50%.

In Figure 30, the case where pooling is applied only to the first layer is explained, but of course, pooling can also be applied to layers after the second layer (Layer #2). If pooling is applied to layers after the second layer, complexity can be greatly reduced, but performance may be degraded. Therefore, the layer to which pooling is applied can be determined considering the trade-off between performance and complexity.

Rate-Matching for higher code-rate method - (Proposal 5)

In the SCNN-AE structure proposed in this specification, the case where the transmitter encoder neural network 101 uses a code with a code-rate of 1/3 has been previously described. If a code-rate greater than 1/3 must be used in the transmitter encoder neural network, a code-rate greater than 1/3 may be generated through punching.

Figure 31 is a diagram showing an example of a rate-matching method for achieving a high code rate proposed in this specification. In Figure 31, punching is performed in the rate-matching (rate-matching) block 3104, and the receiving end decoder 103 receives a rate greater than 1/3 through the de-rate matching block (3105). A transmitted signal encoded with the code-rate of the value can be restored. At this time, punching patterns may be provided to the transmitter encoder 101 and the receiver decoder 103 for rate matching of the transmitter encoder 101/de-rate matching of the receiver decoder 103. Here, the punching pattern may be provided to the transmitting end/receiving end through a predefined signal.

Performance Analysis

Below, we will look at the channel coding performance analysis results when the methods described above are applied to the transmitter/receiver.

Performance analysis when applying the pooling layer configuration method (Proposal 4)

Figure 32 is a diagram showing channel coding performance when applying the pooling layer configuration method proposed in this specification.

More specifically, Figure 32 is a diagram showing channel coding performance when the pooling layer configuration method proposed in this specification is applied and channel coding performance when the pooling layer configuration method is not applied. The y-axis in Figure 32 represents BER, and x represents SNR. The smaller the BER value, the better the channel coding performance. Therefore, the graph located lower at the same SNR value can be seen as having superior channel coding performance.

In Figure 32, 3210 represents channel coding performance when using traditional LTE Turbo codes. 3220 represents the channel coding performance when PCCC-AE is used, and 3230 represents the channel coding performance when SCCC-AE is used. In addition, 3240 represents the channel coding performance when PCCC-AE with pooling applied to the receiving end decoder is used, and 3250 represents the channel coding performance when SCCC-AE with pooling applied to the receiving end decoder is used. Lastly, 3260 shows the channel coding performance when SCCC-AE with pooling applied to the receiving end decoder is used and the pooling layer is located before the activation function (i.e., proposal 4). The pooling refers to average pooling performed by reducing the number of output channels.

Referring to the position of the graph indicated by 3260, although the complexity has been greatly reduced by reducing the number of output channels and activation functions, from the point of 1 to 2 of the SNR value, compared to the channel coding performance in the case of 3220 to 3250, It can be seen that there is almost no loss of performance.

In the graph of FIG. 32, the presence of legends drawn in the same shape above and below can be understood as depending on the number of filters used in the neural network. In other words, the performance difference between legends drawn in the same shape can be understood to be due to the number of filters used in the neural network.

Performance analysis when applying interleaver design method (Proposal 3)

Figures 33 and 34 are diagrams showing channel coding performance when applying the interleaver design method proposed in this specification.

First, Figure 33 is a diagram showing channel coding performance when an interleaver is not used and channel coding performance when an interleaver is used in a transmitter encoder neural network/receiver decoder neural network.

The y-axis in Figure 33 represents BER, and x represents SNR. The smaller the BER value, the better the channel coding performance. Therefore, the graph located lower at the same SNR value can be seen as having superior channel coding performance. In Figure 33, 3310 represents channel coding performance when using traditional LTE Turbo codes. 3320 represents the channel coding performance when PCCC-AE is used, and 3330 represents the channel coding performance when SCCC-AE is used. In addition, 3340 represents the channel coding performance when SCCC-AE with pooling applied to the receiving end decoder is used and the pooling layer is located before the activation function. At this time, an interleaver is used in all 3310 to 3340. Lastly, 3350 is the same as 3340, but represents the channel coding performance when the interleaver is not used. Referring to the results of FIG. 33, it can be seen that as the value of SNR changes, the case where the interleaver is not used shows the lowest channel coding performance compared to other cases. In light of the results in FIG. 33, it can be seen that an interleaver should be used to ensure good channel coding performance.

In the graph of FIG. 33, the legends corresponding to numbers 3320 to 3350 are indicated by dotted/solid lines, respectively, which can be understood as depending on the number of filters used in the neural network. In other words, the fact that the same legend is drawn with a dotted/solid line can be understood as reflecting the difference in the number of filters used in the neural network.

Next, FIG. 34 is a diagram showing channel coding performance according to cases in which various interleaver patterns are set and applied to the case of 3350 in FIG. 33.

In FIG. 34, among the interleaver parameters f0, f1, and f2 previously described in Proposal 3, f0 = 0 is used. As a result of the performance analysis, the channel coding performance was generally similar in the cases of p0 to p7 shown in Figure 34, but it was confirmed that the channel coding (f1, f2) = (11,32) case showed the best channel coding performance. It has been done. In addition, for the case of 3350 in FIG. 33, the channel coding performance when an interleaver set to (f1, f2) = (11,32) is used is compared to the channel coding performance when SCCC-AE without pooling is used. It has been confirmed that it surpasses

Performance analysis according to various activation functions

Figure 35 is a diagram showing channel coding performance according to the use of various activation functions proposed in this specification.

The y-axis in Figure 35 represents BER, and x represents SNR. The smaller the BER value, the better the channel coding performance. Therefore, the graph located lower at the same SNR value can be seen as having superior channel coding performance.

In Figure 35, 3510 represents channel coding performance when using traditional LTE Turbo codes. 3520 represents the channel coding performance when PCCC-AE is used, and 3530 represents the channel coding performance when SCCC-AE is used. At this time, the eLu function is used as an activation function in all 3310 to 3340. Additionally, 3540 represents the channel coding performance when pooling is applied to the decoder at the receiving end and SCCC-AE, in which the pooling layer is located before the activation function, is used. In 3540, the eRelu function is used as an activation function. Additionally, 3550 represents the channel coding performance when pooling is applied to the decoder at the receiving end and SCCC-AE, in which the pooling layer is located before the activation function, is used. In 3550, the Relu function is used as an activation function. 3560 represents the channel coding performance when SCCC-AE, which uses multiple PRelu as an activation function, is used, and 3570 represents the channel coding performance when SCCC-AE, which uses a single PRelu as an activation function, is used.

The ACT functions evaluated were eLu, ReLu, Single-PReLu, and Multiple-PReLu.

Referring to the results of FIG. 35, it is confirmed that the channel coding performance is the best when the eRelu function is used as the activation function (3540), but in this case, the exponential function must be implemented, so the complexity may be the highest. On the other hand, when the Single-PReLu function is used as an activation function (3580), there is a performance loss compared to the 3540 case, but it can be advantageous in terms of complexity.

Figure 36 is a flowchart showing an example of a method for transmitting and receiving signals in a wireless communication system using an auto encoder proposed in this specification.

Referring to FIG. 36, the transmitter encodes input data based on a pre-trained transmitter encoder neural network (S3610).

Thereafter, the transmitting end transmits the signal to the receiving end based on the encoded input data (S3620).

At this time, the transmitter encoder neural network is (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword and interleaves it. , an interleaver that outputs the interleaved first codeword, and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data. Included serially.

Additionally, the first encoder neural network consists of n1 first encoder component neural networks, and the second encoder neural network consists of n2 first encoder component neural networks and n3 second encoder component neural networks, The n1, n2, and n3 may be positive integers other than 0.

Additionally, the code rate of the first encoder neural network may be 1/n1, and the code rate of the second encoder neural network may be n1/(n2+n3).

In addition, the first encoder component neural network and the second encoder component neural network each consist of at least one layer, and the first encoder component neural network and the second encoder component neural network are configured for input values. It may be configured based on a neural network unit including (i) a filtering function for filtering and (ii) an activation function that receives the output of the filtering function.

Here, each of the first encoder component neural network and the second encoder component neural network may be composed of two layers.

Additionally, the values of n1 and n2 may be the same, and n3 may be 1.

Additionally, the first encoder component neural network includes a first neural network unit that receives one input value and a second neural network unit that receives an output of the first neural network unit, and the second neural network unit Linear regression may be applied to the output of the network component, and an eLu (exponential linear unit) function may be applied to the output of the second neural network component to which the linear regression is applied.

And, the second encoder component neural network includes a third neural network unit that receives the n1 input values and a fourth neural network unit that receives the output values of the third neural network unit, and the first The linear regression may be applied to the output value of four neural network constituent units, and the eLu function may be applied to the output value of the fourth neural network constituent unit to which the linear regression has been applied.

In addition, the first neural network structural unit and the third neural network structural unit constitute a first layer, which is one of the two layers, and the second neural network structural unit and the fourth neural network structural unit constitute the second layer. A second layer, which is the remaining one of the layers, can be configured.

Additionally, the number of output channels through which output values of each of the first to fourth neural network units are output may be 100.

And, the i+1th bit in the first codeword bit stream input to the interleaver is interleaved based on the equation below,

Here, L is the size of the interleaver, f0 is a constant offset, f1 is a prime number that satisfies the disjoint relationship with L, and f2 is the power of 2 among the divisors of L. It may be a number corresponding to the power of 2.

Additionally, f0 is 0, and a parallel operation can be performed on the bit string of the first codeword equal to the value of f2.

In addition, the specific neural network unit constituting the first layer among the at least one layer further includes a pooling layer for reducing the number of output channels of the filtering function, and the pooling layer is configured to reduce the number of output channels of the filtering function. It may be located between the filtering function and the activation function.

Additionally, each neural network unit constituting the layers located after the first layer may further include the pooling layer.

Devices used in wireless communication systems

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

Hereinafter, a more detailed example will be provided with reference to the drawings. In the following drawings/descriptions, identical reference numerals may illustrate identical or corresponding hardware blocks, software blocks, or functional blocks, unless otherwise noted.

Figure 37 illustrates a communication system applied to the present invention.

Referring to FIG. 37, the communication system 1 applied to the present invention includes a wireless device, a base station, and a network. Here, a wireless device refers to a device that performs communication using wireless access technology (e.g., 5G NR (New RAT), LTE (Long Term Evolution)) and may be referred to as a communication/wireless/5G device. Although not limited thereto, wireless devices include robots (100a), vehicles (100b-1, 100b-2), XR (eXtended Reality) devices (100c), hand-held devices (100d), and home appliances (100e). ), IoT (Internet of Thing) device (100f), and AI device/server (400). For example, vehicles may include vehicles equipped with wireless communication functions, autonomous vehicles, vehicles capable of inter-vehicle communication, etc. Here, the vehicle may include an Unmanned Aerial Vehicle (UAV) (eg, a drone). XR devices include AR (Augmented Reality)/VR (Virtual Reality)/MR (Mixed Reality) devices, HMD (Head-Mounted Device), HUD (Head-Up Display) installed in vehicles, televisions, smartphones, It can be implemented in the form of computers, wearable devices, home appliances, digital signage, vehicles, robots, etc. Portable devices may include smartphones, smart pads, wearable devices (e.g., smartwatches, smart glasses), and computers (e.g., laptops, etc.). Home appliances may include TVs, refrigerators, washing machines, etc. IoT devices may include sensors, smart meters, etc. For example, a base station and network may also be implemented as wireless devices, and a specific wireless device 200a may operate as a base station/network node for other wireless devices.

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 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, 4G (eg, LTE) network, or 5G (eg, NR) network. Wireless devices 100a to 100f may communicate with each other through the base station 200/network 300, but may also communicate directly (e.g. sidelink communication) without going through the base station/network. For example, vehicles 100b-1 and 100b-2 may communicate directly (e.g. V2V (Vehicle to Vehicle)/V2X (Vehicle to everything) communication). Additionally, an IoT device (eg, sensor) may communicate directly with another IoT device (eg, sensor) or another wireless device (100a to 100f).

Wireless communication/connection (150a, 150b) may be established between the wireless devices (100a~100f)/base station (200) and the base station (200)/wireless devices (100a~100f). Here, wireless communication/connection, uplink/downlink communication 150a and sidelink communication 150b (or D2D communication) may be achieved through various wireless access technologies (e.g., 5G NR). Through wireless communication/connection (150a, 150b), a wireless device and a base station/wireless device can transmit/receive wireless signals to each other. For example, wireless communication/connection 150a, 150b may transmit/receive signals through various physical channels based on all/part of the process of FIG. 1. To this end, based on the various proposals of the present invention, various configuration information setting processes for transmitting/receiving wireless signals, various signal processing processes (e.g., channel encoding/decoding, modulation/demodulation, resource mapping/demapping, etc.) , at least some of the resource allocation process, etc. may be performed.

Figure 38 illustrates a wireless device to which the present invention can be applied.

Referring to FIG. 38, the first wireless device 100 and the second wireless device 200 can transmit and receive wireless signals through various wireless access technologies (eg, LTE, NR). Here, {first wireless device 100, second wireless device 200} refers to {wireless device 100x, base station 200} and/or {wireless device 100x, wireless device 100x) in FIG. 37. } can be responded to.

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 then transmit a wireless signal including the first information/signal through the transceiver 106. Additionally, the processor 102 may receive a wireless 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 may store software code containing 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 wireless communication technology (eg, LTE, NR). Transceiver 106 may be coupled to processor 102 and may transmit and/or receive wireless signals via one or more antennas 108. Transceiver 106 may include a transmitter and/or receiver. The transceiver 106 can be used interchangeably with an RF (Radio Frequency) 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 the information in the memory 204 to generate third information/signal and then transmit a wireless signal including the third information/signal through the transceiver 206. Additionally, the processor 202 may receive a wireless signal including the fourth information/signal through the transceiver 206 and then 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 wireless communication technology (eg, LTE, NR). Transceiver 206 may be coupled to processor 202 and may transmit and/or receive wireless signals via one or more antennas 208. Transceiver 206 may include a transmitter and/or receiver. 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, the hardware elements of the wireless devices 100 and 200 will be described in more detail. Although not limited thereto, 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 (e.g., 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 the functions, procedures, suggestions and/or methods disclosed herein. One or more processors 102, 202 may generate messages, control information, data or information according to the functions, procedures, suggestions and/or methods disclosed herein. One or more processors 102, 202 generate signals (e.g., baseband signals) containing PDUs, SDUs, messages, control information, data or 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 (e.g., baseband signals) from one or more transceivers (106, 206) and transmit a PDU, SDU, or PDU according to the functions, procedures, suggestions, 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. As an 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, suggestions 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, etc. 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) to enable one or more processors (102, 202). 202). The functions, procedures, suggestions and or methods disclosed in this document 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 connected to 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 consist 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 internal to and/or external to one or more processors 102, 202. Additionally, one or more memories 104, 204 may be connected 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, wireless signals/channels, etc. mentioned in the methods and/or operation flowcharts of this document to one or more other devices. One or more transceivers 106, 206 may receive user data, control information, wireless signals/channels, etc. mentioned in the functions, procedures, proposals, methods and/or operational flowcharts disclosed herein, etc. 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 may 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 wireless 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 wireless signals from one or more other devices. In addition, one or more transceivers (106, 206) may be connected to one or more antennas (108, 208), and one or more transceivers (106, 206) may perform the functions and procedures disclosed in this document through one or more antennas (108, 208). , may be set to transmit and receive user data, control information, wireless signals/channels, etc. mentioned in the proposal, method and/or operation flowchart, etc. In this document, one or more antennas may be multiple physical antennas or multiple logical antennas (eg, antenna ports). One or more transceivers (106, 206) process the received user data, control information, wireless signals/channels, etc. using one or more processors (102, 202), and convert the received wireless signals/channels, etc. from the RF band signal. It can be converted to a baseband signal. One or more transceivers (106, 206) may convert user data, control information, wireless signals/channels, etc. processed using one or more processors (102, 202) from baseband signals to RF band signals. For this purpose, one or more transceivers 106, 206 may comprise (analog) oscillators and/or filters.

Figure 39 illustrates a signal processing circuit for a transmission signal.

Referring to FIG. 39, 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 thereto, the operations/functions of Figure 39 may be performed in the processors 102, 202 and/or transceivers 106, 206 of Figure 39. The hardware elements of Figure 39 may be implemented in the processors 102, 202 and/or transceivers 106, 206 of Figure 38. For example, blocks 1010 to 1060 may be implemented in processors 102 and 202 of FIG. 38. Additionally, blocks 1010 to 1050 may be implemented in the processors 102 and 202 of FIG. 38, and block 1060 may be implemented in the transceivers 106 and 206 of FIG. 38.

The codeword can be converted into a wireless signal through the signal processing circuit 1000 of FIG. 39. Here, a codeword is an encoded bit sequence of an information block. The information block may include a transport block (eg, UL-SCH transport block, DL-SCH transport block). Wireless signals may be transmitted through various physical channels (eg, PUSCH, PDSCH) of FIG. 1.

Specifically, the codeword may be converted into a scrambled bit sequence by the scrambler 1010. The scramble sequence used for scrambling is generated based on an initialization value, and the initialization value may include ID information of the wireless device. The scrambled bit sequence may be modulated into a modulation symbol sequence by the modulator 1020. Modulation methods may include pi/2-BPSK (pi/2-Binary Phase Shift Keying), m-PSK (m-Phase Shift Keying), m-QAM (m-Quadrature Amplitude Modulation), etc. The complex modulation symbol sequence may be mapped to one or more transport layers by the layer mapper 1030. The 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 with the precoding matrix W of N*M. 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. Additionally, the precoder 1040 may perform precoding without performing transform precoding.

The resource mapper 1050 can map the modulation symbols of each antenna port to time-frequency resources. A time-frequency resource may include a plurality of symbols (eg, CP-OFDMA symbol, DFT-s-OFDMA symbol) in the time domain and a plurality of subcarriers in the frequency domain. The signal generator 1060 generates a wireless signal from the mapped modulation symbols, and the generated wireless signal can be transmitted to another device through each antenna. For this purpose, 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, etc. .

The signal processing process for the received signal in the wireless device may be configured as the reverse of the signal processing process (1010 to 1060) of FIG. 39. For example, a wireless device (eg, 100 and 200 in FIG. 38) may receive a wireless signal from the outside through an antenna port/transceiver. The received wireless signal can 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. Afterwards, the baseband signal can be restored to a codeword through a resource de-mapper process, postcoding process, demodulation process, and de-scramble process. The codeword can be restored to the original information block through decoding. Accordingly, a signal processing circuit (not shown) for a received signal may include a signal restorer, resource de-mapper, postcoder, demodulator, de-scrambler, and decoder.

Figure 40 shows another example of a wireless device applied to the present invention. Wireless devices can be implemented in various forms depending on usage-examples/services.

Referring to FIG. 40, the wireless devices 100 and 200 correspond to the wireless devices 100 and 200 of FIG. 38 and include various elements, components, units/units, and/or modules. ) can be composed of. 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 and/or one or more memories 104,204 of FIG. 38. For example, transceiver(s) 114 may include one or more transceivers 106, 206 and/or one or more antennas 108, 208 of FIG. 38. 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 the electrical/mechanical operation of the wireless device based on the program/code/command/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 (e.g., another communication device) through the communication unit 110 through a wireless/wired interface, or to the outside (e.g., to another communication device) through the communication unit 110. Information received through a wireless/wired interface from another communication device may be stored in the memory unit 130.

The additional element 140 may be configured in various ways depending on the type of wireless device. For example, the additional element 140 may include at least one of a power unit/battery, an input/output unit (I/O unit), a driving unit, and a computing unit. Although not limited thereto, wireless devices include robots (FIG. 37, 100a), vehicles (FIG. 37, 100b-1, 100b-2), XR devices (FIG. 37, 100c), portable devices (FIG. 37, 100d), and home appliances. (FIG. 37, 100e), IoT device (FIG. 37, 100f), digital broadcasting terminal, hologram device, public safety device, MTC device, medical device, fintech device (or financial device), security device, climate/environment device, It can be implemented in the form of an AI server/device (FIG. 37, 400), a base station (FIG. 37, 200), a network node, etc. Wireless devices can be mobile or used in fixed locations depending on the usage/service.

In FIG. 40 , various elements, components, units/parts, and/or modules within the wireless devices 100 and 200 may be entirely interconnected through a wired interface, or at least some may be wirelessly connected through the communication unit 110. For example, within 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 unit (e.g., 130 and 140) are connected through the communication unit 110. Can be connected wirelessly. Additionally, each element, component, unit/part, and/or module within the wireless devices 100 and 200 may further include one or more elements. For example, the control unit 120 may be comprised of one or more processor sets. For example, the control unit 120 may be comprised of a communication control processor, an application processor, an electronic control unit (ECU), a graphics processing processor, and a memory control processor. As another example, the memory unit 130 includes 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, the implementation example of FIG. 40 will be described in more detail with reference to the drawings.

Figure 41 illustrates a portable device to which the present invention is applied. Portable devices may include smartphones, smartpads, wearable devices (e.g., smartwatches, smartglasses), and portable computers (e.g., laptops, etc.). A mobile device may be referred to as a Mobile Station (MS), user terminal (UT), Mobile Subscriber Station (MSS), Subscriber Station (SS), Advanced Mobile Station (AMS), or Wireless terminal (WT).

Referring to FIG. 41, 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 include. The antenna unit 108 may be configured as part of the communication unit 110. Blocks 110 to 130/140a to 140c correspond to blocks 110 to 130/140 in FIG. 40, respectively.

The communication unit 110 can transmit and receive signals (eg, data, control signals, etc.) with other wireless devices and base stations. The control unit 120 can control the components of the portable device 100 to perform various operations. 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. Additionally, the memory unit 130 can store input/output data/information, etc. The power supply unit 140a supplies power to the portable device 100 and may include a wired/wireless charging circuit, a battery, etc. The interface unit 140b may support connection between the mobile device 100 and other external devices. The interface unit 140b may include various ports (eg, audio input/output ports, video input/output ports) for connection to external devices. The input/output unit 140c may input or output video information/signals, audio information/signals, data, and/or information input from the 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 (e.g., touch, text, voice, image, video) input from the user, and the obtained information/signals are stored in the memory unit 130. It can be saved. The communication unit 110 may convert the information/signal stored in the memory into a wireless signal and transmit the converted wireless signal directly to another wireless device or to a base station. Additionally, the communication unit 110 may receive a wireless signal from another wireless device or a base station and then restore the received wireless signal to the original information/signal. The restored information/signal may be stored in the memory unit 130 and then output in various forms (eg, text, voice, image, video, haptics) through the input/output unit 140c.

Figure 42 illustrates a vehicle or autonomous vehicle to which the present invention is applied. A vehicle or autonomous vehicle can be implemented as a mobile robot, vehicle, train, manned/unmanned aerial vehicle (AV), ship, etc.

Referring to FIG. 42, the vehicle or autonomous vehicle 100 includes an antenna unit 108, a communication unit 110, a control unit 120, a drive unit 140a, a power supply unit 140b, a sensor unit 140c, and an autonomous driving unit. It may include a portion 140d. 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 in FIG. 40.

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

For example, the communication unit 110 may receive map data, traffic information data, etc. from an external server. The autonomous driving unit 140d can create an autonomous driving route and driving plan based on the acquired data. The control unit 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 (e.g., speed/direction control). During autonomous driving, the communication unit 110 may acquire the latest traffic information data from an external server irregularly/periodically and obtain surrounding traffic information data from surrounding vehicles. Additionally, during autonomous driving, the sensor unit 140c can obtain vehicle status and surrounding environment information. The autonomous driving unit 140d may update the autonomous driving route and driving plan based on newly acquired data/information. The communication unit 110 may transmit information about vehicle location, autonomous driving route, driving plan, etc. to an external server. An external server can predict traffic information data in advance using AI technology, etc., based on information collected from vehicles or self-driving vehicles, and provide the predicted traffic information data to the vehicles or self-driving vehicles.

Figure 43 illustrates a vehicle to which the present invention is applied. Vehicles can also be implemented as transportation, trains, airplanes, ships, etc.

Referring to FIG. 43, 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 correspond to blocks 110 to 130/140 of FIG. 40, respectively.

The communication unit 110 may transmit and receive signals (eg, data, control signals, etc.) with other vehicles or external devices such as a base station. The control unit 120 can control components of the vehicle 100 to perform various operations. The memory unit 130 may store data/parameters/programs/codes/commands that support 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 measuring 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 the driving line, acceleration information, and location information with surrounding vehicles. The location measuring unit 140b may include GPS and various sensors.

For example, the communication unit 110 of the vehicle 100 may receive map information, traffic information, etc. 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 control unit 120 creates a virtual object based on map information, traffic information, and vehicle location information, and the input/output unit 140a can display the generated virtual object on the window of the vehicle (1410, 1420). Additionally, the control unit 120 may determine whether the vehicle 100 is operating normally within the travel line based on vehicle location information. If the vehicle 100 deviates from the driving line abnormally, the control unit 120 may display a warning on the window of the vehicle through the input/output unit 140a. Additionally, the control unit 120 may broadcast a warning message regarding driving abnormalities to surrounding vehicles through the communication unit 110. Depending on the situation, the control unit 120 may transmit location information of the vehicle and information about driving/vehicle abnormalities to the relevant organizations through the communication unit 110.

Figure 44 illustrates an XR device applied to the present invention. XR devices can be implemented as HMDs, HUDs (Head-Up Displays) installed in vehicles, televisions, smartphones, computers, wearable devices, home appliances, digital signage, vehicles, robots, etc.

Referring to FIG. 44, 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 correspond to blocks 110 to 130/140 of FIG. 40, respectively.

The communication unit 110 may transmit and receive signals (eg, media data, control signals, etc.) with external devices such as other wireless devices, mobile devices, or media servers. Media data may include video, images, sound, etc. The control unit 120 may perform various operations by controlling the components of the XR device 100a. For example, the control unit 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 generated XR object. The input/output unit 140a may include a camera, microphone, user input unit, display unit, speaker, and/or haptic module. The sensor unit 140b can obtain XR device status, surrounding environment information, user information, etc. The sensor unit 140b may include a proximity sensor, an illumination 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, etc.

As an example, the memory unit 130 of the XR device 100a may include information (eg, data, etc.) necessary for creating an XR object (eg, AR/VR/MR object). The input/output unit 140a can obtain a command to operate the XR device 100a from the user, and the control unit 120 can 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 sends content request information to another device (e.g., mobile device 100b) or It can be transmitted to a media server. The communication unit 130 may download/stream content such as movies and news from another device (eg, mobile 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, and metadata creation/processing for the content, and acquires it through the input/output unit 140a/sensor unit 140b. XR objects can be created/output based on information about surrounding space or real objects.

Additionally, the XR device 100a is wirelessly connected to the mobile device 100b through the communication unit 110, and the operation of the XR device 100a can be controlled by the mobile 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 obtain 3D location information of the mobile device 100b and then generate and output an XR object corresponding to the mobile device 100b.

Figure 45 illustrates a robot applied to the present invention. Robots can be classified into industrial, medical, household, military, etc. depending on the purpose or field of use.

Referring to FIG. 45, 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 driver 140c. Here, blocks 110 to 130/140a to 140c correspond to blocks 110 to 130/140 of FIG. 40, respectively.

The communication unit 110 may transmit and receive signals (e.g., driving information, control signals, etc.) with external devices such as other wireless devices, other robots, or control servers. The control unit 120 can control the components of the robot 100 to perform various operations. The memory unit 130 may store data/parameters/programs/codes/commands that support 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, microphone, user input unit, display unit, speaker, and/or haptic module. The sensor unit 140b can obtain internal information of the robot 100, surrounding environment information, user information, etc. The sensor unit 140b may include a proximity sensor, an illumination 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, etc. The driving unit 140c can perform various physical operations such as moving robot joints. Additionally, the driving unit 140c can cause the robot 100 to run on the ground or fly in the air. The driving unit 140c may include an actuator, motor, wheel, brake, propeller, etc.

Figure 46 illustrates an AI device applied to the present invention. AI devices are 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, vehicles, etc. It can be implemented with any available device.

Referring to FIG. 46, the AI device 100 includes a communication unit 110, a control unit 120, a memory unit 130, an input/output unit (140a/140b), a learning processor unit 140c, and a sensor unit 140d. may include. Blocks 110 to 130/140a to 140d correspond to blocks 110 to 130/140 in FIG. 40, respectively.

The communication unit 110 uses wired and wireless communication technology to communicate with external devices such as other AI devices (e.g., 100x, 200, 400 in Figure 37) or the AI server 200 and wired and wireless signals (e.g., sensor information, user input, learning 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 an external device to the memory unit 130.

The control unit 120 may determine at least one executable operation of the AI device 100 based on information determined or generated using a data analysis algorithm or a machine learning algorithm. And, the control unit 120 can control the components of the AI device 100 to perform the determined operation. For example, the control unit 120 may request, search, receive, or utilize data from the learning processor unit 140c or the memory unit 130, and may select at least one executable operation that is predicted or is determined to be desirable. Components of the AI device 100 can be controlled to execute operations. In addition, the control unit 120 collects history information including the user's feedback on the operation content or operation of the AI device 100 and stores it in the memory unit 130 or the learning processor unit 140c, or the AI server ( It can be transmitted to an external device such as Figure 37, 400). The collected historical information can be used to update the learning model.

The memory unit 130 can 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. Additionally, 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 can obtain various types of data from outside the AI device 100. For example, the input unit 120 may acquire training data for model training and input data to which the learning model will be applied. The input unit 140a may include a camera, microphone, and/or a user input unit. The output unit 140b may generate output related to vision, hearing, or tactile sensation. 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 using various sensors. The sensing unit 140 may include a proximity sensor, an illumination 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 can train a model composed of an artificial neural network using training data. The learning processor unit 140c may perform AI processing together with the learning processor unit of the AI server (FIG. 37, 400). 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. Additionally, 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 embodiments described above combine the components and features of the present invention in a predetermined form. Each component or feature should be considered optional unless explicitly stated otherwise. Each component or feature may be implemented in a form that is not combined with other components or features. Additionally, it is possible to configure an embodiment of the present invention by combining some components and/or features. The order of operations described in embodiments of the present invention may be changed. Some features or features of one embodiment may be included in other embodiments or may be replaced with corresponding features or features of other embodiments. It is obvious that claims that do not have an explicit reference relationship in the patent claims can be combined to form an embodiment or included as a new claim through amendment after filing.

Embodiments according to the present invention may be implemented by various means, for example, hardware, firmware, software, or a combination thereof. In the case of implementation by hardware, an embodiment of the present invention includes one or more application specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), and FPGAs ( It can be implemented by field programmable gate arrays, processors, controllers, microcontrollers, microprocessors, etc.

In the case of implementation by firmware or software, an embodiment of the present invention may be implemented in the form of a module, procedure, function, etc. that performs the functions or operations described above. Software code can be stored in memory and run by a processor. The memory is located inside or outside the processor and can exchange data with the processor through various known means.

It is obvious to those skilled in the art that the present invention can be embodied in other specific forms without departing from the essential features of the present invention. Accordingly, the above detailed description should not be construed as restrictive in all respects and should be considered illustrative. The scope of the present invention should be determined by reasonable interpretation of the appended claims, and all changes within the equivalent scope of the present invention are included in the scope of the present invention.

The present invention has been described focusing on examples of application to 3GPP LTE/LTE-A and 5G systems, but it can be applied to various wireless communication systems in addition to 3GPP LTE/LTE-A and 5G systems.

Claims (19)

The method for the transmitter to transmit a signal in a wireless communication system using an auto encoder is:
Encoding input data based on a pre-trained transmitter encoder neural network; and
Including transmitting the signal to a receiving end based on the encoded input data,
The transmitter encoder neural network is (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword and interleaves it, and the interleaving an interleaver that outputs the first codeword, and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data in series. (serially) A method characterized by including. According to claim 1,
The first encoder neural network is composed of n1 first encoder component neural networks, and the second encoder neural network is composed of n2 first encoder component neural networks and n3 second encoder component neural networks,
Wherein n1, n2 and n3 are positive integers other than 0. According to claim 2,
Based on the length of the input data being k, the code rate of the first encoder neural network is k/n1, the code rate of the second encoder neural network is n1/(n2+n3), A method characterized in that the total code rate of the transmitter encoder neural network is k/(n2+n3). According to claim 2,
Each of the first encoder component neural network and the second encoder component neural network consists of at least one layer,
The first encoder component neural network and the second encoder component neural network include (i) a filtering function for filtering the input value and (ii) an activation function that receives the output of the filtering function. A method characterized in that it is constructed based on a neural network structural unit including. According to claim 4,
The method is characterized in that each of the first encoder component neural network and the second encoder component neural network consists of two layers. According to claim 5,
The method wherein the values of n1 and n2 are the same, and n3 is 1. According to claim 6,
The first encoder component neural network includes a first neural network unit that receives one input value and a second neural network unit that receives an output of the first neural network unit, and the second neural network unit consists of A method characterized by applying linear regression to the output of a unit and applying an eLu (exponential linear unit) function to the output of the second neural network constituent unit to which the linear regression is applied. According to claim 7,
The second encoder component neural network includes a third neural network unit that receives the n1 input values and a fourth neural network unit that receives output values of the third neural network unit, and the fourth neural network unit receives the n1 input values. A method characterized by applying the linear regression to an output value of a network component and applying the eLu function to an output value of the fourth neural network component to which the linear regression is applied. According to claim 8,
The first neural network structural unit and the third neural network structural unit constitute a first layer, one of the two layers,
The method characterized in that the second neural network structural unit and the fourth neural network structural unit constitute a second layer, which is the remaining one of the two layers. According to claim 8,
A method wherein the number of output channels through which output values of each of the first to fourth neural network units are output is 100. According to claim 1,
The i+1th bit in the first codeword bit stream input to the interleaver is interleaved based on the equation below,
[Equation]

Here, L is the size of the interleaver, f0 is a constant offset, f1 is a prime number that satisfies the disjoint relationship with L, and f2 is the power of 2 among the divisors of L. A method characterized in that the number corresponds to the power of 2. According to claim 11,
The method is characterized in that the f0 is 0, and a parallel operation is performed on the bit string of the first codeword as much as the value of f2. According to claim 4,
A specific neural network unit constituting the first layer among the at least one layer further includes a pooling layer for reducing the number of output channels of the filtering function,
The method characterized in that the pooling layer is located between the filtering function and the activation function. According to claim 13,
Each neural network structural unit constituting layers located after the first layer further includes the pooling layer. In a transmitting terminal for transmitting and receiving signals in a wireless communication system using an auto encoder,
A transmitter for transmitting wireless signals;
A receiver for receiving wireless signals;
at least one processor; and
at least one computer memory operably connectable to the at least one processor and storing instructions that, when executed by the at least one processor, perform operations;
The above operations are,
Encoding input data based on a pre-trained transmitter encoder neural network; and
Including transmitting the signal to a receiving end based on the encoded input data,
The transmitter encoder neural network is (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword and interleaves it, and the interleaving an interleaver that outputs the first codeword, and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data in series. A transmitting terminal characterized in that it includes (serially). The method for the receiving end to receive a signal in a wireless communication system using an auto encoder is:
Receiving a signal including data encoded based on a pre-trained transmitter encoder neural network from a transmitter;
Including decoding the encoded data included in the signal based on a pre-trained receiving end decoder neural network,
The receiving end decoder neural network includes at least one basic decoder neural network connected sequentially,
Each of the at least one basic decoder neural network (i) performs decoding on a second codeword output from a second encoder neural network included in the transmitter encoder neural network and outputs a second decoded codeword. A decoder neural network, (ii) a de-interleaver that receives the second decoded codeword, de-interleaves it, and outputs the deinterleaved second decoded codeword, and (iii) a first decoder neural that performs decoding on the de-interleaved second decoded codeword and outputs a first decoded codeword based on the decoding on the de-interleaved second decoded codeword A method characterized in that it includes the network serially. In a transmitting terminal for transmitting and receiving signals in a wireless communication system using an auto encoder,
A transmitter for transmitting wireless signals;
A receiver for receiving wireless signals;
at least one processor; and
at least one computer memory operably connectable to the at least one processor and storing instructions that, when executed by the at least one processor, perform operations;
The above operations are,
Receiving a signal including data encoded based on a pre-trained transmitter encoder neural network from a transmitter;
Including decoding the encoded data included in the signal based on a pre-trained receiving end decoder neural network,
The receiving end decoder neural network includes at least one basic decoder neural network connected sequentially,
Each of the at least one basic decoder neural network (i) performs decoding on a second codeword output from a second encoder neural network included in the transmitter encoder neural network and outputs a second decoded codeword. A decoder neural network, (ii) a de-interleaver that receives the second decoded codeword, de-interleaves it, and outputs the deinterleaved second decoded codeword, and (iii) a first decoder neural that performs decoding on the de-interleaved second decoded codeword and outputs a first decoded codeword based on the decoding on the de-interleaved second decoded codeword A method characterized in that it includes the network serially. In a non-transitory computer readable medium (CRM) storing one or more instructions,
One or more instructions executable by one or more processors are provided by the transmitting end,
Encoding the input data based on a pre-trained transmitter encoder neural network,
The signal is transmitted to the receiving end based on the encoded input data,
The transmitter encoder neural network is (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword and interleaves it, and the interleaving an interleaver that outputs the first codeword, and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data in series. A non-transitory computer-readable medium comprising (serially). A device comprising one or more memories and one or more processors functionally connected to the one or more memories,
The one or more processors allow the device to:
Control the device to encode input data based on a pre-trained transmitter encoder neural network of the device,
Control the device to transmit the signal to the receiving end based on the encoded input data,
The transmitter encoder neural network is (i) a first encoder neural network that receives the input data and outputs a first codeword, (ii) receives the first codeword and interleaves it, and the interleaving an interleaver that outputs the first codeword, and (iii) a second encoder neural network that receives the interleaved first codeword and outputs a second codeword constituting the encoded input data in series. (serially) A device characterized in that it includes.

Images