Constellation diagram determination method and apparatus, and device and storage medium

The pre-trained neural network determines the constellation diagram based on channel characteristics and modulated bit sequences in wireless communication, which solves the constellation diagram design problem in the prior art and improves communication performance and matching degree.

WO2024230255A1PCT designated stage expired Publication Date: 2024-11-14ZTE CORP
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
PCT/CN2024/073939
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-08
Filing Date
2024-01-25
Publication Date
2024-11-14

AI Technical Summary

Technical Problem

The prior art is difficult to effectively utilize neural networks to design and optimize constellation diagrams in wireless communications, resulting in limited communication performance.

Method used

Through a pre-trained neural network, the constellation diagram is determined based on channel feature information and modulated bit sequences, so as to match the constellation diagram and channel features, and improve communication performance.

Benefits of technology

The performance and communication performance of the constellation diagram are improved, making it highly matched with the current channel characteristics and adapted to different channel environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN2024073939_14112024_PF_FP_ABST
    Figure CN2024073939_14112024_PF_FP_ABST
Patent Text Reader

Abstract

A constellation diagram determination method and apparatus, and a device and a storage medium. The constellation diagram determination method comprises: acquiring channel feature information and M modulation bit sequences (S101); and by means of a pre-trained neural network, determining a constellation diagram corresponding to the M modulation bit sequences, wherein M is the number of points in a modulation constellation diagram, and the constellation diagram corresponds to the channel feature information (S102).
Need to check novelty before this filing date? Find Prior Art

Description

Constellation diagram determination method, device, equipment and storage medium Technical Field

[0001] The present application relates to the field of communication technology, for example, to a constellation diagram determination method, apparatus, device and storage medium. Background Art

[0002] Digital communication systems use digital signals to transmit information. In digital communication systems, the information to be transmitted is transmitted on a carrier wave. The information to be transmitted and the information transmitted on the carrier wave are typically mapped using digital modulation. Digital modulation can distinguish signals based on amplitude and phase. Each amplitude and phase combination can be represented as a point in two-dimensional space. All points with amplitude and phase combinations can be viewed as a constellation diagram on a two-dimensional plane. A constellation diagram consists of multiple constellation points, each of which is formed by a combination of amplitude and phase. Generally, the information to be transmitted is encoded and then mapped onto the constellation diagram to achieve digital modulation.

[0003] Furthermore, neural networks excel in fields such as image and natural language processing, and therefore can be applied to the physical layer of wireless communications, such as encoding and decoding, modulation and demodulation, and channel estimation. Therefore, how to use neural networks to design constellations has become a pressing technical challenge for those skilled in the art.

[0004] Summary of the Invention

[0005] Embodiments of the present application provide a constellation diagram determination method, apparatus, device, and storage medium.

[0006] In a first aspect, an embodiment of the present application provides a constellation diagram determination method, including:

[0007] Acquire channel characteristic information and M modulation bit sequences; determine a constellation diagram corresponding to the M modulation bit sequences through a pre-trained neural network; wherein M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0008] In a second aspect, an embodiment of the present application provides a constellation diagram determination device, including:

[0009] An acquisition module is used to obtain channel characteristic information and M modulated bit sequences; a determination module is used to determine the constellation diagram corresponding to the M modulated bit sequences through a pre-trained neural network; wherein M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0010] In a third aspect, an embodiment of the present application provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the constellation diagram determination method provided in the first aspect of the embodiment of the present application is implemented.

[0011] In a fourth aspect, an embodiment of the present application provides a storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the constellation diagram determination method provided in the first aspect of the embodiment of the present application.

[0012] The technical solution provided by the embodiments of the present application obtains channel characteristic information and M modulated bit sequences; a constellation corresponding to the M modulated bit sequences is determined using a pre-trained neural network; where M is the number of points in the modulation constellation, and the constellation corresponds to the channel characteristic information. In this method, the constellation used by both communicating parties during the communication process can be adaptively changed based on the different channel characteristics, resulting in a high degree of match between the determined constellation and the current channel characteristics, thereby improving the performance of the constellation and, consequently, communication performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] FIG1 is a flow chart of a constellation diagram determination method provided in an embodiment of the present application;

[0014] FIG2 is a flow chart of a neural network determination process according to an embodiment of the present application;

[0015] FIG3 is a schematic diagram of a structure of a neural network provided in an embodiment of the present application;

[0016] FIG4 is another schematic diagram of the structure of a neural network provided in an embodiment of the present application;

[0017] FIG5 is another structural diagram of a neural network provided in an embodiment of the present application;

[0018] FIG6 is another structural diagram of a neural network provided in an embodiment of the present application;

[0019] FIG7 is a schematic diagram of a structure of a neural network in conventional technology;

[0020] FIG8 is a schematic diagram of the mutual information empirical cumulative distribution function of the constellation diagram provided by conventional technology and an embodiment of the present application;

[0021] FIG9 is a schematic structural diagram of a constellation diagram determination device provided in an embodiment of the present application;

[0022] FIG10 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0023] The specific embodiments described herein are merely for explaining the present application. The embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0024] The constellation diagram determination method provided in the embodiments of the present application can be applied to various wireless communication systems, such as satellite communication systems, Internet of Things (IoT), narrowband Internet of Things (NB-IoT), global system for mobile communications (GSM), enhanced data rate for GSM evolution (EDGE) system, wideband code division multiple access (WCDMA) system, code division multiple access 2000 (CDMA2000) system, time division-synchronization code division multiple access (TD-SCDMA) system, long term evolution (LTE) system, fourth generation mobile communication technology (4G) system, fifth generation mobile communication technology (5G) system, LTE and 5G hybrid architecture system, 5G New Radio (NR) system, and new communication systems emerging in future communication development, such as sixth generation mobile communication technology (6G) system.

[0025] To facilitate understanding by those skilled in the art, the following describes the relevant concepts involved in the embodiments of the present application:

[0026] Constellation diagrams are used by communication nodes to modulate signals during transmission and demodulate signals during reception. Therefore, both the transmitter and receiver must use the same constellation diagram for communication. Generally, constellations define the amplitude and phase of signal elements. Digital modulation can distinguish signals based on amplitude and phase. In a constellation diagram, each signal element is represented by a constellation point, and each combination of amplitude and phase is represented by a constellation point. For example, the horizontal X-axis of a constellation diagram corresponds to the in-phase carrier, while the vertical Y-axis corresponds to the quadrature carrier.

[0027] The number and coordinates of constellation points in the constellation diagram of different modulation and demodulation methods may be different. For example, the modulation and demodulation methods include Quadrature Phase Shift Keying (QPSK), 16-bit Quadrature Amplitude Modulation (QAM) or 64QAM. When there are x constellation points in a constellation diagram, it means that each constellation point can represent log2(x) bits of binary information. For example, there are 4 constellation points in the QPSK constellation diagram, each constellation point can represent 2 bits of binary information, and each point can also be regarded as a code with 2 bits of binary information. The codes corresponding to the four constellation points are 00, 01, 10 and 11 respectively. The coordinates of these four constellation points can be [1,1], [-1,1], [1,-1] and [-1,-1] respectively. Taking into account energy normalization, the horizontal and vertical coordinates of each constellation point can also be divided by

[0028] During modulation, the binary information to be transmitted is encoded and then mapped to constellation points in a constellation diagram, achieving digital modulation (i.e., constellation modulation). During demodulation, the distance between the received signal and each constellation point in the constellation diagram is used to determine which signal the transmitter sent, thereby correctly demodulating the data. For example, when demodulating a received QPSK signal, the distance between the QPSK signal and the four constellation points in the constellation diagram is used. If the distance to point 00 is closest, the received QPSK signal can be determined to be 00. Alternatively, if the four constellation points in the QPSK constellation diagram are symmetrically distributed across the four quadrants, the receiver can complete signal demodulation by determining which quadrant the received signal is in.

[0029] A neural network has an input layer and an output layer, with at least one hidden layer between them. The hidden layer may be followed by a nonlinear activation function, such as a rectified linear unit (ReLU) or a tanh function. The input layer, output layer, and hidden layer are collectively referred to as network layers, or simply layers. Layers are connected by nodes on each layer. A pair of connected nodes has a weight and a bias. A neural network can be viewed as a nonlinear transformation from input to output. The output can also be calculated using a loss function, which updates the weights and biases of each layer to minimize the loss.

[0030] FIG1 is a flow chart of a constellation diagram determination method provided by an embodiment of the present application. As shown in FIG1 , the method may include:

[0031] S101. Acquire channel characteristic information and M modulated bit sequences.

[0032] S102: Determine a constellation diagram corresponding to M modulated bit sequences through a pre-trained neural network.

[0033] The M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0034] Channel characteristic information describes the channel environment in which a communication node operates and can be the result of statistics collected over a period of time. This information can include channel environment type indicators, such as city, rural, office, street, and home, which can reflect different channel characteristics. This information can also include channel model indicators. Multiple channel models can be stored in a communication system, each corresponding to a different scenario. Providing a channel model indicator can provide information about the channel characteristics of the current scenario. This channel characteristic information can be a statistically derived channel model, a channel model determined using methods such as ray tracing, or a channel model implemented using a neural network.

[0035] The M modulation bit sequences are binary information to be modulated. After obtaining the channel characteristic information and the M modulation bit sequences, the channel characteristic information and the M modulation bit sequences are input into a pre-trained neural network. The trained neural network can then obtain a constellation diagram corresponding to the M modulation bit sequences. For example, assuming M is 4, and the four modulation bit sequences are 00, 01, 10, and 11, respectively, these four modulation bit sequences and the current channel characteristic information are input into the pre-trained neural network to obtain the coordinates of four constellation points, which form the corresponding constellation diagram.

[0036] The constellation diagram determination method provided in an embodiment of the present application obtains channel characteristic information and M modulated bit sequences; and determines the constellation diagram corresponding to the M modulated bit sequences using a pre-trained neural network; where M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information. In this method, the constellation diagram used by the communicating parties during the communication process can be adaptively changed based on the different channel characteristics, so that the determined constellation diagram has a high degree of match with the current channel characteristics, thereby improving the performance of the constellation diagram and thus improving communication performance.

[0037] In one embodiment, the neural network may include a modulation subnetwork and a demodulation subnetwork. In this embodiment, both modulation and demodulation are implemented using neural networks. By performing end-to-end learning on the modulation subnetwork and the demodulation subnetwork, a constellation diagram for any channel can be designed. As shown in FIG2 , the neural network may be determined by:

[0038] S201: Construct a modulation sub-network according to a modulation mode.

[0039] Constructing a modulation subnetwork based on a modulation scheme refers to constructing a modulation subnetwork based on a calculation formula for a modulation scheme. This involves converting the operations in a modulation scheme calculation formula into a combination of network layers. For example, a linear operation y = a1x2 + a2x2 + b1 can be converted into a fully connected layer, where the relationship between the output y of the fully connected layer and the input [x1; x2] is y = W[x1; x2] + b.

[0040] One modulation mode may be a Quadrature Amplitude Modulation (QAM) mode, an Amplitude Phase Shift Keying (APSK) modulation mode, or other modulation modes. This embodiment does not limit the modulation mode.

[0041] S202: Construct a demodulation sub-network according to a demodulation method.

[0042] Constructing a demodulation subnetwork based on a demodulation method refers to constructing a demodulation subnetwork based on a calculation formula for a demodulation method. This involves converting the operations in the calculation formula for a demodulation method into a combination of network layers. For example, a linear operation y = a1x2 + a2x2 + b1 can be converted into a fully connected layer, where the relationship between the output y of the fully connected layer and the input [x1; x2] is y = W[x1; x2] + b.

[0043] A demodulation method may be a boundary method demodulation method, a formula method demodulation method, or other demodulation methods. This embodiment does not limit the demodulation method.

[0044] S203: Initialize network parameters of the modulation sub-network according to a modulation mode.

[0045] Initializing the network parameters of the modulation subnetwork according to a modulation scheme refers to determining the first initial parameters of the modulation subnetwork based on a calculation formula for the modulation scheme, and initializing the parameters of the modulation subnetwork's network layer based on the first initial parameters. This involves converting the operations in the calculation formula into combinations of network layers, and then determining the network layer parameters based on the operations. For example, after converting a linear operation into a fully connected layer, the coefficients of the linear operation can be used to determine the weight coefficient W = [a1, a2] and the bias coefficient b = b1 of the fully connected layer.

[0046] Initializing the modulation subnetwork may include unscrambled initialization and scrambled initialization. When unscrambled initialization is used, the modulation subnetwork is initialized directly based on the first initial parameters. When scrambled initialization is used, the process of initializing the modulation subnetwork based on the first initial parameters may include scrambling the first initial parameters and initializing the modulation subnetwork based on the scrambled first initial parameters.

[0047] The scrambling process can be to add random noise to the first initial parameter, so that the initial state of the modulation sub-network deviates from the one modulation mode, so that more network parameters in the modulation sub-network can be changed, thereby increasing the expression capability of the modulation sub-network. For example, when the network parameters [c1, c2, c3, ..., c n ] so that the modulation sub-network realizes the modulation mode, and the scrambling initialization is [c1+w1,c2+w2,c3+w3,...,c n +w n ] initialize the modulation subnetwork, where [w1,w2,w3,...,w n ] is random noise.

[0048] S204: Initialize network parameters of the demodulation subnetwork according to a demodulation mode.

[0049] Initializing the network parameters of the demodulation subnetwork according to a demodulation method refers to determining the second initial parameters of the demodulation subnetwork based on a calculation formula for the demodulation method, and initializing the parameters of the demodulation subnetwork's network layer based on the second initial parameters. This involves converting the operations in the calculation formula into combinations of network layers, and then determining the network layer parameters based on the operations. For example, after converting a linear operation into a fully connected layer, the coefficients of the linear operation can be used to determine the weight coefficient W = [a1, a2] and the bias coefficient b = b1 of the fully connected layer.

[0050] Initializing the demodulation subnetwork may also include unscrambled initialization and scrambled initialization. When unscrambled initialization is used, the demodulation subnetwork is initialized directly based on the second initial parameters. When scrambled initialization is used, the process of initializing the demodulation subnetwork based on the second initial parameters may include scrambling the second initial parameters and initializing the demodulation subnetwork based on the scrambled second initial parameters.

[0051] The scrambling process can be to add random noise to the second initial parameter, so that the initial state of the demodulation sub-network deviates from the one demodulation mode, so that more network parameters in the demodulation sub-network can be changed, thereby increasing the expression capability of the demodulation sub-network. For example, when the network parameter [c 11 ,c 22 ,c 33 ,...,c nn ] enables the demodulation sub-network to implement said one demodulation mode, and the scrambling initialization is [c 11 +w 11 ,c 22 +w 22 ,c 33 +w 33 ,...,c nn +w nn] initialize the demodulation subnetwork, where [w 11 ,w 22 ,w 33 ,...,w nn ] is random noise.

[0052] S205 , training the initialized modulation sub-network and demodulation sub-network according to the training data set to obtain a trained neural network.

[0053] The training data set includes a plurality of sample channel characteristic information and demodulated bit probabilities corresponding to M modulated bit sequences under each sample channel characteristic information.

[0054] Multiple sample channel characteristic information and M modulated bit sequences are used as the input of the neural network (i.e., the modulation subnetwork and the demodulation subnetwork), and the demodulated bit probability corresponding to the M modulated bit sequences under each sample channel characteristic information is used as the expected output of the neural network. The neural network is end-to-end trained using a preset loss function until the preset convergence condition is reached, thereby obtaining a trained neural network.

[0055] In one embodiment, a modulation subnetwork constructed according to a modulation method may include a bit selection unit, a coordinate calculation unit, a symbol calculation unit and a normalization unit, and a demodulation subnetwork constructed according to a demodulation method may include a log-likelihood ratio (LLR) calculation unit and a bit probability calculation unit.

[0056] A bit selection unit is configured to select a first bit and a second bit from a modulated bit sequence. The bit selection unit is composed of a fully connected layer, and the input of the bit selection unit is the modulated bit sequence. The output of the bit selection unit can be the first bit and the second bit in the modulated bit sequence. The relationship between the output y and the input x of the fully connected layer can be expressed as y = Wx + b, where W is the weight coefficient of the fully connected layer and b is the bias coefficient of the fully connected layer. W and b are collectively referred to as the parameters of the fully connected layer.

[0057] The coordinate calculation unit is used to determine the absolute value of the coordinates of the constellation point corresponding to the modulated bit sequence based on the first bit. The coordinate calculation unit is composed of several fully connected layers and several nonlinear layers. The relationship between the output y and the input x of the nonlinear layer can be expressed as y = f(x), where f represents a nonlinear function. Common nonlinear functions include the ReLU function and the sigmoid function.

[0058] A symbol calculation unit is configured to determine the quadrant in which the constellation point corresponding to the modulated bit sequence is located using the second bit. The symbol calculation unit is composed of several fully connected layers and several nonlinear layers and is configured to determine the sign of the coordinates of the constellation point corresponding to the modulated bit sequence using the second bit, thereby determining the quadrant in which the constellation point is located based on the sign of the coordinates. When the signs of the coordinates are both positive, the corresponding constellation point is located in the first quadrant; when the signs of the coordinates are both negative and positive, the corresponding constellation point is located in the second quadrant; when the signs of the coordinates are both negative, the corresponding constellation point is located in the third quadrant; and when the signs of the coordinates are both positive and negative, the corresponding constellation point is located in the fourth quadrant.

[0059] The number of the bit selection units is N, and N is equal to log2(M) / 2; the number of the coordinate calculation units is Q, and Q is equal to log2(M) / 2-1.

[0060] A normalization unit is configured to normalize the coordinates of the constellation points corresponding to the modulated bit sequence. The normalization unit may comprise a batch normalization layer that normalizes the coordinates of the constellation points corresponding to the modulated bit sequence to obtain transmit coordinates. The relationship between the output y of the batch normalization layer and the input x can be expressed as y = (x - μ) / σ, where μ is the mean value of x and σ is the standard deviation of x.

[0061] The transmission coordinates pass through a channel to generate the reception coordinates. The channel may be an additive white Gaussian noise channel, a fading noise channel, or other channels.

[0062] The received coordinates (i.e., received signals) are received by the demodulation subnetwork, which demodulates the received signals. The LLR calculation unit in the demodulation subnetwork is used to determine the LLRs of the received signals. The LLR calculation unit consists of several fully connected layers and several nonlinear layers.

[0063] The bit probability calculation unit is used to determine the demodulation bit probability based on the LLR of the received signal. The bit probability calculation unit consists of a nonlinear layer, which is a sigmoid function layer. The relationship between its output y and input x can be expressed as y = [exp(x) - exp(-x)] / [exp(x) + exp(-x)].

[0064] The number of the LLR calculation units is P, and P is equal to log2(M) / 2.

[0065] When initializing the modulation sub-network, the method may further include: controlling the training attributes of the bit selection unit and the symbol calculation unit to be non-trainable; or controlling the training attributes of the bit selection unit and the symbol calculation unit to be trainable.

[0066] The above-mentioned bit selection unit and symbol calculation unit have training properties, including trainable or non-trainable. Trainable means that the parameters of the fully connected layer in the bit selection unit and the symbol calculation unit change with the training of the network, and non-trainable means that the parameters of the fully connected layer in the bit selection unit and the symbol calculation unit do not change with the training of the network.

[0067] In one embodiment, the modulation subnetwork may further include a coordinate transformation unit, which is composed of a coordinate transformation layer and is configured to transform the coordinates of the constellation points corresponding to the modulated bit sequence from a first coordinate system to a second coordinate system. The input of the coordinate transformation layer is the first coordinate (x1, x2) in the first coordinate system, and the output is the second coordinate (y1, y2) of the first coordinate in the second coordinate system. For example, the relationship between (x1, x2) and (y1, y2) can be expressed as: y1 = x2cos(kx1), y2 = x2sin(kx1), where k is a constant coefficient. The above equation implements the transformation from a polar coordinate system to a rectangular coordinate system, i.e., it can transform a square constellation diagram into a circular constellation diagram.

[0068] The above-mentioned modulation subnetwork may further include a first identity unit, which is composed of several fully connected layers and several nonlinear layers. The output of the first identity unit is equal to the input, and is used to increase the network parameter amount of the modulation subnetwork, thereby increasing the expression ability of the modulation subnetwork.

[0069] In one embodiment, the demodulation subnetwork may further include an inverse coordinate transformation unit for converting the coordinates of the received signal from the second coordinate system to the first coordinate system. The inverse coordinate transformation unit is composed of an inverse coordinate transformation layer, the input of which is the second coordinate (y1, y2) in the second coordinate system, and the output is the first coordinate (x1, x2) of the second coordinate (y1, y2) in the first coordinate system. For example, the relationship between (y1, y2) and (x1, x2) is expressed as: x1 = arctan (y2 / y1) / k, x2 = sqrt (y1^2 + y2^2), where k is a constant coefficient. The above formula realizes the transformation from the rectangular coordinate system to the polar coordinate system, that is, the circular constellation diagram can be transformed into a square constellation diagram. The coordinate transformation unit in the modulation subnetwork and the inverse coordinate transformation unit in the demodulation subnetwork appear in pairs.

[0070] The above-mentioned demodulation subnetwork may further include a second identity unit, which is composed of several fully connected layers and several nonlinear layers. The output of the second identity unit is equal to the input, and is used to increase the network parameter amount of the demodulation subnetwork, thereby increasing the expressive power of the demodulation subnetwork.

[0071] In this embodiment, by initializing the neural network to a specific modulation mode and demodulation mode, compared with the random initialization method, the accuracy of the initial network parameters of the neural network is improved, and the number of network parameters of the neural network is reduced, thereby ensuring the stability of the training results and further improving the performance of the constellation diagram.

[0072] Next, taking M equal to 64 as an example, a neural network initialized based on a modulation method and a demodulation method is introduced.

[0073] Case 1: A possible structure of the above-mentioned neural network is shown in FIG3 . The modulation subnetwork in the neural network may include a bit selection unit, a coordinate calculation unit, a symbol calculation unit, and a normalization unit. The demodulation subnetwork may include an LLR calculation unit and a bit probability calculation unit. The number of bit selection units is 3, the number of coordinate calculation units is 2, and the number of LLR calculation units is 3.

[0074] Case 2: A possible structure of the aforementioned neural network is shown in Figure 4. The modulation subnetwork in this neural network can include a bit selection unit, a coordinate calculation unit, a coordinate transformation unit, a symbol calculation unit, and a normalization unit. The demodulation subnetwork can include an LLR calculation unit, an inverse coordinate transformation unit, and a bit probability calculation unit. In other words, the addition of a coordinate transformation unit to the modulation subnetwork and an inverse coordinate transformation unit to the demodulation subnetwork implements constellation-type transformations.

[0075] Case 3: A possible structure of the aforementioned neural network is shown in Figure 5. The modulation subnetwork in this neural network may include a bit selection unit, a coordinate calculation unit, a first identity unit, a coordinate transformation unit, a symbol calculation unit, and a normalization unit. The demodulation subnetwork may include an LLR calculation unit, a second identity unit, an inverse coordinate transformation unit, and a bit probability calculation unit. In other words, the addition of the coordinate transformation unit and the first identity unit to the modulation subnetwork, and the addition of the inverse coordinate transformation unit and the second identity unit to the demodulation subnetwork, achieve constellation-type transformation and increase the expressive power of the modulation and demodulation subnetworks.

[0076] The fourth situation: the neural network as shown in FIG3 , FIG4 or FIG5 can be initialized without scrambling, so that the neural network starts training from the one modulation mode and one demodulation mode.

[0077] The fifth case: the neural network as shown in Figure 3, Figure 4 or Figure 5 can be scrambled and initialized so that the initial state of the neural network deviates from the one modulation method and one demodulation method, so that more network parameters in the neural network can be changed, increasing the expression ability of the neural network, thereby improving the performance of the constellation diagram.

[0078] The sixth case: the training attributes of the bit selection unit and the symbol calculation unit in the neural network as shown in Figure 3, Figure 4 or Figure 5 can be controlled to be non-trainable, so that the constellation diagram obtained by training has a certain symmetry, thereby reducing the optimization space of the network parameters.

[0079] The seventh case: the training properties of the bit selection unit and the symbol calculation unit in the neural network as shown in Figure 3, Figure 4 or Figure 5 can be controlled to be trainable, thereby increasing the degree of freedom of the network parameters in the neural network.

[0080] Continuing with the example of M being equal to 64, the determination process of the above-mentioned neural network is introduced. The neural network is designed according to the structure shown in Figure 5 (that is, the modulation subnetwork includes a bit selection unit, a coordinate calculation unit, a first identity unit, a coordinate transformation unit, a symbol calculation unit and a normalization unit, and the demodulation subnetwork includes an LLR calculation unit, a second identity unit, an inverse coordinate transformation unit and a bit probability calculation unit), and the neural network is initialized according to the above-mentioned fourth and seventh cases. At the same time, the modulation subnetwork is constructed and initialized according to the 64QAM modulation method, and the demodulation subnetwork is constructed and initialized according to the 64QAM boundary method demodulation method.

[0081] For 64QAM modulation, when the input modulation bit sequence is b1b2b3b4b5b6, according to the mapping relationship between the 64QAM modulation bit sequence and the constellation points (the following formula (1)), it can be obtained:

[0082] Where [x, y] is the coordinate of the constellation point. The above formula shows that the two bits b1b2 determine the sign of the constellation point coordinate, and the four bits b3b4b5b6 determine the absolute value of the constellation point coordinate. Multiplying the two yields the target coordinate of the constellation point. First, three coordinate selection units sequentially select b1b2, b3b4, and b5b6 from the input modulated bit sequence. Each of these units consists of a fully connected layer. The weight coefficients of the fully connected layers are: [1, 0, 0, 0, 0, 0; 0, 1, 0, 0, 0], [0, 0, 1, 0, 0, 0; 0, 0, 1, 0, 0], [0, 0, 0, 0, 1, 0; 0, 0, 0, 1], and the bias coefficients are: [0; 0], [0; 0], [0; 0].

[0083] The calculation of the absolute value of the coordinate can be divided into two steps. The first step is to calculate |b3+b5-1| and |b4+b6-1| through the four bits b3b4b5b6. The second step is to calculate the absolute value of the coordinate through b3b4, |b3+b5-1| and |b4+b6-1|. The above two steps are implemented by two coordinate calculation units respectively. For the first coordinate calculation unit, the calculation of the absolute value is decomposed into the form of the sum of the relu function (|x|=relu(x)+relu(-x)), so |b3+b5-1|=relu(b3+b5-1)+relu(-(b3+b5-1)). First, b3+b5-1, -(b3+b5-1), b4+b6-1, -(b4+b6-1) are calculated through a fully connected layer. The weight coefficients and bias coefficients of the fully connected layer are: [1, 0, 1, 0; -1 ,0,-1,0;0,1,0,1;0,-1,0,-1],[-1;1;-1;1];Then through the relu function layer to obtain relu(b3+b5-1), relu(-(b3+b5-1)), relu(b4+b6-1), relu(-(b4+b6-1));Finally, through a fully connected layer to calculate the absolute value of the coordinate, the weight coefficient and bias coefficient of the fully connected layer are: [1,1,0,0;0,0,1,1],[0;0]. For the second coordinate calculation unit, it can be directly implemented through a fully connected layer, the weight coefficient and bias coefficient of the fully connected layer are: [4,0,2,0;0,4,0,2],[1;1].

[0084] For the first identity unit, x = relu(x) - relu(-x), the first identity unit can also be implemented using a relu function layer and a fully connected layer. When x is a 4×1 column vector, x and -x are first obtained through a fully connected layer. The weight coefficients and bias coefficients of the fully connected layer are: [1, 0, 0, 0; -1, 0, 0, 0; 0, 1, 0, 0; 0, -1, 0, 0; 0, 0, 1, 0; 0, 0, -1, 0; 0, 0, 0, 1; 0, 0, 0, -1], [0; 0; 0; 0; 0; 0; 0]; then relu(x) and relu(-x) are obtained through the relu function layer; finally, the sum of relu(x) and relu(-x) is realized through a fully connected layer. The weight coefficients and bias coefficients of the fully connected layer are: [1,-1,0,0,0,0,0,0; 0,0,1,-1,0,0,0,0; 0,0,0,0,1,-1,0,0; 0,0,0,0,0,0,1,-1], [0; 0; 0; 0].

[0085] Calculation shows that the coordinates input to the coordinate transformation unit differ from the constellation points of 64QAM modulation by only a multiple of 1 / sqrt(21). For example, when b3b4b5b6 is 0000, the coordinates input to the coordinate transformation unit are (3,3), while the constellation points of 64QAM modulation are (3 / sqrt(21), 3 / sqrt(21)). The coordinates are all positive, so they are all distributed in the first quadrant. The relationship between the input and output of the coordinate transformation layer of the coordinate transformation unit is expressed as: y1 = x2cos(kx1), y2 = x2sin(kx1), where k = π / 16. The coordinate transformation layer transforms the coordinates of the square distributed in the first quadrant into the coordinates of the 1 / 4 circle distributed in the first quadrant. sqrt represents the square root operation.

[0086] The symbol calculation unit is implemented using a fully connected layer with weight coefficients and position coefficients of [-2, 0; 0, -2] and [1; 1], respectively. The symbol calculation unit maps the coordinates of the quarter circle distributed in the first quadrant to the other three quadrants based on the coordinate signs determined by b1b2. This transforms the square constellation distribution of 64QAM modulation into a circular distribution. Since the square constellation distribution is uniform, the circular constellation distribution is also uniform in both angular and radial directions.

[0087] For 64QAM boundary method demodulation, when the received signal is (x, y), the boundary method demodulation scheme (such as the following formula (2)) shows that:

[0088] The LLR calculation of b1b2 can be implemented using a fully connected layer, whose weight coefficient and bias coefficient are [-1, 0; 0, -1] and [0; 0] respectively.

[0089] The relationship between the input and output of the coordinate inverse transformation layer of the coordinate inverse transformation unit is expressed as: x1=arctan(y2 / y1) / k, x2=sqrt(y1^2+y2^2), where k=π / 16.

[0090] For the LLR calculation of b3b4, it can be implemented using a fully connected layer and a relu function layer. Similar to the processing of the absolute value of the coordinates in the modulation part, first, LLR(b1), -LLR(b1), LLR(b2), -LLR(b2) are calculated through the fully connected layer. The weight coefficient and bias coefficient of the fully connected layer are: [1, 0; -1, 0; 0, 1; 0, -1], [0; 0; 0; 0] respectively; then, relu(LLR(b1)), relu(-LLR(b1)), relu(LLR(b2)), relu(-LLR(b2)) are obtained through the relu function layer; finally, it can be obtained through another fully connected layer. The weight coefficient and bias coefficient of the fully connected layer are: [1, -1, 0, 0; 0, 0, 1, -1], [-4 / sqrt(42); -4 / sqrt(42)].

[0091] For the LLR calculation of b5b6, it can be implemented using a fully connected layer and a relu function layer. Similar to the processing of the absolute value of the coordinates in the modulation part, first, LLR(b3), -LLR(b3), LLR(b4), -LLR(b4) are calculated through the fully connected layer, and the weight coefficients and bias coefficients of the fully connected layer are: [1, 0; -1, 0; 0, 1; 0, -1], [0; 0; 0; 0]; then, relu(LLR(b3)), relu(-LLR(b3)), relu(LLR(b4)), relu(-LLR(b4)) are obtained through the relu function layer; finally, it can be obtained through another fully connected layer, and the weight coefficients and bias coefficients of the fully connected layer are: [1, -1, 0, 0; 0, 0, 1, -1], [-2 / sqrt(42); -2 / sqrt(42)].

[0092] For the second identity unit, x = relu(x) - relu(-x) can be obtained. The second identity unit can also be implemented using a relu function layer and a fully connected layer. When x is a 2×1 column vector, x and -x are first obtained through a fully connected layer. The weight coefficient and bias coefficient of the fully connected layer are: [1, 0; -1, 0; 0, 1; 0, -1], [0; 0; 0; 0]; then relu(x) and relu(-x) are obtained through the relu function layer; finally, the sum of relu(x) and relu(-x) is implemented through a fully connected layer. The weight coefficient and bias coefficient of the fully connected layer are: [1, -1, 0, 0; 0, 0, 1, -1], [0; 0].

[0093] The structures of the modulation subnetwork and demodulation subnetwork designed according to the above method are shown in Figure 6. The numbers in the brackets represent the lengths of the input and output vectors, respectively. The coefficients of all fully connected layers are trainable.

[0094] In traditional technology, the neural network is designed as a fully connected neural network, which is implemented through multiple layers of fully connected layers and relu function layers. The fully connected neural network is randomly initialized. The structures of the designed modulation subnetwork and demodulation subnetwork are shown in Figure 7. The numbers in the brackets represent the lengths of the input and output vectors, respectively.

[0095] The parameter amounts of the two methods and the mutual information of the constellation diagrams trained multiple times are compared. In terms of the network parameter amount, the parameter amount of the fully connected neural network in the traditional technology is 4808, while the parameter amount of the neural network constructed and initialized based on the specific modulation method and demodulation method in the embodiment of the present application is only 356. The mutual information empirical cumulative distribution function of the constellation diagram obtained after 300 training of the neural network in the traditional technology and the neural network in the embodiment of the present application is shown in Figure 8. It can be seen from Figure 8 that compared with the mutual information distribution of the fully connected network in the traditional technology, the mutual information distribution range of the neural network provided by the embodiment of the present application is narrower, all of which are within the range of 4.78-4.8, which means that the constellation diagram obtained by training the neural network provided by the embodiment of the present application has better stability.

[0096] FIG9 is a schematic diagram of a structure of a constellation diagram determination device provided in an embodiment of the present application. As shown in FIG9 , the device may include: an acquisition module 901 and a determination module 902 .

[0097] The acquisition module 901 is used to obtain channel characteristic information and M modulated bit sequences; the determination module 902 is used to determine the constellation diagram corresponding to the M modulated bit sequences through a pre-trained neural network; wherein, M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0098] The constellation diagram determination device provided in an embodiment of the present application obtains channel characteristic information and M modulated bit sequences; and determines the constellation diagram corresponding to the M modulated bit sequences using a pre-trained neural network, where M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information. In this method, the constellation diagram used by the communicating parties during the communication process can be adaptively changed based on the different channel characteristics, so that the determined constellation diagram has a high degree of match with the current channel characteristics, thereby improving the performance of the constellation diagram and thus improving communication performance.

[0099] Based on the above embodiment, the neural network includes a modulation subnetwork and a demodulation subnetwork; the device may also include: a construction module, an initialization module and a training module; the construction module is used to construct the modulation subnetwork according to a modulation method; the construction module is also used to construct the demodulation subnetwork according to a demodulation method; the initialization module is used to initialize the network parameters of the modulation subnetwork according to the modulation method; the initialization module is also used to initialize the network parameters of the demodulation subnetwork according to the demodulation method; the training module is used to train the initialized modulation subnetwork and demodulation subnetwork according to a training data set to obtain a trained neural network; wherein, the training data set includes multiple sample channel feature information, and the demodulation bit probabilities corresponding to M modulation bit sequences under each sample channel feature information.

[0100] Based on the above embodiment, the modulation sub-network includes:

[0101] a bit selection unit for selecting a first bit and a second bit from the modulated bit sequence; a coordinate calculation unit for determining, using the first bit, the absolute value of the coordinate of the constellation point corresponding to the modulated bit sequence; a symbol calculation unit for determining, using the second bit, the quadrant in which the constellation point corresponding to the modulated bit sequence is located; and a normalization unit for normalizing the coordinates of the constellation point corresponding to the modulated bit sequence.

[0102] Based on the above embodiment, the number of the bit selection units is N, and N is equal to log2(M) / 2; the number of the coordinate calculation units is Q, and Q is equal to log2(M) / 2-1.

[0103] Based on the above embodiment, the modulation sub-network further includes:

[0104] A coordinate transformation unit is used to transform the coordinates of the constellation points corresponding to the modulated bit sequence from the first coordinate system to the second coordinate system.

[0105] Based on the above embodiment, the modulation sub-network further includes:

[0106] The first identity unit is configured to increase the network parameter quantity of the modulation sub-network.

[0107] Based on the above embodiment, the demodulation subnetwork includes:

[0108] The log-likelihood ratio (LLR) calculation unit is configured to determine the LLR of the received signal; and the bit probability calculation unit is configured to determine the demodulation bit probability according to the LLR of the received signal.

[0109] Based on the above embodiment, the number of the LLR calculation units is P, and P is equal to log2(M) / 2.

[0110] Based on the above embodiment, the demodulation subnetwork further includes:

[0111] The coordinate inverse transformation unit is used to transform the coordinates of the received signal from the second coordinate system to the first coordinate system.

[0112] Based on the above embodiment, the demodulation subnetwork further includes:

[0113] The second identity unit is configured to increase the network parameter quantity of the demodulation subnetwork.

[0114] Based on the above embodiment, the initialization module is used to determine the first initial parameters of the modulation subnetwork according to the calculation formula of the modulation mode; and initialize the modulation subnetwork according to the first initial parameters.

[0115] Based on the above embodiment, the initialization module is further configured to scramble the first initial parameters and initialize the modulation sub-network according to the scrambled first initial parameters.

[0116] Based on the above embodiment, the initialization module is further configured to control the training attributes of the bit selection unit and the symbol calculation unit to be non-trainable; or, control the training attributes of the bit selection unit and the symbol calculation unit to be trainable.

[0117] Based on the above embodiment, the initialization module is further configured to determine a second initial parameter of the demodulation subnetwork according to a calculation formula of the demodulation mode; and initialize the demodulation subnetwork according to the second initial parameter.

[0118] Based on the above embodiment, the initialization module is further configured to scramble the second initial parameters and initialize the demodulation subnetwork according to the scrambled second initial parameters.

[0119] In one embodiment, an electronic device is provided, the internal structure of which may be shown in FIG10 . The electronic device includes a processor, a memory, a network interface, and a database connected via a system bus. The processor of the electronic device is configured to provide computing and control capabilities. The memory of the electronic device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the electronic device is configured to store data generated during the constellation diagram determination process. The network interface of the electronic device is configured to communicate with an external terminal via a network connection. When the computer program is executed by the processor, a constellation diagram determination method is implemented.

[0120] Those skilled in the art will understand that the structure shown in Figure 10 is merely a block diagram of a partial structure related to the scheme of the present application, and does not constitute a limitation on the electronic device to which the scheme of the present application is applied. The electronic device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0121] In one embodiment, an electronic device is provided. The electronic device includes a memory and multiple processors. The memory stores a computer program. When one of the multiple processors executes the computer program, the following is achieved:

[0122] Acquire channel characteristic information and M modulation bit sequences; determine a constellation diagram corresponding to the M modulation bit sequences through a pre-trained neural network; wherein M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0123] The present application also provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the computer program implements the following:

[0124] Acquire channel characteristic information and M modulation bit sequences; determine a constellation diagram corresponding to the M modulation bit sequences through a pre-trained neural network; wherein M is the number of points in the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

[0125] The computer storage medium of the embodiment of the present application may adopt any combination of one or more computer-readable media. The computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium may be, for example, but not limited to: an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or component, or any combination thereof. Computer-readable storage media include (a non-exhaustive list): an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM), a flash memory, an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by an instruction execution system, device or device or used in combination with it.

[0126] A computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, the data signal carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transfer a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0127] The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wire, optical cable, radio frequency (RF), etc., or any suitable combination of the foregoing.

[0128] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages ​​or a combination of multiple programming languages, including object-oriented programming languages ​​(such as Java, Smalltalk, C++, Ruby, Go), and conventional procedural programming languages ​​(such as "C" or similar programming languages). The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0129] It will be appreciated by those skilled in the art that the term user terminal covers any suitable type of wireless user equipment, such as a mobile phone, a portable data processing device, a portable web browser or a vehicle-mounted mobile station.

[0130] In general, various embodiments of the present application may be implemented in hardware or dedicated circuits, software, logic, or any combination thereof. For example, some aspects may be implemented in hardware, while other aspects may be implemented in firmware or software that can be executed by a controller, microprocessor, or other computing device, although the present application is not limited thereto.

[0131] Embodiments of the present application may be implemented by executing computer program instructions by a data processor of a mobile device, for example, in a processor entity, or by hardware, or by a combination of software and hardware. The computer program instructions may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages.

[0132] Any block diagram of a logical flow in the drawings of this application may represent program operations, or may represent interconnected logical circuits, modules and functions, or may represent a combination of program operations and logical circuits, modules and functions. A computer program may be stored on a memory. The memory may be of any type suitable for the local technical environment and may be implemented using any suitable data storage technology, such as, but not limited to, read-only memory (ROM), random access memory (RAM), optical memory devices and systems (digital versatile discs DVD or CD), etc. Computer-readable media may include non-transitory storage media. A data processor may be of any type suitable for the local technical environment, such as, but not limited to, a general-purpose computer, a special-purpose computer, a microprocessor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a programmable logic device (FPGA), and a processor based on a multi-core processor architecture.

Claims

1. A method for determining a constellation diagram, comprising: Acquire channel characteristic information and M modulation bit sequences; Determine the constellation diagram corresponding to the M modulated bit sequences by a pre-trained neural network; Wherein, M is the number of points of the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

2. The method according to claim 1, wherein: The neural network includes a modulation subnetwork and a demodulation subnetwork; The neural network is determined in the following manner: Constructing the modulation subnetwork according to a modulation mode; Constructing the demodulation subnetwork according to a demodulation mode; Initialize the network parameters of the modulation subnetwork according to the modulation mode; Initialize the network parameters of the demodulation subnetwork according to the one demodulation mode; The initialized modulation sub-network and demodulation sub-network are trained according to the training data set to obtain a trained neural network; The training data set includes a plurality of sample channel characteristic information and demodulation bit probabilities corresponding to M modulated bit sequences under the plurality of sample channel characteristic information.

3. The method according to claim 2, wherein: The modulation subnetwork comprises: A bit selection unit, configured to select a first bit and a second bit from the modulated bit sequence; a coordinate calculation unit, configured to determine, by using the first bit, an absolute value of a coordinate of a constellation point corresponding to the modulated bit sequence; a symbol calculation unit, configured to determine, by using the second bit, the quadrant in which the constellation point corresponding to the modulated bit sequence is located; The normalization unit is used to normalize the coordinates of the constellation points corresponding to the modulated bit sequence.

4. The method according to claim 3, wherein: The number of the bit selection units is N, and N is equal to log2(M) / 2; The number of the coordinate calculation units is Q, and Q is equal to log2(M) / 2-1.

5. The method according to claim 3, wherein: The modulation subnetwork further includes: A coordinate transformation unit is used to transform the coordinates of the constellation point corresponding to the modulated bit sequence from the first coordinate system to the second coordinate system.

6. The method according to claim 3, wherein: The modulation subnetwork further includes: The first identity unit is used to increase the network parameter amount of the modulation sub-network.

7. The method according to claim 2, wherein: The demodulation subnetwork comprises: A log-likelihood ratio (LLR) calculation unit, used to determine the LLR of the received signal; The bit probability calculation unit is used to determine the demodulation bit probability according to the LLR of the received signal.

8. The method according to claim 7, wherein: The number of the LLR calculation units is P, and P is equal to log2(M) / 2.

9. The method according to claim 7, wherein: The demodulation subnetwork further includes: The coordinate inverse transformation unit is used to transform the coordinates of the received signal from the second coordinate system to the first coordinate system.

10. The method according to claim 7, wherein: The demodulation subnetwork further includes: The second identity unit is used to increase the network parameter amount of the demodulation subnetwork.

11. The method according to claim 3, wherein: Initializing the network parameters of the modulation subnetwork according to the modulation mode includes: Determining a first initial parameter of the modulation subnetwork according to a calculation formula of the modulation mode; The modulation subnetwork is initialized according to the first initial parameters.

12. The method according to claim 11, wherein: Initializing the modulation subnetwork according to the first initial parameter includes: The first initial parameters are scrambled, and the modulation subnetwork is initialized according to the scrambled first initial parameters.

13. The method according to claim 11, wherein: Also includes: Controlling the training attributes of the bit selection unit and the symbol calculation unit to be non-trainable; or, The training attributes of the bit selection unit and the symbol calculation unit are controlled to be trainable.

14. The method according to claim 2, wherein: The initializing the network parameters of the demodulation subnetwork according to the one demodulation mode includes: Determining a second initial parameter of the demodulation subnetwork according to a calculation formula of the demodulation mode; The demodulation subnetwork is initialized according to the second initial parameters.

15. The method according to claim 14, wherein: Initializing the demodulation subnetwork according to the second initial parameter includes: The second initial parameter is scrambled, and the demodulation subnetwork is initialized according to the scrambled second initial parameter.

16. A constellation diagram determination device, wherein: include: An acquisition module, configured to acquire channel characteristic information and M modulation bit sequences; A determination module, configured to determine the constellation diagram corresponding to the M modulated bit sequences through a pre-trained neural network; Wherein, M is the number of points of the modulation constellation diagram, and the constellation diagram corresponds to the channel characteristic information.

17. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the constellation diagram determination method according to any one of claims 1 to 15 when executing the computer program.

18. A storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the constellation diagram determination method according to any one of claims 1 to 15.

Citation Information

Patent Citations

  • Physical layer security communication method and system based on deep learning

    CN109246048A

  • Data transmission method

    CN114765570A

  • Design of adaptive power line dual-mode communication system based on neural network

    CN115664972A

  • Methods, procedures, appartuses and systems for data-driven wireless transmit / receive unit specific symbol modulation

    WO2022271634A1