Receiving device, receiving method, receiving program, and transmitting device

Abstract

To solve the problem in which the transmission capacity of conventional receiving devices has been unable to be sufficiently increased.SOLUTION: A receiving device according to one embodiment includes: a signal compensation unit 31 for compensating noise information Θ*[n] of at least one of noise and distortion components included in a received signal Y[n], and outputting a received signal Y*[n]; a neural network processing unit 33 for classifying the received signal Y*[n] into one of classes in the same number as the number of signal points; an output layer processing unit 34 for outputting likelihood information p[n] indicating a probability that the received signal Y[n] belongs to each of M classes from output of the neural network processing unit 33; a data conversion unit 36 for outputting an estimation transmission signal X*[n] by referring to the likelihood information p[n]; and a compensation information prediction unit 35 for predicting the magnitude of at least one of noise and distortion remaining in the received signal Y*[n], and generating the noise information Θ*[n] to be given to the signal compensation unit 31 at a next point in time on the basis of the prediction.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present disclosure relates to a receiving device, a receiving method, a receiving program, a transmitting device, etc., and more particularly to a receiving device, a receiving method, a receiving program, a transmitting device, etc. that perform reception processing on a quadrature amplitude modulated signal. [Background technology]

[0002] Quadrature Amplitude Modulation (QAM), which uses both phase and amplitude information of a transmission signal to identify data, is a well-known modulation / demodulation method for high-capacity and efficient data transmission in digital communications. In recent years, there has been a growing demand for higher-capacity transmission systems, resulting in a demand for an increased number of signal levels in QAM modulation. However, as the number of modulation levels increases, resistance to signal quality degradation due to various noises, interference, and the like decreases. Therefore, simply increasing the number of signal points does not lead to an increase in transmission capacity. Meanwhile, in wireless and wired communication systems, phase noise caused by local oscillators used to convert the frequency of transmission signals is known to be a signal degradation factor that makes it difficult to increase transmission capacity. Therefore, a technology for increasing transmission capacity is disclosed in Patent Document 1.

[0003] The demodulation device described in Patent Document 1 demodulates a signal modulated using a method of mapping information bits to signal points on multiple concentric rings by using the lowest n amplitude bits of the signal. a The lowest n bits and phase bits p The error correction decoder is used to perform error correction decoding on the bits, and symbol replica candidates are reduced based on the a posteriori log-likelihood ratios of the information bits and parity check bits output from the error correction decoder. For the reduced symbol replica candidates, the most significant uncoded bits (m a -n a ) bits and the most significant uncoded bits of the phase bits (m p -n p ) is obtained.

[0004] Furthermore, in order to address the problem that simply increasing the number of signal points does not lead to an increase in transmission capacity, the use of machine learning techniques, such as neural networks, has been attracting attention in recent years. For example, Non-Patent Document 1 discloses a method for achieving a good balance between the increase in the number of signal points that accompanies an increase in capacity and tolerance to degradation of signal quality by introducing neural networks into both the transmitter and receiver. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] International Publication No. 2018 / 116411 [Non-patent literature]

[0006] [Non-Patent Document 1] T. O'Shea and J. Hoydis, “An introduction to deep learning for physical layer,” IEEE Transactions on Cognitive Communications and Networking, vol.3, no.4, pp.563-575, Dec. 2017. Summary of the Invention [Problem to be solved by the invention]

[0007] However, there is a problem that even if Patent Document 1 and Non-Patent Document 1 are applied to communication channels in which the classification method of received signals executed on the receiving device side changes dynamically, they cannot be adequately addressed and transmission capacity cannot be increased.

[0008] In view of the above-mentioned problems, an object of the present disclosure is to provide a receiving device and the like that can increase the transmission capacity for a communication path in which the classification method of received signals changes dynamically. [Means for solving the problem]

[0009] A receiving device according to one embodiment includes: a signal correction unit that divides transmission information into a predetermined bit length, selects one of M predetermined signal points (M is a value of 2 to the power of the bit length), and receives the generated transmission signal as a received signal via a communication channel; applies noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; a neural network processing unit that classifies the corrected received signal into one of the same number of classes as the number of signal points; an output layer processing unit that outputs likelihood information indicating the probability that the received signal belongs to each of the M classes from the output of the neural network processing unit; a data conversion unit that refers to the likelihood information to identify the class with the highest probability, converts the estimated transmitted signal into an estimated transmitted signal corresponding to the identified class and having the bit length of the transmission information, and outputs it; and a correction information prediction unit that detects at least one of noise and distortion remaining in the corrected received signal, predicts the magnitude of a residual correction component including at least one of noise and distortion remaining in the corrected received signal at a next time point using the detected residual information, and generates the noise information to be provided to the signal correction unit at the next time point based on the prediction.

[0010] A receiving method according to one embodiment is a receiving method in a transmitting device that divides transmission information into a predetermined bit length, selects one of M predetermined signal points (M is a value of 2 to the power of the bit length) to generate a transmission signal, receives the generated transmission signal via a communication channel as a received signal, and outputs an estimated transmission signal that is an estimate of the transmission information from the received signal by calculation, the receiving method comprising: a signal correction process that applies noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; a neural network process that classifies the corrected received signal into one of the same number of classes as the number of the signal points; and the received signal to be output from the signal correction unit; a data conversion process that refers to the likelihood information to identify the class with the highest probability, converts the signal into an estimated transmitted signal that corresponds to the identified class and has the bit length of the transmitted information, and outputs the result; and a correction information prediction process that detects at least one of noise and distortion remaining in the corrected received signal, predicts the magnitude of a residual correction component containing at least one of noise and distortion that will remain in the corrected received signal at the next time point using the detected residual information, and generates the noise information to be given to the signal correction unit at the next time point based on the prediction.

[0011] a signal correction process for dividing transmission information into a predetermined bit length, selecting one of M predetermined signal points (M is a value of 2 to the power of the bit length), and receiving the generated transmission signal as a received signal via a communication channel; applying noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; a neural network process for classifying the corrected received signal into one of the M classes, the number of which is the same as the number of signal points; an output layer process for outputting likelihood information indicating the probability that the received signal belongs to each of the M classes from the output of the neural network processing unit; a data conversion process for identifying the class with the highest probability by referring to the likelihood information, converting the signal into an estimated transmitted signal that corresponds to the identified class and has the bit length of the transmission information, and outputting the estimated transmitted signal; and a correction information prediction process for detecting at least one of noise and distortion remaining in the corrected received signal, predicting the magnitude of a residual correction component including at least one of noise and distortion that will remain in the corrected received signal at a next time point using the detected residual information, and generating the noise information to be provided to the signal correction unit at the next time point based on the prediction. [Effects of the Invention]

[0012] The receiving device and the like according to one embodiment can reduce reception errors and increase transmission capacity by performing highly accurate correction. [Brief explanation of the drawings]

[0013] [Figure 1] 1 is a block diagram of a communication system according to a first embodiment. [Figure 2] FIG. 2 is a detailed block diagram of a transmitting device according to the first embodiment. [Figure 3] FIG. 1 is a diagram illustrating signal points having a regular arrangement. [Figure 4] FIG. 1 is a diagram illustrating a constellation of signal points obtained by learning a neural network. [Figure 5] FIG. 2 is a detailed block diagram of a receiving device according to the first embodiment. [Figure 6]4 is a flowchart illustrating an operation of the receiving device according to the first embodiment. [Figure 7] FIG. 1 is a block diagram illustrating a hardware configuration of a receiving device according to a first embodiment. [Figure 8] FIG. 10 is a detailed block diagram of a receiving device according to a second embodiment. [Figure 9] FIG. 10 is a block diagram of a communication system according to a third embodiment. [Figure 10] 10 is a flowchart illustrating an operation of a receiving device according to a third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0014] For clarity of explanation, the following description and drawings have been omitted and simplified as appropriate. Furthermore, each element shown in the drawings as a functional block performing various processes can be configured in hardware with a CPU (Central Processing Unit), memory, and other circuits, and in software with a program loaded into memory, etc. Therefore, those skilled in the art will understand that these functional blocks can be realized in various forms using only hardware, only software, or a combination thereof, and are not limited to any one of these. In addition, the same elements are designated by the same reference numerals in each drawing, and redundant explanations are omitted as necessary.

[0015] Furthermore, the above-described program includes a set of instructions (or software code) that, when loaded into a computer, causes the computer to perform one or more functions described in the embodiments. The program may be stored in a non-transitory computer-readable medium or a tangible storage medium. By way of example and not limitation, computer-readable media or tangible storage media include random-access memory (RAM), read-only memory (ROM), flash memory, solid-state drive (SSD) or other memory technology, CD-ROM, digital versatile disc (DVD), Blu-ray disc or other optical disk storage, magnetic cassette, magnetic tape, magnetic disk storage or other magnetic storage device. The program may also be transmitted on a transitory computer-readable medium or communication medium. By way of example and not limitation, transitory computer-readable media or communication media include electrical, optical, acoustic, or other forms of propagated signals. Such embodiments are described below with reference to FIG. 7.

[0016] Embodiment 1 Hereinafter, embodiments of the present invention will be described with reference to the drawings. Fig. 1 shows a block diagram of a communication system according to the first embodiment. As shown in Fig. 1, the communication system 1 according to the first embodiment includes a transmitting device 10, a communication path 20, and a receiving device 30.

[0017] In the communication system 1, quadrature amplitude modulation is performed on the transmission information Di, which is information obtained by dividing the transmission bit sequence into b-bit units (b is an integer representing the number of bits), and M (=2 b ) signal points (constellation) are used for communication. For example, if the transmission bit sequence is divided into 6-bit units, the number of signal points M will be 64.

[0018] The transmitting device 10 receives transmission information Di from a processing device, which may be illustrated, and outputs a transmission signal X[n] by performing quadrature amplitude modulation on the transmission information Di. Here, n is an arbitrary integer and is a symbol indicating the timing point of the processing. n will have the same meaning in the following description. The transmission signal X[n] includes output signal points, which are expressed as pairs of two real values ​​or complex numbers and are illustrated as points on a two-dimensional plane.

[0019] The transmitting device 10 has a data conversion unit 11 and a signal generation unit 12. The data conversion unit 11 receives transmission information Di and converts the value indicated by the transmission information Di into information indicating the location of a signal point corresponding to the transmission information Di. The signal generation unit 12 outputs a signal point based on the information output by the data conversion unit 11.

[0020] Here, a configuration example of the transmitting device 10 will be described. Fig. 2 shows a detailed block diagram of the transmitting device according to the first embodiment. The block diagram shown in Fig. 2 is one configuration example of the transmitting device 10, and it is sufficient if the transmitting device 10 can generate signal points generated from the transmission information Di in accordance with the receiving method of the receiving device 30.

[0021] 2, the data conversion unit 11 has a one-hot vector generation unit 41 and an address generation unit 42. The signal generation unit 12 has a signal point generation processing unit 43 and a storage device 44. The transmission device 10 uses switches SW11 to SW14 to select either a first processing system that combines the one-hot vector generation unit 41 and the signal point generation processing unit 43, or a second processing system that combines the address generation unit 42 and the signal point generation processing unit 43, thereby outputting a transmission signal X[n] corresponding to transmission information Di.

[0022] The one-hot vector generation unit 41 outputs a vector of length M, in which only one position corresponding to the b-bit transmission information Di is 1 and the rest are all zero. The signal point generation processing unit 43 is, for example, a fully connected neural network, and receives the vector output from the one-hot vector generation unit 41 as input and outputs signal point data.

[0023] The address generator 42 generates and outputs an address at which a signal point corresponding to the b-bit transmission information Di is held. The storage device 44 outputs the data of the signal point held at the input address.

[0024] Next, we will explain the communication path 20. The communication path 20 does not only refer to a physical communication path such as wireless or optical fiber, but also includes signal processing devices such as antenna devices, analog-to-digital converters, and equalization filters that are provided before and after the communication path 20. The input to this communication path 20 is a transmission signal X[n], and the output is a reception signal Y[n].

[0025] Next, the receiving device 30 will be described. The receiving device 30 has a signal correction unit 31, a signal estimation processing unit 32, a correction information prediction unit 35, and a data conversion unit 36. The signal correction unit 31 corrects at least one of the noise and distortion components contained in the received signal Y[n] by converting the noise information Θ * [n] is applied to the corrected received signal Y * Outputs [n].

[0026] The signal estimation processing unit 32 includes a neural network processing unit 33 and an output layer processing unit 34. The neural network processing unit 33 calculates the corrected received signal Y * [n] is classified into one of the same number of classes as the number of signal points M. More specifically, the neural network processing unit 33 is configured by a trained fully connected neural network. This fully connected neural network classifies the corrected received signal Y * The output layer processing unit 34 outputs likelihood information p[n] indicating the probability that the received signal Y[n] belongs to each of the M classes from the output of the neural network processing unit 33. The data conversion unit 36 ​​refers to the likelihood information p[n] to identify the most probable class, and converts the estimated transmitted signal X[n] corresponding to the identified class and having the bit length of the transmitted information Di. * Convert to [n] and output.

[0027] The correction information prediction unit 35 also calculates the corrected received signal Y * At least one of noise and distortion remaining in [n] is detected, and the detected residual information is used to correct the received signal Y * The magnitude of the residual correction component including at least one of noise and distortion remaining in [n] is predicted, and noise information Θ is provided to the signal correction unit 31 at the next time based on the prediction. * Generate [n+1].

[0028] Here, we will explain signal points in a communication system 1 that uses a neural network processing unit. Fig. 3 shows a diagram illustrating signal points with a regular arrangement, and Fig. 4 shows a diagram illustrating a signal point arrangement obtained by learning a neural network. The signal point arrangement shown in Fig. 3 is obtained when basic quadrature amplitude modulation is performed when the number of signal points M is 64 (for example, 64QAM). On the other hand, the signal point arrangement shown in Fig. 4 is generated by a trained neural network using 64QAM, as in Fig. 3. In the example shown in Fig. 4, signal points are arranged in a concentric circle.

[0029] When the main cause of signal degradation in communication channel 20 is typical additive white Gaussian noise, the signal quality can be improved by having the neural network learn to suppress the additive white Gaussian noise and applying a learning result generated based on the learning result to the neural network. More specifically, when the main cause of degradation in communication channel 20 is typical additive white Gaussian noise, the communication system 1 including transmitting device 10 and receiving device 30 using a fully connected neural network can achieve the effect of increasing communication capacity.

[0030] This type of learning can prevent signal degradation due to additive white Gaussian noise by, for example, performing learning to minimize equation (1) using the test entropy error and phase noise error as loss functions.

number

[0031] In the communication system 1 according to the first embodiment, learning to minimize equation (1) is performed before the start of operation, and the learning results are applied to all fully connected neural networks. That is, in the communication system 1 according to the first embodiment, once the installation locations of the transmitting device 10, the communication path 20, and the receiving device 30 are determined and operation begins, no neural network training is performed during operation. Note that the learning parameters of each neural network may be updated after operation begins, whether periodically or irregularly.

[0032] In addition to additive white Gaussian noise, the communication channel 20 also contains noise and signal distortion whose magnitude changes over time. For example, in the communication system 1, phase noise caused by frequency fluctuations of a local oscillator (LO) used to convert the frequency of a transmission signal is well known to increase the uncertainty of the phase information of the transmission signal and to degrade signal quality and transmission capacity. In particular, in recent years, signal degradation due to phase noise of the local oscillator has been known to be an issue in the millimeter wave frequency spectrum used in 5G New Radio (NR).

[0033] However, in order to maintain communication quality against such time-varying noise and signal distortion, it is necessary to reconfigure the neural network to follow changes in the magnitude of noise and signal distortion occurring in the communication path, but such adjustments are difficult. In other words, in order to achieve an increase in transmission capacity, an extension of transmission distance, and low power consumption in a communication system, it is necessary to reconfigure the neural network in accordance with time-varying changes in the characteristics of the communication path 20.

[0034] Therefore, in the communication system 1 according to the first embodiment, for noise and signal distortion that change in correlation with time, the correction information prediction unit 35 predicts noise information Θ that is used for correcting the received signal in the signal correction unit 31. * is changed in accordance with the magnitude of noise and signal distortion that change in correlation with time, thereby reducing signal degradation caused by noise and signal distortion that change in correlation with time.

[0035] 5 is a detailed block diagram of the receiving device 30 according to the first embodiment. In the example shown in FIG. 5, a phase rotation processing unit 50 is used as the signal correction unit 31, and the output layer processing unit 34 outputs a likelihood vector p[n] using a softmax function. In this receiving device 30, the correction information prediction unit 35 has a first parameter generation unit 51, a first multiplier 52, a register 53, a residual noise detection unit 54, a second parameter generation unit 55, a second multiplier 56, and an adder 57.

[0036] Register 53 stores noise information Θ * The first parameter generation unit 51 receives the received signal Y[n] and the noise estimation value φ[n] as input and outputs the output result of a trained neural network as a first parameter (for example, a first coefficient K1). The first parameter generation unit 51 includes a signal estimation processing unit 51a and a sigmoid function processing unit 51b. The signal estimation processing unit 51a is a fully connected neural network to which the same training result as that of the neural network processing unit 33 is applied. The sigmoid function processing unit 51b applies a sigmoid function to the output value of the signal estimation processing unit 51a and outputs the first coefficient K1. The first multiplier 52 multiplies the noise estimation value φ[n] and the noise information Θ as the first coefficient K1. * Outputs [n].

[0037] The residual noise detector 54 detects the corrected received signal Y * [n] and estimated transmitted signal X *The second parameter generation unit 55 outputs the difference between the received signal Y[n] and the noise estimation value φ[n] as a residual correction component ΔΘ[n]. The second parameter generation unit 55 outputs the output result of a trained neural network that receives the received signal Y[n] and the noise estimation value φ[n] as input as a second parameter (e.g., a second coefficient K2). The second parameter generation unit 55 includes a signal estimation processing unit 55a and a sigmoid function processing unit 55b. The signal estimation processing unit 55a is a fully connected neural network to which the same training results as those of the neural network processing unit 33 are applied. The sigmoid function processing unit 55b applies a sigmoid function to the output value of the signal estimation processing unit 51a to output a second coefficient K2. The second multiplier 56 multiplies the second coefficient K2 by the residual correction component ΔΘ and outputs an uncorrected noise estimation value (e.g., K2·ΔΘ[n]). The adder 57 adds the noise estimate φ[n] held in the register 53 to the uncorrected noise estimate to generate a new noise estimate φ[n+1] that updates the noise estimate φ[n] held in the register 53.

[0038] Next, we will explain the operation of the receiving device 30. First, the received signal Y[n] is expressed by equation (2).

number

[0039] Then, the signal correction unit 31 calculates the noise information Θ given by the correction information prediction unit 35. * [n] to correct the received signal Y * Calculate [n]. Noise information Θ * [n] is the estimated value of the phase noise estimated from the received signal received before time point n. * [n] is expressed by equation (3), and the corrected received signal Y * [n] is expressed by equation (4).

number

number

[0040] The neural network processing unit 33 calculates the corrected received signal Y * The neural network processor 33 receives signal points [n] as input and outputs a vector (u1, u2, uM) consisting of M numerical values, the same number as the number of signal points. The output layer processor 34 then applies a softmax function to the vector output by the neural network processor 33, and outputs a likelihood vector p[n] indicating the probability (likelihood) that each of the M signal points is the transmitted signal X[n]. This likelihood vector is expressed as (p1, p2, pM). Each value in the likelihood vector is expressed by equation (5).

number

[0041] The data conversion unit 36 ​​outputs a bit representation of the index i that is the largest value in the likelihood vector p[n]. That is, the data conversion unit 36 ​​outputs a value that indicates pi such that pi=max(p1, p2,... pM) as a b-bit digital value. This bit representation of the index i is the estimated transmitted bit, and the i-th signal point is the estimated transmitted signal X * It becomes [n].

[0042] In the receiving device 30, the correction information prediction unit 35 calculates noise information Θ to be given to the signal correction unit 31. * Update [n] to estimate transmitted signal X * This is performed every time [n] is updated. Specifically, the correction information prediction unit 35 calculates j by using the noise information Θ * Update the noise estimate φ[n] used to calculate [n].

[0043] In updating the noise estimate φ[n], first, the residual noise detector 54 calculates the corrected received signal Y * [n] and estimated transmitted signal X *[n]. More specifically, the correction information prediction unit 35 calculates the residual correction component ΔΘ based on the corrected received signal Y * [n] and estimated transmitted signal X * The difference value with [n] is the residual correction component ΔΘ. This residual correction component ΔΘ is expressed by equation (6)

number

[0044] Further, the second parameter generation unit 55 outputs the second coefficient K2. Then, the second multiplier 56 outputs the product of the residual correction component ΔΘ and the second coefficient K2 as the uncorrected noise estimate. The adder 57 adds the uncorrected noise estimate and the noise estimate φ[n] stored in the signal point generation processing unit 43 and outputs the result as the new noise estimate φ[n+1]. The new noise estimate φ[n+1] is expressed by equation (7).

number

[0045] Next, the above operation will be explained using a flowchart. Fig. 6 shows a flowchart explaining the operation of the receiving device 30 according to the first embodiment. The receiving device 30 according to the first embodiment repeatedly executes the operation shown in Fig. 6 for each reception cycle.

[0046] The receiving device 30 receives the received signal Y[n] in synchronization with the start of the receiving cycle (step S1). The receiving device 30 also calculates a first coefficient K1 using the noise estimation value φ[n] stored in the register 53 and the received signal Y[n] received in step S1 (step S2). In FIG. 6, the formula for the calculation performed in step S2 is shown as K1=sigmoid(NN1(Y[n])), φ[n]). The function indicated by NN1 represents the output obtained by inputting the received signal Y[n] and the noise estimation value φ[n] to the signal estimation processing unit 51a, and the function indicated by sigmoid represents the output obtained by inputting the output of the signal estimation processing unit 51a to the sigmoid function processing unit 51b. The correction information prediction unit 35 then calculates the noise information Θ using equation (3). *[n] is calculated (step S3).

[0047] Next, the receiving device 30 calculates the noise information Θ * [n] to remove noise components including at least the phase noise of the received signal Y[n] by correction, and obtain a corrected received signal Y * [n] is calculated (step S4). * [n] is a signal expressed by equation (4). The receiving device 30 then converts the corrected received signal Y * [n] is input to the signal estimation processing unit 32 to calculate the likelihood vector p[n] (step S5). The likelihood vector p[n] is calculated by p[n]=softmax(NN(Y * That is, the output of the neural network processing unit 33, NN(Y * The likelihood vector p[n] is obtained by inputting the likelihood vector p[n] to the softmax function. The likelihood vector p[n] is then encoded by the data converter 36 to obtain the estimated transmitted signal X * [n] is output (step S6). * It can also be expressed as [n]=enc(p[n]).

[0048] The receiving device 30 also calculates a second coefficient K2 using the noise estimation value φ[n] stored in the register 53 and the received signal Y[n] received in step S1 (step S7). In FIG. 6, the formula for the calculation performed in step S7 is K2=sigmoid(NN2(Y[n])), φ[n]). The function indicated by NN2 indicates the output obtained by inputting the received signal Y[n] and the noise estimation value φ[n] to the signal estimation processing unit 55a, and the function indicated by sigmoid indicates the output obtained by inputting the output of the signal estimation processing unit 51a to the sigmoid function processing unit 55b. Thereafter, the residual correction component ΔΘ[n] is calculated using the residual noise detection unit 54 (step S8). In this step S8, the corrected received signal Y * [n] and estimated transmitted signal X *The difference between [n] and [n] is defined as the residual correction component ΔΘ[n]. The processing in step S8 is expressed by equation (6). Then, the adder 57 outputs a new noise estimate φ[n+1] using equation (7), thereby updating the noise estimate φ[n] stored in the register 53 to the noise estimate φ[n+1] (step S9).

[0049] Here, the hardware configuration of the receiving device 30 will be described. Fig. 7 shows a block diagram illustrating the hardware configuration of the receiving device 30 according to the first embodiment. The example shown in Fig. 7 is an example in which a computer 100 is used as the receiving device 30. The computer 100 shown in Fig. 7 has a calculation unit 101, a memory 102, and a communication interface 103. The calculation unit 101, the memory 102, and the communication interface 103 are configured to be able to communicate with each other via a bus.

[0050] The calculation unit 101 is a calculation device capable of executing a program, such as a CPU (Central Processing Unit), and realizes the functions of the receiving device 30 described above by executing a receiving program as a program.

[0051] The memory 102 is configured by at least one of a volatile memory such as a dynamic random access memory (DRAM) and a non-volatile memory such as a flash memory. A part of the storage area of ​​the memory 102 serves as a register 53.

[0052] The communication interface 103 includes a part of the communication path 20, and transmits the signal output by the transmitting device 10 into the computer 100 as a received signal.

[0053] As described above, in the communication system 1 according to the first embodiment, the influence of noise components that are not correlated with the passage of time (for example, additive white Gaussian noise) is reduced using a neural network processing unit. Furthermore, in the communication system 1 according to the first embodiment, noise information Θ used to cancel noise components that are correlated with the passage of time is *By updating X[n] using the transmission signal X[n] calculated from the received signal during operation, the influence of noise components correlated with the passage of time is reduced. As a result, the communication system 1 according to the first embodiment can maintain communication quality against both noise that varies over time and noise that does not vary over time. By maintaining high communication quality, the communication system 1 according to the first embodiment can achieve an increase in transmission capacity, an extension of transmission distance, and low power consumption.

[0054] Embodiment 2 In the second embodiment, a description will be given of a receiving device 30a which is another embodiment of the receiving device 30. In the description of the second embodiment, the same components as those in the first embodiment will be assigned the same reference numerals as those in the first embodiment, and the description thereof will be omitted.

[0055] Fig. 8 shows a detailed block diagram of the receiving device 30a according to the second embodiment. As shown in Fig. 8, the receiving device 30a has a correction information prediction unit 35a instead of the correction information prediction unit 35. The correction information prediction unit 35a is obtained by removing the first parameter generation unit 51, the first multiplier 52, and the second parameter generation unit 55 from the correction information prediction unit 35. Furthermore, the correction information prediction unit 35a inputs a third coefficient K, which is a third parameter that is a fixed value, to the second multiplier 56 instead of the second coefficient K2.

[0056] That is, the correction information prediction unit 35a can be considered as a unit in which the first coefficient K1 of the correction information prediction unit 35 is fixed at 1 and the second coefficient K2 is replaced with a fixed value determined according to the installation location and specifications of the communication system. Also, in the correction information prediction unit 35a, the noise estimated value φ stored in the register 53 is directly converted into the noise information Θ * It becomes [n].

[0057] In the communication system according to the second embodiment, the calculation load is reduced by omitting the first parameter generating unit 51 and the second parameter generating unit 55 from the correction information predicting unit 35. Also, in the communication system according to the second embodiment, the circuit scale of the correction information predicting unit 35a can be made smaller than that of the correction information predicting unit 35.

[0058] Embodiment 3 In the third embodiment, a communication system 2 will be described in which the communication system 1 is adapted to correspond to a communication path of MIMO (Multi Input Multi Output) format. In the description of the third embodiment, the same components as those in the first embodiment will be assigned the same reference numerals as those in the first embodiment, and the description thereof will be omitted.

[0059] Fig. 9 shows a block diagram of a communication system 2 according to a third embodiment. The example shown in Fig. 9 includes a communication path 20a that supports 2x2 MIMO communication. The communication system 2 according to the third embodiment also includes a transmitting device 60 and a receiving device 70.

[0060] The transmitting device 60 has a first transmission processing unit 61 and a second transmission processing unit 62. The first transmission processing unit 61 converts transmission information Di1 into a transmission signal X1[n] and outputs it. The second transmission processing unit 62 converts transmission information Di2 into a transmission signal X2[n] and outputs it. Since the first transmission processing unit 61 and the second transmission processing unit 62 have the same configuration, FIG. 9 shows the internal processing blocks of only the first transmission processing unit 61. As shown in FIG. 8, the first transmission processing unit 61 corresponds to the transmitting device 10 and has a data conversion unit 11 and a signal generation unit 12.

[0061] The receiving device 70 has a first receiving processing unit 71 and a second receiving processing unit 72. The first receiving processing unit 71 receives a received signal Y1[n] as a primary received signal out of a received signal Y1[n] that is transmitted to the receiving device 70 when a transmission signal X1[n] is input to the MIMO communication channel 20a and transmitted to the receiving device 70, and a received signal Y2[n] that is transmitted to the receiving device 70 when a transmission signal X2[n] is input to the MIMO communication channel 20a, and outputs an estimated transmitted signal X1[n] corresponding to the transmitted signal X1[n]. * The second reception processing unit 72 has substantially the same configuration as the first reception processing unit 71, except for the signal that is handled as the main reception signal. Therefore, only the first reception processing unit 71 will be described here.

[0062] The first reception processing unit 71 has a main reception signal calculation unit 80, a main reception signal processing unit 81, and a second correction information prediction unit 82. The first reception processing unit 71 receives a reception signal Y1[n] as a main reception signal to be processed from a plurality of reception signals, and also receives another reception signal that interferes with the main reception signal as a secondary reception signal Y2[n], and applies interference noise information φ2 to the secondary reception signal Y2[n] to calculate corrected interference component information R2 * [n] is generated and interference component information R2 * The main received signal calculation unit 80 uses the first signal adjustment unit 90, the interference removal unit 91, the second signal adjustment unit 92, and the second signal correction unit 93 to remove the interference component from the main received signal Y1[n] and output the result as the received signal Y1'[n].

[0063] First, the transmission signals X1[n], X2[n] and the reception signals Y1[n], Y2[n] are input and output signals of the MIMO communication channel 20a, and are assumed to satisfy the relationship shown in equation (8).

number

[0064] Here, Θ1[n] and Θ2[n] in equation (8) are discrete-time Weiner phase difference noise, and Z1[n] and Z2[n] are complex Gaussian noise. Also, H11, H12, H21, and H22 represent the channel matrices of the MIMO communication channel 20a. In the following explanation, the inverse matrix of the channel matrix is ​​expressed by equation (9).

number

[0065] The first signal adjustment unit 90 adjusts the received signal Y1[n] based on equation (10) to output a signal R11[n].

number

[0066] The second signal adjuster 92 also adjusts the received signal Y2[n] based on equation (11) to output a signal R2[n].

number

[0067] The second signal correction unit 93 performs phase rotation processing in the same manner as the signal correction unit 31. The second signal correction unit 93 calculates interference component information R2[n] using the signal R2[n] and interference noise information φ2[n] acquired from a second correction information prediction unit 82 (described later). * [n] is output. This interference component information R2 * [n] is expressed by equation (12).

number

[0068] The interference removal unit 91 receives the interference component information R2 * [n] and signal R11[n] are used to output a received signal Y1'[n]. The received signal Y1'[n] is calculated based on equation (13).

number

[0069] The main received signal processing unit 81 handles the received signal Y1'[n] as the received signal Y[n] input to the receiving device 30, and performs the same processing as the receiving device 30. In addition, the main received signal calculation unit 80 calculates information (for example, R11) to be transmitted to the second correction information prediction unit 82 from the correction information prediction unit 35. * The first correction information prediction unit 94 has an additional function for generating information R11 ([n]). * [n] is expressed by equation (14).

number

[0070] In FIG. 8, the corrected received signal Y * [n], noise information Θ * [n], the signal or information corresponding to the likelihood vector p[n] is the corrected received signal Y1 * [n], noise information Θ1 * [n], and likelihood vector p1[n].

[0071] The second correction information prediction unit 82 updates the interference noise information φ2 using the difference between the main received signal Y1 and the estimated transmitted signal X*. More specifically, the second correction information prediction unit 82 updates the noise estimation value φ2[n] by performing the following calculation:

number

number

number

[0072] Next, the operation of the receiving device 30a will be described using a flowchart. FIG. 10 shows a flowchart illustrating the operation of the receiving device 30a according to the third embodiment. As shown in FIG. 10, the receiving device 30a according to the third embodiment receives received signals Y1[n] and Y2[n] in synchronization with the start of a receiving cycle (step S11). Next, in the receiving device 30a, the first signal adjustment unit 90 calculates an adjusted first received signal R11 according to equation (10), and the second signal adjustment unit 92 calculates an adjusted second received signal R12 according to equation (11) (step S12). Next, in the receiving device 30a, the second signal correction unit 93 calculates a corrected second received signal R2 according to equation (12). *[n] (step S13). Next, the reception device 30a causes the interference canceller 91 to calculate the first reception signal Y1'[n] from which interference has been cancelled in accordance with equation (13) (step S14).

[0073] Next, the main received signal processing unit 81 of the receiving device 30a performs the processing shown in the flowchart of FIG. 6 on the first received signal Y1′[n] from which interference has been removed, thereby obtaining an estimated transmitted signal X1 * [n] and update the noise estimation value φ1 (step S15). Next, the first correction information prediction unit 94 and the second correction information prediction unit 82 of the main received signal calculation unit 80 update the interference noise information φ2 according to equations (14) to (17) (step S16).

[0074] As described above, in the communication system 2 according to the third embodiment, in MIMO communication, a signal in which phase noise has been corrected for a secondary reception signal that has an interference effect on a primary reception signal is generated, and the secondary reception signal after phase noise correction is used to cancel the interference noise component of the primary reception signal. Then, the primary reception signal from which the interference noise component has been canceled is subjected to the same processing as in the communication system 1 according to the first embodiment. Furthermore, in the communication system 2 according to the third embodiment, the interference noise information φ2 is updated for each reception cycle using the residual phase noise at that time. As a result, in the communication system 2 according to the third embodiment, the interference noise component is also canceled with the phase difference corrected. In other words, in the communication system 2 according to the third embodiment, the influence of phase noise contained in the interference component on the primary reception signal from which the interference component has been canceled is suppressed, so that high communication quality can be maintained in MIMO communication as in the communication system 1 according to the first embodiment.

[0075] The present invention is not limited to the above-described embodiment, and can be modified as appropriate within the scope of the invention. [Explanation of symbols]

[0076] 1, 2 Communication Systems 10, 60 Transmitting device 11 Data conversion section 12 Signal generation unit 20 Communication Channel 20a MIMO channel 30, 70 Receiver 31 Signal correction unit 32 Signal estimation processing unit 33 Neural network processing unit 34 Output layer processing section 35 Correction information prediction unit 36 Data conversion section 41 One-hot vector generator 42 Address generation unit 43 Signal point generation processing unit 44 Storage device 50 Phase rotation processing unit 51 First parameter generation unit 51a Signal estimation processing unit 51b Sigmoid function processing section 52 First Multiplier 53 Registers 54 Residual noise detector 55 Second parameter generation unit 55a Signal estimation processing unit 55b Sigmoid function processing section 56 Second Multiplier 57 Adder 61 First transmission processing unit 62 Second transmission processing unit 71 First receiving processing unit 72 Second receiving processing section 80 Main received signal calculation unit 81 Main receiving signal processing section 82 Second correction information prediction unit 90 First signal conditioning unit 91 Interference removal section 92 Second signal conditioning unit 93 Second signal correction unit 94 First correction information prediction unit

Claims

1. a signal correction unit that divides transmission information into a predetermined bit length, selects one of M predetermined signal points (M is a value of 2 to the power of the bit length), and receives the generated transmission signal via a communication path as a received signal, and applies noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; a neural network processing unit that classifies the corrected received signal into one of the same number of classes as the number of the signal points; an output layer processing unit that outputs likelihood information indicating the probability that the received signal belongs to each of the M classes from the output of the neural network processing unit; a data conversion unit that identifies a class with the highest probability by referring to the likelihood information, converts the signal into an estimated transmission signal that corresponds to the identified class and has the bit length of the transmission information, and outputs the signal; a correction information prediction unit that detects at least one of noise and distortion remaining in the corrected received signal, predicts the magnitude of a residual correction component including at least one of noise and distortion remaining in the corrected received signal at a next time point using the detected residual information, and generates the noise information to be provided to the signal correction unit at the next time point based on the prediction; A receiving device having:

2. The correction information prediction unit a residual noise detector that outputs a difference between the corrected received signal and the estimated transmitted signal as the residual correction component; a register for holding a noise estimate; a first parameter generation unit that outputs an output result of a trained neural network that receives the received signal and the noise estimation value as input as a first parameter; a second parameter generation unit that outputs an output result of a trained neural network that receives the received signal and the noise estimation value as input as a second parameter; a first multiplier that multiplies the noise estimate value by the first parameter and outputs the noise information; a second multiplier that multiplies the second parameter by the residual correction component to output an uncorrected noise estimate; an adder that adds the noise estimate held in the register and the uncorrected noise estimate to generate a new noise estimate that updates the noise estimate held in the register; 2. The receiving device according to claim 1, comprising:

3. The correction information prediction unit a residual noise detector that outputs a difference between the corrected received signal and the estimated transmitted signal as the residual correction component; a register for storing the noise information; a multiplier that multiplies the residual correction component by a third preset parameter and outputs uncorrected noise information; an adder that adds the noise information held in the register and the uncorrected noise information to generate new noise information that updates the noise information held in the register; 2. The receiving device according to claim 1, comprising:

4. The received signal is one of a plurality of received signals transmitted via a MIMO (Multi-Input Multi-Output) communication channel that transmits signals by a MIMO system, a main reception signal calculation unit that receives the reception signal as a main reception signal to be processed from the plurality of reception signals, and receives other reception signals that interfere with the main reception signal as secondary reception signals, applies interference noise information to the secondary reception signal to generate corrected interference component information, and outputs a signal obtained by removing the interference component from the main reception signal using the interference component information as the reception signal; an interference noise prediction unit that updates the interference noise information using a difference between the main received signal and the estimated transmitted signal; 2. The receiving device according to claim 1, comprising:

5. 2. The receiving device according to claim 1, wherein the neural network processing unit is configured by a trained fully connected neural network.

6. A transmitting device that outputs the transmission signal to the receiving device according to claim 1, a data conversion unit that receives the transmission information and converts it into information indicating the location of the signal point corresponding to a value indicated by the transmission information; a signal generation unit that outputs the transmission signal including the signal points based on the information output by the data conversion unit; A transmitting device having:

7. A receiving method in a transmitting device, which divides transmission information into a predetermined bit length, selects one of M predetermined signal points (M is a value of 2 to the power of the bit length) to generate a transmission signal, receives the generated transmission signal as a received signal via a communication path, and outputs an estimated transmission signal that is an estimate of the transmission information from the received signal by calculation, applying noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; classifying the corrected received signal into one of the same number of classes as the number of the signal points using neural network processing; outputting likelihood information indicating the probability that the received signal output by the neural network processing belongs to each of the M classes; Identifying a class with the highest probability by referring to the likelihood information, converting the signal into an estimated transmission signal that corresponds to the identified class and has the bit length of the transmission information, and outputting the signal; detecting at least one of noise and distortion remaining in the corrected received signal, predicting a magnitude of a residual correction component including at least one of noise and distortion remaining in the corrected received signal at a next time point using the detected residual information, and generating the noise information at the next time point based on the prediction; How to receive.

8. a signal correction process for dividing transmission information into a predetermined bit length, selecting one of M predetermined signal points (M is a value of 2 to the power of the bit length), and receiving the generated transmission signal via a communication channel as a received signal, and applying noise information to at least one of noise and distortion components contained in the received signal to output a corrected received signal; a neural network process for classifying the corrected received signal into one of the same number of classes as the number of the signal points; an output layer process that outputs likelihood information indicating the probability that the received signal output by the neural network process belongs to each of the M classes; a data conversion process of identifying a class with the highest probability by referring to the likelihood information, converting the signal into an estimated transmission signal that corresponds to the identified class and has the bit length of the transmission information, and outputting the signal; a correction information prediction process for detecting at least one of noise and distortion remaining in the corrected received signal, predicting the magnitude of a residual correction component including at least one of noise and distortion remaining in the corrected received signal at a next time point using the detected residual information, and generating the noise information to be used in the signal correction process at the next time point based on the prediction; A receiving program that causes a computer to execute the above.

Images