Signal analysis device and signal analysis method

Abstract

A signal analysis device (100) comprising a frequency / sampling rate adjustment unit (101) which adjusts at least one of the frequency and sampling rate of a first received signal and which outputs the result as a second received signal, a frequency domain conversion unit (102) which converts the second received signal from a time domain signal into a frequency domain signal and which outputs the signal that has been converted into the frequency domain signal as a third received signal, an action value calculation unit (103) which is formed from a neural network and which uses the third received signal as an input to output a plurality of actions and a value for each of the actions, the actions adjusting the frequency or sampling rate of the first received signal in the frequency / sampling rate adjustment unit (101), and an action determination unit (104) which uses the value to determine the action to be taken in the frequency / sampling rate adjustment unit (101) from among the plurality of actions and which outputs the determined action to the frequency / sampling rate adjustment unit (101).

Description

Technical Field

[0001] The present disclosure relates to a signal analysis device, a control circuit, a storage medium, and a signal analysis method for analyzing a received signal.

Background Art

[0002] Conventionally, in general wireless communication, communication parameters such as a carrier frequency or a center frequency, a modulation method, and a symbol rate are shared by a transmission device and a reception device. On the other hand, in fields such as spectrum monitoring, these communication parameters are unknown, and a reception device needs to estimate these communication parameters from a received signal and perform synchronization processing, demodulation processing, and the like. Patent Document 1 discloses a technique in which a modulation method estimation device estimates a center frequency and a modulation method of a wireless communication wave with unknown communication parameters using machine learning. The modulation method estimation device of Patent Document 1 uses an IQ (In-Phase / Quadrature-Phase) signal in the time domain as an input to a neural network and obtains a center frequency and a modulation method estimated as an output from the neural network.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, according to the above conventional technology, the input to the neural network for frequency estimation is an IQ signal in the time domain. Therefore, when it is desired to support a large number of modulation methods, there is a problem that the learning time increases according to the number of corresponding modulation methods. In addition, there is a problem that as the number of input-output correspondences increases, the coefficients of the neural network may not converge well during learning, and the estimation accuracy may deteriorate.

[0005] The present disclosure has been made in view of the above, and an object thereof is to obtain a signal analysis device capable of coping with a large number of modulation schemes while suppressing an increase in learning time and a deterioration in estimation accuracy.

Means for Solving the Problems

[0006] In order to solve the above-described problems and achieve the object, the signal analysis device of the present disclosure adjusts at least one of the frequency and the sampling rate with respect to a first received signal and outputs it as a second received signal; a frequency domain conversion unit that converts the second received signal from a time domain signal to a frequency domain signal and outputs the signal after the conversion to the frequency domain as a third received signal; a behavior value calculation unit configured by a neural network, which takes the third received signal as an input and outputs a plurality of behaviors for adjusting the frequency or the sampling rate of the first received signal in the adjustment unit and the value for each behavior; and a behavior determination unit that determines the behavior to be performed by the adjustment unit from the plurality of behaviors based on the value and outputs the determined behavior to the adjustment unit. A modulation method estimation unit that estimates the modulation method of the second received signal, a demodulation unit that demodulates the first received signal based on the second received signal and the estimation result of the modulation method estimated by the modulation method estimation unit, the probability of the modulation method with the highest probability based on the estimation result, the difference between the probabilities of each of the plurality of modulation methods based on the estimation result, the constellation based on the demodulation result of the demodulation unit, the eye pattern based on the demodulation result, the Error Vector Magnitude based on the demodulation result, and an input signal generation unit that outputs at least one of the bit patterns based on the demodulation result to the action value calculation unit comprises . The action value calculation unit further takes the information from the input signal generation unit as input and outputs a plurality of actions and values to the action determination unit characterized in that

Advantages of the Invention

[0007] The signal analysis device according to the present disclosure has an effect of being capable of coping with a large number of modulation schemes while suppressing an increase in learning time and a deterioration in estimation accuracy.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Mode for Carrying Out the Invention

[0009] Hereinafter, a signal analysis device, a control circuit, a storage medium, and a signal analysis method according to embodiments of the present disclosure will be described in detail with reference to the drawings.

[0010] Embodiment 1. FIG. 1 is a diagram showing a configuration example of a signal analysis device 100 according to Embodiment 1. The signal analysis device 100 is mounted on a receiving device (not shown) that receives a received signal with unknown communication parameters, and is a device that analyzes a received signal with unknown communication parameters. The signal analysis device 100 includes a frequency sampling rate adjustment unit 101, a frequency domain conversion unit 102, a behavior value calculation unit 103, a behavior determination unit 104, a modulation method estimation unit 105, and a demodulation unit 106.

[0011] The frequency sampling rate adjustment unit 101 is an adjustment unit that adjusts at least one of the frequency and the sampling rate according to the action determined by the action determination unit 104 for the received signal input to the signal analysis device 100. The frequency sampling rate adjustment unit 101 outputs the adjusted signal to the frequency domain conversion unit 102 and the modulation method estimation unit 105. The received signal input to the signal analysis device 100 is, for example, an IQ signal as described in the background art. In the following description, the received signal input to the signal analysis device 100, that is, the received signal to be adjusted in the frequency sampling rate adjustment unit 101 is referred to as the first received signal, and the signal adjusted by the frequency sampling rate adjustment unit 101 may be referred to as the second received signal.

[0012] For example, immediately after the signal analysis device 100 is started and when no action for adjusting at least one of the frequency and the sampling rate has been determined in the action determination unit 104, the frequency sampling rate adjustment unit 101 outputs the received signal input to the signal analysis device 100 as it is. Note that the frequency sampling rate adjustment unit 101 may be configured to output to the modulation method estimation unit 105 after performing the adjustment to at least one of the frequency and the sampling rate a prescribed number of times according to the action determined by the action determination unit 104. That is, if the adjustment to at least one of the frequency and the sampling rate has not been performed a prescribed number of times according to the action determined by the action determination unit 104, the frequency sampling rate adjustment unit 101 may not output to the modulation method estimation unit 105.

[0013] The frequency domain conversion unit 102 converts the signal adjusted by the frequency sampling rate adjustment unit 101 from a time-domain signal to a frequency-domain signal, and outputs the signal after being converted to a frequency-domain signal to the action value calculation unit 103. The frequency domain conversion unit 102 outputs, as the signal after being converted to a frequency-domain signal, the value on the frequency axis and the power value or amplitude value corresponding to the value on the frequency axis. In the following description, the signal after being converted to a frequency-domain signal by the frequency domain conversion unit 102 may be referred to as the third received signal. The configuration of the frequency domain conversion unit 102 will be described in detail. FIG. 2 is a diagram showing a configuration example of the frequency domain conversion unit 102 included in the signal analysis apparatus 100 according to the first embodiment. The frequency domain conversion unit 102 includes an FFT (Fast Fourier Transform) 201, a power conversion unit 202, and a normalization unit 203. Here, the case where the frequency domain conversion unit 102 outputs, as the signal after being converted to a frequency-domain signal, the value on the frequency axis and the power value corresponding to the value on the frequency axis will be described as an example.

[0014] The FFT 201 converts the signal adjusted by the frequency sampling rate adjustment unit 101 from a time-domain signal to a frequency-domain signal. The FFT 201 outputs, as the frequency-domain signal, the value on the frequency axis and the FFT result corresponding to the value on the frequency axis. For high-precision spectrum, the FFT 201 may store the signal adjusted by the frequency sampling rate adjustment unit 101, which is the input signal, in a memory (not shown), average it using the Welch method or the like, and then perform the conversion process. Further, after the conversion process, the FFT 201 may apply a moving average filter or the like to average in the frequency domain. Further, the FFT 201 may be trained to improve the spectrum accuracy in the frequency domain using a neural network different from the action value calculation unit 103 described later, and then perform the conversion process. Note that the frequency domain conversion unit 102 may use a DFT (Discrete Fourier Transform) as a method for converting the signal adjusted by the frequency sampling rate adjustment unit 101 from a time-domain signal to a frequency-domain signal.

[0015] The power conversion unit 202 converts the FFT result obtained from the FFT 201 into a power value. Note that, as described above, when the frequency domain conversion unit 102 outputs the value on the frequency axis and the amplitude value corresponding to the value on the frequency axis as the signal after being converted into the signal in the frequency domain, instead of the power conversion unit 202, it is configured to convert the FFT result into an amplitude value.

[0016] The normalization unit 203 normalizes the value on the frequency axis obtained from the FFT 201 and the power value obtained from the power conversion unit 202. As the normalization process, the normalization unit 203 may perform normalization processing for stabilizing the learning coefficient in the subsequent action value calculation unit 103, such as the process of making the average of various values 0 and the variance 1, or the process of keeping the values within the range from the maximum value 1 to the minimum value 0. Note that the frequency domain conversion unit 102 may normalize only one of the value on the frequency axis from the FFT 201 and the power value from the power conversion unit 202.

[0017] FIG. 3 is a diagram showing an image of the operation in the frequency domain conversion unit 102 included in the signal analysis device 100 according to Embodiment 1. In the frequency domain conversion unit 102, for modulation schemes such as BPSK (Binary Phase Shift Keying), QPSK (Quadrature Phase Shift Keying), 16QAM (Quadrature Amplitude Modulation), 32QAM, and 64QAM among a plurality of modulation schemes, the FFT 201 can convert the signal in the time domain into the signal in the frequency domain to obtain a similar signal shape. Therefore, the signal analysis device 100 can reduce the input pattern in the configuration after the frequency domain conversion unit 102, that is, in the action value calculation unit 103, and thus can reduce the number of input-output correspondences in the action value calculation unit 103. Note that, as shown in FIG. 3, the frequency domain conversion unit 102 is also applicable to modulation schemes such as MSK (Minimum Shift Keying) and FSK (Frequency Shift Keying) in addition to the above modulation schemes.

[0018] The action value calculation unit 103 is composed of a neural network that takes as input the signal in the frequency domain converted from the frequency domain conversion unit 102. The output of the neural network is a plurality of actions that are candidates for adjustment in the frequency sampling rate adjustment unit 101 and the value of each action. That is, the action value calculation unit 103 takes as input the signal in the frequency domain converted from the frequency domain conversion unit 102, and outputs a plurality of actions for adjusting the frequency or the sampling rate of the received signal in the frequency sampling rate adjustment unit 101 and the value for each action. For the plurality of actions, as shown in Example 1, it may include both the frequency adjustment amount and the sampling rate adjustment amount, or as shown in Example 2 and Example 3, it may include only one of the frequency adjustment amount and the sampling rate adjustment amount, or as shown in Example 4, a simple method such as obtaining the centroid of the spectrum, for example, a frequency rough estimation method, may be included as one of the actions.

[0019] (Example 1) "Action 1: Frequency +x%", "Action 2: Frequency -x%", "Action 3: Sampling rate +y%", "Action 4: Sampling rate -y%"

[0020] (Example 2) "Action 1: Frequency +x%", "Action 2: Frequency -x%", "Action 3: Frequency +10x%", "Action 4: Frequency -10x%"

[0021] (Example 3) "Action 1: Sampling rate +y%", "Action 2: Sampling rate -y%", "Action 3: Sampling rate +10y%", "Action 4: Sampling rate -10y%"

[0022] (Example 4) "Action 1: Frequency +x%", "Action 2: Frequency -x%", "Action 3: Frequency rough estimation method 1"

[0023] Note that in the above example, the adjustment amounts in the + direction and - direction are made the same magnitude with respect to the frequency as +x% or -x%, and with respect to the sampling rate as +y% or -y%, but it is not limited to this. The action value calculation unit 103 may set the adjustment amounts in the + direction and - direction to different magnitudes, such as "Action 1: frequency +x1%" and "Action 2: frequency -x2%". Further, the action value calculation unit 103 may include instructions in both the + and - directions for each of "x1" and "x2" without indicating +- as in "Action 1: frequency x1%" and "Action 2: frequency x2%". Also, when "Action 1: frequency +x%" is selected by the action decision unit 104 during the previous adjustment, the action value calculation unit 103 may use the same "Action 1: frequency +x%" in the + direction for the next adjustment, but may change the magnitude of the adjustment amount in the - direction, such as "Action 2: frequency -x2%". In this way, the action value calculation unit 103 outputs to the action decision unit 104, as a plurality of actions, at least two of one or more actions for adjusting the frequency and one or more actions for adjusting the sampling rate in the frequency sampling rate adjustment unit 101.

[0024] The action value calculation unit 103 obtains in advance, by simulation or the like, the coefficients of the neural network. Regarding the neural network of the action value calculation unit 103, it is pre-trained so as to provide information necessary for the subsequent action decision unit 104 to determine an action, such as the higher the numerical value of the value output to the action decision unit 104, the higher the value of taking the corresponding action. Also, the action value calculation unit 103 determines in advance, that is, at the time of learning of the neural network, the adjustment amount for each action.

[0025] The action decision unit 104 acquires a plurality of actions and values from the action value calculation unit 103. The action decision unit 104 determines the action to be performed by the frequency sampling rate adjustment unit 101 from among the plurality of actions based on the values, that is, determines the action with respect to the received signal input to the signal analysis device 100, and outputs the determined action to the frequency sampling rate adjustment unit 101. Regarding the method of determining the action, as shown in Example 1, the action shown to have the highest value may be selected, or as shown in Example 2, the action with the highest value for each of the frequency and sampling rate may be selected one by one, or as shown in Example 3, it may be probabilistically selected based on the ratio of the output values.

[0026] (Example 1) "Action 1: Frequency +x% Value = 1.0", "Action 2: Frequency -x% Value = 0.2", "Action 3: Sampling rate +y% Value = 0.3", "Action 4: Sampling rate -y% Value = 0.9" In this case, the action decision unit 104 selects Action 1.

[0027] (Example 2) "Action 1: Frequency +x% Value = 1.0", "Action 2: Frequency -x% Value = 0.2", "Action 3: Sampling rate +y% Value = 0.3", "Action 4: Sampling rate -y% Value = 0.9" In this case, the action decision unit 104 selects Action 1 and Action 4.

[0028] (Example 3) "Action 1: Frequency +x% Value = 1.0", "Action 2: Frequency -x% Value = 0.2", "Action 3: Sampling rate +y% Value = 0.3", "Action 4: Sampling rate -y% Value = 0.9" In this case, the action decision unit 104 selects action 1 with a probability of 1.0 / (1.0 + 0.2 + 0.3 + 0.9), selects action 2 with a probability of 0.2 / (1.0 + 0.2 + 0.3 + 0.9), selects action 3 with a probability of 0.3 / (1.0 + 0.2 + 0.3 + 0.9), and selects action 4 with a probability of 0.9 / (1.0 + 0.2 + 0.3 + 0.9). Note that if the action decision unit 104 selects an action with a low probability in the previous selection, it may select an action with a high probability in the next selection.

[0029] In this way, the action decision unit 104 determines one action from a plurality of actions based on the value. Alternatively, the action decision unit 104 determines one action from one or more actions that adjust the frequency based on the value, and determines one action from one or more actions that adjust the sampling rate.

[0030] The modulation method estimation unit 105 estimates the modulation method for a signal in which at least one of the frequency and the sampling rate is adjusted by the frequency sampling rate adjustment unit 101. As for the method of estimating the modulation method, for example, there is a method using a neural network as described in Japanese Patent No. 7130179, but it is not limited to this. Since the modulation method estimation unit 105 is designed assuming to some extent the frequency range of the received signal, the ratio of the sampling rate to the symbol rate of the received signal, that is, the oversampling rate, etc., the processing in the frequency sampling rate adjustment unit 101, the frequency domain conversion unit 102, the action value calculation unit 103, and the action decision unit 104 is repeated a plurality of times so that the frequency and the sampling rate of the received signal are adjusted and fall within the assumption of the modulation method estimation unit 105, the modulation method can be correctly estimated.

[0031] The demodulation unit 106 demodulates the received signal input to the signal analysis device 100 based on the signal in which at least one of the frequency and the sampling rate is adjusted by the frequency sampling rate adjustment unit 101 and the estimation result of the modulation method estimated by the modulation method estimation unit 105. Generally, in wireless communication, demodulation is possible if the frequency, symbol rate, modulation method, etc. of the target signal are known. Therefore, the demodulation unit 106 can demodulate the received signal with unknown communication parameters by using the estimation results up to the previous stage.

[0032] Note that in the signal analysis device 100, it is assumed that the adjustment of the frequency and the sampling rate has not been completed immediately after receiving the received signal, and the estimation accuracy of the modulation method, the demodulation accuracy, etc. are poor. Therefore, in the signal analysis device 100, the frequency sampling rate adjustment unit 101 may manage the adjustment completion flag and control the operations of the modulation method estimation unit 105 and the demodulation unit 106. For example, when the processing in the frequency sampling rate adjustment unit 101, the frequency domain conversion unit 102, the action value calculation unit 103, and the action determination unit 104 is repeated multiple times and the adjustment is performed a specified number of times, assuming that the frequency and the sampling rate have become substantially constant values, the adjustment completion flag is set. While the adjustment completion flag is not set, the frequency sampling rate adjustment unit 101 does not output the signal after adjusting the frequency and the sampling rate to the modulation method estimation unit 105, and when the adjustment completion flag is set, the frequency sampling rate adjustment unit 101 outputs the signal after adjusting the frequency and the sampling rate to the modulation method estimation unit 105.

[0033] FIG. 4 is a flowchart showing the operation of the signal analysis device 100 according to Embodiment 1. In the signal analysis device 100, the frequency sampling rate adjustment unit 101 adjusts at least one of the frequency and the sampling rate of the received signal according to the action determined by the action determination unit 104 (step S1). Note that, as described above, the frequency sampling rate adjustment unit 101 may omit step S1 when the action has not been determined by the action determination unit 104 immediately after the activation of the signal analysis device 100 or the like. The frequency domain conversion unit 102 converts the received signal adjusted by the frequency sampling rate adjustment unit 101 from a time-domain signal to a frequency-domain signal (step S2). The action value calculation unit 103 calculates a plurality of actions that are candidates for adjustment by the frequency sampling rate adjustment unit 101 and the value of each action based on the signal converted into a frequency-domain signal obtained from the frequency domain conversion unit 102 (step S3).

[0034] The action determination unit 104 determines an action based on the plurality of actions and values obtained from the action value calculation unit 103 (step S4). The action determination unit 104 selects, that is, determines, an action from the plurality of actions and outputs the determined action to the frequency sampling rate adjustment unit 101. The modulation method estimation unit 105 estimates the modulation method for the received signal adjusted by the frequency sampling rate adjustment unit 101 and obtained from the frequency sampling rate adjustment unit 101 (step S5). The demodulation unit 106 demodulates the received signal based on the received signal whose frequency and sampling rate have been adjusted and the estimation result of the modulation method (step S6). Note that the signal analysis device 100 may not perform the operations of step S5 and step S6 while the adjustment completion flag is not set by using the adjustment completion flag as described above. The signal analysis device 100 repeats the operations from step S1 to step S4 while the adjustment completion flag is not set.

[0035] Next, the hardware configuration of the signal analysis device 100 will be described. In the signal analysis device 100, the frequency sampling rate adjustment unit 101, the frequency domain conversion unit 102, the action value calculation unit 103, the action decision unit 104, the modulation method estimation unit 105, and the demodulation unit 106 are realized by a processing circuit. The processing circuit may be a processor and a memory that execute a program stored in the memory, or may be dedicated hardware. The processing circuit is also called a control circuit.

[0036] FIG. 5 is a diagram showing a configuration example of a processing circuit 900 when the processing circuit that realizes the signal analysis device 100 according to the first embodiment is composed of a processor 901 and a memory 902. The processing circuit 900 shown in FIG. 5 is a control circuit and includes a processor 901 and a memory 902. When the processing circuit 900 is composed of the processor 901 and the memory 902, each function of the processing circuit 900 is realized by software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in the memory 902. In the processing circuit 900, the processor 901 reads and executes the program stored in the memory 902 to realize each function. That is, the processing circuit 900 includes a memory 902 for storing a program in which the processing of the signal analysis device 100 is ultimately executed. This program can also be said to be a program for causing the signal analysis device 100 to execute each function realized by the processing circuit 900. This program may be provided by a storage medium in which the program is stored, or may be provided by other means such as a communication medium.

[0037] The above program includes an adjustment step in which a frequency sampling rate adjustment unit 101 adjusts at least one of the frequency and the sampling rate for a first received signal and outputs it as a second received signal; a frequency domain conversion step in which a frequency domain conversion unit 102 converts the second received signal from a time domain signal into a frequency domain signal and outputs the converted signal as a third received signal; an action value calculation step in which an action value calculation unit 103 composed of a neural network takes the third received signal as an input and outputs a plurality of actions for adjusting the frequency or the sampling rate of the first received signal in the frequency sampling rate adjustment unit 101 and the values for each action; and an action determination step in which an action determination unit 104 determines an action to be performed by the frequency sampling rate adjustment unit 101 from a plurality of actions based on the values and outputs the determined action to the frequency sampling rate adjustment unit 101. It can also be said that this is a program executed by the signal analysis device 100.

[0038] Here, the processor 901 is, for example, a CPU (Central Processing Unit), a processing device, an arithmetic device, a microprocessor, a microcomputer, or a DSP (Digital Signal Processor). Also, the memory 92 corresponds to, for example, a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable ROM), an EEPROM (registered trademark) (Electrically EPROM), a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini disk, or a DVD (Digital Versatile Disc).

[0039] FIG. 6 is a diagram showing an example of a processing circuit 903 in the case where the processing circuit that realizes the signal analysis device 100 according to Embodiment 1 is configured by dedicated hardware. The processing circuit 903 shown in FIG. 6 corresponds to, for example, a single circuit, a composite circuit, a programmed processor, a parallel-programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or a combination thereof. For the processing circuit, part of it may be realized by dedicated hardware and part of it may be realized by software or firmware. In this way, the processing circuit can realize each of the above functions by dedicated hardware, software, firmware, or a combination thereof.

[0040] As described above, according to the present embodiment, in the signal analysis device 100, the frequency sampling rate adjustment unit 101 adjusts at least one of the frequency and the sampling rate in the time domain according to the action determined by the action determination unit 104 for the received signal. The frequency domain conversion unit 102 converts the signal adjusted by the frequency sampling rate adjustment unit 101 from a time domain signal to a frequency domain signal. The action value calculation unit 103 configured by a neural network takes the signal converted to the frequency domain by the frequency domain conversion unit 102 as an input, and outputs a plurality of actions that are candidates for adjustment by the frequency sampling rate adjustment unit 101 and the value of each action. The action determination unit 104 determines the action by the frequency sampling rate adjustment unit 101 based on the value. Thereby, the signal analysis device 100 can estimate the modulation method of the received signal by the modulation method estimation unit 105 and can demodulate the received signal by the demodulation unit 106 by repeating a series of operations from the frequency sampling rate adjustment unit 101 to the action determination unit 104, and can adjust the frequency and the sampling rate of the received signal to such an extent. The signal analysis device 100 can support a large number of modulation methods while suppressing an increase in learning time and a deterioration in estimation accuracy.

[0041] Generally, when it is desired to support a large number of modulation schemes, the learning time increases according to the number of modulation schemes, and since the number of input-output correspondences is large, the coefficients of the neural network may not converge well during learning, and the estimation accuracy may deteriorate. In contrast, the signal analysis device 100 according to the present embodiment converts the received signal into the frequency domain and inputs it to the neural network, so that different modulation schemes are often represented by the same frequency spectrum, and since the number of input-output correspondences can be reduced, the learning time can be reduced and high accuracy can be expected. Further, the signal analysis device 100 according to the present embodiment can perform frequency adjustment in the time domain and then convert it into the frequency domain, so that it does not depend on the resolution of the conversion into the frequency domain and can handle fine frequency offsets.

[0042] Also, in the case of estimating and decoding the modulation scheme of the received signal and directly outputting the frequency estimation value as in Patent Document 1 of the prior art, when the SNR (Signal to Noise Ratio) drops instantaneously or the output value deviates greatly from the true value, separate determination processing, correction processing, etc. are required. In contrast, the signal analysis device 100 according to the present embodiment sets the output of the neural network as "action" and "value", and restricts the content of the "action" to the adjustment of communication parameters such as "adjusting the frequency or symbol rate to a specified value", thereby suppressing performance degradation due to instantaneous deterioration of the communication environment.

[0043] Also, the signal analysis device 100 according to the present embodiment can adjust the sampling rate of the received signal and convert it into an oversampling rate, which is the ratio of the sampling rate to the symbol rate assumed during learning, so that the operations in the subsequent modulation scheme estimation unit 105 and demodulation unit 106 become possible.

[0044] Embodiment 2. In Embodiment 2, a case will be described in which the estimation result of the modulation scheme estimation unit 105 and the demodulation result of the demodulation unit 106 are input to the neural network of the action value calculation unit 103.

[0045] FIG. 7 is a diagram showing a configuration example of the signal analysis device 100a according to Embodiment 2. The signal analysis device 100a is mounted on a receiving device (not shown) that receives a received signal with unknown communication parameters, and is a device that analyzes a received signal with unknown communication parameters. The signal analysis device 100a of Embodiment 2 shown in FIG. 7 is obtained by adding an input signal generation unit 107 to the signal analysis device 100 of Embodiment 1 shown in FIG. 1.

[0046] In Embodiment 2, the modulation method estimation unit 105 outputs an estimation result of the modulation method to the input signal generation unit 107. As shown in Example 1, the modulation method estimation unit 105 may output the modulation method with the highest probability and its probability value, or as shown in Example 2, may output a plurality of modulation methods and the accuracy for each modulation method. Note that the modulation method estimation unit 105 may be the same as that in Embodiment 1 in terms of the estimation method itself of the modulation method, as long as the number of output destinations of the estimation result increases.

[0047] (Example 1) "QPSK 80%"

[0048] (Example 2) "QPSK 70%", "BPSK 20%", "FSK 10%"

[0049] In Embodiment 2, the demodulation unit 106 outputs a demodulation result to the input signal generation unit 107. As the demodulation result, for example, a constellation, an eye pattern, an EVM (Error Vector Magnitude), a bit pattern, etc. are output. Note that the demodulation unit 106 may be the same as that in Embodiment 1 in terms of the demodulation method itself, as long as the number of output destinations of the demodulation result increases.

[0050] The input signal generation unit 107 generates a signal to be output to the action value calculation unit 103 by using the estimation result obtained from the modulation method estimation unit 105 and the demodulation result obtained from the demodulation unit 106. FIG. 8 is a diagram showing a configuration example of the input signal generation unit 107 included in the signal analysis device 100a according to the second embodiment. The input signal generation unit 107 includes a maximum probability selection unit 301, a probability difference calculation unit 302, a constellation generation unit 303, an eye pattern generation unit 304, an EVM calculation unit 305, a bit pattern analysis unit 306, and a normalization unit 307.

[0051] When the estimation result obtained from the modulation method estimation unit 105 is as in Example 1 described in the modulation method estimation unit 105 of the second embodiment, the maximum probability selection unit 301 outputs "QPSK 80%", and when it is as in Example 2 described in the modulation method estimation unit 105 of the second embodiment, it selects and outputs "QPSK 70%".

[0052] When the estimation result obtained from the modulation method estimation unit 105 is as in Example 2 described in the modulation method estimation unit 105 of the second embodiment, the probability difference calculation unit 302 calculates and outputs a dispersion value or the like representing how far apart the probabilities of a plurality of modulation methods are using "QPSK 70%" and "FSK 10%".

[0053] When the demodulation result obtained from the demodulation unit 106 is a constellation, the constellation generation unit 303 generates and outputs an image of the constellation.

[0054] When the demodulation result obtained from the demodulation unit 106 is an eye pattern, the eye pattern generation unit 304 generates and outputs an image of the eye pattern.

[0055] The EVM calculation unit 305 may output the demodulation result obtained from the demodulation unit 106 as it is when the demodulation result is the EVM, or may calculate and output the EVM from the constellation when the demodulation result obtained from the demodulation unit 106 is a constellation.

[0056] When the demodulation result obtained from the demodulation unit 106 is a bit pattern, the bit pattern analysis unit 306 calculates and outputs an average value, a variance value, etc., representing the randomness of the bit pattern.

[0057] The normalization unit 307 normalizes the output from the bit pattern analysis unit 306 with respect to the output from the maximum probability selection unit 301. As normalization processing, the normalization unit 307 may perform normalization processing for stabilizing the learning coefficient in the subsequent action value calculation unit 103, such as processing to make the average of various values 0 and the variance 1, or processing to confine the values within the range from the maximum value 1 to the minimum value 0. Note that the normalization unit 307 may normalize only a part of the output from the bit pattern analysis unit 306 with respect to the output from the maximum probability selection unit 301.

[0058] In this way, the input signal generation unit 107 outputs at least one of the probability of the modulation method with the highest probability based on the estimation result of the modulation method estimation unit 105, the difference between the probabilities of a plurality of modulation methods based on the estimation result of the modulation method estimation unit 105, the constellation based on the demodulation result of the demodulation unit 106, the eye pattern based on the demodulation result of the demodulation unit 106, the EVM based on the demodulation result of the demodulation unit 106, and the bit pattern based on the demodulation result of the demodulation unit 106 to the action value calculation unit 103. The action value calculation unit 103 further takes the information from the input signal generation unit 107 as input and outputs a plurality of actions and values to the action determination unit 104.

[0059] The hardware configuration of the signal analysis device 100a will be described. In the signal analysis device 100a, the input signal generation unit 107 is also realized by a processing circuit. The processing circuit may be a processor and a memory that execute a program stored in the memory, or may be dedicated hardware.

[0060] As described above, according to this embodiment, in the signal analysis device 100a, the input signal generation unit 107 generates a signal to be output to the action value calculation unit 103 by using the estimation result obtained from the modulation method estimation unit 105 and the demodulation result obtained from the demodulation unit 106. The action value calculation unit 103 composed of a neural network takes as inputs the signal converted into the frequency domain by the frequency domain conversion unit 102 and the signal generated by the input signal generation unit 107, and outputs a plurality of actions that are candidates for adjustment in the frequency sampling rate adjustment unit 101 and the value of each action. The action determination unit 104 determines the action in the frequency sampling rate adjustment unit 101 based on the value.

[0061] Thereby, the signal analysis device 100a can handle not only the received signal but also the estimation result from the subsequent modulation method estimation unit 105 and the demodulation result of the demodulation unit 106 by repeating a series of operations from the frequency sampling rate adjustment unit 101 to the demodulation unit 106, and can improve the accuracy of the actions and values calculated by the action value calculation unit 103. Since the action is for improving the performance of the subsequent processing, these subsequent processes, that is, the results of the estimation result of the modulation method estimation unit 105 and the demodulation result of the demodulation unit 106 are fed back. The signal analysis device 100a can also use the signal generated by the input signal generation unit 107 to improve the accuracy of the actions and values calculated by the action value calculation unit 103 as compared with the signal analysis device 100 of Embodiment 1.

[0062] Embodiment 3. In Embodiment 3, a case will be described in which a reward based on the estimation result of the modulation method estimation unit 105 and the demodulation result of the demodulation unit 106 is input to the neural network of the action value calculation unit 103.

[0063] FIG. 9 is a diagram showing a configuration example of the signal analysis device 100b according to Embodiment 3. The signal analysis device 100b is mounted on a receiving device (not shown) that receives a received signal with unknown communication parameters, and is a device that analyzes a received signal with unknown communication parameters. The signal analysis device 100b of Embodiment 3 shown in FIG. 9 is obtained by adding a reward calculation unit 108 to the signal analysis device 100a of Embodiment 2 shown in FIG. 7.

[0064] In Embodiment 3, the modulation method estimation unit 105 outputs an estimation result of the modulation method to the reward calculation unit 108. The modulation method estimation unit 105 may be the same as in Embodiment 2 in terms of the estimation method of the modulation method itself and the content of the estimation result, as long as the number of output destinations of the estimation result increases.

[0065] In Embodiment 3, the demodulation unit 106 outputs a demodulation result to the reward calculation unit 108. The demodulation unit 106 may be the same as in Embodiment 2 in terms of the demodulation method itself and the content of the demodulation result, as long as the number of output destinations of the demodulation result increases.

[0066] The reward calculation unit 108 calculates a reward for updating the coefficients of the neural network constituting the action value calculation unit 103 based on the action selection result obtained from the action determination unit 104, the estimation result of the modulation method obtained from the modulation method estimation unit 105, and the demodulation result obtained from the demodulation unit 106.

[0067] For example, based on the action selection result obtained from the action decision unit 104, when the frequency becomes within the desired frequency error or the sampling rate becomes within the desired rate range as a result of taking an action to adjust the frequency or the sampling rate, the reward calculation unit 108 outputs a high reward, for example, a reward of "1". When, conversely, the result of taking an action to adjust the frequency or the sampling rate is significantly deviated from the true value, the reward calculation unit 108 outputs a low reward, for example, a reward of "-1". When the result of taking an action to adjust the frequency or the sampling rate is that the frequency or the sampling rate neither reaches the desired state nor significantly deviates from the true value, the reward calculation unit 108 outputs a reward of "0" as if there is nothing. Such a method is difficult when the parameters of the received signal are unknown, but is effective when learning by inputting a signal with known parameters such as a test signal.

[0068] Also, based on the estimation result of the modulation method obtained from the modulation method estimation unit 105, when the probability value of the most probable modulation method exceeds a certain threshold, the reward calculation unit 108 outputs a high reward, and when the probability value of the most probable modulation method is below a certain threshold, the reward calculation unit 108 outputs a low reward. Based on the estimation result of the modulation method obtained from the modulation method estimation unit 105, when the variance value representing how much the probabilities of a plurality of modulation methods are separated exceeds a certain threshold, the reward calculation unit 108 outputs a high reward, and when the variance value representing how much the probabilities of a plurality of modulation methods are separated is below a certain threshold, the reward calculation unit 108 outputs a low reward.

[0069] Also, based on the demodulation result obtained from the demodulation unit 106, when the variation of the center of the eye pattern is within the specified range, that is, small, the reward calculation unit 108 outputs a high reward, and when the variation of the center of the eye pattern is outside the specified range, that is, large, the reward calculation unit 108 outputs a low reward. Based on the demodulation result obtained from the demodulation unit 106, when the EVM is below a certain threshold, the reward calculation unit 108 outputs a high reward, and when the EVM exceeds a certain threshold, the reward calculation unit 108 outputs a low reward.

[0070] As described above, the reward calculation unit 108 may determine the reward based on each of the action selection result obtained from the action determination unit 104, the estimated modulation method result obtained from the modulation method estimation unit 105, and the demodulation result obtained from the demodulation unit 106, or may determine the reward based on two or more of the results.

[0071] Compared with the action value calculation unit 103 in Embodiment 2, the action value calculation unit 103 further obtains the reward output from the reward calculation unit 108. The action value calculation unit 103 updates, that is, learns, each coefficient of the neural network that calculates the action value from the reward calculated by the reward calculation unit 108 and its own output result. For example, if the obtained reward is high, the action and value are correct, and if the obtained reward is low, the action and value are incorrect. The action value calculation unit 103 learns to output actions and values that can obtain higher rewards. Such learning is generally a framework called reinforcement learning, and learning can be performed using a general-purpose algorithm. In Embodiments 1 and 2, it was assumed that the coefficients of the neural network were obtained in advance by simulation or the like, but by adding the reward calculation unit 108 as in Embodiment 3, it is also possible to learn from the received signal.

[0072] As described above, the reward calculation unit 108 calculates the reward for the action based on the estimated modulation method result obtained from the modulation method estimation unit 105 and the demodulation result obtained from the demodulation unit 106, and outputs it to the action value calculation unit 103. The action value calculation unit 103 learns by updating each coefficient of the neural network so that a high reward can be obtained from the reward calculation unit 108 based on the plurality of actions and values output to the action determination unit 104 and the reward obtained from the reward calculation unit 108.

[0073] The hardware configuration of the signal analysis device 100b will be described. In the signal analysis device 100b, the reward calculation unit 108 is also realized by a processing circuit. The processing circuit may be a processor and a memory that execute a program stored in the memory, or may be dedicated hardware.

[0074] As described above, according to this embodiment, in the signal analysis device 100b, the reward calculation unit 108 calculates the reward to be given to the action value calculation unit 103 based on the action selection result acquired from the action determination unit 104, the estimated result of the modulation method acquired from the modulation method estimation unit 105, and the demodulation result acquired from the demodulation unit 106. The action value calculation unit 103 performs learning so as to obtain a high reward from the reward calculation unit 108. As a result, the signal analysis device 100b can perform online learning from the received signal as well as offline learning by prior simulation or the like, and even when the characteristics of the target received signal change, it can perform estimation of the modulation method adapted to the environment, demodulation of the received signal, and the like.

[0075] The configurations shown in the above embodiments are merely examples, and it is possible to combine them with other known techniques, combine the embodiments with each other, and omit or change a part of the configuration without departing from the gist.

Explanation of Reference Numerals

[0076] 100, 100a, 100b Signal analysis device, 101 Frequency sampling rate adjustment unit, 102 Frequency domain conversion unit, 103 Action value calculation unit, 104 Action determination unit, 105 Modulation method estimation unit, 106 Demodulation unit, 107 Input signal generation unit, 108 Reward calculation unit, 201 FFT, 202 Power conversion unit, 203, 307 Normalization unit, 301 Maximum probability selection unit, 302 Probability difference calculation unit, 303 Constellation generation unit, 304 Eye pattern generation unit, 305 EVM calculation unit, 306 Bit pattern analysis unit, 900, 903 Processing circuit, 901 Processor, 902 Memory.

Claims

1. An adjustment unit that adjusts at least one of a frequency and a sampling rate with respect to a first received signal and outputs the adjusted signal as a second received signal; A frequency domain conversion unit that converts the second received signal from a time domain signal into a frequency domain signal and outputs the converted signal as a third received signal; An action value calculation unit configured by a neural network, which takes the third received signal as an input and outputs a plurality of actions for adjusting the frequency or the sampling rate of the first received signal in the adjustment unit and values for each action; An action determination unit that determines an action to be performed by the adjustment unit from the plurality of actions based on the value and outputs the determined action to the adjustment unit; A modulation method estimation unit that estimates a modulation method of the second received signal; A demodulation unit that demodulates the first received signal based on the second received signal and an estimation result of the modulation method estimated by the modulation method estimation unit; An input signal generation unit that outputs at least one of a probability of the modulation method with the highest probability based on the estimation result, a difference between probabilities of a plurality of modulation methods based on the estimation result, a constellation based on a demodulation result of the demodulation unit, an eye pattern based on the demodulation result, an Error Vector Magnitude based on the demodulation result, and a bit pattern based on the demodulation result to the action value calculation unit; comprising The action value calculation unit further takes information from the input signal generation unit as an input and outputs the plurality of actions and the values to the action determination unit; A signal analysis device characterized by the above.

2. An adjustment unit that adjusts at least one of a frequency and a sampling rate with respect to a first received signal and outputs the adjusted signal as a second received signal; A frequency domain conversion unit that converts the second received signal from a time domain signal into a frequency domain signal and outputs the converted signal as a third received signal; An action value calculation unit configured by a neural network, which takes the third received signal as an input and outputs a plurality of actions for adjusting the frequency or the sampling rate of the first received signal in the adjustment unit and values for each action; An action determination unit that determines an action to be performed by the adjustment unit from the plurality of actions based on the value and outputs the determined action to the adjustment unit; A modulation method estimation unit that estimates a modulation method of the second received signal; A demodulation unit that demodulates the first received signal based on the second received signal and the estimation result of the modulation scheme estimated by the modulation scheme estimation unit; A reward calculation unit that calculates a reward for the action based on the estimation result and the demodulation result of the demodulation unit and outputs the reward to the action value calculation unit; Comprising; The action value calculation unit performs learning by updating each coefficient of the neural network so that a high reward can be obtained from the reward calculation unit based on the plurality of actions, values, and rewards output to the action decision unit. A signal analysis device characterized by the above.

3. The action value calculation unit includes, as the plurality of actions, at least two of one or more actions for adjusting the frequency and one or more actions for adjusting the sampling rate, and outputs the actions to the action decision unit. The signal analysis device according to claim 1 or 2, characterized in that.

4. The action decision unit determines one action from the plurality of actions based on the value. The signal analysis device according to claim 3, characterized in that.

5. The action decision unit determines one action from one or more actions for adjusting the frequency based on the value, and determines one action from one or more actions for adjusting the sampling rate. The signal analysis device according to claim 3, characterized in that.

6. The frequency domain conversion unit outputs, as the third received signal, a value on the frequency axis and a power value or an amplitude value corresponding to the value on the frequency axis. The signal analysis device according to claim 1 or 2, characterized in that.

7. An adjustment step in which an adjustment unit adjusts at least one of the frequency and the sampling rate with respect to the first received signal and outputs the adjusted signal as a second received signal; A frequency domain conversion step in which a frequency domain conversion unit converts the second received signal from a time domain signal to a frequency domain signal and outputs the converted signal as a third received signal; An action value calculation step in which an action value calculation unit configured by a neural network takes the third received signal as an input and outputs a plurality of actions for adjusting the frequency or the sampling rate of the first received signal in the adjustment unit and values for each action; An action decision step in which an action decision unit determines an action to be performed by the adjustment unit from the plurality of actions based on the value and outputs the determined action to the adjustment unit; a modulation method estimation step in which a modulation method estimation unit estimates the modulation method of the second received signal; a demodulation step in which a demodulation unit demodulates the first received signal based on the second received signal and the estimation result of the modulation method estimated by the modulation method estimation unit; an input signal generation step in which an input signal generation unit outputs at least one of the probability of the modulation method with the highest probability based on the estimation result, the difference between the probabilities of a plurality of modulation methods based on the estimation result, the constellation based on the demodulation result of the demodulation unit, the eye pattern based on the demodulation result, the Error Vector Magnitude based on the demodulation result, and the bit pattern based on the demodulation result to the action value calculation unit; comprising; in the action value calculation step, the action value calculation unit further takes the information from the input signal generation unit as an input and outputs the plurality of actions and the values to the action determination unit; A signal analysis method characterized by the above.

8. An adjustment step in which an adjustment unit adjusts at least one of the frequency and the sampling rate with respect to a first received signal and outputs it as a second received signal; a frequency domain conversion step in which a frequency domain conversion unit converts the second received signal from a time domain signal to a frequency domain signal and outputs the signal after conversion to a frequency domain signal as a third received signal; an action value calculation step in which an action value calculation unit configured by a neural network takes the third received signal as an input and outputs a plurality of actions for adjusting the frequency or the sampling rate of the first received signal in the adjustment unit and the values for each action; an action determination step in which an action determination unit determines an action to be performed by the adjustment unit from the plurality of actions based on the value and outputs the determined action to the adjustment unit; a modulation method estimation step in which a modulation method estimation unit estimates the modulation method of the second received signal; a demodulation step in which a demodulation unit demodulates the first received signal based on the second received signal and the estimation result of the modulation method estimated by the modulation method estimation unit; a reward calculation step in which a reward calculation unit calculates a reward for the action based on the estimation result and the demodulation result of the demodulation unit and outputs it to the action value calculation unit; comprising; In the action value calculation step, the action value calculation unit performs learning by updating each coefficient of the neural network so that a high reward can be obtained from the reward calculation unit based on the plurality of actions, the values, and the reward output to the action determination unit. A signal analysis method characterized by the above.

Images