Signal specification determination method and signal specification determination device

Abstract

To provide a signal specification determination device that can align the center frequencies of the spectrograms of multiplied signals. [Solution] According to the embodiment, the signal specification determination device comprises a nonlinear processing unit that transforms an input signal using a nonlinear function, a frequency data generation unit that generates frequency data from the transformed signal, and a signal specification determination unit that performs a class classification process on the generated frequency data to classify it into one of the classes associated with predetermined signal specifications, determines the signal specifications of the input signal based on the classified class, and estimates the center frequency and duration of the signal.

Description

Technical Field

[0001] The present invention relates to a signal parameter determination method and a signal parameter determination apparatus for determining parameters of a wireless signal.

Background Art

[0002] Conventionally, a signal parameter determination apparatus that receives a wireless communication signal and determines the signal parameters of the received signal has been used. A conventional signal parameter determination apparatus determines signal parameters using a deterministic method that identifies a signal from feature quantities such as the shape of a spectrum, the presence or absence of a peak, the dispersion of an amplitude, and the signal bandwidth. In recent years, signal parameter determination apparatuses that determine signal parameters using machine learning methods such as a support vector machine and a neural network have also been proposed.

[0003] Patent Document 1 discloses a signal parameter determination apparatus using spectrograms of signals after multiple multiplication processes. A spectrogram represents the time change of a frequency spectrum obtained as a result of each Fourier transform by applying the Fourier transform a plurality of times while shifting an input signal in the time direction, in a two-dimensional image. For example, the horizontal direction represents the change in frequency, and the vertical direction represents the change in time. And the value of each pixel of this image represents the amplitude of the corresponding frequency component at each time.

[0004] Patent Document 2 discloses a signal parameter determination apparatus that identifies a modulation method regardless of different signal characteristics (for example, communication speed, signal length, etc.). There are many examples of using a neural network for signal parameter determination, including these.

[0005] In addition, Non-Patent Document 1 proposes an object detection type neural network that regresses the position in a general image while classifying the class of an object, as a neural network in the field of image identification for general images such as animals and humans.

Prior Art Documents

Patent Documents

[0006] [Patent Document 1] Japanese Patent Publication No. 2019-165320 [Patent Document 2] Japanese Patent Publication No. 2021-136493 [Non-patent literature]

[0007] [Non-Patent Document 1] SSD: Single Shot MultiBox Detector, https: / / arxiv.org / pdf / 1512.02325 (searched on September 5, 2020) [Overview of the Initiative] [Problems that the invention aims to solve]

[0008] The frequency multiplication process exemplified in Patent Document 1 is known to be effective in extracting features such as phase-modulated signals. However, when a spectrogram is generated by performing frequency multiplication without using a tuning function, the signal center frequency in the spectrogram shifts according to the multiplication factor.

[0009] However, when using a neural network, such as an object detection neural network, to determine signal parameters and signal center frequency from a spectrogram, if the input signal center frequency is shifted due to multiple multiplication processes, the signal center frequency cannot be determined. Therefore, during the training of the classifier, it is necessary to adjust the classifier using the correct data provided during the training process so that the signal center frequencies match. This makes training more difficult, leading to increased training time or decreased discrimination performance.

[0010] Therefore, in determining and detecting the signal parameters of wireless communication signals, it is necessary to align the center frequencies of the spectrograms of the multiplied signals.

[0011] The problem that this invention aims to solve is to provide a signal specification determination method and a signal specification determination device that can align the center frequencies of the spectrogram of a multiplied signal. [Means for solving the problem]

[0012] The signal specification determination device of the embodiment includes a nonlinear processing unit that transforms an input signal using a nonlinear function, a frequency data generation unit that generates frequency data from the transformed signal, and a signal specification determination unit that performs a class classification process on the generated frequency data to classify it into one of the classes associated with predetermined signal specifications, determines the signal specifications of the input signal based on the classified class, and estimates the center frequency and duration of the signal. [Brief explanation of the drawing]

[0013] [Figure 1] Figure 1 is a functional block diagram showing an example of the configuration of a signal parameter determination device to which the signal parameter determination method of an embodiment of the present invention is applied. [Figure 2] Figure 2 is a functional block diagram showing an example of the configuration of the nonlinear processing unit in this embodiment. [Figure 3] Figure 3 is a functional block diagram showing an example of the configuration of the frequency data generation unit in this embodiment. [Figure 4] Figure 4 is a flowchart showing an example of the operation of a signal parameter determination device to which the signal parameter determination method of the present invention is applied. [Figure 5] Figure 5 is a conceptual diagram illustrating the spectrogram obtained by doubling the frequency range and then performing a Fourier transform, and the spectrogram obtained by further reducing it by half in the frequency direction. [Figure 6A] Figure 6A is an illustrative diagram showing a spectrogram obtained when the sample is processed without multiplication. [Figure 6B] Figure 6B is an illustrative diagram showing the spectrogram obtained when the spectrogram exemplified in Figure 6A is doubled. [Figure 7]FIG. 7 is a flowchart illustrating a process flow that is executed to align the center positions of the spectrogram after multiplication and the spectrogram without multiplication. [Figure 8] FIG. 8 is an image diagram for explaining the positive / negative inversion process of the spectrogram. [Figure 9] FIG. 9 is an image diagram for explaining the reduction process. [Figure 10] FIG. 10 is an image diagram for explaining the concatenation process. [Figure 11] FIG. 11 is an image diagram showing an example of the signal spectrogram obtained when doubling the signal after sampling. [Figure 12] FIG. 12 is an image diagram showing an example of the signal spectrogram when doubling the signal after sampling. [Figure 13] FIG. 13 is a functional block diagram showing a configuration example of a signal parameter determination device to which the signal parameter determination method in Modified Example 1 is applied. [Figure 14] FIG. 14 is a functional block diagram showing a configuration example of the frequency data generation unit in Modified Example 1. [Figure 15] FIG. 15 is a flowchart showing an operation example of a signal parameter determination device to which the signal parameter determination method of Modified Example 1 is applied. [Figure 16] FIG. 16 is a functional block diagram showing a configuration example of a signal parameter determination device to which the signal parameter determination method in Modified Example 2 is applied. [Figure 17] FIG. 17 is a functional block diagram showing a configuration example of the frequency data generation unit in Modified Example 2. [Figure 18] FIG. 18 is a flowchart showing an operation example of a signal parameter determination device to which the signal parameter determination method of Modified Example 2 is applied. BEST MODE FOR CARRYING OUT THE INVENTION

[0014] A signal parameter determination device to which the signal parameter determination method of an embodiment of the present invention is applied will be described below with reference to the drawings. The drawings are schematic or conceptual, and the relationship between the thickness and width of each part, the ratio of the sizes between parts, etc., are not necessarily the same as those of actual objects. Furthermore, even when representing the same part, the dimensions and ratios may be represented differently in the drawings. In this specification and each drawing, elements similar to those described in previously shown drawings are denoted by the same reference numerals, and detailed explanations and redundant explanations are omitted as appropriate.

[0015] (Device configuration) First, the configuration of a signal parameter determination device to which the signal parameter determination method of the embodiment of the present invention is applied will be described.

[0016] Figure 1 is a functional block diagram showing an example of the configuration of a signal parameter determination device to which the signal parameter determination method of an embodiment of the present invention is applied.

[0017] The signal specification determination device 10 comprises a nonlinear processing unit 100, a frequency data generation unit 200, and a signal specification determination unit 300. The signal specification determination device 10 can be implemented using a computer such as a personal computer. The nonlinear processing unit 100, the frequency data generation unit 200, and the signal specification determination unit 300 may be implemented by software executed by a processor such as a CPU, or by dedicated hardware.

[0018] Figure 2 is a functional block diagram showing an example configuration of the nonlinear processing unit 100.

[0019] The nonlinear processing unit 100 includes a signal distribution unit 110 and a plurality of conversion units 120.

[0020] The signal distribution unit 110 receives a digital radio signal r(t) transmitted from an external device, such as a receiver (not shown), and distributes it to each conversion unit 120.

[0021] Figure 2 shows, as an example, a nonlinear processing unit 100 having four conversion units 120 (#1 to #4), but the number of conversion units 120 is not limited to four. The number of conversion units 120 can be appropriately determined according to the number and type of signal parameters determined by the signal parameter determination device 10.

[0022] Each of the conversion units 120 (#1 to #4) is configured with a function for converting the digital radio signal r(t) sent from the signal distribution unit 110.

[0023] Although the functions set in each transformation unit 120 (#1 to #4) are different, any multiple transformation units 120 (for example, #2 to #4) among the transformation units 120 (#1 to #4) are set to nonlinear functions.

[0024] This is the case when the function f1(x) is set in the transformation unit 120(#1), the function f2(x) is set in the transformation unit 120(#2), the function f3(x) is set in the transformation unit 120(#3), and the function f4(x) is set in the transformation unit 120(#4). For example, f1(x)=x, f2(x)=x 2 f3(x)=x 4 f4(x)=x 8 This can be achieved by doing so. In this case, f1(x) is a linear function, but f2(x), f3(x), and f4(x) are nonlinear functions. Note that such nonlinear functions are just examples and are not limited to this case.

[0025] Each conversion unit 120 (#1 to #4) converts the digital radio signal r(t) using a set function f(x), and outputs the converted digital radio signal f(r(t)) to the frequency data generation unit 200. That is, in the configuration shown in Figure 2, conversion unit 120 (#1) converts the digital radio signal r(t) using the function f1(x), and outputs the converted digital radio signal f1(r(t)) to the frequency data generation unit 200.

[0026] Similarly, the conversion unit 120 (#2) outputs the digital radio signal f2(r(t)) obtained by converting the digital radio signal r(t) using the function f2(x) to the frequency data generation unit 200, the conversion unit 120 (#3) outputs the digital radio signal f3(r(t)) obtained by converting the digital radio signal r(t) using the function f3(x) to the frequency data generation unit 200, and the conversion unit 120 (#4) outputs the digital radio signal f4(r(t)) obtained by converting the digital radio signal r(t) using the function f4(x) to the frequency data generation unit 200.

[0027] In this way, the nonlinear processing unit 100 transforms the digital radio signal r(t) with a function that includes multiple different nonlinear functions, thereby outputting a digital radio signal f(r(t)) to the frequency data generation unit 200, in which different characteristics of the digital radio signal r(t) are emphasized.

[0028] Figure 3 is a functional block diagram showing an example configuration of the frequency data generation unit 200.

[0029] The frequency data generation unit 200 includes multiple systems in which an expanded frequency conversion unit 210 and a frequency regeneration processing unit 220 are arranged in series. Each system is provided in a one-to-one correspondence with each conversion unit 120. Therefore, the number of systems is equal to the number of conversion units 120.

[0030] The frequency data generation unit 200 illustrated in Figure 3 shows an example where four systems are provided, corresponding to the four conversion units 120 (#1 to #4) provided in the nonlinear processing unit 100 illustrated in Figure 2.

[0031] Specifically, the first system is provided in correspondence with the conversion unit 120(#1) and consists of an expanded frequency conversion unit 210(#1) and a frequency regeneration processing unit 220(#1). The second system is provided in correspondence with the conversion unit 120(#2) and consists of an expanded frequency conversion unit 210(#2) and a frequency regeneration processing unit 220(#2). The third system is provided in correspondence with the conversion unit 120(#3) and consists of an expanded frequency conversion unit 210(#3) and a frequency regeneration processing unit 220(#3). The fourth system is provided in correspondence with the conversion unit 120(#4) and consists of an expanded frequency conversion unit 210(#4) and a frequency regeneration processing unit 220(#4).

[0032] The frequency expansion unit 210 performs frequency resolution expansion processing on the digital radio signal f(r(t)) output from the conversion unit 120, as described later, and outputs the obtained frequency data F to the frequency regeneration processing unit 220 of the same system. Each frequency expansion unit 210 (#1 to #4) of each system may perform different or the same processing as the frequency expansion processing.

[0033] The frequency regeneration processing unit 220 performs a regeneration process to regenerate the frequency information of the frequency data F output from the frequency expansion conversion unit 210, and outputs the obtained frequency data G to the signal specification determination unit 300. Each frequency regeneration processing unit 220 (#1 to #4) performs a regeneration process determined according to the processing performed by the corresponding conversion unit 120 (#1 to #4).

[0034] In the frequency data generation unit 200 as illustrated in Figure 3, in the first system, the augmented frequency conversion unit 210(#1) performs augmented frequency processing on the digital radio signal f1(r(t)) output from the conversion unit 120(#1), and outputs the obtained frequency data F1 to the frequency regeneration processing unit 220(#1). The frequency regeneration processing unit 220(#1) performs frequency information regeneration processing on the frequency data F1 output from the augmented frequency conversion unit 210(#1), and outputs the obtained frequency data G1 to the signal specification determination unit 300.

[0035] Similar processing is performed in other systems, and ultimately, frequency data G2 is output from frequency regeneration processing unit 220(#2), frequency data G3 from frequency regeneration processing unit 220(#3), and frequency data G4 from frequency regeneration processing unit 220(#4) to the signal specification determination unit 300.

[0036] In some systems, depending on the algorithm used in the signal specification determination unit 300 (described later), it is possible to disable either or both of the processing of the expanded frequency conversion unit 210 and the frequency reproduction processing unit 220.

[0037] For example, if the first frequency amplification unit 210(#1) is disabled, the digital radio signal f1(r(t)) input to the frequency amplification unit 210(#1) undergoes frequency conversion by setting N to 1 in the N-fold amplification process and 1 / N reduction process, and outputs it to the frequency regeneration processing unit 220(#1) as frequency data F1 (the amplification and reduction processes will be described later).

[0038] If the first frequency regeneration processing unit 220(#1) is disabled, the frequency data F1 input to the frequency regeneration processing unit 220(#1) is output directly to the signal specification determination unit 300 as frequency data G1 without being regenerated by the frequency regeneration processing unit 220(#1).

[0039] The signal specification determination unit 300 determines the signal specifications of the digital radio signal r(t) from the frequency data G output from all frequency data generation units 200 and estimates the center frequency and duration of the digital radio signal r(t). As illustrated in Figure 3, if the frequency data generation unit 200 has four systems, the signal specifications of the digital radio signal r(t) are determined from the frequency data G1 to G4 from all four frequency regeneration processing units 220 (#1 to #4) and the center frequency and duration of the digital radio signal r(t) are estimated.

[0040] To determine the signal characteristics, the signal characteristic determination unit 300 classifies each frequency data G1 to G4 into one of several classes associated with predetermined signal characteristics, and then uses, for example, an object detection neural network to regress the signal occurrence position in the frequency data. Each class is pre-associated with signal characteristics. The signal occurrence position is the center frequency of the signal if the frequency data is a spectrum, or the center frequency and duration of the signal if it is a spectrogram.

[0041] A spectrogram is a two-dimensional representation of the time evolution of the frequency spectrum obtained by applying the discrete Fourier transform multiple times to an input signal while shifting it in the time direction. For example, the horizontal direction represents the change in frequency, and the vertical direction represents the change in time. A spectrogram is two-dimensional discrete data, and the coordinate values ​​correspond to the amplitude of the corresponding frequency component at each time point.

[0042] In this way, the signal specification determination unit 300 determines the signal specifications of the digital radio signal r(t) and estimates the center frequency and time length in the frequency-time space.

[0043] (Operation instructions) Next, the operation of a signal specification determination device to which the signal specification determination method of the embodiment of the present invention is applied will be described.

[0044] Figure 4 is a flowchart showing an example of the operation of a signal parameter determination device to which the signal parameter determination method of the present invention is applied.

[0045] First, a digital radio signal r(t) whose signal parameters are to be determined is input to the signal parameter determination device 10 (S1). The digital radio signal r(t) is, for example, a PSK signal or an FSK signal, and is input to the signal parameter determination device 10 from an external device such as a receiver (not shown), and is received by the signal distribution unit 110 of the nonlinear processing unit 100.

[0046] The digital radio signal r(t) received by the signal distribution unit 110 is distributed from the signal distribution unit 110 to each conversion unit 120. For example, as illustrated in Figure 2, if four conversion units 120 (#1 to #4) are provided, the digital radio signal r(t) is distributed by the signal distribution unit 110 to the four conversion units 120 (#1 to #4).

[0047] The digital radio signal r(t) distributed to each conversion unit 120 (#1 to #4) is converted in each conversion unit 120 (#1 to #4) using a pre-set function (S2).

[0048] Although the functions set in each transformation unit 120 (#1 to #4) are different, any multiple transformation units 120 (for example, #2 to #4) among the transformation units 120 (#1 to #4) have nonlinear functions set.

[0049] Here, we will explain using an example where different power functions with different multipliers are set in the transformation unit 120 (#1~#4). If the multipliers are 1, 2, 4, and 8 respectively, the functions (f1(x), f2(x), f3(x), f4(x)) set in each transformation unit 120 (#1~#4) can be as follows. f1(x) = x f2(x) = x 2 f3(x) = x 4 f4(x)=x 8 Of the four functions f(x) above, f1(x) is a linear function, while f2(x), f3(x), and f4(x) are nonlinear functions.

[0050] In each conversion unit 120 (#1 to #4), the digital radio signal r(t) is converted using the function f(x) set for each unit, and the converted digital radio signal f(r(t)) is output to the frequency data generation unit 200.

[0051] Furthermore, when the function set in the conversion unit 120 is a power function, the multiple of the power function is not limited to 1, 2, 4, and 8, but can also be a multiple of any other value.

[0052] When a digital radio signal r(t) is a PSK signal, if the frequency multiplication factor is an integer multiple of the modulation level, it exhibits the characteristic of having peaks in the spectrum at the center frequency and the bandwidth frequency. Therefore, in this case, the digital radio signals f(r(t)) output from each conversion unit 120 (#1~#4) exhibit different characteristics depending on the modulation level.

[0053] Generally, it is difficult to determine the modulation scheme of a FSK signal with a small number of modulation levels, or to classify FSK signals with small differences in frequency shift width, from the spectrum of a digital radio signal r(t). Therefore, it is preferable for the conversion unit 120 (#1~#4) to convert the digital radio signal r(t) using power functions with different frequency multiplication values, as described above. As a result, the amount of frequency shift of the digital radio signal f(r(t)) output from each conversion unit 120 (#1~#4) will be different, and different characteristics will be emphasized. This makes it easier to classify the frequency data performed by the signal specification determination unit 300.

[0054] Furthermore, QAM signals, APSK signals, OFDM signals, etc., also have their own unique spectral characteristics, and these are emphasized by the conversion unit 120 (#1~#4).

[0055] Next, returning to Figure 4, we will explain the augmented frequency processing (S3) performed by the augmented frequency conversion unit 210.

[0056] The digital radio signals f(r(t)) (in this case, f1(r(t)), f2(r(t)), f3(r(t)), and f4(r(t))) output from each conversion unit 120 (#1 to #4) to the frequency data generation unit 200 are received in the frequency data generation unit 200 by the augmented frequency conversion units 210 corresponding to each conversion unit 120 (#1 to #4). That is, the digital radio signal f1(r(t)) output from conversion unit 120 (#1) is received by the augmented frequency conversion unit 210 (#1).

[0057] Similarly, the digital radio signal f2(r(t)) output from the conversion unit 120(#2) is received by the augmented frequency conversion unit 210(#2), the digital radio signal f3(r(t)) output from the conversion unit 120(#3) is received by the augmented frequency conversion unit 210(#3), and the digital radio signal f4(r(t)) output from the conversion unit 120(#4) is received by the augmented frequency conversion unit 210(#4).

[0058] Then, each of the frequency expansion units 210 (#1 to #4) performs frequency expansion processing, such as frequency resolution expansion processing, on the received digital radio signal f(r(t)), and as a result, frequency data F (in this case, F1 to F4), such as a spectrogram, is obtained. The obtained frequency data F1 to F4 are output from each of the frequency expansion units 210 (#1 to #4) to the corresponding frequency regeneration processing units 220 (#1 to #4).

[0059] While not limited to, Fourier transforms can be applied to frequency expansion processing. When a Fourier transform is applied for frequency expansion processing, each digital radio signal f1(r(t)) to f4(r(t)) is Fourier transformed in each frequency expansion transformer 210 (#1 to #4) using N times the number of points (time samples) of the desired frequency resolution, and the resulting output spectrogram is reduced to 1 / N in the frequency direction.

[0060] Here, N can be any positive integer. Such frequency-direction reduction is not limited to reduction based on the maximum value of neighboring pixels, but may also be based on, for example, the mean, median, etc. Alternatively, it may be a decimation process using a filter that extracts peaks.

[0061] The augmented frequency conversion unit 210 performs augmented and reduced frequency processing, which increases the number of points in the Fourier transform by a factor of N to enhance the features. In addition, the 1 / N reduction process shortens the learning time and discrimination time performed by the signal specification determination unit 300.

[0062] In Fourier transforms, the sample size is generally set to a power of two, and the Fast Fourier Transform (FFT) is performed. However, this is not the only option; zero-padding can be performed on sample sizes other than powers of two, and the FFT can be performed to match the sample size to a power of two, or the Discrete Fourier Transform (DFT) can be performed directly. Furthermore, the probability density function of the spectrum can be estimated based on an autoregressive model or the like.

[0063] Furthermore, when performing a Fourier transform, a window function is generally applied to the input signal, and similarly, a window function can be applied in this embodiment. Moreover, in this case, there are no particular limitations on the type of window function, and any window function can be applied.

[0064] Furthermore, while it is common practice to apply a Fourier transform to generate a spectrogram so that the time intervals of the digital radio signals r(t) overlap, in this embodiment, the amount of overlap is not limited when generating a spectrogram as a result of the expanded frequency processing, and any overlap amount may be set. Moreover, the overlap amount may be adaptively controlled according to the duration of the signal.

[0065] Each of the frequency expansion units 210 (#1 to #4) in each system may perform different or the same processing as part of this frequency expansion processing.

[0066] Next, returning to Figure 4, we will explain the frequency information regeneration process (S4) performed by the frequency regeneration processing unit 220.

[0067] Each frequency regeneration processing unit 220 (#1~#4) performs regeneration processing of frequency data F1~F4 output from the same system's expanded frequency conversion unit 210 (#1~#4), and the regenerated frequency data G1~G4 is output to the signal specification determination unit 300. The regeneration processing performed in each system's frequency regeneration processing unit 220 (#1~#4) is determined according to the processing performed by the same system's conversion unit 120 (#1~#4).

[0068] As an example of the frequency regeneration processing performed by the frequency regeneration processing unit 220 (#1~#4), we will explain using a spectrogram obtained by a single multiplication process using a nonlinear function. Here, the digital radio signal r(t) is assumed to be a signal sampled with a center frequency of -fc and a sampling period of fs.

[0069] Figure 5(a) is a conceptual diagram showing an example of a spectrogram obtained by Fourier transforming a digital radio signal r(t) after it has been doubled. In Figure 5(a), the shaded area represents the signal portion, the horizontal axis represents frequency, and the vertical axis represents time.

[0070] In the frequency regeneration processing performed by the frequency regeneration processing unit 220 (#1~#4), DFT or FFT are used for digital radio signal processing. However, for the sake of simplicity, we will explain using the Fourier transform, which transforms the frequency domain into an infinitely continuous frequency domain, as an example. In this case, the spectrogram is repeated with a sampling period fs by the Fourier transform, and the regions R1 and R2 shown in Figure 5(a) have the same spectrogram.

[0071] Figure 5(b) is a conceptual diagram showing the spectrogram of Figure 5(a) when it is reduced to half its original size in the frequency direction, with 0 Hz as the boundary. In Figure 5(b), the shaded area represents the signal portion, the horizontal axis represents frequency, and the vertical axis represents time.

[0072] When the spectrogram in Figure 5(a) is reduced by half in the frequency direction with 0 Hz as the boundary, a signal appears at the same frequency position as the center frequency -fc of the original signal, as shown in region R3 of Figure 5(b).

[0073] The frequency regeneration processing unit 220 uses the frequency data F, which has been amplified and frequency-converted by the augmented frequency conversion unit 210, to output a spectrogram equivalent to the spectrogram in the range of -fs / 2 to fs / 2, as shown in region R4 of Figure 5(b).

[0074] In addition, the following description of the frequency regeneration processing unit 220 will also use a spectrogram image, but this is a spectrogram image that assumes the DFT or FFT used in the actual processing, where the frequency is discrete and the range is the Nyquist frequency (-fs / 2 to fs / 2). This is different from the spectrograms shown in Figures 5(a) and 5(b), but it is used to clearly convey the image of where the signal spectrogram appears in the frequency-time space.

[0075] Figure 6A is an illustrative diagram showing a spectrogram obtained when a signal with a center frequency of -fc is sampled at a sampling frequency of fs and processed without frequency multiplication. The center of the frequency axis is 0 Hz, and the frequency range is -fs / 2 to fs / 2.

[0076] Figure 6B is an illustrative diagram showing the spectrogram obtained when the spectrogram exemplified in Figure 6A is doubled. In both Figure 6A and Figure 6B, the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0077] The examples shown in FIGS. 6A and 6B are cases where fs / 4 < |fc| < fs / 2, and show cases where aliasing has occurred.

[0078] The purpose of the frequency reproduction process performed in step S4 is to reproduce the frequency information of the original digital radio signal r(t). The operation of such a frequency reproduction process will be described below using the flowchart of FIG. 7.

[0079] FIG. 7 is a flowchart illustrating the flow of a process executed in the frequency reproduction processing unit 220 (#1 to #4) so as to match the center positions of the multiplied spectrogram and the spectrogram without multiplication.

[0080] When frequency data F, which is a spectrogram of the multiplied digital radio signal, is input from the same-system up-conversion unit 210 (#1 to #4) to the frequency reproduction processing unit 220 (#1 to #4) (S11), a sign inversion process for inverting the positive and negative of the frequency is performed on the spectrogram of this frequency data F in each frequency reproduction processing unit 220 (#1 to #4) (S

[0081] FIG. 8 is an image diagram for explaining the sign inversion process of the spectrogram. FIG. 8(a) shows the spectrogram before sign inversion, and FIG. 8(b) shows the spectrogram after sign inversion. In both FIGS. 8(a) and 8(b), the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0082] In the sign inversion process, as illustrated from FIG. 8(a) to FIG. 8(b), a process of inverting the positive and negative of the frequency is performed with 0 Hz as the boundary. When the sign inversion process is performed, a spectrogram corresponding to the region R1 in FIG. 5(a) is obtained as illustrated in FIG. 8(b).

[0083] The spectrogram after the positive / negative swapping process is then reduced in the frequency direction by the frequency multiplier in each frequency regeneration processing unit 220 (#1~#4) (S13). The reduction ratio of the reduction process is determined according to the frequency multiplier of the corresponding conversion unit 120 (#1~#4) in each frequency regeneration processing unit 220 (#1~#4).

[0084] Specifically, if the spectrogram is a spectrogram that has been doubled in the conversion unit 120, the reduction ratio is determined to be 1 / 2. Then, this spectrogram is reduced to 1 / 2 by the frequency regeneration processing unit 220 to create a reduced spectrogram.

[0085] Figure 9 is an illustrative diagram explaining the reduction process. Figure 9(a) shows an illustrative diagram of the spectrogram before reduction, and Figure 9(b) shows an illustrative diagram of the reduced spectrogram after reduction. In both Figures 9(a) and (b), the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0086] From a spectrogram like the one exemplified in Figure 9(a), a reduction process is performed by each frequency regeneration processing unit 220 (#1~#4) to create a reduced spectrogram like the one exemplified in Figure 9(b). The reduced spectrogram exemplified in Figure 9(b) corresponds to region R3 in Figure 5(b).

[0087] Furthermore, the frequency-based reduction process may be based on the maximum value of nearby pixels, as well as the average value, median, etc. Alternatively, it may be performed using a filter to extract peaks.

[0088] As the final step of step S5, each frequency regeneration processing unit 220 (#1 to #4) concatenates the reduced spectrograms in the frequency direction by the number of frequency multiplications of the corresponding conversion units 120 (#1 to #4) (S14).

[0089] Figure 10 is an illustrative diagram explaining the concatenation process. Figure 10(a) shows an illustrative diagram of the reduced spectrogram before concatenation, and Figure 10(b) shows an illustrative diagram of the reduced spectrogram after concatenation. In both Figures 10(a) and (b), the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0090] The number of concatenations is determined according to the multiplication factor of the conversion unit 120 (#1~#4). For example, if this reduced spectrogram originates from the spectrogram of frequency data F multiplied by 2 in the conversion unit 120, the number of concatenations is determined to be 2, and two reduced spectrograms are concatenated.

[0091] The reduced spectrogram after concatenation, illustrated in Figure 10(b), corresponds to region R4 in Figure 5(b). As a result of the series of processes, it can be seen that the center frequencies of the spectrogram without multiplication and the spectrogram after multiplication coincide.

[0092] In this embodiment, the frequency regeneration processing unit 220 (#1~#4) shows an example where multiplication processing is used as the nonlinear processing in the conversion unit 120. However, the processing performed by the frequency regeneration processing unit 220 (#1~#4) is not limited to this example. Any other method may be used to correct the frequency data obtained by the nonlinear processing of the conversion unit 120 and align it with the linearly processed frequency data and the center position.

[0093] For example, the frequency regeneration processing unit 220 (#1~#4) can convert high-frequency components from the transmission processing back into the frequency components of the original signal. As a result, the center frequencies of the signal spectrograms, which are the frequency data output by the frequency regeneration processing unit 220 (#1~#4), all match. Similarly, even if a one-dimensional spectrum is input, the center frequencies will all match.

[0094] As another example, the processing performed by the frequency regeneration processing unit 220 will be explained when the center frequency of the digital radio signal r(t) is different.

[0095] FIG. 11 is an image diagram showing an example of a signal spectrogram obtained by each step of the frequency reproduction processing unit 220 when a digital radio signal r(t) sampled at a sampling rate fs is subjected to a doubling process by the non-linear processing unit 100 for a signal having a center frequency fc (fs / 4 < |fc| < fs / 2). The frequency range in FIG. 11 is -fs / 2 to fs / 2. In FIG. 11, the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0096] FIG. 11(a) illustrates an image of a signal spectrogram without doubling, FIG. 11(b) illustrates an image of a signal spectrogram after doubling, FIG. 11(c) illustrates an image of a signal spectrogram obtained by the processing in step S12, FIG. 11(d) illustrates an image of a signal spectrogram obtained by the processing in step S13, and FIG. 11(e) illustrates an image of a signal spectrogram obtained by the processing in step S14.

[0097] FIG. 12 is an image diagram showing an example of a signal spectrogram obtained by each step of the frequency reproduction processing unit 220 when a digital radio signal r(t) sampled at a sampling rate fs is subjected to a doubling process by the non-linear processing unit 100 for a signal having a center frequency of -fs / 4. The frequency range in FIG. 12 is -fs / 2 to fs / 2. In FIG. 12, the horizontal axis represents frequency (Hz) and the vertical axis represents time (seconds).

[0098] FIG. 12(a) illustrates an image diagram of a signal spectrogram without doubling, FIG. 12(b) illustrates an image diagram of a signal spectrogram after doubling, FIG. 12(c) illustrates an image diagram of a signal spectrogram obtained by the processing in step S12, FIG. 12(d) illustrates an image diagram of a signal spectrogram obtained by the processing in step S13, and FIG. 12(e) illustrates an image diagram of a signal spectrogram obtained by the processing in step S14.

[0099] The frequency regeneration processing unit 220 can match the center frequencies of the signal before and after frequency multiplication on the spectrogram, regardless of the center frequency of the input digital radio signal. Conventionally, matching the spectrogram of a signal required tuning the digital radio signal. If the center frequency of the original signal is unknown, the frequency of the digital radio signal after frequency multiplication is also unknown, making tuning impossible. The frequency regeneration processing unit 220 takes frequency data as input and, through the processing shown in the flowchart in Figure 7, can match the center frequencies of the signal before and after frequency multiplication even without information on the signal's center frequency. Furthermore, this can be achieved with any positive integer multiplication factor, not just 1, 2, 4, or 8.

[0100] The frequency data G obtained by these regeneration processes performed by each frequency regeneration processing unit 220 (#1 to #4) is sent to the signal specification determination unit 300.

[0101] Next, returning to Figure 4, we will explain the identification process (S5) and signal specification determination (S6) performed by the signal specification determination unit 300.

[0102] The signal specification determination unit 300 performs identification processing (S5) on the frequency data G1 to G4 from each frequency regeneration processing unit 220 (#1 to #4). The identification processing includes class classification processing and position regression.

[0103] This section describes an example of using a neural network for specification determination and location regression. Specification determination using a neural network can be performed, for example, using an object detection network based on a CNN (Convolutional Neural Network), as described in Patent Document 1. The object detection network extracts features from the input data and performs classification and location regression. Here, "location" refers to the coordinates on the input spectrogram.

[0104] For example, if the frequency data output from the frequency data generation unit 200 is a spectrogram, the four spectrograms output from the frequency data generation unit 200 can be superimposed in the channel direction, as in an image. In that case, the system can be configured so that all channels have the signal features at the same frequency as the center frequency of the original signal. The output of the object detection network's classification and position regression is a three-dimensional tensor. The horizontal direction represents the change in feature quantities in the frequency direction, the vertical direction represents the change in feature quantities in the time direction, and the channel direction outputs the results of the classification and position information.

[0105] In a CNN's convolutional filter, sum-of-products operations are performed in the channel direction. Therefore, the alignment of the signal features from each multiplication process in the channel direction improves the learning efficiency of the signal specification determination unit 300 (the learning process is not shown in the diagram). In other words, if there were no frequency reconstruction processing unit 220 and the spectrogram of the multiplied signal were used as input to the signal specification determination unit 300, the learning algorithm would need to be adjusted during the learning process so that the signal features from the multiplication process become the center frequency of the original signal, which would increase the learning time. Furthermore, the discrimination performance during discrimination, i.e., the accuracy of discriminating signal specifications and estimating frequency / signal length, would be reduced.

[0106] The signal specification determination unit 300 may use various machine learning methods, such as support vector machines, not limited to CNNs, to classify the frequency data and detect signals appropriately, as long as the method is capable of doing so. In addition, an attention function, such as a transformer, may be provided to mask other frequency bands so that only the frequencies in which signals appear from the frequency data G output by the frequency reproduction processing unit 220 are focused on.

[0107] The modulation parameters of the digital wireless communication signal r(t), which is the target of signal parameter determination and signal detection, include analog modulation such as AM modulation and FM modulation, digital modulation such as PSK modulation and FSK modulation, modulation scheme such as Morse code OOK modulation, modulation depth, modulation index, modulation level, communication speed, frequency shift amount, number of multi-carriers, symbol sequence, and bit pattern, and can be arbitrarily set according to the parameters to be identified.

[0108] The signal specification determination unit 300 determines the signal specifications from the classified class. It also calculates the signal frequency and time from the signal position coordinates on the spectrogram obtained as a result of position regression (S6). As mentioned above, since the class is linked to the signal specifications, the signal specifications can be determined from the class. The frequency and time from the signal position coordinates can be calculated based on the information used by the frequency data generation unit 200 (sampling frequency, FFT shift width, number of FFT points, etc.).

[0109] If the frequency data output from the frequency data generation unit 200 is a spectrum, the signal specification determination unit 300 determines the signal specifications from the classified class. Since the spectrum contains only one-dimensional frequency information, the signal frequency is calculated from the signal position coordinates on the spectrum obtained as a result of position regression. The frequency from the signal position coordinates can be calculated based on the information used by the frequency data generation unit 200 (sampling frequency, number of FFT points, etc.).

[0110] The signal specification determination unit 300 may perform only class classification processing, and a nonlinear processing unit 100 and a frequency data generation unit 200 may be used instead.

[0111] As described above, the signal specification determination device 10 to which the signal specification determination method of this embodiment is applied can align the center frequencies of the multiplied signal spectrograms.

[0112] Furthermore, the augmented frequency conversion unit 210 can further emphasize the characteristics of the signal and maintain the input data size to the signal specification determination unit 300, thereby improving identification performance while reducing the learning and identification time.

[0113] Furthermore, the frequency regeneration processing unit 220 converts the high-frequency components generated by nonlinear processing back into the frequency components of the original digital radio signal. This ensures that signal features are present at the signal appearance frequencies in all of the multiple output frequency data, thereby improving both learning efficiency and discrimination performance.

[0114] Furthermore, a common effect of these methods is that they reduce the memory size required for learning and recognition. This means that the signal parameter determination device 10 does not require high GPU performance and can be implemented using a CPU or embedded FPGA. In addition, the data size used for learning can be reduced, which also reduces the storage capacity of the PC used for learning. The time required to adjust the hyperparameters used for learning can also be shortened.

[0115] (Variation 1) A modified example of the present invention will be described with reference to the drawings.

[0116] Figure 13 is a functional block diagram showing an example of the configuration of a signal specification determination device to which the signal specification determination method in Modification Example 1 is applied.

[0117] Figure 14 is a functional block diagram showing an example configuration of the frequency data generation unit in Modification Example 1.

[0118] The signal specification determination device 10A has a configuration in which the frequency data generation unit 200 of the signal specification determination device 10 is replaced by a frequency data generation unit 200A. The frequency data generation unit 200A has a configuration in which the expanded frequency conversion unit 210 is replaced by a frequency conversion unit 215. Since the other components of the signal specification determination device 10A are the same as those of the signal specification determination device 10, in Figures 13 and 14, the same components as in Figures 1 and 3 are represented by the same reference numerals, and redundant explanations of components with the same reference numerals will be avoided below.

[0119] As illustrated in Figure 14, if the frequency data generation unit 200A has four systems, the frequency data generation unit 200A includes frequency conversion units 215 (#1~#4) and frequency regeneration processing units 220 (#1~#4) provided one-to-one with each conversion unit 120 (#1~#4) in order to convert each digital radio signal output from each conversion unit 120 (#1~#4) into frequency data.

[0120] Next, we will explain an example of the operation of the signal specification determination device 10A.

[0121] Figure 15 is a flowchart showing an example of the operation of a signal specification determination device to which the signal specification determination method of Modification Example 1 is applied.

[0122] The flowchart shown in Figure 15 is a modified version of the flowchart in Figure 4, with step S3 replaced by step S31. Since the processes of the other steps are the same as those in Figure 4, the same step numbers are used to avoid redundant explanations.

[0123] In other words, in step S31, each frequency conversion unit 215 (#1 to #4) generates a spectrogram as frequency data H (H1 to H4) from the digital radio signals f(r(t)) (f1(r(t)) to f4(r(t))) output from the corresponding conversion unit 120 (#1 to #4).

[0124] In the expanded frequency conversion unit 210, the number of points in the Fourier transform is expanded by a factor of N and then reduced to 1 / N, but the frequency conversion unit 215 performs the FFT with a predetermined number of points.

[0125] Furthermore, in the Fourier transform of digital radio signals, the number of samples is generally set to a power of two before performing the FFT. However, in this modified example, we are not limited to this; we may also zero-fill samples other than powers of two and then perform the FFT to match the number of samples to a power of two, or we may perform the DFT directly. Moreover, we may estimate the probability density function of the spectrum based on an autoregressive model or the like.

[0126] Furthermore, when performing a Fourier transform, a window function may be applied to the input signal. A window function can also be applied in this modified example. In this case, the type of window function is not limited, and any window function may be applied.

[0127] Furthermore, when generating a spectrogram, a Fourier transform may be applied so that the time intervals of the digital radio signals overlap. In this modified example, the amount of overlap is not limited, and any overlap amount may be set. Moreover, the amount of overlap may be adaptively controlled according to the duration of the signals.

[0128] As explained above, in the signal specification determination device 10A to which the signal specification determination method of Modified Example 1 is applied, the frequency reproduction processing unit 220 converts the high-frequency components by nonlinear processing into the frequency components of the original signal. As a result, signal features are present at the signal appearance frequencies in all of the multiple output frequency data, leading to high learning efficiency and improved discrimination performance.

[0129] (Modification 2) A modified example of the present invention, Part 2, will be described with reference to the drawings.

[0130] Figure 16 is a functional block diagram showing an example configuration of a signal specification determination device to which the signal specification determination method in Modification Example 2 is applied.

[0131] Figure 17 is a functional block diagram showing an example configuration of the frequency data generation unit in Modification Example 2.

[0132] The signal specification determination device 10B has a configuration in which the frequency data generation unit 200 of the signal specification determination device 10 is replaced by a frequency data generation unit 200B. The frequency data generation unit 200B does not have a frequency regeneration processing unit 220, and is configured to have only an expanded frequency conversion unit 210. Since the other components of the signal specification determination device 10B are the same as those of the signal specification determination device 10, in Figures 16 and 17, the same components as in Figures 1 and 3 are represented by the same reference numerals, and redundant explanations of components with the same reference numerals will be avoided below.

[0133] As illustrated in Figure 17, when the frequency data generation unit 200B has four systems, the frequency data generation unit 200B includes an expanded frequency conversion unit 210 (#1 to #4) provided one-to-one with each conversion unit 120 (#1 to #4) in order to convert each digital radio signal output from each conversion unit 120 (#1 to #4) into frequency data.

[0134] Each augmented frequency conversion unit 210 (#1 to #4) may perform different processing or the same processing for each conversion unit 120 (#1 to #4), and the output obtained by the processing is input to the signal specification determination unit 300.

[0135] Next, an example of the operation of the signal parameter determination device 10B will be described.

[0136] Figure 18 is a flowchart showing an example of the operation of a signal specification determination device to which the signal specification determination method of Modification Example 2 is applied.

[0137] The signal specification determination device 10B does not have a frequency data generation unit 200B with a frequency regeneration processing unit 220. Therefore, the flowchart shown in Figure 18 omits the processing of step S4 in the flowchart shown in Figure 4. The processing of the other steps is the same as the processing of the steps shown in Figure 4, so the same step numbers are used to avoid redundant explanations.

[0138] Since the signal specification determination device 10B does not have a frequency regeneration processing unit 220 in the frequency data generation unit 200B, the output obtained by the processing of the augmented frequency conversion unit 210 is input to the signal specification determination unit 300 without undergoing frequency regeneration processing.

[0139] In response, the signal specification determination unit 300 classifies the image data with each characteristic emphasized into classes as an identification process (S5). Each class is associated with predetermined signal specifications. Therefore, the signal specification determination unit 300 further determines the signal specifications of the digital wireless communication signal r(t) according to the classified class (S6).

[0140] This section describes an example using a neural network for classification. Any method capable of classifying multiple image data sets can be used for classification using a neural network, such as a Deep Neural Network (DNN) using a Convolutional Neural Network (CNN). Furthermore, to handle multiple image data sets, a multi-channel neural network can be applied, or multiple image data sets can be cascaded and processed as a single image data set by the neural network. Other processing steps are the same as those explained using the flowchart in Figure 4.

[0141] As explained above, the signal specification determination device to which the signal specification determination method of Modified Example 2 is applied allows for greater emphasis on signal characteristics by the augmented frequency conversion unit 210, and maintains the input data size to the signal specification determination unit 300. This has the effect of improving identification performance while reducing learning and identification time.

[0142] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of Symbols]

[0143] 10,10A,10B signal specification judgment device 100 Nonlinear Processing Unit 110 Signal distribution section 120 Conversion Unit 200, 200A, 200B Frequency Data Generation Unit 210 Enlarged frequency conversion unit 215 Frequency conversion section 220 Frequency Regeneration Processing Unit 300 Signal specification judgment section

Claims

1. A nonlinear processing unit that transforms the input signal using a nonlinear function, A frequency data generation unit generates frequency data from the converted signal, A signal specification determination unit performs a class classification process on the generated frequency data to classify it into one of the classes associated with predetermined signal specifications, determines the signal specifications of the input signal based on the classified class, and estimates the center frequency and duration of the signal. A signal specification determination device equipped with the following features.

2. The signal parameter determination device according to claim 1, wherein the transformation using the aforementioned nonlinear function includes a multiplication process.

3. The aforementioned nonlinear function includes multiple power functions with different multiplication factors, The nonlinear processing unit includes a conversion unit that converts the input signal using the plurality of power functions to obtain the frequency data. The signal parameter determination device according to claim 2.

4. The frequency data generation unit, An augmented frequency conversion unit performs an augmentation process on the frequency data obtained by the conversion unit to increase the frequency resolution, The frequency regeneration processing unit includes a frequency regeneration processing unit that performs a regeneration process to regenerate the frequency information of the frequency data that has undergone the augmentation processing by the augmentation frequency conversion unit, and outputs the regenerated frequency data to the signal specification determination unit for use in the class classification processing. The signal parameter determination device according to claim 3.

5. The aforementioned frequency expansion unit performs the expansion process as follows: The process involves performing a Fourier transform on the frequency data obtained by the conversion unit using time samples N times the desired frequency resolution, The output spectrogram obtained by the Fourier transform is then reduced to 1 / N in the frequency direction, where N is an arbitrary positive integer. The signal parameter determination device according to claim 4.

6. The frequency regeneration processing unit performs the regeneration process as follows: The frequency data that has been augmented by the augmented frequency conversion unit is subjected to a process of swapping the sign of the frequency, The process of creating a reduced spectrogram by reducing the frequency data that has undergone the aforementioned swapping process in the frequency direction by a reduction ratio determined according to the multiplication factor by the conversion unit, The reduced spectrogram is then processed to concatenate in the frequency direction by the amount of the multiplication factor by the conversion unit. The signal parameter determination device according to claim 4.

7. The signal parameter determination unit performs the class classification process using a neural network. The signal parameter determination device according to claim 4.

8. The signal parameter determination unit uses the neural network to regress the position of the input signal in frequency-time space. The signal parameter determination device according to claim 7.

9. The processor of the signal specification determination device The input signal is transformed using a nonlinear function. Frequency data is generated from the converted signal. The generated frequency data is subjected to a class classification process that classifies it into one of the classes associated with predetermined signal parameters. Based on the classified class, the signal parameters of the input signal are determined, and the center frequency and duration of the signal are estimated. Signal specification determination method.

Images