METHOD AND SYSTEM FOR DETECTING AND/OR CLASSIFYING A WANTED SIGNAL

Abstract

A method for at least one of detecting and classifying a wanted signal in an electromagnetic signal is described. The method includes the following steps: the electromagnetic signal is received; a spectrogram of the electromagnetic signal is determined; at least one correlation parameter, for example a correlation multi-dimensional algebraic object including several correlation parameters, is determined based on the determined spectrogram; the at least one correlation parameter is used as an input for a machine learning module; the wanted signal in the electromagnetic signal is detected and/or classified via the machine learning module based at least partially on the at least one correlation parameter. Further, a signal detection and/or classification system and a computer program are described.

Claims

1. A method for at least one of detecting and classifying a wanted signal in an electromagnetic signal, comprising: receiving the electromagnetic signal; determining a spectrogram of the electromagnetic signal; determining at least one correlation parameter based on the determined spectrogram; using said at least one correlation parameter as an input for a machine learning module; and at least one of detecting and classifying the wanted signal in the electromagnetic signal via the machine learning module based at least partially on the at least one correlation parameter.

2. The method of claim 1, wherein a correlation multi-dimensional algebraic object comprising several correlation parameters is determined.

3. The method of claim 1, wherein a time evolution of the spectrogram of the electromagnetic signal is determined.

4. The method according to claim 1, wherein the correlation parameter is a frequency correlation parameter.

5. The method according to claim 1, wherein the correlation parameter is a frequency-to-frequency correlation coefficient.

6. The method according to claim 5, wherein the frequency-to-frequency correlation coefficient is a Pearson correlation coefficient.

7. The method according to claim 1, wherein a correlation multi-dimensional algebraic object of at least third order is determined.

8. The method according to claim 7, wherein entries of the correlation multi-dimensional algebraic object are determined that correspond to frequency-to-frequency correlation coefficients.

9. The method according to claim 8, wherein at least one of an observed frequency range of the electromagnetic signal is divided into several frequency portions in order to determine the correlation multi-dimensional algebraic object, and an observed time interval is divided into several time intervals in order to determine the correlation multi-dimensional algebraic object.

10. The method according to claim 7, wherein the correlation multi-dimensional algebraic object is unfolded, thereby obtaining at least one unfolded correlation multi-dimensional algebraic object.

11. The method of claim 10, wherein the at least one unfolded correlation multi-dimensional algebraic object is used as an input for the machine learning module.

12. The method according to claim 11, wherein higher order singular values of the correlation multi-dimensional algebraic object are determined based on the at least one unfolded correlation multi-dimensional algebraic object, and wherein the higher order singular values of the correlation multi-dimensional algebraic object are used as an input for the machine learning module.

13. The method according to claim 1, wherein an image processing technique is applied to the at least one correlation parameter in order to at least one of detect and classify the wanted signal in the electromagnetic signal.

14. The method according to claim 1, wherein an image processing technique is applied to the determined spectrogram in order to at least one of detect and the wanted signal in the electromagnetic signal, and wherein said at least one correlation parameter serves as metadata for the machine learning module.

15. The method according to claim 1, wherein the machine learning module comprises an artificial neural network that is trained to at least one of detect and classify the wanted signal in the electromagnetic signal based on the determined spectrogram.

16. The method according to claim 15, wherein the artificial neural network is a convolutional neural network.

17. The method according to claim 16, wherein the artificial neural network is a deep convolutional neural network.

18. A signal detection and/or classification system, comprising a receiver and an analysis circuit, wherein the receiver is configured to receive electromagnetic signals, wherein the analysis circuit is configured to process the electromagnetic signals, and wherein the signal detection and/or classification system is configured to perform the method according to claim 1.

19. A computer readable medium comprising executable computer instructions adapted to cause a signal detection and/or classification system to perform the method according to claim 1 when the computer instructions are executed on a one or more computing devices of the signal detection and/or classification system.

Description

DESCRIPTION OF THE DRAWINGS

[0046] The foregoing aspects and many of the attendant advantages of the claimed subject matter will become more readily appreciated as the same become better understood by reference to the following detailed description, when taken in conjunction with the accompanying drawings, wherein:

[0047] FIG. 1 schematically shows a block diagram of a signal detection and/or classification system according to an embodiment of the present disclosure;

[0048] FIG. 2 shows a flow chart of a representative method for detecting and/or classifying a wanted signal according to an embodiment of the present disclosure;

[0049] FIG. 3 shows a first spectrogram of a first electromagnetic signal;

[0050] FIG. 4 shows a plot of a frequency-to-frequency correlation parameter of the spectrogram of FIG. 3;

[0051] FIG. 5 shows a second spectrogram of a second electromagnetic signal;

[0052] FIG. 6 shows a plot of a frequency-to-frequency correlation parameter of the spectrogram of FIG. 5; and

[0053] FIG. 7 shows an illustration of a correlation multi-dimensional algebraic object.

DETAILED DESCRIPTION

[0054] The detailed description set forth below in connection with the appended drawings, where like numerals reference like elements, is intended as a description of various embodiments of the disclosed subject matter and is not intended to represent the only embodiments. Each embodiment described in this disclosure is provided merely as an example or illustration and should not be construed as preferred or advantageous over other embodiments. The illustrative examples provided herein are not intended to be exhaustive or to limit the claimed subject matter to the precise forms disclosed.

[0055] FIG. 1 schematically shows a block diagram of a signal detection and/or classification system 10 that comprises a receiver 12 and an analysis circuit or module 14. Therein and in the following, the term “module” is understood to describe suitable hardware, suitable software, or a combination of hardware and software that is configured to have a certain functionality. The hardware may, inter alia, comprise a CPU, a GPU, an FPGA, an ASIC, or other types of electronic circuitry. The signal detection and/or classification system 10 may be part of a measurement instrument such as an oscilloscope or another type of measurement instruments that is used for signal detection and/or classification.

[0056] In the embodiment shown, the receiver 12 comprises a radio frequency (RF) antenna 16. Generally speaking, the receiver 12 or rather the antenna 16 is configured to receive an electromagnetic signal and to forward the received electromagnetic signal to the analysis module 14. The receiver 12 may further be configured to digitize the electromagnetic signal, such that a digitized electromagnetic signal is generated and forwarded to the analysis module 14.

[0057] The analysis module 14 comprises a central processing unit (CPU) 18 and a memory 20. Moreover, the term “central processing unit” is understood to comprise all suitable types of machine processing units. In some embodiments, the central processing unit 18 may be established as a graphics processing unit (GPU).

[0058] A computer program comprising a machine learning module 21 is stored in the memory 20. The computer program comprises program code means that are adapted to cause the signal detection and/or classification system 10 to perform a method for detecting and/or classifying a wanted signal in the electromagnetic signal described in the following with reference to FIG. 2, when the computer program is executed on the central processing unit 18.

[0059] Therein and in the following, the term “program code means” is understood to comprise machine-readable instructions in compiled and/or uncompiled form being provided in any suitable programming language and/or machine language.

[0060] More specifically, the machine learning module 21 comprises an artificial neural network that is trained to detect and/or classify the wanted signal in the electromagnetic signal in the way described in the following.

[0061] Therein and in the following, the term “detect the wanted signal” is understood to mean that a signal position regarding time and frequency in the spectrogram is localized.

[0062] Moreover, the term “classify the wanted signal” is generally understood to mean that certain pieces of information about the signal are determined that are associated with a certain class of electromagnetic signals. For example, a modulation type of the wanted signal and/or a symbol rate of the wanted signal are determined.

[0063] In some embodiments, the artificial neural network is established as a convolutional neural network, preferably as a deep convolutional neural network.

[0064] First, the electromagnetic signal is received by the antenna 16 and the receiver 12 and is forwarded to the analysis module 14 (step S1).

[0065] The analysis module 14 determines a spectrogram of the received electromagnetic signal (step S2). More precisely, a time evolution of the spectrogram is determined.

[0066] In the following, the term “spectrogram” is understood to comprise both a “snapshot” of the spectrum of the electromagnetic signal at a particular time and a time evolution of the spectrum. Put differently, the spectrogram may be the spectrum of the electromagnetic signal at a particular time or the time evolution of the spectrum.

[0067] Generally speaking, the spectrogram is a function of frequency and time that associates a logarithmic power level of the electromagnetic signal with the respective frequency bin and the respective time bin.

[0068] The spectrogram is determined for a predetermined frequency range. More specifically, the spectrogram is determined in a frequency range where the wanted signal is supposed or at least assumed to be.

[0069] The result of step S2 for an exemplary electromagnetic signal is shown in FIG. 3, where the spectrogram is visualized as a heat map, which is also called “waterfall diagram”.

[0070] In the example of FIG. 3, the spectrogram is shown for frequency bins ranging from 1500 to 2200. Each frequency bin represents a certain frequency range. More precisely, each frequency bin is associated with a certain frequency range having a certain bandwidth in the complex baseband. For example, each frequency bin may have a bandwidth of 30 to 50 Hz, for example of 35 to 45 Hz.

[0071] In FIG. 3, the time is plotted on the vertical axis, the frequency is plotted on the horizontal axis and the respective amplitude of the electromagnetic signal is visualized by the color of the respective points in the frequency-time diagram.

[0072] In some embodiments, the amplitude is plotted on a logarithmic scale. In the example shown in FIG. 3, the scale is 20.Math.log.sub.10 (Amplitude) [dB].

[0073] In other words, the spectrogram shows a logarithmic power trend of the (digitized) electromagnetic signal over time, for example over N temporal realizations, and over the predetermined frequency range.

[0074] The analysis module 14 then determines at least one correlation parameter or rather a correlation multi-dimensional algebraic objects that comprises several correlation parameters based on the determined spectrogram (step S3).

[0075] Some specific examples for the correlation multi-dimensional algebraic object will be given in the following. However, it is to be understood that every other suitable type of correlation parameter could be used as well.

[0076] One example for the correlation multi-dimensional algebraic object is a frequency-to-frequency correlation matrix ρ.sub.ij=ρ(A(f.sub.i),A(f.sub.j)) that is calculated for several, for example all frequency pairs f.sub.i, f.sub.j in the predetermined frequency range. Therein, A(f.sub.i) is the amplitude of the electromagnetic signal at frequency f.sub.i.

[0077] Thus, a correlation multi-dimensional algebraic object of order two is determined, namely the frequency-to-frequency correlation matrix ρ.sub.ij.

[0078] The frequency-to-frequency correlation matrix is calculated according to the following formula:

[00001]${\rho }_{i.Math.j}=\frac{1}{\left(N-1\right)}.Math.\underset{k=1}{\overset{N}{.Math.}}.Math.\frac{\left(A_k\left(f_i\right)-{\mu }_{A\left(f_i\right)}\right).Math.\left(A_k\left(f_j\right)-{\mu }_{A\left(f_j\right)}\right)}{{\sigma }_{A\left(f_i\right)}.Math.{\sigma }_{A\left(f_j\right)}}.$

[0079] Like above, N is the number of temporal realizations. Further, μ.sub.A(f.sub.i) is the expectation value of the amplitude of the electromagnetic signal at frequency f.sub.i and σ.sub.A(f.sub.i) is the variance of the amplitude of the electromagnetic signal at frequency f.sub.i.

[0080] In other words, a frequency-to-frequency correlation coefficient, more precisely the Pearson correlation coefficient ρ is determined for several, for example for all frequency pairs f.sub.i, f.sub.j.

[0081] The result of step S3 is shown in FIG. 4, which shows a plot or rather a heat map of the determined frequency-to-frequency correlation matrix. Therein, the frequencies f.sub.i and f.sub.j are plotted on the horizontal and on the vertical axis, respectively. The respective value ρ.sub.ij of the correlation coefficient is depicted by the color of the respective point in the diagram.

[0082] Generally speaking, the determined correlation multi-dimensional algebraic object is then used as an input for the machine learning module 21 (step S4)

[0083] In some embodiments, the determined correlation parameters, i.e. the entries of the determined frequency-to-frequency correlation matrix ρ.sub.ij are then used as an input for the machine learning module 21.

[0084] Finally, the machine learning module 21, for example the artificial neural network, detects and/or classifies the wanted signal in the electromagnetic signal at least partially based on the determined correlation multi-dimensional algebraic object (step S5).

[0085] In the example described above, the artificial neural network detects and/or classifies the wanted signal based on the determined correlation parameters, i.e. based on the determined frequency-to-frequency correlation matrix p.sub.ij.

[0086] Generally speaking, there are two possible variants how the machine learning module 21 or rather the artificial neural network can detect and/or classify the wanted signal in step S5.

[0087] According to a first variant of the signal detection and/or classification method, an image is generated based on the determined correlation parameters, for example a heat map of the correlation coefficients. An example for such a heat map is depicted in FIG. 4, as already discussed above.

[0088] The machine learning module 21 or rather the artificial neural network then applies image processing and/or image recognition techniques to the generated image in order to detect and/or classify the wanted signal comprised in the electromagnetic signal.

[0089] It has turned out that several different classes of wanted signals exhibit similar correlation patterns, i.e. the image generated based on the determined correlation coefficients are similar to each other.

[0090] Accordingly, the number of different correlation patterns that the artificial neural network has to be trained for is reduced. Thus, smaller artificial neural networks can be used, which results in an enhanced processing speed for the detection and/or classification according to this variant of the method.

[0091] Thus, this method is particularly suitable for detecting the wanted signal or for classifying the wanted signal if a rough classification is sufficient.

[0092] Moreover, the correlation parameters and thus also the respective generated image comprise information on signal collisions, i.e. about a superposition of two or more signals in the same frequency band. An example of a spectrogram of an electromagnetic signal comprising a signal collision is shown in FIG. 5.

[0093] Such signal collisions usually are not visible in the spectrogram. However, it turned out that such signal collisions can be identified based on the corresponding correlation parameters of the spectrogram as shown in FIG. 6.

[0094] An example for a typical structure corresponding to colliding signals is depicted in the magnified area in FIG. 6.

[0095] According to a second variant of the signal detection and/or classification method, an image is generated based on the determined spectrogram, for example a heat map of the spectrogram like exemplarily shown in FIGS. 3 and 5.

[0096] The machine learning module 21 or rather the artificial neural network then applies image processing and/or image recognition techniques to the generated image in order to detect and/or classify the wanted signal comprised in the electromagnetic signal. Additionally, the determined correlation multi-dimensional algebraic object is used as additional metadata for the machine learning module 21, for example for the artificial neural network.

[0097] Thus, in this variant, the image processing and/or recognition techniques are applied directly to the spectrogram instead of to the image generated based on the correlation parameters. The determined correlation parameters serve as additional information.

[0098] As already explained above, the correlation parameters inter alia comprise information on signal collisions, which might not be resolvable in the spectrogram alone.

[0099] However, by combining the image analysis of the spectrogram with the additional information comprised in the correlation parameters, the information on signal collisions comprised in the electromagnetic signal is recovered.

[0100] In principal, the correlation parameters described above can be used as metadata.

[0101] It turned out that correlation multi-dimensional algebraic objects of third order are particularly suitable for use as metadata.

[0102] An example for such a correlation multi-dimensional algebraic objects of third order is illustrated in FIG. 7. In some embodiments, the correlation multi-dimensional algebraic objects of third order may be a tensor of third order.

[0103] Instead of averaging over all N time realizations as in the definition of ρ.sub.ij given above, the sum or rather the time average is only performed over r time realizations for each entry of the correlation multi-dimensional algebraic object of third order, wherein r is a natural number. In other words, the N time realizations are divided into several time slices that each contain r time realizations.

[0104] Moreover, the predetermined frequency range is divided into several frequency portions that each contain s frequency bins of a total number of F frequency bins, wherein s and F are natural numbers. In other words, each entry of the correlation multi-dimensional algebraic object of third order is associated with s frequency bins. For example, for each entry a sum or an average over the respective s frequency bins is taken.

[0105] The resulting correlation multi-dimensional algebraic object of third order ρ.sub.ij,k has three indices, namely indices i and j each indicating the respective frequency portion, and an index k indicating the respective time slice.

[0106] For example, the entry “ρ.sub.12,3” of the correlation multi-dimensional algebraic object of third order is associated with the frequency portions (f.sub.1 . . . f.sub.s) and (f.sub.s+1 . . . f.sub.2s) as well as with the time slice (t.sub.2r+1 . . . t.sub.3r).

[0107] The determined correlation multi-dimensional algebraic object of third order may then be used as an input for the machine learning module 21, for example as metadata for the machine learning module 21.

[0108] Alternatively, the determined correlation multi-dimensional algebraic object of third order may be unfolded with respect to one of the indices i, j and k. In the unfolding, one dimension of the correlation multi-dimensional algebraic object of third order is kept constant while the other two dimensions are stacked.

[0109] More specifically, the indices i and j each run from 1 to F/s, while the index k runs from 1 to N/r.

[0110] Accordingly, if the multi-dimensional algebraic object of third order is unfolded with respect to index i, the result is a the multi-dimensional algebraic object of second order, i.e. a matrix, with F/s entries for the first index and with F/s.Math.N/r entries for the second index.

[0111] Similarly, if the multi-dimensional algebraic object of third order is unfolded with respect to index j, the result is a the multi-dimensional algebraic object of second order, i.e. a matrix, with F/s.Math.N/r entries for the first index and with F/s entries for the second index.

[0112] Finally, if the multi-dimensional algebraic object of third order is unfolded with respect to index k, the result is a the multi-dimensional algebraic object of second order, i.e. a matrix, with F.sup.2/s.sup.2 entries for the first index and with N/r entries for the second index.

[0113] The unfolded multi-dimensional algebraic object may then be used as an input for the machine learning module, for example as metadata for the machine learning module 21.

[0114] Alternatively, a higher order singular values of the correlation multi-dimensional algebraic object are determined based on at least one, for example based on all of the unfolded correlation multi-dimensional algebraic objects described above.

[0115] For this purpose, a singular value decomposition may be determined for each of the unfolded correlation multi-dimensional algebraic objects and the correlation multi-dimensional algebraic objects of third order may be multiplied with the respective singular vector of every unfolded correlation multi-dimensional algebraic object.

[0116] Then, the higher order singular values may be used as metadata for the machine learning module 21, for example for the artificial neural network.

[0117] The signal detection and/or classification system 10, including such components as, for example, the analysis module 14, the machine learning module 21, etc., is configured to perform one or more steps schematically shown, for example, in FIG. 2. In some embodiments, one or more of these components includes one or more computer-readable media containing computer readable instructions embodied thereon that, when executed by one or more computer circuits (contained in or associated with the signal detection and/or classification system 10 or components thereof), cause the one or more computer circuits to perform one or more steps of the method of FIG. 2. In some embodiments, the one or more computer circuits includes a microprocessor, a microcontroller, a central processing unit, a graphics processing unit (GPU), a digital signal processor (DSP), etc.

[0118] In some embodiments, the one or more computer-readable media contains computer readable instructions embodied thereon that, when executed by the one or more computer circuits, sometimes referred to as computing devices, cause the one or more computer circuits to perform one or more steps of any of the methods of Claims 1-17.

[0119] As described briefly above, certain embodiments disclosed herein utilize circuitry (e.g., one or more circuits) in order to implement standards, protocols, methodologies or technologies disclosed herein, operably couple two or more components, generate information, process information, analyze information, store information, display information, generate signals, encode/decode signals, convert signals, transmit and/or receive signals, control other devices, etc. Circuitry of any type can be used. It will be appreciated that the term “information” can be use synonymously with the term “signals” in this paragraph.

[0120] In an embodiment, circuitry includes, among other things, one or more computing devices or computer circuits such as a processor (e.g., a microprocessor), a central processing unit (CPU), a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), a system on a chip (SoC), or the like, or any combinations thereof, and can include discrete digital or analog circuit elements or electronics, or combinations thereof. In an embodiment, circuitry includes hardware circuit implementations (e.g., implementations in analog circuitry, implementations in digital circuitry, and the like, and combinations thereof).

[0121] In an embodiment, circuitry includes combinations of circuits and computer program products having software or firmware instructions stored on one or more computer readable memories that work together to cause a device to perform one or more protocols, methodologies or technologies described herein. In an embodiment, circuitry includes circuits, such as, for example, microprocessors or portions of microprocessor, that require software, firmware, and the like for operation. In an embodiment, circuitry includes one or more processors or portions thereof and accompanying software, firmware, hardware, and the like.

[0122] In some examples, the functionality described herein can be implemented by special purpose hardware-based computer systems or circuits, etc., or combinations of special purpose hardware and computer instructions.

[0123] Of course, in some embodiments, two or more of the modules, units, etc., described above, or parts thereof, can be integrated or share hardware and/or software, circuitry, etc. In some embodiments, these components, or parts thereof, may be grouped in a single location or distributed over a wide area. In circumstances were the components are distributed, the components are accessible to each other via communication links.

[0124] The principles, representative embodiments, and modes of operation of the present disclosure have been described in the foregoing description. However, aspects of the present disclosure which are intended to be protected are not to be construed as limited to the particular embodiments disclosed. Further, the embodiments described herein are to be regarded as illustrative rather than restrictive. It will be appreciated that variations and changes may be made by others, and equivalents employed, without departing from the spirit of the present disclosure. Accordingly, it is expressly intended that all such variations, changes, and equivalents fall within the spirit and scope of the present disclosure, as claimed.