Co-channel signal classification using deep learning

Abstract

One or more aspects of the present disclosure are directed to a method, device, and computer-readable medium for receiving, at a receiver, a signal, the signal including a cover signal and an embedded co-channel anomaly signal; performing, at the receiver, signal processing on the signal to determine one or more features of the signal; inputting, at the receiver, the one or more features into one or more trained neural networks; and receiving, as an output of the trained neural networks, a classification of the signal, wherein the classification distinguishes between the cover signal and the embedded co-channel anomaly signal.

Description

[Technical Field]

[0001] (Related Applications) This application claims priority to U.S. Provisional Application No. 68 / 407,367, filed September 16, 2022, entitled "Co-Channel Signal Classification Using Cyclostationary Signal Processing and Deep Learning," which claims priority to U.S. Application No. 18 / 369,586, filed September 18, 2023, entitled "Co-Channel Signal Classification Using Deep Learning," the entire contents of which are incorporated herein by reference.

[0002] (Statement Regarding Federally Sponsored Research or Development) This application was made with government support under Intelligence Advanced Research Projects Activity (IARPA) Contract No. 2021-21062400004. The U.S. Government has certain rights in this invention.

[0003] The subject matter of this disclosure relates generally to the field of wireless network operation, and more particularly to signal classification in the presence of co-channel interference using deep learning techniques, which may also be referred to as signal-on-signal detection and characterization. [Background technology]

[0004] Wireless broadband represents a key element of economic growth, job creation, and global competitiveness as consumers increasingly use wireless broadband services to support their daily lives. Demand for wireless broadband services and the network capacity associated with those services is rapidly increasing, resulting in the development of a variety of systems and architectures capable of meeting this demand.

[0005] In congested airspace, multiple different signals may be transmitted simultaneously over the same channel, and separating the desired signal from unwanted interfering signals at a receiver device is an ever-present challenge that needs to be addressed. Summary of the Invention

[0006] One or more aspects of the present disclosure are directed to identifying anomalous signals that attempt to use existing signals as cover or operate in close proximity to such cover signals (e.g., Snugglers). As further described, signal processing combined with machine learning techniques is proposed to classify a given signal and identify co-channel anomalous signals. The power spectral density and cyclostationary signal processing features of the captured signal are calculated and fed into a neural network to generate a classification decision.

[0007] In one aspect, a method includes receiving, at a receiver, a signal including a cover signal and an embedded co-channel anomaly signal; performing, at the receiver, signal processing on the signal to determine one or more features of the signal; inputting, at the receiver, the one or more features to one or more trained neural networks; and receiving, as an output of the trained neural networks, a classification of the signal, the classification distinguishing between the cover signal and the embedded co-channel anomaly signal.

[0008] In another embodiment, the one or more signal features include a power spectral density of the signal, a conjugate cycle frequency of the signal, and a non-conjugate cycle frequency of the signal.

[0009] In another embodiment, at least one of the one or more features is input into a trained neural network.

[0010] In another embodiment, all of the one or more features are input into a trained neural network.

[0011] In another aspect, the method further includes performing multimodal fusion to combine outputs of at least two of the trained neural networks to determine an output. Fusion of outputs from different neural networks can be performed to improve results.

[0012] In another aspect, the coverage signal is one of a Long Term Evolution (LTE), a 3GPP 5G signal, Wi-Fi, a Digital Video Broadcast (DVB), or an Advanced Television Systems Committee-Digital Television (ATSC-DTV) signal.

[0013] In another aspect, the co-channel anomaly signal is one of a direct sequence spread spectrum (DSSS) signal, a single carrier signal using binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), quadrature amplitude shift keying (QAM), amplitude phase shift keying (APSK) modulation, a chirp modulated signal, a frequency modulation (FM) signal, a frequency shift keying (FSK) signal, an orthogonal frequency division multiplexing (OFDM) signal, a burst signal, a frequency hopping spread spectrum signal (FHSS), or a Gaussian minimum shift keying (GMSK) signal.

[0014] In another aspect, the trained neural network is trained using a combination of over-the-air captured signals injected with synthetic co-channel anomaly signals.

[0015] In one aspect, a wireless network receiver includes one or more memories containing computer-readable instructions and one or more processors configured to execute the computer-readable instructions to receive signals, the signals including a cover signal and an embedded co-channel anomaly signal, perform signal processing on the signals to determine one or more features of the signals, input the one or more features to one or more trained neural networks, and receive as an output of the trained neural networks a classification of the signals, the classification distinguishing between the cover signal and the embedded co-channel anomaly signal.

[0016] In one aspect, one or more non-transitory computer-readable media include computer-readable instructions that, when executed by one or more processors of a wireless network receiver, cause the wireless network receiver to receive a signal including a cover signal and an embedded co-channel anomaly signal, perform signal processing on the signal to determine one or more features of the signal, input the one or more features to one or more trained neural networks, and receive as an output of the trained neural networks a classification of the signal, wherein the classification distinguishes between the cover signal and the embedded co-channel anomaly signal. [Brief explanation of the drawings]

[0017] The details of one or more aspects of the subject matter described in this disclosure are set forth in the accompanying drawings and the following description. However, the accompanying drawings illustrate only some typical aspects of the disclosure and therefore should not be considered limiting of its scope. Other features, aspects, and advantages will become apparent from the description, drawings, and claims.

[0018] To explain how the above-mentioned and other advantages and features of the present disclosure can be obtained, a more particular description of the principles briefly described above will be made by reference to specific embodiments thereof that are illustrated in the accompanying drawings. With the understanding that these drawings depict only exemplary embodiments of the present disclosure and therefore should not be considered limiting of its scope, the principles herein will be described and explained with added specificity and detail through the use of the accompanying drawings in which:

[0019] [Figure 1] FIG. 1 illustrates an exemplary environment in which wireless communication may occur in accordance with some aspects of the present disclosure.

[0020] [Figure 2] FIG. 2 provides a visual depiction of a hybrid signal processing and machine learning method for signal classification according to some aspects of the present disclosure.

[0021] [Figure 3] FIG. 3 is a visual representation of a composite signal including a cover signal and a co-channel anomaly signal in accordance with some aspects of the present disclosure.

[0022] [Figure 4] FIG. 4 shows representative uncoupled and coupled cycle domain profiles (CDPs) of four exemplary class types of signals, according to some embodiments of the present disclosure;

[0023] [Figure 5] FIG. 5 illustrates parameter distributions for synthetic anomalies according to some embodiments of the present disclosure;

[0024] [Figure 6] FIG. 6 shows accuracy results of a trained neural network for signal classification according to some embodiments of the present disclosure;

[0025] [Figure 7A]7A-C illustrate an example architecture with multimodal fusion and associated accuracy results according to some aspects of the present disclosure; [Figure 7B] 7A-C illustrate an example architecture with multimodal fusion and associated accuracy results according to some aspects of the present disclosure; [Figure 7C] 7A-C illustrate an example architecture with multimodal fusion and associated accuracy results according to some aspects of the present disclosure;

[0026] [Figure 8] FIG. 8 illustrates improvement in classifier accuracy according to some aspects of the present disclosure;

[0027] [Figure 9] FIG. 9 shows accuracy results for a persistent GMSK snagler according to some aspects of the present disclosure;

[0028] [Figure 10] FIG. 10 illustrates an example neural network that may be trained to perform interference signal detection and classification and / or interference mitigation schemes according to some aspects of the present disclosure;

[0029] [Figure 11] FIG. 11 is an exemplary flowchart of a method of signal classification according to some aspects of the present disclosure.

[0030] [Figure 12] FIG. 12 illustrates an exemplary computing system according to some aspects of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0031] (Detailed explanation) Various embodiments of the present disclosure are described in detail below. While specific embodiments are discussed, it should be understood that this is done for illustrative purposes only. Those skilled in the art will recognize that other components and configurations may be used without departing from the spirit and scope of the present disclosure. Accordingly, the following description and drawings are illustrative and should not be construed as limiting. Numerous specific details are set forth to provide a thorough understanding of the present disclosure. However, in some instances, well-known or conventional details are not described in order to avoid obscuring the description.

[0032] References to "one embodiment" or "an embodiment" in this disclosure may be to the same embodiment or any embodiment, and such references refer to at least one of the embodiments. References to "one embodiment" or "an embodiment" mean that a particular feature, structure, or characteristic described in connection with an embodiment is included in at least one embodiment of the present disclosure. The appearances of the phrase "in one embodiment" in various places in the specification are not necessarily all referring to the same embodiment, nor are they separate or alternative embodiments mutually exclusive of other embodiments. Furthermore, various features are described that may be exhibited by some embodiments and not by others.

[0033] The terms used herein generally have their ordinary meaning in the art, within the context of this disclosure and in the specific context in which each term is used. Alternative language and synonyms may be used for any one or more of the terms discussed herein, and no particular importance should be placed on whether a term is recited or discussed herein. In some cases, synonyms for a particular term are provided. The listing of one or more synonyms does not exclude the use of other synonyms. The use of examples anywhere in this specification, including examples of any term discussed herein, is illustrative only and is not intended to further limit the scope and meaning of the disclosure or any example term. Similarly, the present disclosure is not limited to the various embodiments provided herein.

[0034] While not intended to limit the scope of the present disclosure, examples of instruments, devices, methods, and their related results according to embodiments of the present disclosure are provided below. It should be noted that titles or subtitles may be used in the examples for the convenience of the reader, and should not be used to limit the scope of the present disclosure. Unless otherwise defined, technical and scientific terms used herein have the meanings commonly understood by those skilled in the art to which this disclosure belongs. In case of conflict, the present specification, including definitions, will control.

[0035] Additional features and advantages of the present disclosure will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by the practice of the principles disclosed herein. The features and advantages of the present disclosure may be realized and obtained by means of the instruments and combinations particularly pointed out in the appended claims. These and other features of the present disclosure will become more fully apparent from the following description and the appended claims, or may be learned by the practice of the principles described herein.

[0036] As noted above, in congested airspace, multiple different signals may be transmitted simultaneously over the same channel (or over finite and limited available bandwidth), but separating the desired signal from unwanted interfering signals at the receiver device is an ever-present challenge that needs to be addressed.

[0037] An anomalous co-channel signal can attempt to use an existing signal (the intended or desired signal) as cover. There are at least two ways this can be done. One approach is to transmit the anomalous signal at lower power within the full bandwidth of the cover signal and counter the interference from the cover signal by using spreading. Another approach is to transmit a narrowband signal, called a "snuggler," placed at a frequency at the edge of the cover signal's spectral occupancy. The logic behind this is that low-resolution energy detection methods will not easily identify the presence of the anomalous signal because it will appear to merge with the cover signal.

[0038] The cover signal may be a signal including, but not limited to, a Long Term Evolution (LTE) signal, an Advanced Television System Committee Digital Television (ATSC-DTV) signal, etc. These signals may be continuously transmitted at a known frequency in a given geographic location. The underlay signal (e.g., the anomaly signal) may be a direct sequence spread spectrum (DSSS) signal with phase shift keying / quadrature amplitude modulation (PSK / QAM) modulation, pulse-shaping filters, and many options for spreading sequences. The underlay signal remains co-channel with the cover signal and occupies a significant percentage of the bandwidth of the cover signal. The snagler signal may be any narrowband signal with modulation such as PSK / QAM or Gaussian minimum shift keying (GMSK) modulation and with a bandwidth that is a significant portion of the cover signal. The underlay signal may be placed as close as possible to the cover signal to prevent any spectral leakage from causing performance-reducing interference to the snagler signal.

[0039] As described below, one or more aspects of the present disclosure are directed to identifying the presence and type of anomalous signal (e.g., underlay / DSSS signal, snagler, narrowband signal, etc.) for a given known cover signal.

[0040] FIG. 1 illustrates an example environment in which wireless communications according to some aspects of the present disclosure may occur. The non-limiting example environment 102 may be any medium or environment in which terrestrial and / or non-terrestrial wireless communications may occur. The types of wireless communications may include, but are not limited to, satellite or radar communications, cellular technology-based wireless communications (e.g., 4G, LTE, 5G, etc.), known or to-be-developed WiFi-based communications, etc. As known, any one unknown or to-be-developed type of wireless communications may utilize licensed and / or unlicensed bands for the transmission and reception of signals. Each wireless communication system may operate according to an associated standard established and agreed upon for such wireless communication system (e.g., the IEEE 802.11x standard for WiFi).

[0041] The frequency spectrum available for wireless communication does not grow linearly with the ever-increasing number of devices and systems that communicate using wireless communication methods. Thus, as spectrum availability becomes scarcer and more limited, multi-system or multi-user communication becomes more prevalent, where a given frequency band and channel is used to simultaneously transmit multiple signals (operating according to the same or different types of communication methods).

[0042] For example, multiple exemplary wireless communication systems may operate in the environment 102 of Figure 1. Different types of transmitters (grouped as transmitters 104) may be present. The transmitters 104 may include satellites 106, eNode-Bs 108, and WiFi routers 110. The number and types of transmitters 104 are not limited to those shown in Figure 1. There may be more than one of each of the types of transmitters shown as part of the transmitter 104 (e.g., two or more satellites 106, two or more eNode-Bs 108, two or more WiFi routers 110, etc.).

[0043] Environment 100 may further include receiver 112. Receiver 112 may include satellite receiver 114 capable of transmitting or receiving radar signals to and from satellite 106. Receiver 112 may further include mobile devices 116, receivers 118, etc., each of which may be capable of receiving and / or transmitting wireless signals according to any one or more wireless communication protocols. The types and number of receivers are not limited to those shown in FIG. 1 and may include any number of the same types of receivers shown, and / or any other types of known or to-be-developed devices capable of transmitting and receiving wireless signals.

[0044] Any one of the example receivers 112 may be configured to operate based on more than one type of wireless communication scheme. For example, the mobile device 116 may operate using cellular and WiFi technologies, and the receiver 118 may operate based on radar, cellular, and / or WiFi technologies.

[0045] Various transmitted radio signals transmitted in the environment 102 by any one of the transmitters 104 for reception by an intended one or more of the receivers 112 are shown as exemplary signals 120, 122, and 124.

[0046] In one example, any one of the transmitters 104 can also act as a receiver, and similarly, any one of the receivers 112 can act as a transmitter.

[0047] As mentioned above, as frequency spectrum becomes more limited and scarce due to increasing demand, a single frequency channel may be utilized by more than one system for signal transmission, thus resulting in simultaneous use of the channel, which may lead to multi-user interference (MUI). Various techniques have been introduced to avoid MUI (e.g., channel sniffing to determine if a particular frequency channel is available, and if not, performing a random backoff until the channel becomes available).

[0048] In wireless communications, signal classification is a challenging problem that, at its core, involves mapping vectors of captured quadrature (IQ) data to labels. Methods for doing this include both statistical signal processing and machine learning techniques.

[0049] Signal processing techniques such as cyclostationary signal processing (CSP) allow for well-conditioned statistical estimation and signal counts, and perform well for signal classification when their complexity is low. CSP takes a high-dimensional IQ vector and maps it to a lower-dimensional set of CSP features, including cycle frequency (CF) and spectral correlation and coherence values. These CSP features are then interpreted in several ways to assign a signal type label. Another common machine learning approach to signal classification is to use raw IQ data and pass it through a neural network, such as a convolutional neural network (CNN), that has been trained on prior data.

[0050] Aspects of the present disclosure propose a hybrid signal processing and machine learning methodology for signal classification by computing cyclostationary signal processing features and using them as input to a trained neural network architecture instead of raw IQ data.

[0051] FIG. 2 provides a visual depiction of a hybrid signal processing and machine learning method for signal classification, according to some aspects of the present disclosure.

[0052] In one aspect, the parameter space underlying the signals being considered is lower than the dimension of the captured IQ space, which enables preservation of most of the relevant parameter information and, at the same time, reduces the dimensional size of the feature space through power spectral density (PSD) and CSP processing.

[0053] Flow 200 includes several stages for generating hybrid signal processing and machine learning solutions for signal classification. Stage 202 is the dataset generation stage. In stage 202, a dataset of IQ files can be generated using both the captured signals and synthetic signals controlled by underlying parameterization (e.g., modulation, power, noise floor, symbol rate, center frequency, etc.). And a high-dimensional IQ space of a given length N can be generated. The dataset can be transmitted and received in a wireless receiver (e.g., one of receivers 112). In stage 204, signal processing techniques can be applied to generate a lower-dimensional feature space. This lower-dimensional feature space may be generated by calculating a PSD having a configurable frequency resolution that results in M points in the frequency space, where M << N. And non-conjugate and conjugate (CSP) features (CFs) can be calculated. In stage 206, the PSD and conjugate and non-conjugate CFs can be fed to a trained neural network (e.g., a neural network having fully connected layers), and its output can provide a classification decision (signal classification) that classifies the signal as DTV, DTV+DSSS, DTV+snagla (GMSK), DTV+DSSS+snagla (GMSK), etc.

[0054] The generation of the dataset in stage 202 can be as follows. A combination of over-the-air (OTA) captured ATSC-DTV signals can be used. These signals are normalized to unit power and have a length of 2 nThe ATSC-DTV signal and the composite signal may be combined to generate a composite signal according to selected parameter selections. The composite signal may be represented as follows: S(t)= SDTV (t)+ SDSSS(t, α~DSSS)+ SGMSK(t, α~GMSK)+ N(t) (1)

[0055] In equation (1), SDTV is the normalized OTA ATSC-DTV signal, SDSSS is the DSSS signal controlled by the parameter α ∼ DSSS, SGMSK is the GMSK signal with controllable parameter α ∼ GMSK, and N is a controllable noise signal.

[0056] 3 is a visual representation of a composite signal including a cover signal and a co-channel anomaly signal according to some embodiments of the present disclosure. Signal 300 in FIG. 3 is a non-limiting example of signal S(t) of Equation (1) in the frequency domain. There are seven controllable parameters for generating the composite signal, including noise power, DSSS power, DSSS bandwidth, DSSS center frequency, GMSK power, GMSK bandwidth, and GMSK center frequency. By setting the power of either or both of the DSSS and GMSK signals, four classes of signals can be generated: DTV, DTV+DSSS, DTV+GMSK, and DTV+DSSS+GMSK.

[0057] Signal 302 in FIG. 3 shows power over the frequencies of a cover signal (eg, an ATSC-DTV signal) with lower power anomalous signals 304 and snagler signals 306 at the edges of the cover signal's spectral occupancy.

[0058] The signal processing of step 204 may be as follows: Using the data set of IQ data files generated as step 202, signal processing may be applied to each file to generate features to be used as input to a neural network for signal classification. First, the PSD of the signal S(t) may be determined using, for example, Welch's method with some selected frequency resolution Δf, resulting in M ​​PSD points. Next, a list of CSP4 tuples of the following form may be generated: (f*, α*, s*, c*).

[0059] For a cyclostationary stochastic process x(t), the spectral correlation is defined as: S x α (f) = ∫ -∞ ∞ R x α (τ)e -i2πfτ dτ (2) where Rαx(τ) is the circular autocorrelation function of the stochastic process x(t). Yet another related function is the spectral coherence function (Coh), which is defined in terms of the SCF as: TIFF2025538916000002.tif18158 has the property that its modulus is always less than or equal to 1.

[0060] One of the defining properties of a cyclostationary stochastic process is that the spectral correlation is non-zero only for a set of α values ​​known as the cycle frequency (CF). For a given CF α*, TIFF2025538916000003.tif9157 This produces a CSP 4 tuple as described above.

[0061] For all of the above, these functions are known as the non-conjugate (NC) versions of the SCF and Coh. There are associated second-order cyclostationary functions known as the conjugate (C) spectral correlation function and the conjugate (C) spectral coherence function. These functions have similar properties to the non-conjugate versions.

[0062] For realized cyclostationary stochastic processes, which are many communication signals, the SCF and CF can be estimated using, for example, a strip spectrum correlation analyzer (SSCA). From this, it can be determined that the finite lists of CSP 4-tuples, one list of unconjugated values ​​and another list of conjugated values, do not necessarily have to have the same length.

[0063] In one example, after obtaining these 4-tuple matrices, the 4-tuples can be ordered by a third factor, e.g., coherence, since greater coherence represents more significant cycle frequencies. Then, given that the list of 4-tuples can potentially be of any size, the L most significant 4-tuples can be selected based on their importance, as measured by coherence.

[0064] One approach to visualizing CSP characteristics is to use a cyclic domain profile (CDP), which plots either SCF or Coh as a function of CF. Figure 4 shows representative cyclic domain profiles for four exemplary class-type signals according to some embodiments of the present disclosure. Plot 400 shows the CDPs (NC and C versions of SCF) for each of the four class types: DTV, DTV+DSSS, DTV+GMSK, and DTV+DSSS+GMSK. DTV is shown to have a small number of CFs in the C region and no CFs in the NC region. In plot 402, DTV has no CFs in the NC region and a small number of CFs in the C region. Plot 404 shows that when combined with DTV, GMSK (Snagler) has no CFs in the NC region and a small number of CFs in the NC region. DSSS is known to have many CFs in both the NC and C regions. For DTV+DSSS and DTV+DSSS+GMSK, this is evident as shown in plots 406 and 408. However, in the case of DTV+DSSS+GMSK, the classification problem can be thought of as trying to find the GMSK CF "needle" in the DSSS CF "haystack."

[0065] With step 202 (dataset generation) and step 204 (signal processing) completed, step 206 (neural network training and machine learning) is now described.

[0066] In some examples, two types of architectures may be utilized, both with separate and combined CSP and PSD inputs. However, the present disclosure is not limited to only these two types of architectures, and other architectures may be used as well. The first type may use only one form of input feature, while the second type may be a fusion model that takes both features as input. In another example, extensions to these models that use self-attention to improve training time and locate anomalies in the input may also be utilized.

[0067] In these architectures, C can be the number of classes characterizing the input signal. Let the dimension of the input be n. Each model is a parameterized class of functions fθ : Rn -+ Rc. For x E Rn, define fθ(x)= z as follows: TIFF2025538916000004.tif14158 where a1 = n, b1, a2 = 128, b2 = 64, k1 = n, k2 = a2. ReLU denotes the standard rectified linear unit function (ReLU(x) = 0 for x<0; x for x>0), and LayerNormi ​​is the layer normalization defined for z E Rk as follows: TIFF2025538916000005.tif28158The softmax function normalizes the input into a probability vector as follows: TIFF2025538916000006.tif18158

[0068] The model parameters θ consist of a weight matrix Wi and LayerNorm parameters γi and βi. The input vector x to the model can be thought of as representing either PSD features or CSP features of the signal. In the latter case, the CSP features can be represented as an m × 4 matrix, with each row containing a single 4-dimensional FACS cyclostationary feature, and the input can be flattened to a single dimension.

[0069] Thus, the above architecture gives rise to two distinct deep learning models: the first type that takes PSD features as input, and the second type that takes CSP features. However, another model may be a feature fusion model that takes both PSD and CSP features as input. In this fusion model, the input is the concatenation of the PSD x E Rn and CSP y E Rn features (x,y): TIFF2025538916000007.tif30158Here a1 = n, b1, a2 = 128, b2 = 64, k1 = n, k2 = a2, and a3 = n', b3, a4 = 128, b4 = 64, k3 = n', k4 = a4.

[0070] In some embodiments, to not only identify anomalies but also to specify where in the cover signal the anomalies are, the above model may be modified by applying a multi-head self-attention layer to the input vector before proceeding as described above for each architecture. TIFF2025538916000008.tif26158 where W3 and Li are as above. Similarly, replace each input component x and y with MultiHeadAttn(x) and MultiHeadAttn(y) from the combined model.

[0071] To train any of the above-described neural network architectures, the above-described IQ datasets can be generated by taking over-the-air captures of ATSC-DTV signals and injecting various anomalies into these captures. In a non-limiting example, SignalHound can be used to generate six usable signals of 6.25 Mhz ATSC-DTV, taking ten 100 ms captures at 575 Mhz with a sampling of 61.44 Mhz. These signals can then be split into 2000 seg- ments of length 2. 18 The input DTV signal and the output composite signal may both be normalized to unit power. The number of signals, their duration and frequency, and the size of the blocks are not limited to these examples and may vary based on experimentation and / or empirical research.

[0072] Also, as described above, two non-limiting exemplary anomaly classes can be artificially injected into the captured signal using GNU Radio. These two non-limiting example anomalies can be DSSS with a gain of 1k and GMSK with a BT=0.35. These anomalies are inserted by randomly sampling the parameters described above (e.g., modulation, power, noise floor, symbol rate, center frequency, etc.). For DSSS and GMSK, the power, bandwidth, frequency center, and noise floor can be used. The parameters were biased toward DSSS with full bandwidth. The DSSS and GMSK center frequencies were selected according to their bandwidths. In one example, these parameters are pushed slightly beyond the expected realistic scenario to improve neural network performance.

[0073] 5 illustrates parameter distributions for synthetic anomalies according to some aspects of the present disclosure. Plot 500 shows various parameter (e.g., PSD) distributions for different anomaly signals.

[0074] The PSD may be calculated using Welch's method for 1024 points. The SSCA may be determined for block lengths of 218 and 64 channels. However, the method and associated parameters used for PSD calculation are not limited to these examples and may be modified according to experimentation and / or empirical studies.

[0075] In total, in one non-limiting example, the final dataset consists of approximately 1000 ATSC-DTV signals split approximately evenly across the classes: DTV, DTV+DSSS, DTV+Snagler, and DTV+DSSS+Snagler. The neural network may be trained for 40 epochs at 6 seconds per epoch using an Intel Xenon 6230R CPU (2.10 GHz) and an NVIDIA RTX A5000 (24 GB memory).

[0076] Next, the results of the training signal classification neural network described above are provided.

[0077] The performance of the trained CSP-only model was analyzed and tested on a dataset with realistic anomaly parameters. Figure 6 shows accuracy results of a trained neural network for signal classification according to some embodiments of the present disclosure. As shown in table 600, the CSP-only model (table 602) achieves 96% accuracy, while the combined model (table 604) improves to 99% accuracy. Furthermore, to better understand the generalization of the model, a more challenging dataset was considered with much more aggressive DSSS anomaly parameters.

[0078] 7A-C illustrate example architectures with multimodal fusion and associated accuracy results according to some aspects of the present disclosure. Model 702 in FIG. 7A illustrates a CSP-only neural network architecture according to one non-limiting example, with fully connected layers (FC) 128 and 64 trained to distinguish four signal classes (FC 4). Table 706 in FIG. 7B shows the accuracy results of the CSP-only model. As shown, the CSP-only model achieves an accuracy of 84%.

[0079] Model 704 in FIG. 7A illustrates a combined CSP-PSD neural network architecture according to one non-limiting example, where the PSD and CSP portions of the combined model each have 128 and 64 fully connected layers, respectively, as shown, and are combined into another 128 and 64 fully connected layers trained to ultimately identify four signal classes. Table 708 in FIG. 7B shows the accuracy results of the combined PSD-CSP model. As shown, the combined model achieves near-perfect accuracy of 98%. However, the model can also distinguish between different types of the same anomaly. In particular, the classifier can distinguish between the presence of snags to the left or right of the center frequency and between large and small bandwidths of DSSS anomalies. The term classifier may be used interchangeably with the terms trained neural network and / or trained machine learning model.

[0080] 7C provides non-limiting examples of neural network architectures for fusing the outputs of combined PSD-CSP models. Exemplary architecture 710 of FIG. 7C may utilize the exemplary approach of feature concatenation, while exemplary architecture 712 of FIG. 7C may utilize last-layer softmax averaging as described with reference to the mathematical equations above. While feature concatenation and last-layer softmax averaging are mentioned, the disclosure is not limited thereto, and other known or to-be-developed approaches for multi-model feature fusion may be used instead and / or in combination with each other.

[0081] Exemplary neural network architectures that may be utilized for the CSP and / or combined PSD-CSP models may include any other neural or deep learning network, such as a convolutional neural network (CNN), an autoencoder, a deep belief net (DBN), a recurrent neural network (RNN), etc.

[0082] To test the generalization of these models, further accuracy tests of the CSP-only classifier were performed on a series of ATSC-DTV signal files injected with various amounts of DSSS power. Figure 8 illustrates the improvement in classifier accuracy, according to some aspects of the present disclosure. Plot 800 shows the continued improvement in accuracy, with the model continuing to provide good accuracy outside the training range, particularly on the right (plot 802). Figure 9 illustrates the accuracy results for a persistent GMSK snugler, according to some aspects of the present disclosure. The experiment is repeated using the persistent GMSK snugler, as shown in plot 900. In this case, the classifier is slightly less effective because the DSSS anomaly eventually becomes too strong and the classifier is unable to see the snugler.

[0083] In the exemplary embodiment described above, a trained neural network (classifier) ​​determines a specific type of anomaly in a cover signal (e.g., DSSS, GMSK, etc.). However, the type of anomaly may be unknown / new, such that the neural network has not been trained to identify such new anomalies. In other words, it is possible that the anomaly may fall into the "none of the above" category.

[0084] In one example, machine learning techniques such as autoencoders can be utilized that can compress (using an encoder) and decompress (using a decoder) a received input signal so that such "none of the above" anomalies can be identified. The encoder can reduce the input signal to smaller dimensions, which can then be decoded back to the original input dimensions. The difference (error) between the input signal and the decoder's output can be used as a test of how similar the input signal is to the training data used to train the neural network. If such an error or the like is within a configurable threshold, the output of the trained neural network can indicate the presence of an anomaly (and possibly indicate that the anomaly is not one of the specific anomalies the neural network was trained to identify).

[0085] FIG. 10 illustrates an example neural network that may be trained to perform interference signal detection and classification and / or interference mitigation schemes according to some aspects of the present disclosure.

[0086] The architecture 1000 includes a neural network 1010 defined by an exemplary neural network description 1001 within a rendering engine model (neural controller) 1030. The neural network description 1001 may include a complete specification of the neural network 1010. For example, the neural network description 1001 may include a description or specification of the architecture of the neural network 1010 (e.g., layers, layer interconnections, number of nodes in each layer, etc.); input and output descriptions indicating how inputs and outputs are formed or processed; activation functions in the neural network, signals such as operations or filters in the neural network; neural network parameters such as weights, biases, etc.

[0087] In this example, neural network 1010 can be any of the neural networks described above that have been trained for signal classification.

[0088] The neural network 1010 includes an input layer 1002 that can receive input data, including but not limited to the above-mentioned IQ dataset, signal features determined for a given signal (e.g., PSD, conjugate CF, non-conjugate CF), etc. The neural network 1010 includes hidden layers 1004A-1004N (hereinafter collectively referred to as "1004"). The hidden layers 1004 can include n hidden layers, where n is an integer greater than or equal to 1. The number of hidden layers can include as many layers as necessary for the desired processing results and / or rendering intent. The neural network 1010 further includes an output layer 1006 that provides as output identified anomalous signals (co-channel signals) within a cover signal, such as a DSSS signal embedded with the above-mentioned DTV signal, a Snuggler signal, etc.

[0089] The neural network 1010 in this example is a multi-layer neural network of interconnected nodes. Each node can represent a piece of information. The information associated with a node is shared between different layers, and each layer retains the information as it is processed. In some cases, the neural network 1010 can include a feedforward neural network, in which there are no feedback connections where the output of the neural network is fed back to itself. In other cases, the neural network 1010 can include a recurrent neural network, which can have loops that allow information to be carried across nodes while reading the input.

[0090] Information may be exchanged between nodes via interconnections between the various layers. A node in the input layer 1002 can activate a set of nodes in the first hidden layer 1004A. For example, as shown, each of the input nodes in the input layer 1002 is connected to each of the nodes in the first hidden layer 1004A. The nodes in the hidden layer 1004A can transform the information of each input node by applying an activation function to the information. The information derived from the transformation can then be passed to and activated by nodes in the next hidden layer (e.g., 1004B), which can perform their own specified function. Exemplary functions include convolution, upsampling, data transformation, pooling, and / or any other suitable function. The output of a hidden layer (e.g., 1004B) can then activate nodes in the next hidden layer (e.g., 1004N), and so on. The output of the last hidden layer can activate one or more nodes in the output layer 1006, at which point an output is provided. Although in some cases nodes in neural network 1010 (e.g., nodes 1008A, 1008B, 1008C) are shown as having multiple output lines, the nodes have a single output, and all lines shown as outputting from the node represent the same output value.

[0091] In some cases, each node or interconnection between nodes may have a weight, which is a set of parameters derived from the training neural network 1010. For example, the interconnections between nodes may represent some of the information learned about the interconnected nodes. The interconnections may have numerical weights that can be adjusted (e.g., based on a training data set), allowing the neural network 1010 to adapt to inputs and learn as more data is processed.

[0092] The neural network 1010 may be pre-trained to process features from the data in the input layer 1002 using different hidden layers 1004 to provide an output via the output layer 1006. This training may be performed as described above with reference to step 206 of FIG. 2. In examples where the neural network 1010 is used to predict usage of a shared band, the neural network 1010 may be trained using training data including past transmissions and operations in the shared band by the same UE or UEs of similar systems (e.g., radar systems, RAN systems, etc.). For example, past transmission information may be input to the neural network 1010, which may be processed by the neural network 1010 to generate an output that may be used to adjust one or more aspects of the neural network 1010, such as weights, biases, etc.

[0093] In some cases, the neural network 1010 may adjust node weights using a training process called backpropagation. Backpropagation may include a forward pass, a loss function, a backward pass, and a weight update. The forward pass, loss function, backward pass, and parameter update are performed for one training iteration. This process may be repeated for a number of iterations for each set of training media data until the layer weights are accurately adjusted.

[0094] For the first training iteration for the neural network 1010, the output may include values ​​that do not favor any particular class because the weights are randomly selected during initialization. For example, if the output is a vector with probabilities that the object contains different product(s) and / or different users, the probability values ​​for each of the different products and / or users may be equal or at least very similar (e.g., for 10 possible products or users, each class may have a probability value of 0.1). With the initial weights, the neural network 1010 cannot determine low-level features and therefore cannot make an accurate determination of what classifications of the object are possible. A loss function may be used to analyze the error in the output. Any suitable loss function definition may be used.

[0095] The loss (or error) may be high for the first training data set (e.g., images) because the actual values ​​differ from the predicted outputs. The goal of training is to minimize the amount of loss so that the predicted outputs match the target or ideal outputs. The neural network 1010 can perform a backward pass by determining which inputs (weights) contributed most to the neural network's 1010 loss, and can adjust the weights so that the loss is reduced and eventually minimized.

[0096] The derivative of the loss with respect to the weights may be calculated to determine the weights that contributed most to the loss of the neural network 1010. After the derivative is calculated, weight updates may be performed by updating the filter weights. For example, the weights may be updated to change in the opposite direction of the gradient. The learning rate may be set to any suitable value, with a high learning rate indicating larger weight updates and a lower value indicating smaller weight updates.

[0097] The neural network 1010 can include any suitable neural or deep learning network. One example includes a convolutional neural network (CNN) that includes an input layer and an output layer with multiple hidden layers between the input and output layers. The hidden layers of a CNN include a series of convolutional, nonlinear, pooling (for downsampling), and fully connected layers. In other examples, the neural network 1010 can represent any other neural or deep learning network, such as an autoencoder, a deep belief net (DBN), a recurrent neural network (RNN), etc.

[0098] FIG. 11 is an example flowchart of a method of signal classification according to some aspects of the present disclosure. The process of FIG. 11 may be performed by any signal receiver operating in a wireless communication environment, including any one of the receivers 112 of FIG. 1. Such a receiver may have a trained neural network and logic for signal processing to perform signal classification as described above. The steps of FIG. 11 are described from the perspective of receiver 112 (as representative of receivers 114, 116, and 118). However, the present disclosure is not limited thereto, and the steps of FIG. 11 may be performed by any other network element capable of receiving wireless communication signals that may have co-channel anomalies and undesired signals embedded therein.

[0099] As described above, once the neural network is trained for signal classification, the following steps can be implemented for real-time reception, processing, and classification of co-channel signals.

[0100] In step 1100, the receiver 112 may receive a signal. The signal may be received at the air interface of the receiver 112. The signal may include a cover signal and an embedded co-channel anomaly signal (e.g., DSSS, Snuggler, or an otherwise unknown type of anomaly).

[0101] In step 1102, the receiver 112 may perform signal processing on the received signal to determine one or more signal features to be used as input to a neural network trained for signal classification. The type of signal processing performed on the received signal may be as described above and / or any other known or to-be-developed signal processing technique that enables extraction of features that can be used for signal classification and detection of anomalies embedded within the cover signal.

[0102] As mentioned above, the one or more signal characteristics may include, but are not limited to, the PSD of the signal received in step 1100, the conjugate CF of the signal received in step 1100, the unconjugate CF of the signal received in step 1100, etc.

[0103] In step 1104, the receiver 112 may provide one or more signal features as inputs to one or more trained neural networks that have been trained as described above with reference to stage 206 of FIG. 2 and FIGS. 7A-C. In one example, all of the determined signal features may be provided as inputs to the trained neural network. In another example, only one or a subset of the determined signal features may be provided as inputs to the trained neural network model.

[0104] At 1106, the receiver 112 may perform multimodal feature fusion to combine outputs from the trained neural networks. Such multimodal feature fusion may be performed using techniques such as feature concatenation or last layer softmax averaging, as described above with respect to FIGS. 7A-C.

[0105] In step 1108, the trained neural network model can provide as output a classification of the signal received in step 1100. The classification can identify cover signals (e.g., DTV signals) and anomalies (e.g., DSSS and / or GMSK, or otherwise unknown anomalies when an autoencoder as described above is utilized).

[0106] In step 1110, the receiver 112 may output the results of the signal classification. The output may be in any desired format. For example, the output may be a visual representation of the signal with anomalies identified therein (e.g., similar to visual plot 300 of FIG. 3), or may simply be text identifying the signal and the anomalies.

[0107] 12 illustrates an exemplary computing system according to some aspects of the present disclosure. The computing system 1200 can be any computing device suitable for performing signal classification and co-channel anomaly detection, for example, as described above with respect to FIGS. 1-11 , including, but not limited to, the transmitter 104, the receiver 112, etc., and / or any components thereof in which components of the system communicate with each other using a connection 1202. The connection 1202 can be a physical connection via a bus or a direct connection to a processor 1210, such as in a chipset architecture. The connection 1202 can also be a virtual connection, a network connection, or a logical connection.

[0108] In some embodiments, computing system 1200 is a distributed system in which the functionality described in this disclosure may be distributed across a data center, multiple data centers, a peer network, etc. In some embodiments, one or more of the system components described represent many such components, each performing some or all of the functionality for which that component is described. In some embodiments, the components may be physical or virtual devices.

[0109] Exemplary computing system 1200 includes at least one processing unit (CPU or processor) 1210 and connections 1202 coupling various system components to processor 1210, including system memory 1215, read-only memory (ROM) 1220, and random access memory (RAM) 1225. Computing system 1200 may include a cache of high-speed memory 1212 directly, closely connected to, or integrated as part of, processor 1210.

[0110] Processor 1210 can include any general-purpose processor, hardware or software services, such as services 1232, 1234, and 1236, stored in storage device 1230, configured to control processor 1210, and special-purpose processors where software instructions are incorporated into the actual processor design. Processor 1210 may essentially be a completely self-contained computing system, including multiple cores or processors, buses, memory controllers, caches, etc. Multi-core processors can be symmetric or asymmetric.

[0111] To enable user interaction, computing system 1200 includes input devices 1245, which can represent any number of input mechanisms, such as a microphone for audio, a touch-sensitive screen for gesture or graphical input, a keyboard, a mouse, motion input, voice, etc. Computing system 1200 also includes output devices 1235, which can be one or more of several output mechanisms known to those skilled in the art. In some examples, a multimodal system may enable a user to provide multiple types of input / output for communicating with computing system 1200. Computing system 1200 may include a communications interface 1240, which can generally manage and manage user input and system output. There is no restriction to operating on any particular hardware configuration, and thus the basic features herein may readily substitute for improved hardware or firmware configurations as they are developed.

[0112] The storage device 1230 may be a non-volatile memory device, and may be a hard disk or other type of computer-readable medium capable of storing data accessible by a computer, such as a magnetic cassette, a flash memory card, a solid-state memory device, a digital versatile disk, a cartridge, random access memory (RAM), read-only memory (ROM), and / or some combination of these devices.

[0113] The storage devices 1230 may include software services, servers, services, etc., where code defining such software, when executed by the processor 1210, causes the system to perform a function. In some embodiments, a hardware service that performs a particular function may include software components stored on a computer-readable medium in association with the necessary hardware components, such as the processor 1210, connections 1202, output devices 1235, etc., to perform that function.

[0114] For clarity of explanation, in some instances, the technology may be presented as including individual functional blocks, including functional blocks comprising devices, device components, method steps or routines implemented in software or a combination of hardware and software.

[0115] Any of the steps, operations, functions, or processes described herein may be performed or implemented by a combination of hardware and software services, alone or in combination with other devices. In some embodiments, a service may be software that resides in memory of a client device and / or one or more servers of a content management system and performs one or more functions when a processor executes the software associated with the service. In some embodiments, a service is a program or collection of programs that perform a particular function. In some embodiments, a service may be considered a server. Memory may be a non-transitory computer-readable medium.

[0116] In some embodiments, computer-readable storage devices, media, and memories may include cables or wireless signals containing bitstreams, etc. However, when referred to, non-transitory computer-readable storage media explicitly excludes media such as energy, carrier signals, electromagnetic waves, and signals themselves.

[0117] The methods according to the above-described examples may be implemented using computer-executable instructions stored on or otherwise available from a computer-readable medium. Such instructions may include, for example, instructions and data that cause or otherwise configure a general-purpose computer, a special-purpose computer, or a special-purpose processing device to perform a particular function or group of functions. Some of the computer resources used may be accessible over a network. The executable computer instructions may be, for example, intermediate format instructions such as binary, assembly language, firmware, or source code. Examples of computer-readable media that may be used to store instructions, information used, and / or information generated during methods according to the described examples include magnetic or optical disks, solid-state memory devices, flash memory, USB devices with non-volatile memory, networked storage devices, etc.

[0118] Devices implementing methods according to these disclosures can include hardware, firmware, and / or software and can take any of a variety of form factors. Typical examples of such form factors include servers, laptops, smartphones, small form factor personal computers, personal digital assistants, etc. The functionality described herein can also be embodied in peripheral devices or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes running in a single device, as further examples.

[0119] The instructions, media for communicating such instructions, computing resources for executing them, and other structures for supporting such computing resources are means for providing the functionality described in these disclosures.

[0120] While various examples and other information have been used to describe aspects within the appended claims, those skilled in the art can use these examples to derive a wide variety of implementations, and therefore, limitations on the claims should not be implied based on the specific features or configurations in such examples. Moreover, while some subject matter may be described in language specific to example structural features and / or method steps, it should be understood that the subject matter defined in the appended claims is not necessarily limited to these described features or operations. For example, such functionality may be differently distributed or performed in components other than those identified herein. Rather, the described features and steps are disclosed as example components of systems and methods within the appended claims. Claim language or other language stating "at least one of" a set and / or "one or more" of a set indicates that one member of the set or multiple members of the set (in any combination) satisfies the claim.

[0121] For example, a claim reciting "at least one of A and B" or "at least one of A or B" can mean A, B, or A and B. In other examples, a claim reciting "at least one of A, B, and C" or "at least one of A, B, or C" can mean A, B, C, or A and B, or A and C, or B and C, or A, B, and C. The language of "at least one" of a set and / or "one or more" of a set does not limit the set to the items listed in the set. For example, claim language reciting "at least one of A and B" or "at least one of A or B" can mean A, B, or A and B, and can further include items not listed in the set of A and B.

Claims

1. 1. A method comprising: receiving, at a receiver, a signal, the signal including a cover signal and an embedded co-channel anomaly signal; performing signal processing on the signal at the receiver to determine one or more characteristics of the signal; inputting the one or more features into one or more trained neural networks at the receiver; and receiving a classification of the signal as an output of the trained neural network, the classification distinguishing between the cover signal and the embedded co-channel anomaly signal.

2. The method of claim 1 , wherein the one or more signal features include a power spectral density of the signal, a conjugate cycle frequency of the signal, and a non-conjugate cycle frequency of the signal.

3. The method of claim 1 , wherein at least one of the one or more features is input to the trained neural network.

4. The method of claim 1 , further comprising performing multimodal fusion to combine outputs of at least two of the trained neural networks to determine the output.

5. 10. The method of claim 1, wherein the cover signal is one of a Long Term Evolution (LTE), a 3GPP 5G signal, Wi-Fi, a Digital Video Broadcast (DVB), or an Advanced Television Systems Committee Digital Television (ATSC-DTV) signal.

6. 2. The method of claim 1, wherein the co-channel anomaly signal is one of one of a direct sequence spread spectrum (DSSS) signal, a single carrier signal using binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), quadrature amplitude shift keying (QAM), amplitude phase shift keying (APSK) modulation, a chirp modulated signal, a frequency modulation (FM) signal, a frequency shift keying (FSK) signal, an orthogonal frequency division multiplexing (OFDM) signal, a burst signal, a frequency hopping spread spectrum signal (FHSS), or a Gaussian minimum shift keying (GMSK) signal.

7. The method of claim 1 , wherein the trained neural network is trained using a combination of wireless capture signals injected with a synthetic co-channel anomaly signal.

8. 1. A wireless network receiver, comprising: one or more memories containing computer readable instructions; Executing the computer-readable instructions, receiving a signal, the signal including a cover signal and an embedded co-channel anomaly signal; performing signal processing on the signal to determine one or more characteristics of the signal; inputting the one or more features into a trained neural network; 1. A wireless network receiver comprising: one or more processors configured to receive a classification of the signal as an output of the trained neural network, the classification distinguishing between the cover signal and the embedded co-channel anomaly signal.

9. The wireless network receiver of claim 8 , wherein the one or more signal characteristics include a power spectral density of the signal, a conjugate cycle frequency of the signal, and a non-conjugate cycle frequency of the signal.

10. The wireless network receiver of claim 8 , wherein at least one of the one or more features is input to the trained neural network.

11. 10. The wireless network receiver of claim 8, wherein the one or more processors are further configured to perform multimodal fusion to combine outputs of at least two of the trained neural networks to determine the output.

12. 9. The method of claim 8, wherein the cover signal is one of a Long Term Evolution (LTE), a 3GPP 5G signal, Wi-Fi, a Digital Video Broadcast (DVB), or an Advanced Television Systems Committee Digital Television (ATSC-DTV) signal.

13. 9. The wireless network receiver of claim 8, wherein the co-channel anomaly signal is one of a direct sequence spread spectrum (DSSS) signal, a single carrier signal using binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), quadrature amplitude shift keying (QAM), amplitude phase shift keying (APSK) modulation, a chirp modulation signal, a frequency modulation (FM) signal, a frequency shift keying (FSK) signal, an orthogonal frequency division multiplexing (OFDM) signal, a burst signal, a frequency hopping spread spectrum signal (FHSS), or a Gaussian minimum shift keying (GMSK) signal.

14. The wireless network receiver of claim 8 , wherein the trained neural network is trained using a combination of wireless capture signals injected with synthetic co-channel anomaly signals.

15. When executed by one or more processors of a wireless network receiver, the wireless network receiver: receiving a signal, the signal including a cover signal and an embedded co-channel anomaly signal; performing signal processing on the signal to determine one or more characteristics of the signal; inputting the one or more features into a trained neural network and receiving as an output of the trained neural network a classification of the signal; One or more non-transitory computer-readable media containing computer-readable instructions, wherein the classification distinguishes between the cover signal and the embedded co-channel anomaly signal.

16. 16. The one or more non-transitory computer-readable media of claim 15, wherein the one or more signal features include a power spectral density of the signal, a conjugate cycle frequency of the signal, and a non-conjugate cycle frequency of the signal.

17. 16. The one or more non-transitory computer-readable media of claim 15, wherein the execution of the computer-readable instructions by the one or more processors causes the wireless network receiver to perform multi-modal fusion that combines outputs of at least two of the trained neural networks to determine the output.

18. 16. The one or more non-transitory computer-readable media of claim 15, wherein the cover signal is one of a Long Term Evolution (LTE), a 3GPP 5G signal, Wi-Fi, a Digital Video Broadcast (DVB), or an Advanced Television Systems Committee Digital Television (ATSC-DTV) signal.

19. 16. The one or more non-transitory computer-readable media of claim 15, wherein the co-channel anomaly signal is one of a direct sequence spread spectrum (DSSS) signal, a single-carrier signal using binary phase shift keying (BPSK), quadrature phase shift keying (QPSK), quadrature amplitude shift keying (QAM), amplitude phase shift keying (APSK) modulation, a chirp modulated signal, a frequency modulation (FM) signal, a frequency shift keying (FSK) signal, an orthogonal frequency division multiplexing (OFDM) signal, a burst signal, a frequency hopping spread spectrum signal (FHSS), or a Gaussian minimum shift keying (GMSK) signal.

20. 16. The one or more non-transitory computer-readable media of claim 15, wherein the trained neural network is trained using a combination of wireless capture signals injected with a synthetic co-channel anomaly signal.

Images