公开(公告)号：US11558810B2

公开(公告)日：2023-01-17

申请号：US17142800

申请日：2021-01-06

Applicant: BAE SYSTEMS Information and Electronic Systems Integration Inc.

Inventor： Joshua W. Robinson ,  Joseph M. Carmack ,  Scott A Kuzdeba ,  James M. Stankowicz, Jr.

IPC: H04W48/16 ,  G06N20/00

Abstract: A system whereby individual RF emitter devices are distinguished in real-world environments through deep-learning comprising an RF receiver for receiving RF signals from a plurality of individual devices; a preprocessor configured to produce complex-valued In-phase (I) and Quadrature (Q) IQ signal sample representations; a two-stage Augmented Dilated Causal Convolution (ADCC) network comprising a stack of dilated causal convolution layers and traditional convolutional layers configured to process I and Q components of the complex IQ samples; transfer learning comprising a classifier and a cluster embedding dense layer; unsupervised clustering whereby the RF signals are grouped according to a device that transmitted the RF signal; and an output identifying the individual RF emitter device whereby the individual RF emitter device is distinguished in the real-world environment.

公开(公告)号：US20230130863A1

公开(公告)日：2023-04-27

申请号：US17358153

申请日：2021-06-25

Applicant: BAE SYSTEMS Information and Electronic Systems Integration Inc.

Inventor： Scott A. Kuzdeba ,  Joseph M. Carmack ,  James M. Stankowicz, JR.

IPC: G06N3/0464 ,  G06N3/063

Abstract: A Deep-Learning (DL) system for representation and construction or reconstruction of signals includes an encoder stage; an encoding; an optional modification stage; a decoder stage, and a (re)construction stage. The encoder stage includes layers of dilated convolutions, and the encoder maps from an input representation into a latent embedded representation. It learns a set of features that encode the input signals. The encoding stage comprises latent space; the decoder stage maps from latent features back to an output of the same size as the input signals, whereby the output has the same dimensionality and representation as the input signals. Modification to the signal can be conducted within the latent representation to alter the (re)construction for specific tasks, such as increasing a device's RF fingerprint.

公开(公告)号：US20230004763A1

公开(公告)日：2023-01-05

申请号：US17503205

申请日：2021-10-15

Applicant: BAE SYSTEMS Information and Electronic Systems Integration Inc.

Inventor： James M. Stankowicz, JR. ,  Joseph M. Carmack ,  Scott A Kuzdeba ,  Steven Schmidt

IPC: G06K9/62 ,  G06K9/00

Abstract: A Deep-Learning (DL) explainable AI system for Radio Frequency (RF) machine learning applications with expert driven neural explainability of input signals combines three algorithms (A1, A2, and A3). A1 is a neural network that learns to classify spectrograms. During training, A1 learns to map a spectrogram to its paired label. It outputs a label estimate from a spectrogram. Labels account for device number and spectrum utilization. The neural network is built on two-dimensional dilated causal convolutions to account for frequency and time dimensions of spectrogram data. A2 is a user-defined function that converts an input spectrogram into a vector that quantifies human-identifiable elements of the spectrogram. A3 is a random forest feature extraction algorithm. It takes as input the outputs of A2 and A1. From these, A3 learns which elements in the vector output by A2 were most important for choosing the labels output from A1.

公开(公告)号：US20220217619A1

公开(公告)日：2022-07-07

申请号：US17142800

申请日：2021-01-06

Applicant: BAE SYSTEMS Information and Electronic Systems Integration Inc.

Inventor： Joshua W. Robinson ,  Joseph M. Carmack ,  Scott A. Kuzdeba ,  James M. Stankowicz, JR.

IPC: H04W48/16 ,  G06N20/00

Abstract: A system whereby individual RF emitter devices are distinguished in real-world environments through deep-learning comprising an RF receiver for receiving RF signals from a plurality of individual devices; a preprocessor configured to produce complex-valued In-phase (I) and Quadrature (Q) IQ signal sample representations; a two-stage Augmented Dilated Causal Convolution (ADCC) network comprising a stack of dilated causal convolution layers and traditional convolutional layers configured to process I and Q components of the complex IQ samples; transfer learning comprising a classifier and a cluster embedding dense layer; unsupervised clustering whereby the RF signals are grouped according to a device that transmitted the RF signal; and an output identifying the individual RF emitter device whereby the individual RF emitter device is distinguished in the real-world environment.

公开(公告)号：US11342946B1

公开(公告)日：2022-05-24

申请号：US17208304

申请日：2021-03-22

Applicant: BAE SYSTEMS Information and Electronic Systems Integration Inc.

Inventor： Amit Bhatia ,  Joseph M. Carmack ,  Scott A Kuzdeba ,  Joshua W. Robinson

IPC: H04B1/00 ,  G06N3/04 ,  H04B1/10

Abstract: An artifact-suppressing neural network (NN) kernel comprising at least one neural network, implemented in replacement of a DSP, provides comparable or better performance under non-edge conditions, and superior performance under edge conditions, due to the ease of updating the NN kernel training without enlarging its computational footprint or latency to address a new edge condition. In embodiments, the NN kernel can be implemented in a field programmable gate array (FPGA) or application specific integrated circuit (ASIC), which can be configured as a direct DSP replacement. In various embodiments, the NN kernel training can be updated in near real time when a new edge condition is encountered in the field. The NN kernel can include DCC lower layers and dense upper layers. Initial NN kernel training can require fewer examples. Example embodiments include a noise suppression NN kernel and a modem NN kernel.