1© 2015 The MathWorks, Inc.

2© 2015 The MathWorks, Inc.

Machine Learning for Radar & EW

서기환과장

3

Agenda

▪ Radar and EW modeling

▪ Synthesizing data for Machine Learning workflows

▪ Machine Learning Examples

Target

Classification

Signal

ProcessingTracking and

Sensor Fusion

Scheduling

and Control

Antenna/Array/RF

4

Radar and EW Simulation and Modeling Architecture

Waveform

Generator

λ

Transmitter

Pt

Transmit

Array

Gt

Signal

Processing

Receiver

Gr

Receive

Array

Gr

Environment,

Targets, &

Interference (L, 𝝈, 𝑹𝒓 ,𝑹𝒕)

Radar

Scheduler and

Tracker

▪ Functions for calculations and analysis

▪ Apps for common workflows

▪ Parameterized components for system modeling

▪ Easy path to increased fidelity for antenna and RF design

▪ Code generation for deployment

𝑃𝑟 =𝑃𝑡𝐺𝑡𝐺𝑟𝜆

2𝜎

4𝜋 3𝑅𝑡2𝑅𝑟

2𝐿

5

Design Phased Array Antennas

Design subarrays

Model failuresSynthesize arrays Model mutual coupling

Model imperfections Import antenna patterns

Taper/thin arrays

Design an array

6

Design Radar and EW Systems

Data cube processing

Spatial signal processing

Polarization

Code generation

HDL

Scenario visualization

Wideband

Detections

Targets & Environment

7

Agenda

▪ Radar and EW modeling

▪ Synthesizing data for Machine Learning workflows

▪ Machine Learning Examples

Target

Classification

Signal

ProcessingTracking and

Sensor Fusion

Scheduling

and Control

Antenna/Array/RF

8

Machine Learning Workflow

Integrate Analytics with

Systems

Desktop Apps

Enterprise Scale

Systems

Embedded Devices

and Hardware

Files

Databases

Sensors

Access and Explore

Data

Develop Predictive

Models

Model Creation e.g.

Machine Learning

Model

Validation

Parameter

Optimization

Preprocess Data

Working with

Messy Data

Data Reduction/

Transformation

Feature

Extraction

9

Synthesize Radar Data for Machine Learning

Collect data

Synthesize data

Create data set Learn

Train on data

Validate

Measure accuracy

10

Synthesize Received Radar Signals

11

Define a Backscatter Target with Angle and Frequency

rcs_cyl = cylinderrcs(r1,r2,H,c,fc,az,el);

12

Model Basic Shapes

13

Model Extended Targets with Multiple Scatters

14

Agenda

▪ Radar and EW modeling

▪ Synthesizing data for Machine Learning workflows

▪ Machine Learning Examples

Target

Classification

Signal

ProcessingTracking and

Sensor Fusion

Scheduling

and Control

Antenna/Array/RF

15

Machine Learning for Radar Examples

Synthesize returns (radar cross section) Synthesize micro-Doppler (Time-frequency)

Statistics and Machine Learning Toolbox

Signals

Features

Time-frequency

Etc.

Classification

Synthesize waveforms

16

Signal Processing and Wavelets for Feature Extraction

▪ Signal Manipulation

– Signal Analysis App

▪ Time-Frequency Analysis Capabilities

– Short Time Fourier Transform

– Continuous Wavelet Transform

– Synchrosqueezing

▪ Multiresolution Analysis Capabilities

– Discrete Wavelet Analysis

– Wavelet Packets

17

Signal Analyzer App

▪ With the Signal Analyzer you can :

– Import multichannel signals

– Explore signals jointly in time-frequency domain

– Zoom and pan signals

Analyze signals in time, frequency and time-frequency domains

18

Identifying Features in Real World Signals

▪ Characterizing signal features in spectral domain is often

challenging as one needs to appropriate tools

▪ Accurate time-frequency measurements are possible

using wavelet based time frequency analysis techniques

▪ Features once identified, can be extracted from signals

for further processing

▪ In this demo, we will characterize features in EKG signals

using Continuous Wavelet Transform

Quantify time-varying signals in frequency domain

» Demo

19

Filtering Frequency Localized Components

▪ Sometimes unwanted signals can get captured

during signal acquisition process

▪ A traditional filter cannot be used if the frequency

range of the interference lies within the frequency

range of the signal

▪ Unwanted components can be localized jointly in

time and frequency using wavelets and removed

20

Wavelet Synchrosqueezing

▪ For certain non-stationary signals, wavelet Synchrosqueezing can be

used to identify and extract signal modes

▪ Wavelet Synchrosqueezing helps extract signal components from

localized regions of time frequency plane

21

Example 1: Radar Echoes from Cylinder and Cone

22

Synthesize

data

Create data set Learn

Train on

data

Randomize parameters

Generate many data sets

Generate law radar data from models

23

Extract features: 700 samples/object -> 8 samples/object with Wavelet Transform

Feature Extraction

24

Testing Against Training Data

25

Algorithms for training

MATLAB code gen

Classification metrics

Classification Learner App

26

Range-Doppler from Parrot Quadcopter

We can identify:

• Rotation rate

• Number of blades

• Tip velocity

• Blade length

Micro-Doppler returns

Example 2: Micro-Doppler for Drones

27

Synthesize Micro-Doppler Motion

Rotation rate

Number of blades

Tip velocity

Blade length

Radar return Micro-Doppler Time-frequencyRange-Doppler of blade

28

Micro-Doppler in Time-Frequency Domain

Detection of small UAV helicopters using micro-Doppler David Tahmoush

29

Example 3: Waveform Modulation ID for RWR

30

Radar & EW Classification – Workflow

Integrate Analytics with

Systems

Desktop Apps

Enterprise Scale

Systems

Embedded Devices

and Hardware

Files

Databases

Sensors

Access and Explore

Data

Develop Predictive

Models

Model Creation e.g.

Machine Learning

Model

Validation

Parameter

Optimization

Preprocess Data

Working with

Messy Data

Data Reduction/

Transformation

Feature

Extraction

MATLAB

Signal Processing Toolbox

Wavelet Toolbox

Statistics & Machine Learning Toolbox

Statistics & Machine Learning Toolbox

MATLAB CoderPhased Array System

Toolbox

31

Key Takeaways

▪ Radar and EW modeling

▪ Synthesizing data for Machine Learning workflows

▪ Machine Learning Examples

– Synthesize Return, Micro-Doppler, Waveform

– Classification: Target, Radar

32

Resources to Help You Get Started

매트랩과머신러닝 (eBook)