SlideShare a Scribd company logo
Radio Signal Classification with Deep
Neural Networks
2018 Army Signal Classification Challenge
6th place solution
Kachi Odoemene
29 Aug 2018
Competition
• Army Rapid Capabilities Office (RCO)
– April 30th – August 13th 2018
– $150,000 in prizes
• Radio signal modulation classification
– Automatically identify modulation type of
received radio signal
– 24 modulation classes, including noise
class
Competition
• Finished 6th place (of 161 participants)
Radio Signal Modulation Schemes
Binary Frequency Shift Keying (BFSK)
Quadrature Signals
• I/Q: In-phase and Quadrature components
• 90° shift between identical periodic signals
– e.g. sine and cosine wave
Q
I
I/Q Modulation
• Summation of I/Q pairs results in any modulation
(frequency, amplitude, phase, etc)
• Transmitted & received radio signal represented
as I/Q components
Modulated
RF Signal
I channel, I(t)
Q channel, Q(t)
Inputs
(Modulating waves)
Carrier
wave
I/Q Time Domain Examples – high SNR
PSK: phase shift keying
FSK: frequency shift keying
ASK: amplitude shift keying
MSK: Minimum (frequency) shift keying
QAM: quadrature amplitude modulation
CP: continuous phase
OQ: offset quadrature
SNR: Signal to Noise
I/Q Time Domain Examples – low SNR
Dataset
• > 30 GB
– 4.32 million I/Q instances
• Dimension: 2 x 1024
– Short time window, real world conditions
– 24 modulation classes
– White noise added to signals
• 6 signal-to-noise (SNR) levels
• Test datasets (public)
– 2 sets of 100000 I/Q instances (unlabeled)
Evaluation Metric
• Derived from multiclass log loss
Score =
Goal
• Automatically identify the modulation type
of the received radio signal
– What modulation format was employed?
– Eg. Military applications:
• Spectrum surveillance, electronic warfare, and
threat analysis
– Identify modulation type of intercepted enemy
communication
Traditional approach
• Hand-crafted features (feature engineering)
– Higher order statistics
– Autocorrelation and spectral correlation functions
– Measures derived from instantaneous frequency,
amplitude, phase
• Mean, standard deviation, kurtosis, etc
– Small number of features (28-32)
• SVM, decision trees, ensemble, neural
networks
Deep Neural Networks for Radio Modulation
Recognition
(2016)
Convolutional networks outperform
expert feature-based classifiers
• Proposed 3 additional architectures for
modulation recognition:
– Inception
– ResNet
– Hybrid of Convolutional, Long short term memory
(LSTM), and Fully Connected (FC) Deep Neural
Network (CLDNN)
• No source code provided
• Sparse details on architecture hyperparameters
(2017)
Residual Unit
Layer Input
Model 1: ResNet (original)
• Image classification
• CNNs with skip (residual or
shortcut) connections
– Feed previous representations
(activations) into downstream layers
– Prevents information loss
• Enables training of deeper
networks
– 100s to 1000s of layers
He et al Deep Residual Learning for Image Recognition (ArXiv 2015)
Skipconnection
Input
(2x1024)
Conv
(1x128)
D-Conv
(2x1)
Avg. Pool
“ResBlock” x4
Global Avg
Pool
FC (128) x2
Softmax (24)
(Temporal)
(Spatial)
Model 1: ResNet (modified)
• Temporal convolution on each
IQ channel separately
• Depth-wise convolution (D-
Conv)
• Batch normalization after
convolutional and fully
connected (FC) layers
• Multiple residual units within
ResBlock
• # Parameters: 255,944
He et al Deep Residual Learning for Image Recognition (ArXiv 2015)
Model 2: CLDNN (original)
• Speech recognition
• Unified model: CNN, LSTM, FC
• CNN: reduce spectral variations of
input data
• LSTM: learn temporal structure
• FC: transform LSTM features into
output easy to classify
Sainath et al Convolutional, Long Short-Term Memory, fully
connected Deep Neural Networks (IEEE, 2015)
Convolutional
layers
Linear
layer
LSTM
layers
Fully
Connected
layers
Output targets
(1)
(2)
C
C
L
L
D
D
Dim.
red.Xt
[Xt-l,..,Xt,…,Xt+r]
Model 2: CLDNN (modified)
• Temporal convolution on each IQ
channel separately
• Depth-wise convolution (D-Conv)
• Batch normalization after each
convolutional and fully connected
(FC) layers
• “ConvBlock”: Conv + BatchNorm +
ReLU
• Dropout between FC layers
• # Parameters: 147,480
Sainath et al Convolutional, Long Short-Term Memory, fully
connected Deep Neural Networks (IEEE, 2015)
Input
(2x1024)
Conv
(1x128)
D-Conv
(2x1)
Avg. Pool
ConvBlock x2
Concatenate
LSTM (48)
x2FC (128)
Softmax (24)
(Temporal)
(Spatial)
Data preparation
• Raw data, no preprocessing
• Trained from scratch
• Data split
– Train: 80%, Valid: 13.33%, Holdout: 6.67%
Implementation details
• Hyper-parameter selection
– Temporal filter kernel size
– Number of FC units
• Maximum of 25 epochs, early stopping
• Adam optimizer
• Learning rate: 1e-3
• Keras (Tensorflow)
• Hardware (Personal)
– GTX 1080Ti (11 GB) GPU
– 16 GB RAM
Loss curves
Accuracy per modulation class: all SNRs
Accuracy per modulation class: high SNR
Accuracy per SNR: pooled across classes
Accuracy per SNR: individual classes
PSK: phase shift keying
FSK: frequency shift keying
ASK: amplitude shift keying
MSK: Minimum (frequency) shift keying
QAM: quadrature amplitude modulation
CP: continuous phase
OQ: offset quadrature
Leaderboard performance
Model TestSet1 Score TestSet2 Score Final Score
CLDNN 42.09 54.22 50.58
ResNet-18 33.27 58.27 50.77
Combined 42.09 58.27 53.41
Final Score = 0.3*TestSet1 + 0.7* TestSet2
Ensembling
Ensembling
• Trained meta-learner to combine
predictions from multiple models
Leaderboard performance: Ensembling
Model TestSet1 Score TestSet2 Score Final Score
CLDNN 42.09 54.22 50.58
ResNet-18 33.27 58.27 50.77
Combined 42.09 58.27 53.41
Logistic
Regression
45.17 63.25 57.83
Light GBM 46.40 64.54 59.10
Final Score = 0.3*TestSet1 + 0.7* TestSet2
Challenges
• Structural (competition organization)
– Multiple changes to scoring procedure and test set,
submission site shutdown, etc
• Time constraint
– 1 submission/day (10am) & leaderboard update
(5pm)
– On Kaggle: up to 5 submissions/day, immediate LB
update
• Technical
– Hardware failure final week of competition
• Backup: Google Colab
Future Efforts
• Incorporate additional features:
– Amplitude
– Phase difference
– Magnitude of Fourier transform
– Spectrogram
• Explore other architectures & ensembling methods
– Inception-like architecture: process and combine multiple
frequency scales
• Model interpretability
– DeepLIFT (Deep Learning Important FeaTures)
– LIME (Local Interpretable Model-agnostic Explanations)
– Ablation studies
Acknowledgements
Thank You!
Questions?

More Related Content

What's hot

Image Processing: Spatial filters
Image Processing: Spatial filtersImage Processing: Spatial filters
Image Processing: Spatial filters
A B Shinde
 
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and SystemsDSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
Amr E. Mohamed
 
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATIONCOMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
Journal For Research
 
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal ProcessingDSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
Amr E. Mohamed
 
Cognitive radio networks
Cognitive radio networksCognitive radio networks
Cognitive radio networks
Ameer Sameer
 
Lecture 11
Lecture 11Lecture 11
Lecture 11
Wael Sharba
 
Channel equalization
Channel equalizationChannel equalization
Channel equalization
Munnangi Anirudh
 
Image Filtering in the Frequency Domain
Image Filtering in the Frequency DomainImage Filtering in the Frequency Domain
Image Filtering in the Frequency Domain
Amnaakhaan
 
Unit iv wcn main
Unit iv wcn mainUnit iv wcn main
Unit iv wcn main
vilasini rvr
 
Attenuators and phase shifters 24
Attenuators and phase shifters 24Attenuators and phase shifters 24
Attenuators and phase shifters 24
HIMANSHU DIWAKAR
 
Multirate DSP
Multirate DSPMultirate DSP
Multirate DSP
@zenafaris91
 
Basics of Digital Filters
Basics of Digital FiltersBasics of Digital Filters
Basics of Digital Filters
op205
 
OFDM for LTE
OFDM for LTEOFDM for LTE
OFDM for LTE
Madhumita Tamhane
 
floating point multiplier
floating point multiplierfloating point multiplier
floating point multiplierBipin Likhar
 
Intelligent Networks
Intelligent NetworksIntelligent Networks
Intelligent NetworksFaraz Shahid
 
Qudrature Amplitude Modulation by Krishna Teja & Sunil
Qudrature Amplitude Modulation by Krishna Teja & SunilQudrature Amplitude Modulation by Krishna Teja & Sunil
Qudrature Amplitude Modulation by Krishna Teja & Sunil
krishnateja407
 
HFSS using FEM
HFSS using FEMHFSS using FEM
HFSS using FEM
Sai Saketh
 
DSP_2018_FOEHU - Lec 06 - FIR Filter Design
DSP_2018_FOEHU - Lec 06 - FIR Filter DesignDSP_2018_FOEHU - Lec 06 - FIR Filter Design
DSP_2018_FOEHU - Lec 06 - FIR Filter Design
Amr E. Mohamed
 

What's hot (20)

Image Processing: Spatial filters
Image Processing: Spatial filtersImage Processing: Spatial filters
Image Processing: Spatial filters
 
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and SystemsDSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
DSP_2018_FOEHU - Lec 03 - Discrete-Time Signals and Systems
 
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATIONCOMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
COMPARISON OF SISO & MIMO TECHNIQUES IN WIRELESS COMMUNICATION
 
Design of Filters PPT
Design of Filters PPTDesign of Filters PPT
Design of Filters PPT
 
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal ProcessingDSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
DSP_2018_FOEHU - Lec 1 - Introduction to Digital Signal Processing
 
Cognitive radio networks
Cognitive radio networksCognitive radio networks
Cognitive radio networks
 
MIMO.ppt (2) 2
MIMO.ppt (2) 2MIMO.ppt (2) 2
MIMO.ppt (2) 2
 
Lecture 11
Lecture 11Lecture 11
Lecture 11
 
Channel equalization
Channel equalizationChannel equalization
Channel equalization
 
Image Filtering in the Frequency Domain
Image Filtering in the Frequency DomainImage Filtering in the Frequency Domain
Image Filtering in the Frequency Domain
 
Unit iv wcn main
Unit iv wcn mainUnit iv wcn main
Unit iv wcn main
 
Attenuators and phase shifters 24
Attenuators and phase shifters 24Attenuators and phase shifters 24
Attenuators and phase shifters 24
 
Multirate DSP
Multirate DSPMultirate DSP
Multirate DSP
 
Basics of Digital Filters
Basics of Digital FiltersBasics of Digital Filters
Basics of Digital Filters
 
OFDM for LTE
OFDM for LTEOFDM for LTE
OFDM for LTE
 
floating point multiplier
floating point multiplierfloating point multiplier
floating point multiplier
 
Intelligent Networks
Intelligent NetworksIntelligent Networks
Intelligent Networks
 
Qudrature Amplitude Modulation by Krishna Teja & Sunil
Qudrature Amplitude Modulation by Krishna Teja & SunilQudrature Amplitude Modulation by Krishna Teja & Sunil
Qudrature Amplitude Modulation by Krishna Teja & Sunil
 
HFSS using FEM
HFSS using FEMHFSS using FEM
HFSS using FEM
 
DSP_2018_FOEHU - Lec 06 - FIR Filter Design
DSP_2018_FOEHU - Lec 06 - FIR Filter DesignDSP_2018_FOEHU - Lec 06 - FIR Filter Design
DSP_2018_FOEHU - Lec 06 - FIR Filter Design
 

Similar to Radio Signal Classification with Deep Neural Networks

4G-Fourth Generation Mobile Communication System
4G-Fourth Generation Mobile Communication System4G-Fourth Generation Mobile Communication System
4G-Fourth Generation Mobile Communication System
Safaet Hossain
 
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
Berna Bulut
 
Advanced_DSP_J_G_Proakis.pdf
Advanced_DSP_J_G_Proakis.pdfAdvanced_DSP_J_G_Proakis.pdf
Advanced_DSP_J_G_Proakis.pdf
HariPrasad314745
 
Course-Notes__Advanced-DSP.pdf
Course-Notes__Advanced-DSP.pdfCourse-Notes__Advanced-DSP.pdf
Course-Notes__Advanced-DSP.pdf
ShreeDevi42
 
Digital modulation basics(nnm)
Digital modulation basics(nnm)Digital modulation basics(nnm)
Digital modulation basics(nnm)
nnmaurya
 
Chapter 6m
Chapter 6mChapter 6m
Chapter 6mwafaa_A7
 
Positioning techniques in 3 g networks (1)
Positioning techniques in 3 g networks (1)Positioning techniques in 3 g networks (1)
Positioning techniques in 3 g networks (1)kike2005
 
SPREAD SPECTRUM MODULATION.pptx
SPREAD SPECTRUM MODULATION.pptxSPREAD SPECTRUM MODULATION.pptx
SPREAD SPECTRUM MODULATION.pptx
DeepakBaghel50
 
fundamentals of radio RF frequency concept.pdf
fundamentals of radio RF frequency concept.pdffundamentals of radio RF frequency concept.pdf
fundamentals of radio RF frequency concept.pdf
ThobiusJoseph3
 
Lte overview titus
Lte overview titusLte overview titus
Lte overview titus
Nabil Al_Mutawakel
 
LTE-Advanced Physical Layer
LTE-Advanced Physical LayerLTE-Advanced Physical Layer
LTE-Advanced Physical Layer
Praveen Kumar
 
Multiband Transceivers - [Chapter 3] Basic Concept of Comm. Systems
Multiband Transceivers - [Chapter 3]  Basic Concept of Comm. SystemsMultiband Transceivers - [Chapter 3]  Basic Concept of Comm. Systems
Multiband Transceivers - [Chapter 3] Basic Concept of Comm. Systems
Simen Li
 
cdma-technology_compress.pdf
cdma-technology_compress.pdfcdma-technology_compress.pdf
cdma-technology_compress.pdf
HODece24
 
Cisco crs1
Cisco crs1Cisco crs1
Cisco crs1wjunjmt
 
TCAM Design using Flash Transistors
TCAM Design using Flash TransistorsTCAM Design using Flash Transistors
TCAM Design using Flash Transistors
Viacheslav Fedorov
 

Similar to Radio Signal Classification with Deep Neural Networks (20)

4G-Fourth Generation Mobile Communication System
4G-Fourth Generation Mobile Communication System4G-Fourth Generation Mobile Communication System
4G-Fourth Generation Mobile Communication System
 
LTE Air Interface
LTE Air InterfaceLTE Air Interface
LTE Air Interface
 
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
Cross-Layer Design of Raptor Codes for Video Multicast over 802.11n MIMO Chan...
 
Advanced_DSP_J_G_Proakis.pdf
Advanced_DSP_J_G_Proakis.pdfAdvanced_DSP_J_G_Proakis.pdf
Advanced_DSP_J_G_Proakis.pdf
 
Course-Notes__Advanced-DSP.pdf
Course-Notes__Advanced-DSP.pdfCourse-Notes__Advanced-DSP.pdf
Course-Notes__Advanced-DSP.pdf
 
Thesis
ThesisThesis
Thesis
 
Thesis
ThesisThesis
Thesis
 
Digital modulation basics(nnm)
Digital modulation basics(nnm)Digital modulation basics(nnm)
Digital modulation basics(nnm)
 
Chapter 6m
Chapter 6mChapter 6m
Chapter 6m
 
Positioning techniques in 3 g networks (1)
Positioning techniques in 3 g networks (1)Positioning techniques in 3 g networks (1)
Positioning techniques in 3 g networks (1)
 
SPREAD SPECTRUM MODULATION.pptx
SPREAD SPECTRUM MODULATION.pptxSPREAD SPECTRUM MODULATION.pptx
SPREAD SPECTRUM MODULATION.pptx
 
fundamentals of radio RF frequency concept.pdf
fundamentals of radio RF frequency concept.pdffundamentals of radio RF frequency concept.pdf
fundamentals of radio RF frequency concept.pdf
 
Lte overview titus
Lte overview titusLte overview titus
Lte overview titus
 
LTE-Advanced Physical Layer
LTE-Advanced Physical LayerLTE-Advanced Physical Layer
LTE-Advanced Physical Layer
 
IMT Advanced
IMT AdvancedIMT Advanced
IMT Advanced
 
Multiband Transceivers - [Chapter 3] Basic Concept of Comm. Systems
Multiband Transceivers - [Chapter 3]  Basic Concept of Comm. SystemsMultiband Transceivers - [Chapter 3]  Basic Concept of Comm. Systems
Multiband Transceivers - [Chapter 3] Basic Concept of Comm. Systems
 
cdma-technology_compress.pdf
cdma-technology_compress.pdfcdma-technology_compress.pdf
cdma-technology_compress.pdf
 
Cisco crs1
Cisco crs1Cisco crs1
Cisco crs1
 
Tdm fdm
Tdm fdmTdm fdm
Tdm fdm
 
TCAM Design using Flash Transistors
TCAM Design using Flash TransistorsTCAM Design using Flash Transistors
TCAM Design using Flash Transistors
 

Recently uploaded

Q1’2024 Update: MYCI’s Leap Year Rebound
Q1’2024 Update: MYCI’s Leap Year ReboundQ1’2024 Update: MYCI’s Leap Year Rebound
Q1’2024 Update: MYCI’s Leap Year Rebound
Oppotus
 
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
dwreak4tg
 
一比一原版(BU毕业证)波士顿大学毕业证成绩单
一比一原版(BU毕业证)波士顿大学毕业证成绩单一比一原版(BU毕业证)波士顿大学毕业证成绩单
一比一原版(BU毕业证)波士顿大学毕业证成绩单
ewymefz
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
u86oixdj
 
Empowering Data Analytics Ecosystem.pptx
Empowering Data Analytics Ecosystem.pptxEmpowering Data Analytics Ecosystem.pptx
Empowering Data Analytics Ecosystem.pptx
benishzehra469
 
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
pchutichetpong
 
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
ukgaet
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
Subhajit Sahu
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
TravisMalana
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
ewymefz
 
一比一原版(UofS毕业证书)萨省大学毕业证如何办理
一比一原版(UofS毕业证书)萨省大学毕业证如何办理一比一原版(UofS毕业证书)萨省大学毕业证如何办理
一比一原版(UofS毕业证书)萨省大学毕业证如何办理
v3tuleee
 
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
Timothy Spann
 
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
2023240532
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP
 
Opendatabay - Open Data Marketplace.pptx
Opendatabay - Open Data Marketplace.pptxOpendatabay - Open Data Marketplace.pptx
Opendatabay - Open Data Marketplace.pptx
Opendatabay
 
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
vcaxypu
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Subhajit Sahu
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
jerlynmaetalle
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
ewymefz
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
nscud
 

Recently uploaded (20)

Q1’2024 Update: MYCI’s Leap Year Rebound
Q1’2024 Update: MYCI’s Leap Year ReboundQ1’2024 Update: MYCI’s Leap Year Rebound
Q1’2024 Update: MYCI’s Leap Year Rebound
 
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
一比一原版(BCU毕业证书)伯明翰城市大学毕业证如何办理
 
一比一原版(BU毕业证)波士顿大学毕业证成绩单
一比一原版(BU毕业证)波士顿大学毕业证成绩单一比一原版(BU毕业证)波士顿大学毕业证成绩单
一比一原版(BU毕业证)波士顿大学毕业证成绩单
 
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
原版制作(Deakin毕业证书)迪肯大学毕业证学位证一模一样
 
Empowering Data Analytics Ecosystem.pptx
Empowering Data Analytics Ecosystem.pptxEmpowering Data Analytics Ecosystem.pptx
Empowering Data Analytics Ecosystem.pptx
 
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
Data Centers - Striving Within A Narrow Range - Research Report - MCG - May 2...
 
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
一比一原版(UVic毕业证)维多利亚大学毕业证成绩单
 
Adjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTESAdjusting primitives for graph : SHORT REPORT / NOTES
Adjusting primitives for graph : SHORT REPORT / NOTES
 
Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)Malana- Gimlet Market Analysis (Portfolio 2)
Malana- Gimlet Market Analysis (Portfolio 2)
 
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
一比一原版(IIT毕业证)伊利诺伊理工大学毕业证成绩单
 
一比一原版(UofS毕业证书)萨省大学毕业证如何办理
一比一原版(UofS毕业证书)萨省大学毕业证如何办理一比一原版(UofS毕业证书)萨省大学毕业证如何办理
一比一原版(UofS毕业证书)萨省大学毕业证如何办理
 
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
06-04-2024 - NYC Tech Week - Discussion on Vector Databases, Unstructured Dat...
 
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
Quantitative Data AnalysisReliability Analysis (Cronbach Alpha) Common Method...
 
Criminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdfCriminal IP - Threat Hunting Webinar.pdf
Criminal IP - Threat Hunting Webinar.pdf
 
Opendatabay - Open Data Marketplace.pptx
Opendatabay - Open Data Marketplace.pptxOpendatabay - Open Data Marketplace.pptx
Opendatabay - Open Data Marketplace.pptx
 
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
一比一原版(ArtEZ毕业证)ArtEZ艺术学院毕业证成绩单
 
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
Levelwise PageRank with Loop-Based Dead End Handling Strategy : SHORT REPORT ...
 
The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...The affect of service quality and online reviews on customer loyalty in the E...
The affect of service quality and online reviews on customer loyalty in the E...
 
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
一比一原版(UofM毕业证)明尼苏达大学毕业证成绩单
 
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
一比一原版(CBU毕业证)不列颠海角大学毕业证成绩单
 

Radio Signal Classification with Deep Neural Networks

  • 1. Radio Signal Classification with Deep Neural Networks 2018 Army Signal Classification Challenge 6th place solution Kachi Odoemene 29 Aug 2018
  • 2. Competition • Army Rapid Capabilities Office (RCO) – April 30th – August 13th 2018 – $150,000 in prizes • Radio signal modulation classification – Automatically identify modulation type of received radio signal – 24 modulation classes, including noise class
  • 3. Competition • Finished 6th place (of 161 participants)
  • 4. Radio Signal Modulation Schemes Binary Frequency Shift Keying (BFSK)
  • 5. Quadrature Signals • I/Q: In-phase and Quadrature components • 90° shift between identical periodic signals – e.g. sine and cosine wave Q I
  • 6. I/Q Modulation • Summation of I/Q pairs results in any modulation (frequency, amplitude, phase, etc) • Transmitted & received radio signal represented as I/Q components Modulated RF Signal I channel, I(t) Q channel, Q(t) Inputs (Modulating waves) Carrier wave
  • 7. I/Q Time Domain Examples – high SNR PSK: phase shift keying FSK: frequency shift keying ASK: amplitude shift keying MSK: Minimum (frequency) shift keying QAM: quadrature amplitude modulation CP: continuous phase OQ: offset quadrature SNR: Signal to Noise
  • 8. I/Q Time Domain Examples – low SNR
  • 9. Dataset • > 30 GB – 4.32 million I/Q instances • Dimension: 2 x 1024 – Short time window, real world conditions – 24 modulation classes – White noise added to signals • 6 signal-to-noise (SNR) levels • Test datasets (public) – 2 sets of 100000 I/Q instances (unlabeled)
  • 10. Evaluation Metric • Derived from multiclass log loss Score =
  • 11. Goal • Automatically identify the modulation type of the received radio signal – What modulation format was employed? – Eg. Military applications: • Spectrum surveillance, electronic warfare, and threat analysis – Identify modulation type of intercepted enemy communication
  • 12. Traditional approach • Hand-crafted features (feature engineering) – Higher order statistics – Autocorrelation and spectral correlation functions – Measures derived from instantaneous frequency, amplitude, phase • Mean, standard deviation, kurtosis, etc – Small number of features (28-32) • SVM, decision trees, ensemble, neural networks
  • 13. Deep Neural Networks for Radio Modulation Recognition (2016) Convolutional networks outperform expert feature-based classifiers
  • 14. • Proposed 3 additional architectures for modulation recognition: – Inception – ResNet – Hybrid of Convolutional, Long short term memory (LSTM), and Fully Connected (FC) Deep Neural Network (CLDNN) • No source code provided • Sparse details on architecture hyperparameters (2017)
  • 15. Residual Unit Layer Input Model 1: ResNet (original) • Image classification • CNNs with skip (residual or shortcut) connections – Feed previous representations (activations) into downstream layers – Prevents information loss • Enables training of deeper networks – 100s to 1000s of layers He et al Deep Residual Learning for Image Recognition (ArXiv 2015) Skipconnection
  • 16. Input (2x1024) Conv (1x128) D-Conv (2x1) Avg. Pool “ResBlock” x4 Global Avg Pool FC (128) x2 Softmax (24) (Temporal) (Spatial) Model 1: ResNet (modified) • Temporal convolution on each IQ channel separately • Depth-wise convolution (D- Conv) • Batch normalization after convolutional and fully connected (FC) layers • Multiple residual units within ResBlock • # Parameters: 255,944 He et al Deep Residual Learning for Image Recognition (ArXiv 2015)
  • 17. Model 2: CLDNN (original) • Speech recognition • Unified model: CNN, LSTM, FC • CNN: reduce spectral variations of input data • LSTM: learn temporal structure • FC: transform LSTM features into output easy to classify Sainath et al Convolutional, Long Short-Term Memory, fully connected Deep Neural Networks (IEEE, 2015) Convolutional layers Linear layer LSTM layers Fully Connected layers Output targets (1) (2) C C L L D D Dim. red.Xt [Xt-l,..,Xt,…,Xt+r]
  • 18. Model 2: CLDNN (modified) • Temporal convolution on each IQ channel separately • Depth-wise convolution (D-Conv) • Batch normalization after each convolutional and fully connected (FC) layers • “ConvBlock”: Conv + BatchNorm + ReLU • Dropout between FC layers • # Parameters: 147,480 Sainath et al Convolutional, Long Short-Term Memory, fully connected Deep Neural Networks (IEEE, 2015) Input (2x1024) Conv (1x128) D-Conv (2x1) Avg. Pool ConvBlock x2 Concatenate LSTM (48) x2FC (128) Softmax (24) (Temporal) (Spatial)
  • 19. Data preparation • Raw data, no preprocessing • Trained from scratch • Data split – Train: 80%, Valid: 13.33%, Holdout: 6.67%
  • 20. Implementation details • Hyper-parameter selection – Temporal filter kernel size – Number of FC units • Maximum of 25 epochs, early stopping • Adam optimizer • Learning rate: 1e-3 • Keras (Tensorflow) • Hardware (Personal) – GTX 1080Ti (11 GB) GPU – 16 GB RAM
  • 22. Accuracy per modulation class: all SNRs
  • 23. Accuracy per modulation class: high SNR
  • 24. Accuracy per SNR: pooled across classes
  • 25. Accuracy per SNR: individual classes PSK: phase shift keying FSK: frequency shift keying ASK: amplitude shift keying MSK: Minimum (frequency) shift keying QAM: quadrature amplitude modulation CP: continuous phase OQ: offset quadrature
  • 26. Leaderboard performance Model TestSet1 Score TestSet2 Score Final Score CLDNN 42.09 54.22 50.58 ResNet-18 33.27 58.27 50.77 Combined 42.09 58.27 53.41 Final Score = 0.3*TestSet1 + 0.7* TestSet2
  • 28. Ensembling • Trained meta-learner to combine predictions from multiple models
  • 29. Leaderboard performance: Ensembling Model TestSet1 Score TestSet2 Score Final Score CLDNN 42.09 54.22 50.58 ResNet-18 33.27 58.27 50.77 Combined 42.09 58.27 53.41 Logistic Regression 45.17 63.25 57.83 Light GBM 46.40 64.54 59.10 Final Score = 0.3*TestSet1 + 0.7* TestSet2
  • 30. Challenges • Structural (competition organization) – Multiple changes to scoring procedure and test set, submission site shutdown, etc • Time constraint – 1 submission/day (10am) & leaderboard update (5pm) – On Kaggle: up to 5 submissions/day, immediate LB update • Technical – Hardware failure final week of competition • Backup: Google Colab
  • 31. Future Efforts • Incorporate additional features: – Amplitude – Phase difference – Magnitude of Fourier transform – Spectrogram • Explore other architectures & ensembling methods – Inception-like architecture: process and combine multiple frequency scales • Model interpretability – DeepLIFT (Deep Learning Important FeaTures) – LIME (Local Interpretable Model-agnostic Explanations) – Ablation studies

Editor's Notes

  1. Radio Signal Modulation Recognition Radio Signal Classification with Neural Networks
  2. Blind signal classification- little to no prior knowledge of signal being detected
  3. Waves consists of phase, amplitude, and frequeny Input signal is the data/information you wish to transmit. This information is mixed with a carrier wave. Analog modulation: input signal varies continuously Digital modulation: input signal are discrete values of 1’s and 0’s. analog data such as speech (sampled at some rate) is first compressed and then converted into bit stream (1s and 0s). This is used as the input signal. Also include phase modulation and combinations of all three Image from https://www.taitradioacademy.com/topic/how-does-modulation-work-1-1/
  4. Radio signal represented as I/Q component Modulating inputs are the data to be sent RF carrier waves are in quadrature, 90deg shift ----- Meeting Notes (8/28/18 04:45) ----- Right down equation of received signal from OShea paper Explain that this is transmitted signal
  5. Panel of images (n=24) Received I/Q samples sampled from a desired carrier frequency
  6. Panel of images (n=24)
  7. Hardware (Personal) GTX 1080Ti (11 GB) GPU 16 GB RAM
  8. Minimize logloss Probability assigned to each class Heavily penalizes confident(high probability) and wrong predictions more than it rewards confident (high probability) and correct predictions the sum of log loss values for each class prediction in the observation Nuanced view of model performance, takes into account uncertainty of predictions.
  9. Traditional approach to automatic modulation classification Expensive and tedious to develop analytically, also not flexible Digital modulations e.g. HOS (moments and cumulants) and Analog modulations for measures derived from instantaneous phase, amplitude, frequency, General example of expert feature: compute the m-th order statistic on n-th power of the instantaneous or time delayed received signal max value of spectral power density of normalized instantaneous amplitude, standard deviation of absolute value of nonlinear component of instantaneous phase of non ----- Meeting Notes (8/28/18 04:45) ----- extract 28-32 features
  10. Fortunately the hard work of figuring out whether neural networks can be successfully applied to the problem of modulation recognition had been worked out by Tim O’Shea and colleagues Harder these days to find a problem where deep learning has not already been applied Learn features from the data directly ----- Meeting Notes (8/28/18 18:35) ----- Replace with deep neural networks
  11. Proposal based on well performing architectures in computer vision Like most papers posted on Arxiv, the paper did not provide source code for the models and was very light on details, hence I had to reverse engineer and run test experiments Settled on two models Done with 8 to 11 modulation classes, not clear whether it would scale to 24 modulations
  12. I made modifications to original ResNet architecture, namely adding the temporal convolution Skip connection or shortcut, which feed activations from one layer into the next layer Allows training of deep networks Plain networks without skip connections tend to struggle when layers are too deep, unable to choose parameters, leading to overfitting. Residual networks on the other hand, learn the identity function because of the skip connections, which leads to better training ----- Meeting Notes (8/28/18 18:35) ----- fix skip connection
  13. Parameters less than the number of samples, helps to regularize the model
  14. CLDNN model first developed by Google for speech recognition applications leverage advantages of convolution and rnn layer CNNs to learn spectral features and reduce spectral variations in the input LSTMs to learn temporal variations FC to transform LSTM features into easily classifiable format LSTMs on their own need better input features, and easily classifiable outputs, hence need help from
  15. CLDNN model first developed by Google for speech recognition applications leverage advantages of convolution and rnn layer One LSTM layer
  16. Hold out set for local evaluation ----- Meeting Notes (8/29/18 05:41) ----- no augmentation
  17. ----- Meeting Notes (8/28/18 04:45) ----- complete learning rate statement ----- Meeting Notes (8/28/18 18:35) ----- Skip!
  18. ----- Meeting Notes (8/28/18 18:35) ----- Fix this figure legend!!
  19. ----- Meeting Notes (8/28/18 04:45) ----- Pooled across SNRs
  20. Might have trouble with high and low frequency
  21. Suggests CLDNN might generalize better to unseen classes
  22. If you’ve ever participated in a machine learning competition on platforms like Kaggle, you know that ensembling predictions from multiple models
  23. ----- Meeting Notes (8/28/18 04:45) ----- calculate percentages final submission bold
  24. Multiple changes to judging criteria and testset Submission site failure 1 submission per day, compared to multiple (up to 5 on Kaggle) ----- Meeting Notes (8/28/18 18:35) ----- update 5pm
  25. Things that I didn’t get to try Magnitude of the square of the ----- Meeting Notes (8/28/18 04:45) ----- Magnitude of the frequency spectrum ----- Meeting Notes (8/28/18 18:35) ----- add ablation!
  26. for organizing and hosting the Army Signal Classification Challenge
  27. Powerful application to real world problems Moving beyond classifiers