compressed sensing mask feature in time … poster_0.pdfa xt x t e dt wdt fe dtdf ambiguity function...

RESEARCH POSTER PRESENTATION DESIGN © 2015

www.PosterPresentations.com

Motivation: Civil Flights Recognition Feature selection: a mask of ambiguity function ConclusionsIn this paper, a compressed sensing based feature ambiguity function mask is proposed to serve as a time-frequency feature for emitter recognition. The sparse CS-mask owns abundant time frequency information for radarprint feature extraction. This approach has some merits:• Cover more time varying information of signals • Avoid high dimension redundancy• Possess better accuracy in civil flight meteorological radar identification• own higher physical representation

References[1] X. H. Ru, Z. Liu, W. L. Jiang, Z. T. Huang, “Recog-nition performance analysis of instantaneous phase and its transformed features for radar emitter identification,” Radar, Sonar & Navigation, IET, vol. 10, no. 5, pp. 945-952, 2016.[2]L. Wang, H. B. Ji,Y. Shi, “Feature extraction and optimization of representative-slice in ambiguity function for moving radar emitter recognition,” ICASSP, IEEE, Dallas, 2010.[3] Z. Yang, W. Qiu, H. Sun, N. Arumugam, “Robust radar emitter recognition based on the three-dimensional distribution feature and transfer learning,” Sensors, vol. 16, no. 3, 2016. [4] S. Stankovic, I. Orovic, M. Amin, “Compressed sensing based robust time-frequency representation for signals in heavy-tailed noise,” ISSPA, IEEE, Montreal, 2012.

AcknowledgementThis work is supported by the National Natural Science Foundation of China (No.61701374)We gratitude to the anonymous reviewers whose valuable comments helped to improve the quality of this paper

Email: [email protected]

Mingzhe Zhu, Xinliang Zhang, Yue Qi, Hongbing JiSchool of Electronic Engineering, Xidian University, Xi’an 710071, China

COMPRESSED SENSING MASK FEATURE IN TIME-FREQUENCY DOMAIN FOR CIVIL FLIGHT RADAR EMITTER RECOGNITION

Merits of CS-mask

Recognition System Structure

DataPreprocessing

Feature ExtractionClassificationRecognition

IdentificationFeature

OptimizingFeatureFusion

Local Radar waveDatabase

Local Radar waveDatabase

Feed Back for Optimization

Pulse Input

Result Output

Masked Ambiguity Function: Containing IF info

time

frequ

ency

AF based WVD

0.2 0.4 0.6 0.8 1

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

Delay

Dop

pler

Ambiguity Function

-0.4 -0.2 0 0.2 0.4

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

Delay

Dop

pler

Ambiguity Function Mask

-0.015 -0.01 -0.005 0 0.005 0.01 0.015

-0.01

-0.005

0

0.005

0.01

Chirp

WVD AF Mask Invert

Compressed Sensing

Signal

Delay

Dop

pler

Ambiguity Function

-0.4 -0.2 0 0.2 0.4

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

time

frequ

ency

AF based WVD

0.2 0.4 0.6 0.8 1

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

Delay

Dop

pler

Ambiguity Function Mask

-0.015 -0.01 -0.005 0 0.005 0.01 0.015

-8

-6

-4

-2

0

2

4

6

8

10

x 10-3

time

frequ

ency

AF-Mask based WVD

0.2 0.4 0.6 0.8 1

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

Multi component

Mask or Slice?

time

frequ

ency

CS-Mask based WVD

50 100 150 200 250 300 350 400 450

0

0.05

0.1

0.15

0.2

time

frequ

ency

AF-RS based WVD

50 100 150 200 250 300 350 400 450

0

0.05

0.1

0.15

0.2

time

frequ

ency

CS-Mask based WVD

50 100 150 200 250 300 350 400 450

0

0.05

0.1

0.15

0.2

time

frequ

ency

AF-RS based WVD

50 100 150 200 250 300 350 400 450

0.05

0.1

0.15

0.2

Invert WVDby AF Mask

Invert WVDby AF Slice

More:

time varying information

Proposed in:

ICASSP 2010By Wang Lei

Appendix

Radar Wave Viewing: TF domain

1. Time domain 2. TF domain 3. IF Estimation 4. Database operation

GUI function:

Processing Batch

transfer the radar signal pulses into ambiguity function

Feature ExtractionSetting ε

* 2

-

2 ( )

-

( , ) ( ) ( )2 2

( , )

j t

j t f

A x t x t e dt

WD t f e dtdf

Ambiguity function is defined as:

If considering WVD and ambiguity as finite length N points sequences, then the matrix correlation of them as follows:

( 1) ( ) ( 1)f N N N D NA W

Ψ is 2D Fourier transform matrix. Φ is measurement matrix. CS-mask feature based on AF is defined as：

)D(NN)(NNM)f(NNMMaskMf WΨΦAΦA 11)1(

The sparse feature can be obtained according to CS theory, which can be optimized as:

1minMask

f fAA A

12 ( , )

1 1

1. . ( - )N M

-d f D

k n

s t A WN M

F

0 10 20 30 40 50 60 70 80 90 100

010

2030

4050

6070

8090

100-6

-4

-2

0

2

4

6

8

10

12

N

3D ambiguity function

f/Hz

pow

er/d

B

Classification Result

Civil Flight Base 01

Ground Radar Base 02


• Four databases including two ground radar bases and two civil flight bases.

• CS-mask owns higher accuracy and robustness in data sets I and IV.

• CS-mask and PSE both performs higher accuracy in data sets II and III.

• CS-mask preforms much more prevalent and stable in general.

Big Radar Wave DatabaseOur Lab own a big Radar waveform database, include measured signals from different kinds of Radar emitter.

Building Civil Flights DatabaseOur Lab own a big Radar waveform database, include measured signals from different kinds of Radar emitter.

Radar Pulse Database

1 2 3 4 5 6 7 8 9 100

10

20

30

40

50

类别

每类

样本

数

LFM数据库类别统计图

1 2 3 4 5 6 7 8 9 100

10

20

30

40

50

类别

每类

样本

数

PM数据库类别统计图

1 2 3 4 5 6 7 8 9 100

10

20

30

40

50

类别

每类

样本

数

SF数据库类别统计图Signal Generator Mode 01 Signal Generator Mode 02 Signal Generator Mode 03

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

1 2 3 4 5 6 70

50

100

150

X = 1Y = 90

类别

每类

样本

数

民航数据库（一）类别统计图

X = 2Y = 76

X = 3Y = 76

X = 4Y = 30

X = 5Y = 144

X = 6Y = 78

X = 7Y = 114

0 5 10 15 20 25 300

10

20

30

40

50

60

类别

每类

样本

数

民航数据库（二）类别统计图Civil Flight Base 01 Civil Flight Base 02

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

1 2 3 4 5 6 7 8 9 100

10

20

30

40

50

60

70

80

90

类别

每类

样本

数

十一五数据库（一）类别统计图

1 2 3 4 5 6 7 8 90

20

40

60

80

100

120

140

160

180

类别

每类

样本

数

十一五数据库（二）类别统计图

1 2 3 4 5 6 7 8 90

5

10

15

20

25

30

类别

每类

样本

数

十一五数据库（三）类别统计图

1 2 3 4 5 60

50

100

150

200

类别

每类

样本

数

十一五数据库（四）类别统计图

1 2 3 4 5 6 7 8 9 10 11 12 130

50

100

150

200

类别

每类

样本

数

十一五数据库（五）类别统计图

0 5 10 15 20 25 30 35 400

50

100

150

200

类别

每类

样本

数

十一五数据库（六）类别统计图

Ground Radar Base 01 Ground Radar Base 02

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et

Sam

ple

Num

ber i

n D

atas

et


Ground Radar Base 03 Ground Radar Base 04


Radar Pulse Database

Moving Radar

Static Radar

Signal Generator Record

Record Amount in each Subclass More than 4GB Data and 10,000 Records.

点数

数据

编号

十一五数据库六

50 100 150 200

500

1000

1500

2000

2500-600

-400

-200

0

200

400

600

点数

数据

编号

十一五数据库五

50 100 150 200

100

200

300

400

500

600

700

800

-300

-200

-100

0

100

200

300

400

点数

数据

编号

LFM数据库

500 1000 1500 2000 2500

50

100

150

200

250

300

350

400

450

500-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

Points Points Points

Puls

e In

dex

Ground Radar Base 06Ground Radar Base 05Signal Generator Mode 01

Planform of Data Sets

Original Data

Detected Pulse

1st 2nd 3th

When Antenna Aim at Flight

Wave Records Pre-Processing

Mean removal De-noise

PulseAligning

PulseDetection

Removebad class

Method:

Mathematic Model:

Algorithm:

Information Representation:

Identification accuracy:

transfer signals from ambiguity to WVD

optimize feature using formulas (4)-(5)

(1)

(2)

(3)

(4)(5)

0 20 40 60 80 100-1

-0.5

0

0.5

1

N

Am

plitu

de

10 20 30 40 50 60 700

10

20

30

40

50

60

70

80

90

Training percent(%)

Corre

ct R

ate(

%)

Recognition result of Data set 4

AF-RSAF-sliceCyc-spPSECS-mask


10 20 30 40 50 60 7010

20

30

40

50

60

70

80

90

100

Training percent(%)

Corre

ct R

ate(

%)



10 20 30 40 50 60 700

10

20

30

40

50

60

70

80

90

100

Training percent(%)

Corre

ct R

ate(

%)




10 20 30 40 50 60 7010

20

30

40

50

60

70

80

90

100

Training percent(%)

Corre

ct R

ate(

%)


AF-RSAF-sliceCyc-spPSECS-mask Training rate 10% 20% 30% 40% 50% 60% 70%

I

CS-mask 93.42 95.59 95.64 96.62 94.64 96.48 98.33

PSE 39.34 27.82 29.73 40.78 49.62 51.57 61.08

Cyc-sp 56.14 53.79 66.57 69.09 69.52 69.91 71.75

AF-RS 80.10 80.62 82.10 81.82 80.86 73.01 76.00

II

CS-mask 92.01 90.75 92.87 90.57 90.03 93.17 90.16

PSE 93.10 92.18 90.61 91.78 92.80 94.01 89.84

Cyc-sp 92.25 83.82 85.15 66.35 63.93 56.80 51.02

AF-RS 88.69 91.27 89.96 91.06 88.59 89.58 87.70

III

CS-mask 96.74 99.39 96.67 97.54 98.72 99.51 98.72

PSE 98.63 99.10 98.68 98.95 98.28 100.0 98.11

Cyc-sp 83.24 81.23 82.91 85.82 77.16 83.59 82.78

AF-RS 82.91 78.20 79.28 81.09 83.13 83.25 83.02

IV

CS-mask 82.07 83.22 79.41 81.61 82.13 80.74 80.10

PSE 10.45 12.64 13.18 15.49 12.30 12.98 10.24

Cyc-sp 18.70 17.33 19.70 29.01 23.81 32.95 34.63

AF-RS 78.83 79.44 77.53 78.07 78.50 75.97 68.78

compressed sensing mask feature in time … poster_0.pdfa xt x t e dt wdt fe dtdf ambiguity function...

Documents