plans for stochastic hardware injection

25
Plans for stochastic hardware injection Eric Thrane, Mandy Pihlaja and Vuk Mandic

Upload: inoke

Post on 31-Jan-2016

34 views

Category:

Documents


0 download

DESCRIPTION

Plans for stochastic hardware injection. Eric Thrane, Mandy Pihlaja and Vuk Mandic. Overview. Motivation Injection flow chart Relationship to reviewed code Testing and validation. Summary of injection. spherical harmonic coefficients and spectrum specified - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Plans for stochastic hardware injection

Plans for stochastic hardware injection

Eric Thrane, Mandy Pihlaja and Vuk Mandic

Page 2: Plans for stochastic hardware injection

2

Overview

• Motivation

• Injection flow chart

• Relationship to reviewed code

• Testing and validation

Page 3: Plans for stochastic hardware injection

3

Summary of injection

• spherical harmonic coefficients and spectrum specified

• User sets parameters of GPS, random seed, injection duration

• Parent routine, SphMapToTimeSeries, calls simulateSkyMapTimeDomain, which returns time series [d1, d2] and overlapping time series [Ovrlp1, Ovrlp2] for {LHO,LLO}.

• simulateSkyMapTimeDomain = reviewed stochastic code

Page 4: Plans for stochastic hardware injection

4

Summary cont.• Then…

– FFT both time series– Apply frequency pendulum transfer function– Window time series using sin function– Combine time series and overlapping time series

• Write combined h(t) to text file• sbaniso.c (based on standard injection code,

sb.c) reads in text file and converts to binary • Sent to psinject• Sent to awg

Page 5: Plans for stochastic hardware injection

5

Injection flowchart

SpHMapToTimeSeries*

SpH map file: l, m, Plm spectrum file: f, H(f)user-specified params: random seeds, seg dur, etc.

simulateSkyMapTimeDomain

(reviewed)

d1,d2

gets time seriesgets overlapping time seriesFFT time seriesApply pendulum T(f)IFFTWindowCombine overlapping seriesWrite combined h(t) to files.

*new

sbaniso.cpsinjectawgbinary out 10110010 pg. 23

pg. 6

Page 6: Plans for stochastic hardware injection

6

simulateSkyMapTimeSeries.m

• Returns [d1(t),d2(t)] for LHO, LLO given user-specified P() and H(f).

• Written by Stefan Balmer and Joe Romano

• Reviewed during SpH review.

Page 7: Plans for stochastic hardware injection

7

Validation and testing- fft

• The next four slides show the process of applying FFT, the pendulum transfer function, and the iFFT to the time series.

• Then there is a consistency check to make sure this process could be reversed

Page 8: Plans for stochastic hardware injection

8

time(s)

strain

Original Time Series

Page 9: Plans for stochastic hardware injection

9

Original time series

strain

time(s)

Page 10: Plans for stochastic hardware injection

10

FFT of time series

frequency(Hz)

fft

Page 11: Plans for stochastic hardware injection

11

• Next slide is the frequency domain Pendulum transfer function is given by sb.c

• tfH=ffrequency^2/constants

• Constants = (2.05e-13)(0.764)(0.764)

Page 12: Plans for stochastic hardware injection

12

Pendulum Transfer Function

frequency(Hz)

P(f)

Page 13: Plans for stochastic hardware injection

13

FFT multiplied by tfH

frequency(Hz)

Time series

Page 14: Plans for stochastic hardware injection

14

ifft back to time series

time(s)

Time series

Page 15: Plans for stochastic hardware injection

15

• In order to check that the pendulum transfer function was applied correctly, the transformed time series had the following inverse process applied– FFT, inverse pendulum transfer function,

iFFT

• The plot of the ratio of the ‘inverted’ time series to the original time series is on the next slide.

• Within very small rounding, the original time series is recovered (~1.0)

Page 16: Plans for stochastic hardware injection

16

Ratio of two time series

time(s)

1

ratio

1.07

Page 17: Plans for stochastic hardware injection

17

Windowing

• Each time series segment is multiplied by the sine function in order for beginning and end to scale to zero.

Page 18: Plans for stochastic hardware injection

18

Windowed time series segment

time(s)

counts

Page 19: Plans for stochastic hardware injection

19

Overlapping and combination

• The overlapping time series lags the original time series by half of segment duration.

• The second half of the current overlapping time series and the first half of the previous overlapping time series are concatenated to make the combined overlapping time series

• The combined series is added to the original time series to form the final time series

Page 20: Plans for stochastic hardware injection

20

Final time series

time (s)

counts

Page 21: Plans for stochastic hardware injection

21

• No modulation of the envelope

• Rms value of the pre-windowed time series = 2.2348

• Rms value of the combined time series = 2.2334

• The close rms values of the two time series shows apparent agreement

Page 22: Plans for stochastic hardware injection

22

Checking code at LHO and LLO

• The code was run at LLO and LHO given the same random seed to check that the time series were the same

• Data in Xcel sheet (next page)

Page 23: Plans for stochastic hardware injection

23

Set with “LHO”

Set with “LLO”

LHO LLO LHO LLO

2.148124 2.148124 -1.29882 -1.29882

-0.082698

-0.082698

1.384901 1.384901

-0.611360

-0.611360

2.570094 2.570094

0.864168 0.864168 2.751956 2.751956

1.978097 1.978097 2.107497 2.107497

Page 24: Plans for stochastic hardware injection

24

sb.c• Originally generates time series, applies

transfer function, overlaps segments and sends to AWG

• Modified to read in text files created by SphMapToTimeSeries.m, send to AWG, and delete file after use (sbaniso.c)

• Already used for S5 (w/o Vuk’s hacks) and reviewed

Page 25: Plans for stochastic hardware injection

25

Handshake

• Day long injection generated in one minute segments

• In order to not create a 1TB injection file, SphMapToTimeSeries.m only generates three files at a time and waits for sbaniso.c to delete a file before continuing.