the spooks behind you cs 791v final project sam delaney

26
The Spooks Behind You CS 791V Final Project Sam Delaney

Upload: kathlyn-norton

Post on 30-Dec-2015

219 views

Category:

Documents


0 download

TRANSCRIPT

The Spooks Behind YouCS 791V Final Project

Sam Delaney

The Rundown

• Review of Project• How to detect a rouge signal transmission.• How we are going to detect a signal transmission• Literature Review• Serial Implementation• Parallel Plan of Attack• Questions

THE REVIEWSignal Detection

Detection Theory

• Detection theory, or signal detection theory, is a means to quantify the ability to discern between information-bearing patterns (called stimulus in humans, signal in machines) and random patterns that distract from the information (called noise, consisting of background stimuli and random activity of the detection machine and of the nervous system of the operator). -wikipedia

Detection Theory Review

• Determine a threshold for detection.• If a signal breaks that threshold, flag the transmission.• The threshold for noise must be tuned.• Too strict or too lax, brains get eaten .

THE MAGICJazz hands

The Data!

• The Data provided is a binary file containing a series of numbers, one real and one imaginary.

• These numbers represent the signal we are receiving.• But they are useless as is. USELESS.

How to be useful

• First we take each set of numbers• One real, one imaginary

• We square each value and sum them the real and imaginary values• This gives us a positive point

• Rinse and Repeat as desired or until you run out of values• (You should use all the values)

But we can make it useful!

Points on a plane, err line

• So great we have a big jagged line of data• MISSION ACCOMPLISHED!

• Oh? We have to do more. Lame• While we now have a signal we can graph, it’s still pretty

messy. • Lots of noise, noise leads to anger, anger leads to hate, hate

leads to the dark side.

• While we can’t completely destroy the noise. (The Force must maintain balance) We can reduce it.

FILTER, Filter, filter

• Make this big jagged line not so jagged

• Many options for filters• Average neighbor• Median• About a million others• No really, you can spend

hours on Wikipedia

Let’s role with Median

• Well, that escalated quickly• Why Median?• For signals it’s considered poor form to lose edges while

filtering• Filters like average neighbor will help reduce noise but at

the cost of edge detection• While median filters are not the only filter we could use it is

perhaps the most easy to implement• I vote easy

How median keeps its edge

• An example of a signal (blue)

• Signal after an average filter is applied (center)

• Signal after a median filter is applied (right)

Step 67

• Now the Easy/Hard part• It’s how you look at it really

• With a clean…ish curve we are now ready to find some spies!

• The Easy Way• After a section of known non transmission, find highest

value and set as threshold. All values above are zombie spies transmitting in your space. (See cheating)

• The Hard Way• Find a section that over time seems to remain relatively

constant. Use that as base point for threshold. However, monitor relative noise level and adjust threshold as need be. Receive PhD for being super smart.

• Constant False Alarm Rate

ALARM

• Constant False Alarm Rate• Adaptive threshold• Many ways to implement• The most advanced use statistics and probability based on past

hits• But for the sake of argument lets use the most simple

SIGNALS A LA SHAKESPEARELiterature Review

What I read on the interwebs

Love

• So many articles!• Computer science has

many problems concerning SDT

• Interesting uses of SDT

Hate

• ~90% where in psychology

• None of them were my problem

• Not interesting uses of SDT

Articles worth reading

• Wikipedia• You saw nothing!

• Clairvoyant and adaptive signal detection in non-Gaussian clutter: A data-dependent threshold interpretation

• Spam filter optimality based on signal detection theory• A behavioral theory of insider-threat risks: A system

dynamics approach• Comparison of anomaly signal quality in common

detection metrics

SEQUENTIAL IMPLEMENTATIONSegmentation Fault

Feed me data yum yum

• Current Implementation• Reads a file with a real and imaginary integer• Uses thread to read in data and store to queue• Also does math on the values to give us the positive point• Sum the squares

• Nothing to write home about

Tap me out!

• Next thread• Take that queue and run through a median filter• The median filter implementation• Fill “window” 32 values• Sort values in window• Median value is the value assigned• Pop off last value, push on new one.

• Keep going while the data thread has work for you.

Let’s do this!

• Main thread• Takes data from median filter• Over first x entrees of the entries we assume no

transmission• Where x is user defined

• Find and retain highest value, set as threshold• After x entrees, any value over the threshold causes flag to

be triggered and inform user with message.

Future Add On’s

• Things that would be nice to have• Implement Constant False Alarm Rate• Make a cool interface, not the command line• CIA endorsement

PARALLEL PLAN OF ATTACKOr PPA for short

HIT THE WEAK POINT!

• Reading data• Streams will be my friend• Need to do tests whether it is worth while to do point

calculation on CPU and pass up, or pass twice the data and do math on card. TESTING!

• Filtering• Median filter on GPU, honestly this could go either way

• Threshold• While the easy pick the highest way should be trivial to

implement need to keep CFAR in mind

Review!

• Signal Detection Theory (SDT)• How to detect rouge transmission• In this problems scope

• Literature Review• Serial Implementation• Parallel Plan of Attack• Review

QUESTIONS?