22
Dr. Weisi Guo Inferring Digital and Physical Environment Knowledge from Mobile Phone Signals

Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

  • Upload
    others

  • View
    0

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Dr. Weisi Guo

Inferring Digital and Physical Environment

Knowledge from Mobile Phone Signals

Page 2: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Brief Bio

2

I am an Assistant Professor in Engineering, at the University of

Warwick, and associated to CUSP.

• Academic Background 1. University of Cambridge: MEng (2005), MA, PhD (2011)

2. Published over 40 papers in wireless communications

3. Research interests: wireless sensing, urban communication networks.

• Industrial Background 1. 2 years as radio-engineer at T-Mobile International (2005-07)

2. 2 years undertaking joint academic-industrial research with

Vodafone, NEC and Fujitsu (2011-12)

3. Author of industrial copyright wireless network simulator

VCEsim

Page 3: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Idea

3

To allow each user to passively monitor its local environment, and

link this to Maria Liakata’ s research on human emotions.

• Passive 1. Extract information using existing ambient radio signals, i.e., without

purpose built sensor hardware or additional electronic signals

2. Extract information in continuously, without human triggers

• Environment 1. Digital Environment in terms of data activity of yourself, and the people

around you.

2. Physical Environment in terms of terrain type and vibrancy of physical

objects (i.e., people, cars) around you.

Page 4: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Motivation

4

Why is passive monitoring of the environment of benefit?

• Digital Activity Level Cross operator understanding of total data activity level has significant

commercial and social implications. 1. Long term trends can reveal where to improve the wireless network

2. Short term trends can advice users on where to seek greatest data service

• Physical Environment Dynamic environment changes that is not available from quasi-static

databases such as Google maps. 1. Understanding how the city environment changes over time

2. Understanding how other human and city factors are associated with physical

environment dynamism

Page 5: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Dr. Weisi Guo

In collaboration with S. Wang of South Australia University

Part 1: Digital Environment

Page 6: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Background Knowledge

6

In wireless communications, data is transmitted on certain

frequency bands.

• Multiple Access System Finite frequency bands (spectrum) forces us to reuse the

spectrum across a large area, which means: 1. Multiple transmissions co-exist on same frequency

2. These co-frequency transmissions interfere with each other

• Interference Noise Tolerating the level of interference noise at the receiver underpins

design performance. Interference has properties: 1. Non-coherent and Aggregated: the noise is aggregated from one or

multiple sources, with no ability to discern which set of transmitters.

2. Measurement: Improved hardware allows us now to measure this very

cheaply ($200), compared to $10,000+ many years ago.

Note: incoherent inference channels are not to be confused with coherent pilot channels used to identify cells

Page 7: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Example Cellular Network

7

Research Challenge: 1. How do we know which set of transmitters are transmitting data?

2. How do we know how much data is transmitted?

-3000 -2000 -1000 0 1000 2000 3000-2500

-2000

-1500

-1000

-500

0

500

1000

1500

2000

2500

X

Y

-150

-145

-140

-135

-130

-125

-120

-115

-110

Page 8: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Benefits over Current Methods

8

There are a number of current methods used to extract data usage:

Physical

Extraction per

Device

Postcode and

Census Data

Sensing

Interference

(very novel!)

Accuracy Very High Low

(no outdoor)

High

Cost Medium Low

(offline)

Medium

(very complex)

Resolution High Medium

(residential)

Medium

(no mobile ID)

Privacy / Commercial

Sensitivity

Extremely

High

Low Low

Page 9: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

How is it done?

9

We have had our initial concept accepted for publication later this year [1]. Whilst it is

plausible to map every interference power pattern to a geographical location,

computationally it is too complex. Per spatial grid, need 105 x 2N unique patterns, where N

is the number of interference sources.

Mathematical model of network interference distribution

Using Stochastic Geometry [2], which requires knowledge of:

1. Spatial distribution of network nodes

2. Statistical radio-wave propagation model

[1] “Mobile Crowd-Sensing Wireless Activity using Measured Interference Power”, W. Guo & S. Wang, IEEE Wireless

Communications Letters, to appear, Sep 2013

[2] “Stochastic Geometry for Wireless Communications”, M. Haenggi, Cambridge University Press, Aug 2012

Page 10: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Expected Interference Power

10

We will flash some key equations, but the paper explains their origins in

detail.

1. Let Ir denote a particular value of the aggregate interference power random variable IR:

2. Given knowledge of the way a network is distributed (random uniform in this case), the

probability distribution of IR can be found. The expectation of the interference power

received at any point in space is given by:

Page 11: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Relation to Traffic Intensity

11

We define the traffic intensity concept and how it relates to the data

volume.

1. We define the traffic intensity (A) as a normalized value between 0 and 1, where A =

Traffic Load / Maximum Capacity of Cell. For a multi-cell network, this corresponds directly

to the ratio between the number of transmitting cells λ, and the total number of deployed

cells χ, at some particular frequency band f0.

2. Given we know the mean interference power as a function of the number of transmitting

cells, we can find the sensed traffic intensity:

Page 12: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Validation Results

12

Below are some validation results

Traffic sensing error as a function of sample size

We need ~50 interference power samples per frequency band. The advantage

of this technique is that the spatial sample resolution only needs to be within a

cell location area (no GPS data is needed). We are working on an alternative

version that improves spatial resolution of output and input.

Page 13: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Single Network Test Results

13

We simulated an example urban network

This figure shows a cellular network’s coverage area. Some accuracy is lost on

the borders of the map. This is a temporal-spatial heat map of wireless activity

level for a single wireless network.

-4000

-2000

0

2000

4000

-3000

-2000

-1000

0

1000

2000

3000

-140

-120

-100

-80

XY

Page 14: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

-3000

-2000

-1000

0

1000

2000

3000

-3000

-2000

-1000

0

1000

2000

3000

0

0.5

1

1.5

XY

Single Network Test Results

14

Single network results are only for validation purposes. It is unlikely to reveal data that

is more information than what the network operator already knows. Nonetheless, here

are some maps for a single operator.

Page 15: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Across Network Test Results

15

We now expand the method to cover 5 commercial operators (Vodafone, EE, O2, 3, BT),

and across their different services Wi-Fi + 3G + 4G.

-3000

-2000

-1000

0

1000

2000

3000 -3000

-2000

-1000

0

1000

2000

3000

2

3

4

5

YX

-3000

-2000

-1000

0

1000

2000

3000

-3000

-2000

-1000

0

1000

2000

3000

0

0.5

1

1.5

XY

1

2

3

Page 16: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Discovering New Hotspots

16

We now expand the method to cover 5 commercial operators (Vodafone, EE, O2, 3, BT),

and across their different services Wi-Fi + 3G + 4G.

-3000

-2000

-1000

0

1000

2000

3000 -3000

-2000

-1000

0

1000

2000

3000

2

3

4

5

YX

1

2

3

1 Regent Park “People in parks are wirelessly very active,

but who is providing this service and what is

their usage pattern?”

2

New Business & Commerce

Centre “People in shopping centre is very active,

but who is providing their wireless

network?”

3

Westminster Uni. and BT “Not as active as imagined, but this area

has proportionally more customers on

competitor networks.”

Page 17: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Urban Science Application

17

Clearly, what has been demonstrated is from an engineering and industrial

perspective. There is however strong urban science application:

• Understanding the Digital Economy 1. Understand the economic relationship between wireless information flow and local

economic growth.

2. Quantify the benefit of wireless infrastructure investment

• Understanding the Human Emotion 1. Our thoughts and emotions are increasingly connected with digital information

availability. What is the relationship between digital activity and our emotions?

2. We communicate our thoughts and observations wirelessly through the internet,

can this mobile traffic volume uncover patterns in the city?

Page 18: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Dr. Weisi Guo

Part 2: Physical Environment

Page 19: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Some Basic Knowledge

19

In wireless communications, data is transmitter on certain frequency

bands, and the signal propagates through a complex environment.

• Multipath Different time of arrival due to multiple terrain paths in cities: 1. Causes constructive and destructive combining at receiver

2. Causes phase / frequency shifts

• Multipath Models Different mathematical models describe the nature of the signal’s propagation: 1. Long range Manhattan model: Rayleigh Distribution

2. Short range model: Rician Distribution

Page 20: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Experimental Setup

20

Recently in an Mobicom paper [3], researchers said that they can accurately

detect body movement using Wi-Fi signals, rather than cameras.

• Novelty If the signal transmitter and receiver is positioned for

this single purpose, then this is not new (Radar)

• Our Experiment

We aim to use ambient signals that already exist in the world, and analyse their multipath

and other property (i.e., Doppler shift).

[3] “Whole-Home Gesture Recognition Using Wireless Signals”, Q. Pu and S. Gupta and S. Gollakota and S. Patel, Mobile

Computing and Networking Conference, to appear, 2013

Page 21: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Urban Science Application

21

The question is, why would you want to monitor urban vibrancy and why can’t

it be observed from alternative methods (e.g. cameras):

• Why 1. Our emotions are affected by what happens around us. Continuous high resolution

data of the volume of vibrancy around us is of interest.

2. It also serves as a city wide data collection service to detect human movement on a

statistical level, without intruding into their privacy.

• Methodology 1. Passive: no additional hardware is required. It is possible to do this with

smartphones.

2. Private: no user identity is sensed in the process. The data is statistical.

Page 22: Inferring Digital and Physical Environment Knowledge from Mobile … · link this to Maria Liakata’ s research on human emotions. • Passive 1. Extract information using existing

Dr. Weisi Guo

Thank You for Listening and

Thank you Maria for Organising