smart contracts for crowdsourced spectrum sensing · 2020. 7. 16. · spass: spectrum sensing as a...

Smart Contracts for Crowdsourced Spectrum Sensing

Aalto CS Forum Talk,

March 1st, 2019

Suzan Bayhan

Telecommunication Networks Group, TU Berlin

suzanbayhan.github.io

Collaborators: Anatolij Zubow, Piotr Gawlowicz, Adam Wolisz

http://suzanbayhan.github.io

Typical centralized crowdsourcing solutions: requester, workers, crowdsourcing platform

CROWDSOURCING PLATFORM

(Trusted Third Party)

Workers

Requester

!2




Upload Task

Workers

Requester

!2




Upload Task

Register worker

Workers

Requester

!2




Upload Task

Select Workers

Register worker

Workers

Requester

!2




Upload Task

Perform Task

Select Workers

Register worker

Workers

Requester

!2




Upload Task

Perform Task

Evaluate Workers

Select Workers

Register worker

Workers

Requester

!2




Upload Task

Perform Task

Evaluate Workers

Select Workers

Register worker

Workers

Pay Workers

Requester

!2




Upload Task

Perform Task

Evaluate Workers

Select Workers

Register worker

Evaluate Requester

Workers

Pay Workers

Requester

!2

But, several shortcomings



Workers

Requester

!3




Workers

Requester

!3

• Trust• Service fees• Single point of failure

• Dispute resolving (repudiation of payments)• Prone to attacks or fraud (free-riders)• Heterogeneity (e.g., uncalibrated, low-precision

sensors)


Workers

Requester

!3

• Trust• Service fees• Single point of failure


sensors)


Workers

Requester

!4

Distributed ledger solutions

• Trust• Service fees • Single point of failure


sensors)

Distributed Ledgers for Crowdsourcing • a decentralized crowdsourcing system• smart contracts (SC) to implement functions of a crowdsourcing platform• requires a secure environment which is decentralized, unalterable and

programmable: provided by Blockchain networks• transaction fee for miners who confirm the transaction and support blockchain

running persistently [Li 2019]

!5

- Lu, Yuan, Qiang Tang, and Guiling Wang. "Zebralancer: Private and anonymous crowdsourcing system atop open blockchain." IEEE ICDCS 2018.

- M. Li et al., "CrowdBC: A Blockchain-based Decentralized Framework for Crowdsourcing," in IEEE Transactions on Parallel and Distributed Systems, 2019.

Challenge: SCs storage and processing limited

BLOCKCHAIN BASED CROWDSOURCING

Upload Task

Perform Task

Workers

Requester

!6

Evaluate Workers

Select Workers

Register Worker

Pay Workers

Spass: Spectrum Sensing as a Service via Smart Contracts

presented at IEEE DySPAN 2018(under review, IEEE Transactions on Cognitive Communications and Networking)

!7

Network Layer

Consensus Layer

Transaction Layer

APP Layer SPASS

Need for more network capacity

!8


!8

Time

Amount


!8

DEMAND

Time

Amount


!8

DEMAND

Time

Amount

SUPPLY


!8

DEMAND

Time

Amount

SUPPLYOPERATOR’S REVENUE

PER BIT

We love flat-rates

Source: https://tefficient.com/ Unlimited moves the needle – but it’s when mobile addresses slow fixed internet that something happens


!8

DEMAND

Time

Amount


PER BIT

We love flat-rates


MIND THE GAP


!8

DEMAND

Time

Amount


PER BIT

We love flat-rates


SUPPLY WITH OPPORTUNISTIC SPECTRUM USE

MIND THE GAP


!8

DEMAND

Time

Amount


PER BIT

We love flat-rates



MIND THE GAP Licensed

SpectrumOpportunistic

spectrum+


!8

DEMAND

Time

Amount


PER BIT

We love flat-rates





spectrum+

Unlicensed spectrum

Licensed spectrum


!8

DEMAND

Time

Amount


PER BIT

We love flat-rates





spectrum+

On Practical Coexistence Gaps in Space for LTE-U/WiFi Coexistence, European Wireless 2018.Coexistence Gaps in Space: Cross-Technology Interference-Nulling for Improving LTE-U/WiFi Coexistence, IEEE WoWMoM 2018.

Unlicensed spectrum

Licensed spectrum

!9

DEMAND

Time

Amount


PER BIT

We love flat-rates



MIND THE GAP

This talk


Licensed Spectrum

Opportunistic spectrum+

Unlicensed spectrum

Licensed spectrum

!10

Mobile Network Operator (MNO) #1

SU customers

Licensed Spectrum

!10


SU customers

Licensed Spectrum


Licensed spectrum

!10


Normalized Load1.0

SU customers

Licensed Spectrum


Licensed spectrum

!10


Normalized Load1.0

SU customers

Licensed Spectrum


Licensed spectrum

Underutilized spectrum

!11


Normalized Load1.0

SU customers

Licensed Spectrum


Licensed spectrum

Opportunistic spectrum

Secondary User Network (SUN)

Primary User (PU) Network

Crowdsourced spectrum discovery

!12

SUN

SU customers

PU traffic


!12

SUN

SU customers

PU traffic

Helpers


!12

SUN

SU customers

PU traffic

Helpers

1(synchronous)


!12

SUN

SU customers

PU traffic

Helpers

1(synchronous)

SPASS

2(asynchronous)


!12

Crowdsourced spectrum discovery • Pay sensors for sensing • Malicious sensors excluded from participation • Meet regulatory constraints • Profitable business

SUN

SU customers

PU traffic

Helpers

1(synchronous)

SPASS

2(asynchronous)

/56

A Spass contract

!13

Spass contract { center_frequency, sensing_rate, min_pd_accuracy, max_pfa, function helper_selection() function malicious_helper_identify() function payment() function sensing_report()} Candidate helpers for spectrum

sensing service(malicious and honest helpers)

SU networkEthereum blockchain

network

• Smart contract usage is not free • Write operations are expensive! (internal storage and manipulation of the contract)• Higher cost with increasing complexity of computation

/56

A Spass contract

!13

Spass contract { center_frequency, sensing_rate, min_pd_accuracy, max_pfa, function helper_selection() function malicious_helper_identify() function payment() function sensing_report()} Candidate helpers for spectrum

sensing service(malicious and honest helpers)

SU networkEthereum blockchain

network

• Smart contract usage is not free • Write operations are expensive! (internal storage and manipulation of the contract)• Higher cost with increasing complexity of computation

Gas costs: https://docs.google.com/spreadsheets/d/1m89CVujrQe5LAFJ8-YAUCcNK950dUzMQPMJBxRtGCqs/edit#gid=0

Our proposal: Smart Contracts for Spectrum Sensing (Spass)

(i) what is the cost of running smart contract based spectrum discovery?

(ii) given that both the helpers and the miners have to be paid, under which conditions an MNO can sustain a profitable business via smart-contract based spectrum discovery?

(iii) How can the smart-contract catch free-riders among the sensing participants?

!14

/56

• Design goals

• Contract functionality

• Optimal contract parameters

• Malicious helper identification

• Performance

• Business feasibility of Spass

!15

Spass

/56

For a feasible business model

!16

/56


!16

SU MNO Regulators Helper nodes (sensors)

/56


!16


Customers Spass cost

/56


!16


HelpersSensing

Ethereum Smart Contract


/56


!16

High PU detection accuracy

Low false alarm in sensing


HelpersSensing



/56


!16



Attracting honest helpers Able to catch

malicious helpers (free riders)


HelpersSensing



/56


!16



Attracting honest helpers Able to catch

malicious helpers (free riders)


HelpersSensing



This talk

/56

• Design goals




• Performance


!17

Spass

Malicious helper identification

Helper selection

Payment

/56

Message flow• Contract duration: a certain time period

• SUN uploads the contract to Ethereum and receives a unique ETH address

• SUN broadcasts the contract address over the air

!18Backhaul

/56




!19Backhaul

/56




!19Backhaul

Contract0

/56




!19Backhaul

Contract0

Short-range unlicensed

/56




!19Backhaul

Contract0


1Announce

/56

Message flow

!20Backhaul

Contract01Announce


/56

Message flow

!20Backhaul

Contract01Announce


Retrieve contract

2

/56

Message flow

!20Backhaul

Contract01Announce


3 accept?

Retrieve contract

2

/56

Message flow

!20Backhaul

Contract01Announce


Register4

3 accept?

Retrieve contract

2

/56

Message flow• Helpers register:

• accuracy in PU detection, false alarms

• Sensing service price, location, etc.

• Helper selection:

• Global sensing accuracy fulfils regulatory requirements

• SUN’s profit is maximised

• cost, sensing reliability, reputation etc

• Optimal number of helpers for maximum profit

!20Backhaul

Contract01Announce


Register4

3 accept?

Retrieve contract

2

/56





• Goal: SUN’s profit is maximised

• Input parameters: cost, sensing reliability, reputation etc.


!21Backhaul

Contract01Announce


Register4

3 accept?

Retrieve contract

2

/56





• Goal: SUN’s profit is maximised

• Input parameters: cost, sensing reliability, reputation etc.


!21Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers

Retrieve contract

2

/56

Message flow

!22Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected

Retrieve contract

2

/56

Message flow

!22Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected7

if selected, sense spectrum

Retrieve contract

2

/56

Message flow

!22Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected7


PU activity report

Retrieve contract

2

/56

Message flow• Spectrum sensing

• Energy detection (hard decision: 0 or 1)

• PU activity report (binary array)

• SUN sensing decision fusion: majority voting

• Helper accuracy verification

• Compressed reports

• Malicious helper detection and payment

• Goal: High malicious helper detection accuracy, low # blacklisted honest helpers

• Payment to only whitelisted-helpers!22

Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected7


PU activity report

Retrieve contract

2

/56










Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected7


PU activity report

Retrieve contract

2

Compressed PU activity report

8

/56










Backhaul

Contract01Announce


Register4

3 accept?

5 Select helpers6

Check if selected7


PU activity report

Retrieve contract

2

Compressed PU activity report

8

9Verification and clearing

/56!23

Verification round k Verification round k+1 Verification round k+2

Malicioushelper

identification

Spass contract

Clearing Update

blacklisted

helpers list

Helper

selection

Collect

sensing

reports

SU network

spectrum

access

decision

Spectrum

sensing

reports

Rs

V

Helper selection

Sensing reports

Malicious helper det.

Update blacklist Clearing

/56

• Design goals




• Performance


!24

Spass

Cost of smart contracts

How many helpers

Profit model

Verification round duration

How to design a smart contract so that SUN maximises its profit?

/56

Assumptions

!25

• Primary User (PU)

• active with probability p1 and inactive with p0

• known by SUN and helpers

• Homogenous helpers with identical sensing cost and accuracy

• Probability of detection (Pd)

• Probability of false alarm (Pf)

• Malicious helpers with identical sensing accuracy: free riders (no energy consumption for sensing or colluding)

• do not sense the spectrum but generate data using their statistical knowledge of p0

• Fraction of malicious helper population known by the SUN

• Malicious helpers do not collude and do not change their strategy

/56

SUN profit • Income: Monetary gain accumulated over all verification rounds by serving its

customers over the opportunistic spectrum

• Payment: Helpers sensing service and the smart contract cost

!26

SU income

Spass cost

Helpers

Smart contract

/56

SUN profit in a verification round

!27

/56


!27

client pays μ euros/bit for second

Technology dependent spectral efficiency

Discovered Bandwidth

{

/56


!27




{

Assumption: write operation is the dominant cost in Ethereum usage

/56


!27

cost of sensing

Compression ratio

number of helpers

rate of sensing

cost of Ethereum




{


/56


!27

We find the operation range of an SU network in which it can make profit from Spass: net profit>0

cost of sensing

Compression ratio

number of helpers

rate of sensing

cost of Ethereum




{


/56

When is Spass profitable?

!28

Required min. spectral efficiency

# OF MALICIOUS HELPERS

/56

• Utility: probability of detecting the spectrum opportunity

!29

/56


!29

Expected utility if H helpers sense

Expected false alarm probability

False alarm probability under j malicious helpers and total H helpers

/56


• Regulatory requirements on sensing accuracy are satisfied

!30

/56



• Helpers detected as malicious are not paid (false and true detection)

!31

/56



• Helpers detected as malicious are not paid (false and true detection)

!31

Fraction of malicious helpers in all populationqd, qf: true and false detections

/56



• Malicious helpers are not paid

• For sustainable operation, honest helpers are not blacklisted and malicious helpers are detected with high probability

!32

/56





• Sensing report size under compression factor ß

!33

https://en.wikipedia.org/wiki/%C3%9F

/56





• Sensing report size under compression factor ß

!33

Rs : Rate of sensing (bps)ß: compression factor

https://en.wikipedia.org/wiki/%C3%9F

/56





• Sensing report size under compression factor beta

!34

But, we do not have a closed formula for malicious helper detection accuracy!

/56

Simplified problem: two phase operation

!35

Malicious helper detection (MHD)

Contracts expires

Contracts published

/56


!35

Might have malicious helpers H0

T0


Contracts expires

Contracts published

/56


!35


T0

Only honest helpers H1

T1


Contracts expires

Contracts published

/56


• Requirement: After the first phase, all malicious helpers are (must be) detected by our MHD algorithm

• Malicious helpers do not change their strategy

!35


T0

Only honest helpers H1

T1


Contracts expires

Contracts published

/56


!36

• Now, decision on H0, H1, T0

• T0: the minimum duration that the MHD algorithm needs for detection of all malicious helpers

• We find T0 via Monte Carlo simulations

• H0, H1: exhaustive search O(N2) where N is number of candidates

/56

How to decrease cost of Spass?

!37

/56


!37

Our approach: Decrease the amount of data written to the contract

/56


!37

Lossless compression (variable length codes)


/56


!37

qi: probability of symbol i



/56


!37




Desirable! But, limited compression depending on qi (occupancy state of

the channel, 0 or 1)

/56


!37




Lossy compression (removing bits randomly)



/56


!37

Compression ratio: β







/56


!37








Might affect malicious helper detection algorithm’s accuracy

/56


!38


Lossless compression (optimal block codes)


/56


• Higher compression when low/high PU idle!38


Lossless compression (optimal block codes)


/56

• Design goals




• Performance


!39

Spass

/56

• Design goals




• Performance


!39

SpassHon

est

Hones

t

Hones

t

Malicio

us

/56

Clustering-based malicious Helper Identification (CHI)

!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001(compressed) Sensing reports

Honest helper model: Identical sensing accuracyPhd, Phfa

Malicious helper model: Knows P0, Generates fake sensing bits without performing sensing1 with prob α1 = (1 − p0). Pmf = p0α1 = p0(1−p0) Pmd = (1−p0)2.

/56


• Distance between two helpers • normalized Hamming distance of two helper reports

!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001(compressed) Sensing reports



/56



!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001

0.2(compressed)

Sensing reports



/56



!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001

0.50.2(compressed)

Sensing reports



/56



!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001

0.50.2 0.3

(compressed) Sensing reports



/56



!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001

0.50.2 0.3


• Helper score: x-percentile of its distance from other helpers



/56



!40

A

1010001000

B

1010101001

C

0111101001

D

0010101001

0.50.2 0.3


• Helper score: x-percentile of its distance from other helpers

If 50-percentile: Score-A = [0.2, 0.3, 0.5]



/56

Key insight of CHI• Honest helpers with similar sensing

reports

• Short distance from each other

• Similar and low scores

• Malicious helpers (do not collude and do not sense the spectrum)

• High distance from each other and from honest helpers

• Two clusters

• Honest helper cluster and malicious helper cluster

!41

/56

CHI: one or two clusters?

!42

/56


!42

Helper scores0 1

/56


!42

Helper scores0 1

K-means clustering for K=1

/56


!42

Helper scores0 1

Helper scores0 1

K-means clustering for K=2K-means clustering for K=1

/56


• Is the clustering accuracy better above a threshold for K=2 compared to K=1?!42

Helper scores0 1

Helper scores0 1


/56



Helper scores0 1

Helper scores0 1


ALL HELPERS ARE HONEST

No

Honest helpers

/56



Helper scores0 1

Honest helpers

Helper scores0 1


Malicious helpers


ALL HELPERS ARE HONEST

No

CLUSTER WITH HIGHER SCORES ARE MALICIOUS HELPERS

Yes

/56

CHI clusters for K=1 and K=2

!45

Honest helpers

Malicious helpers

/56


!45

Honest helpers

Malicious helpers

Δ inertia

/56


!45

Honest helpers

Malicious helpers

High inertia difference indicates the existence of malicious helper(s)

Δ inertia

/56

• Design goals




• Performance


!46

Spass

/56

• Design goals




• Performance


!46

Spass

• Robustness of CHI to malicious helpers

• Impact of lossy compression

• Sensing accuracy

• Optimal number of helpers

/56

Optimal number of helpers

• Higher P0, higher number of helpers

• More malicious helpers expected, more helpers need to be selected!47

/56!48

Report size in bits

p0=0.7Malicious

helper detection accuracy

/56!49

Report size in bits

p0=0.7

CHI achieves high malicious helper detection accuracy

Malicious helper

detection accuracy

/56!49

Report size in bits

p0=0.7

10-percentile

CHI achieves high malicious helper detection accuracy

Malicious helper

detection accuracy

/56!50

p0=0.7

Prob. of blacklisting

honest helpers

/56

Almost always no false alarms if report size is bigger than 100 bits

!51

p0=0.7

• Preferred sensing report size

Prob. of blacklisting

honest helpers

/56

Minimum report size needed

!52

CHI accuracy

/56

Spectrum discovery with very high accuracy

!53

Almost 100% detection accuracy

/56

In a nutshell• Spass:

• Spectrum sensing as a service using smart contracts (Ethereum)

• Cost of smart contract usage and helpers sensing service

• Spass becomes profitable under some conditions

• Optimal number of helpers and verification round durations

• Possible directions:

• More sophisticated malicious users

• Pricing and reward assignment based on helper capabilities

• Computation cost should also be considered

!54

https://github.com/zubow/Spass_contract

Crowdsourcing for spectrum sensing• Spectrum monitoring for policy making and misuse detection• SpecGuard

• Radio Environment Map construction (spectrum sensing provider)• SpecSense

• Pricing for crowdsensing• Han et al.

• Malicious sensor identification• Catch Me If You Can

• Privacy aspects• DPSense

!55

- Jin, Xiaocong, et al. "Specguard: Spectrum misuse detection in dynamic spectrum access systems." IEEE TMC 2018

- Chakraborty, Ayon, et al. "Specsense: Crowdsensing for efficient querying of spectrum occupancy." IEEE INFOCOM 2017

- DPSense: Differentially Private Crowdsourced Spectrum Sensing, ACM CCS 2018

- Ying, Xuhang, Sumit Roy, and Radha Poovendran. "Pricing mechanisms for crowd-sensed spatial-statistics-based radio mapping." IEEE Transactions on Cognitive Communications and Networking, 2017.

- Han, Kai, He Huang, and Jun Luo. "Quality-Aware Pricing for Mobile Crowdsensing." IEEE/ACM TON 2018

- Li, Husheng, and Zhu Han. "Catch me if you can: An abnormality detection approach for collaborative spectrum sensing in cognitive radio networks." IEEE Transactions on Wireless Communications 2010.

/56!56

The future is unlicensed, diverse, and decentralized

Spectrum

Network

Content

Spectrum sharing

Caching and resource allocation

for video

functionality placement on edge/fog/cloud networks

Decentralised solutions New coexistence

metrics

Content placement /edge/

fog/cloud

Information-centric networks

COLLABORATORS:

/56!56

The future is unlicensed, diverse, and decentralized

Spectrum

Network

Content

Spectrum sharing

Caching and resource allocation

for video

functionality placement on edge/fog/cloud networks

Decentralised solutions New coexistence

metrics

Content placement /edge/

fog/cloud

Information-centric networks

COLLABORATORS:

Thank you! https://suzanbayhan.github.io/

Talk Abstract

!57

To cope with the huge increase in cellular data traffic, mobile network operators (MNO) consider using also other spectrum bands which are under-utilized spatiotemporally. To discover such spectrum opportunities, several studies suggest exploiting the ubiquity of mobile devices to perform spectrum sensing. However, current solutions for crowdsourced spectrum sensing overlook the fact that mobile devices lack the incentives to sense without certain benefits as spectrum sensing results in energy and CPU overhead. To encourage participation, we explore the use of smart contracts running on a distributed ledger (e.g., Ethereum) for an MNO to announce its need for spectrum sensing, the interested sensors to register themselves as candidate sensors, and MNO to pay the selected sensors for their service. Despite the simplicity of the idea, the realization requires more thoughts as smart-contracts are limited in their storage and processing capacity, and incurs a high cost for the write operation. I will introduce our proposal Spass: spectrum sensing as a service via smart contracts and address the following questions: (i) what is the cost of running smart contract based spectrum discovery? and (ii) given that both the helpers and the miners have to be paid, under which conditions an MNO can sustain a profitable business via smart-contract based spectrum discovery? (iii) How can the smart-contract catch free-riders among the sensing participants?

Bio:Suzan Bayhan is a senior researcher at TU Berlin and a docent in Computer Science at the University of Helsinki. She got her Ph.D. from Bogazici University in 2012 and worked as a post-doc at the University of Helsinki between 2012-2016. She was on N2Women board between 2017-2018. Her current research interests are resource allocation for wireless networks, spectrum sharing, and edge/fog/cloud computing. Suzan will join the University of Twente on September 2019 as an Assistant Professor.

smart contracts for crowdsourced spectrum sensing · 2020. 7. 16. · spass: spectrum sensing as a...

Documents