Finding Common Ground: Channel Analysis and Receiver Models …€¦ · 1 Background 2 Prior...

Finding Common Ground:Channel Analysis and Receiver Models for

Diffusive Molecular Communication

Adam Noel

School of Electrical Engineering and Computer ScienceUniversity of Ottawa

University of TorontoJuly 27, 2016

Channels and RXers for Diffusive MC Noel 1/40

1 Background

2 Prior ContributionsChannel ModelingCommunications Analysis

3 Current WorkMultiuser CommunicationSimulator DevelopmentFinding Common Ground: Active and Passive Channel Models

4 Where We’re Going

5 Conclusions


1 Background




5 Conclusions


What?Molecular Communication

• How do biological systems share information?• Can synthetic networks use natural communication systems?• What are the practical communication limits?

Image: US Centers for Disease Control and Prevention http://www.cdc.gov/tb/education/corecurr/pdf/chapter2.pdf


Why?

Edited Image From: Nakano, Eckford, Haraguchi, Molecular Communication. Cambridge University Press, 2013.


How?

Diffusion is motion of a molecule colliding with other molecules

• No external energy or infrastructure required• Very fast over “short” distances (≤ 1µm)• Used by many cellular processes

Image: Nakano, Eckford, Haraguchi, Molecular Communication. Cambridge University Press, 2013.


How?

Diffusion is motion of a molecule colliding with other molecules• No external energy or infrastructure required• Very fast over “short” distances (≤ 1µm)• Used by many cellular processes

Image: Nakano, Eckford, Haraguchi, Molecular Communication. Cambridge University Press, 2013.


ExampleNeuromuscular Junction

30 nmPresynaptic

Cleft(Neuron)

PostsynapticCleft

(Muscle)

• Connect motor neuron tomuscle fiber

• Molecules released (∼ 104)• Reception leads to muscle

contraction• Up to ∼ 50 times per second


1 Background




5 Conclusions


Channel Modeling

• Fluid

• Receiver (RX)• Transmitter

(TX)• U sources• A molecules• Diffusion D


Channel Modeling

RX

VRX

• Fluid• Receiver (RX)

• Transmitter(TX)

• U sources• A molecules• Diffusion D


Channel Modeling

TX

RX

VRX

Source u = 2

Source u = 3

• Fluid• Receiver (RX)• Transmitter

(TX)• U sources

• A molecules• Diffusion D


Channel Modeling

xTX {−d, 0, 0}

y

z

RX

VRX

Source u = 2

Source u = 3


(TX)• U sources

• A molecules• Diffusion D


Channel Modeling

xTX {−d, 0, 0}

y

z

RX

VRX

12

- A moleculeSource u = 2

Source u = 3


(TX)• U sources• A molecules• Diffusion D


Channel ModelingSimplified Receiver (Uniform Concentration)

3D Point Receiver Observation (Point TX)

NRX (t) =NVRX

(4πDt)3/2 exp(− d2

4Dt

)

3D Spherical Receiver Observation (Point TX)

NRX (t) =N2

[erf(

rRX − d2√

Dt

)+ erf

(rRX + d2√

Dt

)]+

Nd

√Dtπ

[exp

(−(d + rRX)

2

4Dt

)− exp

(−(d − rRX)

2

4Dt

)]

Noel, Cheung, Schober, Proc. IEEE ICC MoNaCom, Jun. 2013.


Channel ModelingSimplified Transmitter (Point Source)

1D Receiver Observation (Point TX)

NRX (t) =N2

(erf(

rRX + d2√

Dt

)− erf

(d − rRX

2√

Dt

))

1D Receiver Observation (Volume TX)

NRX (t) =N

2rTX

{√Dtπ

[exp(−(xf + rRX)

2

4Dt

)− exp

(−(xf − rRX)

2

4Dt

)− exp

(−(xi + rRX)

2

4Dt

)

+ exp(−(xi − rRX)

2

4Dt

)]+

12

[(xf + rRX) erf

(xf + rRX

2√

Dt

)

− (xi + rRX) erf(

xi + rRX

2√

Dt

)− (xf − rRX) erf

(xf − rRX

2√

Dt

)+ (xi − rRX) erf

(xi − rRX

2√

Dt

)]}

Noel, Makrakis, Hafid, Proc. CSIT BSC, Jun. 2016.


Channel ModelingAccuracy of Point-to-Point Model

Dimensionless Time

10−1 100 101

%DeviationofPointRX

-10

-8

-6

-4

-2

0

2

4

6

Increasing size of RX

Dimensionless Time

10−1 100 101%

DeviationofPointTX

-15

-10

-5

0

5

Increasing distance to RX

Noel, Cheung, Schober, Proc. IEEE ICC MoNaCom, Jun. 2013.

Noel, Makrakis, Hafid, Proc. CSIT BSC, Jun. 2016.


Channel ModelingObservation Independence

Time [s]0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04

Number

ofmoleculesat

RX

0

5

10

15

20

25

30

35

40

Time Between Samples

0 1 2 3 4 5MutualInform

ationBetweenRX

Samples

10−3

10−2

10−1

100

Analytical

Simulation

Noel, Cheung, Schober, IEEE Trans. NanoBiosci., Sept. 2014.


Channel ModelingChanging the Channel

A E EA EAP

k1

k-1

k2

Noel, Cheung, Schober, IEEE Trans. NanoBiosci., Mar. 2014.


Channel ModelingChanging the Channel

A E EA EAP

k1

k-1

k2

Time

0 50 100 150 200

#ofMoleculesatRX

0

0.5

1

1.5

2

2.5

3

3.5

Baseline

With Degradation

Impulse Response with Degradation

NRX (t) =NVRX

(4πDt)3/2 exp(−kt − d2

4Dt

)Noel, Cheung, Schober, IEEE Trans. NanoBiosci., Mar. 2014.


Channel ModelingUniform Flow in Any Direction

xTX {−d, 0, 0}

y

zv‖

v⊥v

RX

VRX

12


Source u = 3


Channel ModelingUniform Flow in Any Direction

Time

0 50 100 150 200

#ofMoleculesatRX

0

1

2

3

4

Baseline

Flow Perpendicular to RX

Flow Towards RX

Impulse Response with Flow

NRX (t) =NVRX

(4πDt)3/2 exp(−kt − (d − v‖t)2 + (v⊥t)2

4Dt

)Noel, Cheung, Schober, IEEE Trans. NanoBiosci., Sept. 2014.


Channel ModelingParameter Estimation

5 10 15 20 25 30 35 40

10−3

10−2

10−1

100

Number of Samples M

NormalizedErrorondEstim

ation

CRLBML, d UnknownML, d, ttx,0 UnknownML, d, ttx,0, v‖ Unknown

d Unknown

Can we estimate underlyingparameters? (e.g., d)• Take M samples of

impulse response• Cramer Rao Lower

Bound vs. MaximumLikelihood

• Bound increases asfewer parameters known

• ML is asymptoticallyefficient

Noel, Cheung, Schober, IEEE Trans. Mol. Biol. Multi-Scale Commun., Mar. 2015.



5 10 15 20 25 30 35 40

10−3

10−2

10−1

100

Number of Samples M


ation


d Unknown d, ttx,0 Unknown









5 10 15 20 25 30 35 40

10−3

10−2

10−1

100

Number of Samples M


ation


d Unknown

d, ttx,0, v‖ Unknown

d, ttx,0 Unknown









5 10 15 20 25 30 35 40

10−3

10−2

10−1

100

Number of Samples M


ation


d Unknown

d, ttx,0, v‖ Unknown

d, ttx,0 Unknown

d, ttx,0, v‖, v⊥ Unknown








Channel ModelingReversible Adsorption

Molecules adhere to RX surface and can detach

RX

0 0.150.05 0.1Time (s)

0.20

1

Net

Num

ber

of A

dsor

bed

Mol

ecul

es d

urin

g E

ach

Sam

plin

g T

ime

2

3

4

5

6

7

8

9

10

k -1= 5

2010

Sim. Anal.

Deng, Noel, Elkashlan, Nallanathan, Cheung, to appear in IEEE Trans. Mol. Biol. Multi-Scale Commun., 2016.


Communications Analysis

RX

10110 ...

How to design communication system?• Different ways to modulate (time of release, # of molecules, type

of molecules)• Choose impulsive binary ON/OFF keying• Detect using multiple RX samples per symbol interval


Communications AnalysisReceiver Design

• Energy detector and matchedfilter with constant decisionthresholds

• Maximum Likelihood detectorbased on Viterbi algorithm

• We can approach MLperformance with moleculedegradation and/or strong flow(not shown)

# of Samples Per Interval M

5 10 15 20 25 30 35 40

Receiver

ErrorProbability

10−4

10−3

10−2

10−1

Energy Detector

Maximum Likelihood

Detector

Matched Filter Detector

Noel, Cheung, Schober, IEEE Trans. NanoBiosci., Sep. 2014.


Communications AnalysisRelaying

• Motivation: Reach further destinations faster• Decode-and-forward, amplify-and-forward• Various schemes for re-using molecule types in different hops• Identified self-interference (SI) and backward ISI (BI)

Ahmadzadeh, Noel, Schober, IEEE Trans. Mol. Biol. Multi-Scale Commun., Jun. 2015.

Ahmadzadeh, Noel, Burkovski, Schober, Proc. IEEE GLOBECOM, Dec. 2015.


Communications AnalysisRelaying Results

• Optimized decisionthreshold or number ofmolecules

• Adaptive relay adjustsdecision threshold

• There is an optimalnumber of relays

0 1 2 3 4 5 610

−2

10−1

100

Number of Relays Q

AverageErrorProbabilityP

eQ+1

Simulation, FD-A-BI-SI

Expected, FD-A-BI-SI

Simulation, HD-A-BI

Expected, HD-A-BI

Simulation, FD-A-SI

Expected, FD-A-SI

Simulation, HD

Expected, HD

Baseline

T = 400µs

T = 200µs

Ahmadzadeh, Noel, Schober, IEEE Trans. Mol. Biol. Multi-Scale Commun., Jun. 2015.


1 Background




5 Conclusions


Multiuser CommunicationCooperative Communication

ξRX

0

QFC

10-3

10-2

10-1

Expected Simulated

Soft Fusion

OR rule

Baseline

Majority rule

AND rule

10 20 30 40 50 60

• Limited work about MC devices actually “cooperating”• Derived performance in symmetric and asymmetric networks• Convex optimization of thresholds (in preparation)

Fang, Noel, Yang, Eckford, Kennedy, to be presented at IEEE GLOBECOM, Dec. 2016.


Multiuser CommunicationLarge-Scale Systems

• Consider “large” number ofTXs defined by density

• What is capability tocommunicate with closest TX?

• Derived channel response ofclosest vs remaining TXs

Time

10−1 100

Net

#ofMoleculesatRX

0

20

40

60

80

100

120

140

160

180

Due to Nearest TX

Due to Remaining TXs

Deng, Noel, Guo, Nallanathan, Elkashlan, to be presented at IEEE GLOBECOM, Dec. 2016.


Simulator DevelopmentMotivate Sandbox

• Reaction-diffusion solvers• Physical-chemistry community• Generic “sandbox” tools with flexible accuracy• Not designed for communications (channel statistics, data

modulation)

• Molecular communication simulators• Communications engineering community• Designed for communications research• Constrained environmental design options

• Need: A “sandbox” simulator for communications research inreaction-diffusion systems


Simulator DevelopmentMotivate Sandbox

• Reaction-diffusion solvers• Physical-chemistry community• Generic “sandbox” tools with flexible accuracy• Not designed for communications (channel statistics, data

modulation)• Molecular communication simulators

• Communications engineering community• Designed for communications research• Constrained environmental design options

• Need: A “sandbox” simulator for communications research inreaction-diffusion systems


Simulator DevelopmentAcCoRD

AcCoRD (Actor-based Communication via Reaction-Diffusion)• Flexible environmental design (accuracy vs efficiency)• Generate “many” independent realizations• Release molecules based on modulated data• Output local molecule counts or specific locationsNoel, Cheung, Schober, Makrakis, Hafid, in preparation, 2016.


Finding Common GroundActive vs Passive

Passive RX - No affect on molecule propagation• Easier to simulate• Easier to analyze• Some biological justification


Finding Common GroundActive vs Passive

Passive RX - No affect on molecule propagation• Easier to simulate• Easier to analyze• Some biological justification

Active RX - Model chemical detection of molecules• More realistic• Harder to simulate• Less convenient analysis


Finding Common GroundCompare 3D Impulse Responses

Passive RX (Sampling)

NRX (t) |PA =NVRX

(4πDt)3/2 exp(− d2

4Dt

)

Absorbing RX (Accumulating)

NRX (t) |AB =NrRX

derfc

(d − rRX√

4Dt

)


Finding Common GroundCompare 3D Impulse Responses

Fundamentally different RX models ... can we unify them?

Absorbing Signal ?= S(Passive Signal)

Idea to Unify ModelsCompare RX models such that they are either both accumulating orboth sampling instantaneous behavior

Why should we bother?• Unify literature that has chosen one RX over the other• Selection of RX model is less critical


Finding Common GroundSample Transform

Integrate passive RX signal to get energy detector

ED (t) |PA =NVRX

4πDderfc

(d√4Dt

)

Use approximation of error function to separate terms inside erfc

erfc (x) ≈ exp(−16

23x2 − 2√

πx)

Write absorbing signal as function of passive energy detector

NRX (t) |AB ≈ 3DA(t)r2RX

ED (t) |PA

where

A(t) = exp(

rRX√Dt

(8(2d − rRX)

23√

4Dt+

1√π

))




ED (t) |PA =NVRX

4πDderfc

(d√4Dt

)Use approximation of error function to separate terms inside erfc


23x2 − 2√

πx)



ED (t) |PA

where

A(t) = exp(

rRX√Dt

(8(2d − rRX)

23√

4Dt+

1√π

))




ED (t) |PA =NVRX

4πDderfc

(d√4Dt

)Use approximation of error function to separate terms inside erfc


23x2 − 2√

πx)



ED (t) |PA

where

A(t) = exp(

rRX√Dt

(8(2d − rRX)

23√

4Dt+

1√π

))Channels and RXers for Diffusive MC Noel 31/40

Finding Common GroundResults

Consider signal of 3D absorbingRX

• Analytical transform no lessaccurate than simulation

• Can also transform passivesimulations

• Similar results with passiveRX and in 1D

Time [s]10−3 10−2

#of

Absorbed

Molecules(N

ormalized)

10−2

10−1

100

Analytical

Analytical via Transform

Simulation

Simulation via Transform

Noel, Deng, Makrakis, Hafid, to be presented at IEEE GLOBECOM, Dec. 2016.



Consider signal of 3D absorbingRX• Analytical transform no less

accurate than simulation

• Can also transform passivesimulations


Time [s]10−3 10−2

#of

Absorbed

Molecules(N

ormalized)

10−2

10−1

100

Analytical


Simulation






accurate than simulation• Can also transform passive

simulations


Time [s]10−3 10−2

#of

Absorbed

Molecules(N

ormalized)

10−2

10−1

100

Analytical


Simulation






accurate than simulation• Can also transform passive

simulations• Similar results with passive

RX and in 1D

Time [s]10−3 10−2

#of

Absorbed

Molecules(N

ormalized)

10−2

10−1

100

Analytical


Simulation




1 Background




5 Conclusions


Where We’re Going

General Directions of MC Research:1 Improving physical and simulation models2 Obtaining experimental data

AcCoRD simulator is in on-going development and publicly available


Expanding Physical Model

xTX {−d, 0, 0}

y

zv‖

v⊥v

RX

VRX

12


Source u = 3


Oscillating Laminar Flow

System BoundaryMotion of

Fluid Layers

Solvent in laminar flow moves as sliding layers with different velocities• Small cross-sections lead to laminar flow• E.g., small blood vessels• Blood flow oscillates


1 Background




5 Conclusions


Final Words

• Molecular communication is a promising interdisciplinary field

• Interesting variations on “traditional” communications analysistechniques

• Inspiration between biology and engineering is bidirectional


Final Words

• Molecular communication is a promising interdisciplinary field• Interesting variations on “traditional” communications analysis

techniques

• Inspiration between biology and engineering is bidirectional


Final Words

• Molecular communication is a promising interdisciplinary field• Interesting variations on “traditional” communications analysis

techniques• Inspiration between biology and engineering is bidirectional


Acknowledgments

• Supervisors• Robert Schober, FAU Erlangen-Nuremberg• Karen C. Cheung, UBC• Dimitrios Makrakis, University of Ottawa• Abdelhakim Hafid, University of Montreal

• “Horizontal” Collaboration• Arman Ahmadzadeh, FAU Erlangen-Nuremberg• Yansha Deng, King’s College London• Jonas Yang & Yuting Fang, Australian National University

• Funding• Natural Sciences and Engineering Research Council


The EndThat’s All, Folks

Thanks for your time!Papers: adamnoel.ca

Simulator code: github.com/adamjgnoel/accord


Slide Backup


Finding Common Ground: Channel Analysis and Receiver Models …€¦ · 1 Background 2 Prior...

Documents

Transcript of Finding Common Ground: Channel Analysis and Receiver Models …€¦ · 1 Background 2 Prior...