Ph.D. Thesis Defense June 2005alumni.media.mit.edu/~aggelos/papers/def.pdf · Ph.D. Thesis Defense June 2005 Aggelos Bletsas aggelos@media.mit.edu Viral Communications Group, MIT
Post on 16-Jul-2020
3 Views
Preview:
Transcript
Intelligent Antenna Sharing in Cooperative Diversity WirelessNetworks
Ph.D. Thesis DefenseJune 2005
Aggelos Bletsas
aggelos@media.mit.edu
Viral Communications Group, MIT Media Laboratory
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 1/51
Thesis Committee
Andy Lippman,Principal Research Scientist, MIT Media Lab.
Joe Paradiso,Associate Professor, MIT Media Lab.
Moe Win,Associate Professor, MIT LIDS.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 2/51
Motivation and Inspirations
You are (probably) here because you have all:
experience bad reception while using you cell phone...
left without battery because you forgot to recharge your cell phone the previous night...
been unable to use your cell phone in large gatherings (such as 4th of July celebration,alongside Charles river!)
Could we fix all the above problems?
Inspirations:
Gupta and Kumar IT 2000 result: local communication through other users could bebeneficial...
Multiple Antennas at each radio, in combination with the richness of the wirelesschannel, could be beneficial...
Could we merge the two above? More users ?= better wireless communication?
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 3/51
Motivation and Inspirations
You are (probably) here because you have all:
experience bad reception while using you cell phone...
left without battery because you forgot to recharge your cell phone the previous night...
been unable to use your cell phone in large gatherings (such as 4th of July celebration,alongside Charles river!)
Could we fix all the above problems?
Inspirations:
Gupta and Kumar IT 2000 result: local communication through other users could bebeneficial...
Multiple Antennas at each radio, in combination with the richness of the wirelesschannel, could be beneficial...
Could we merge the two above? More users ?= better wireless communication?
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 3/51
Motivation and Inspirations
You are (probably) here because you have all:
experience bad reception while using you cell phone...
left without battery because you forgot to recharge your cell phone the previous night...
been unable to use your cell phone in large gatherings (such as 4th of July celebration,alongside Charles river!)
Could we fix all the above problems?
Inspirations:
Gupta and Kumar IT 2000 result: local communication through other users could bebeneficial...
Multiple Antennas at each radio, in combination with the richness of the wirelesschannel, could be beneficial...
Could we merge the two above? More users ?= better wireless communication?
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 3/51
Additional Problem Constraint: Low Complexity and Implementation
SourceRelay
Relay
Destination
In general, multi-antenna systems could in-crease:
reliability (diversity gain).
spectral efficiency bps/Hz (multiplexinggain)
Objective: we would like to explore use of multiple antennas in the Relay channel, viacooperative relays.
IMPLEMENTATION TODAY, with existing RF-front ends.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 4/51
Additional Problem Constraint: Low Complexity and Implementation
SourceRelay
Relay
Destination
In general, multi-antenna systems could in-crease:
reliability (diversity gain).
spectral efficiency bps/Hz (multiplexinggain)
Objective: we would like to explore use of multiple antennas in the Relay channel, viacooperative relays.
IMPLEMENTATION TODAY, with existing RF-front ends.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 4/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Main Difficulties
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
Information is not a priori known at the relays.
Number of participating antennas is unknown.
Number of useful participating antennas is unknown.
Coordination and Group formation ought to be distributed, not "genie-aided".
MIMO ST-coding 6= coding for the Relay channel
Radio transceiver complexity.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 5/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 6/51
Assumptions and System Model
Inline with prior art in the field:
Half-duplex radios.
Simple RF-front ends:
Half-duplex radios.
No phased arrays (No beamforming).
No rate adaptation (no CSI at the source).
Discrete, baseband, flat fading signal model: yd = asd xs + nd.
Neighboring interfering streams will be treated as noise.
(Mostly) Rayleigh fading, E[|asd|2] ∝ 1dsd
.
Slow Fading (most difficult communication problem).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 7/51
Assumptions and System Model
Inline with prior art in the field:
Half-duplex radios.
Simple RF-front ends:
Half-duplex radios.
No phased arrays (No beamforming).
No rate adaptation (no CSI at the source).
Discrete, baseband, flat fading signal model: yd = asd xs + nd.
Neighboring interfering streams will be treated as noise.
(Mostly) Rayleigh fading, E[|asd|2] ∝ 1dsd
.
Slow Fading (most difficult communication problem).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 7/51
Approaches
Non-cooperative communication.
Cooperative Repetition.
Simultaneous transmissions (Space-TimeCoding).
Our Approach.
Proactive single relay selection (beforeany message is transmitted fromsource).
Selection based on instantaneouschannel conditions (instead of average).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 8/51
Approaches
Non-cooperative communication.
Cooperative Repetition.
Simultaneous transmissions (Space-TimeCoding).
Our Approach.
Proactive single relay selection (beforeany message is transmitted fromsource).
Selection based on instantaneouschannel conditions (instead of average).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 8/51
Approaches
Non-cooperative communication.
Cooperative Repetition.
Simultaneous transmissions (Space-TimeCoding).
Our Approach.
Proactive single relay selection (beforeany message is transmitted fromsource).
Selection based on instantaneouschannel conditions (instead of average).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 8/51
Approaches
Non-cooperative communication.
Cooperative Repetition.
Simultaneous transmissions (Space-TimeCoding).
Our Approach.
Proactive single relay selection (beforeany message is transmitted fromsource).
Selection based on instantaneouschannel conditions (instead of average).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 8/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 9/51
Wireless Channel Observations
v = 3.98
Distance d
Receiver cares about signal strength (not distance).
Selection based on distance or average SNR... is suboptimal.
Relays as wireless channel sensors in a fast and distributed way: instantaneous channelconditions matter!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 10/51
Wireless Channel Observations
v = 3.98
Distance d
Receiver cares about signal strength (not distance).
Selection based on distance or average SNR... is suboptimal.
Relays as wireless channel sensors in a fast and distributed way: instantaneous channelconditions matter!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 10/51
Wireless Channel Observations
v = 3.98
Distance d
Receiver cares about signal strength (not distance).
Selection based on distance or average SNR... is suboptimal.
Relays as wireless channel sensors in a fast and distributed way: instantaneous channelconditions matter!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 10/51
Our Approach: Opportunistic Relaying
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Policy I : hi = min{|asi|2, |aid|
2} Policy II : hi =2
1|asi|2
+ 1|aid|
2
=2 |asi|
2 |aid|2
|asi|2 + |aid|2
Ti =λ
hi(1)
Here λ has the units of time. For the discussion in this work, λ has simply values of µsecs.
hb = max{hi}, ⇐⇒ (2)
Tb = min{Ti}, i ∈ [1..M ]. (3)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 11/51
Our Approach: Opportunistic Relaying
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
RTS
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Policy I : hi = min{|asi|2, |aid|
2} Policy II : hi =2
1|asi|2
+ 1|aid|
2
=2 |asi|
2 |aid|2
|asi|2 + |aid|2
Ti =λ
hi(4)
Here λ has the units of time. For the discussion in this work, λ has simply values of µsecs.
hb = max{hi}, ⇐⇒ (5)
Tb = min{Ti}, i ∈ [1..M ]. (6)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 11/51
Our Approach: Opportunistic Relaying
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
CTS
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Policy I : hi = min{|asi|2, |aid|
2} Policy II : hi =2
1|asi|2
+ 1|aid|
2
=2 |asi|
2 |aid|2
|asi|2 + |aid|2
Ti =λ
hi(7)
Here λ has the units of time. For the discussion in this work, λ has simply values of µsecs.
hb = max{hi}, ⇐⇒ (8)
Tb = min{Ti}, i ∈ [1..M ]. (9)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 11/51
Our Approach: Opportunistic Relaying
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Policy I : hi = min{|asi|2, |aid|
2} Policy II : hi =2
1|asi|2
+ 1|aid|
2
=2 |asi|
2 |aid|2
|asi|2 + |aid|2
Ti =λ
hi(10)
Here λ has the units of time. For the discussion in this work, λ has simply values of µsecs.
hb = max{hi}, ⇐⇒ (11)
Tb = min{Ti}, i ∈ [1..M ]. (12)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 11/51
Discussion: a note on CSI and time synchronization
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
RTS/CTS exchange is only needed at the relays to estimate uplink/downlink channel.
CTS reception is not exploited at the source.
No beamforming or rate adaptation at the relays.
No need for an explicit time sync protocol.
It is a multi-hop scheme.
We do know that the term "Opportunistic" has been used before...
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 12/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 13/51
Outage Performance (1)
Outage event between source s and destination d:
log(1 + |asd|2 SNR) ≤ ρ⇔ |asd|
2 ≤ (2ρ − 1)/SNR⇔ γsd ≤ Θ
"Best" opportunistic relay is chosen, according to instantaneous, end-to-end channelconditions:
b = arg︸︷︷︸i
max{min{γsi, γid}}, i ∈ [1..M ] (13)
Probability of outage via "best" relay:
Pr(γsb < Θ2⋃
γbd < Θ2), Θ2 = 2 (22ρ − 1)/SNR (14)
Notice the factor of 2 loss in spectral efficiency. Total tx power is fixed.
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 14/51
Outage Performance (2)
The above outage probability of opportunistic relaying is calculated for the case ofRayleigh Fading:
Pr(γsb < Θ2⋃
γbd < Θ2) =
M∏
i=1
(1− exp(−Θ2 (1
γsi+
1
γid))) (15)
Taking into account the direct path between source and destination, the overall outageprobability becomes:
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 15/51
Outage Performance (2)
The above outage probability of opportunistic relaying is calculated for the case ofRayleigh Fading:
Pr(γsb < Θ2⋃
γbd < Θ2) =
M∏
i=1
(1− exp(−Θ2 (1
γsi+
1
γid))) (16)
Taking into account the direct path between source and destination, the overall outageprobability becomes:
P outr = (1− exp(−Θ2/γsd))︸ ︷︷ ︸
direct
M∏
i=1
(1− exp(−Θ2 (1
γsi+
1
γid)))
︸ ︷︷ ︸relaying
(17)
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 15/51
Outage Performance (3)
1 2 3 4 5 6 7 810-2
10-1
100
101
Number of relays
bps/
Hz
Capacity for outage prob.=0.01 and FIXED total transmissionpower (SNR=10)
cooperative with dsd
=dsr
+ drd
(dsr
=drd
)cooperative with d
sd=d
sr=d
rdNon-cooperative, direct
A single relay doesn’t help... [has been shown before...]
Opportunistic relays do help, even under a total tx power constraint!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 16/51
Outage Performance (3)
1 2 3 4 5 6 7 810-2
10-1
100
101
Number of relays
bps/
Hz
Capacity for outage prob.=0.01 and FIXED total transmissionpower (SNR=10)
cooperative with dsd
=dsr
+ drd
(dsr
=drd
)cooperative with d
sd=d
sr=d
rdNon-cooperative, direct
v = 4
v = 3
A single relay doesn’t help... [has been shown before...]
Opportunistic relays do help, even under a total tx power constraint!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 16/51
Outage Performance (4)
0 5 10 15 20 25 30 35 400
1
2
3
4
5
6
7
SNR [dB]
bps/
Hz
Direct communication1 relay2 Opportunistic Relays4 Opportunistic Relays6 Opportunistic Relays8 Opportunistic Relays
Capacity for outage probability=0.01, dsd= dsr = drd
0 5 10 15 20 25 30 35 40
0
1
2
3
4
5
6
7
8
SNR [dB]
bps/
Hz
Direct communication1 relay2 Opportunistic Relays4 Opportunistic Relays6 Opportunistic Relays8 Opportunistic Relays
v = 4
Capacity for outage probability=0.01, dsd= dsr + drd
Prout = δ.
ρopport =1
2log2(1− ln(1− δ1/M )
SNR
2γsid) (18)
ρdirect = log2(1− ln(1− δ) SNR γsid) (19)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 17/51
Diversity-Multiplexing Tradeoff (1)
d∆= − lim
SNR→∞
logPe(ρ)
logSNRr∆= lim
SNR→∞
ρ(SNR)
logSNR
Diversity-Multiplexing Gain tradeoff tool averages out geometry.
cooperative diversity 6= multihop communication. This tool can reveal associatedgains/losses.
Theorem 0: The achievable diversity multiplexing tradeoff for the decode and forwardstrategy with M intermediate relay nodes is given by d(r) = (M + 1)(1− 2r) forr ∈ (0, 0.5).
Theorem 1∗: Under opportunistic relaying, the decode and forward protocol with Mintermediate relays achieves the same diversity multiplexing tradeoff, as in Theorem 0.
Theorem 2∗: Opportunistic amplify and forward achieves the same diversity multiplexingtradeoff stated in Theorem 0.
*: In cooperation with Ashish Khisti.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 18/51
Diversity-Multiplexing Tradeoff (1)
d∆= − lim
SNR→∞
logPe(ρ)
logSNRr∆= lim
SNR→∞
ρ(SNR)
logSNR
Diversity-Multiplexing Gain tradeoff tool averages out geometry.
cooperative diversity 6= multihop communication. This tool can reveal associatedgains/losses.
Theorem 0: The achievable diversity multiplexing tradeoff for the decode and forwardstrategy with M intermediate relay nodes is given by d(r) = (M + 1)(1− 2r) forr ∈ (0, 0.5).
Theorem 1∗: Under opportunistic relaying, the decode and forward protocol with Mintermediate relays achieves the same diversity multiplexing tradeoff, as in Theorem 0.
Theorem 2∗: Opportunistic amplify and forward achieves the same diversity multiplexingtradeoff stated in Theorem 0.
*: In cooperation with Ashish Khisti.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 18/51
Diversity-Multiplexing Tradeoff (1)
d∆= − lim
SNR→∞
logPe(ρ)
logSNRr∆= lim
SNR→∞
ρ(SNR)
logSNR
Diversity-Multiplexing Gain tradeoff tool averages out geometry.
cooperative diversity 6= multihop communication. This tool can reveal associatedgains/losses.
Theorem 0: The achievable diversity multiplexing tradeoff for the decode and forwardstrategy with M intermediate relay nodes is given by d(r) = (M + 1)(1− 2r) forr ∈ (0, 0.5).
Theorem 1∗: Under opportunistic relaying, the decode and forward protocol with Mintermediate relays achieves the same diversity multiplexing tradeoff, as in Theorem 0.
Theorem 2∗: Opportunistic amplify and forward achieves the same diversity multiplexingtradeoff stated in Theorem 0.
*: In cooperation with Ashish Khisti.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 18/51
Diversity-Multiplexing Tradeoff (2)
d(r)
r10.51/(M+1)
1
M+1Ideal
Space Time Coding
OpportunisticRelaying
Direct
Repetition
Opportunistic, single relay selection is as good as space-time coding simultaneoustransmissions!
This result holds for decode/forward as well as amplify/forward!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 19/51
Diversity-Multiplexing Tradeoff (2)
d(r)
r10.51/(M+1)
1
M+1Ideal
Space Time Coding
OpportunisticRelaying
Direct
Repetition
Opportunistic, single relay selection is as good as space-time coding simultaneoustransmissions!
This result holds for decode/forward as well as amplify/forward!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 19/51
Diversity-Multiplexing Tradeoff (2)
d(r)
r10.51/(M+1)
1
M+1Ideal
Space Time Coding
OpportunisticRelaying
Direct
Repetition
Opportunistic, single relay selection is as good as space-time coding simultaneoustransmissions!
This result holds for decode/forward as well as amplify/forward!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 19/51
Diversity-Multiplexing Tradeoff (2)
d(r)
r10.51/(M+1)
1
M+1Ideal
Space Time Coding
OpportunisticRelaying
Direct
Repetition
Opportunistic, single relay selection is as good as space-time coding simultaneoustransmissions!
This result holds for decode/forward as well as amplify/forward!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 19/51
Diversity-Multiplexing Tradeoff (2)
d(r)
r10.51/(M+1)
1
M+1Ideal
Space Time Coding
OpportunisticRelaying
Direct
Repetition
Opportunistic, single relay selection is as good as space-time coding simultaneoustransmissions!
This result holds for decode/forward as well as amplify/forward!
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 19/51
Results: Transmission Energy Gains
0 5 10 15 20 25 30 35 4010-7
10-6
10-5
10-4
10-3
10-2
10-1
100
E = E1 + E2
Symbol Error Probability for 8-PSK
noncooperativecooperative digital
v = 2
v = 3
v = 4
v = 5
Energy Gains
0 1 2 3 4 5 6 7
5
10
15
20
25
30
35
40
v
Ratio of Energy without cooperation vs (Total energy with cooperation)
Energy gains could counterbalance the decrease of rate by a factor of 2.
For the example above, 50% throughput increase is possible (8-PSK uncodedcooperative vs 2-PSK uncoded direct).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 20/51
Results: Reception Energy Gains
Existing cooperative techniques ignore RECEPTION energy for communication.
Cooperative reception of M relays scales reception energy cost... by a factor of M .
This is not small : Reception energy can become comparable to transmission energy inmodern transceivers [R. Min 2003].
Opportunistic relaying does not have this disadvantage: reception energy is fixed.
This is because best relay is chosen proactively, before message transmission. All otherrelays could go to sleep.
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 21/51
Results: Power Allocation Optimality (1)
What if TOTAL power allocated to the relays was fixed?
For amplify and forward networks, the equivalent system equation can be shown to be:
It can be shown that opportunistic relaying is superior to other approaches in the field.
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 22/51
Results: Power Allocation Optimality (1)
What if TOTAL power allocated to the relays was fixed?
For amplify and forward networks, the equivalent system equation can be shown to be:
yD,1
yD,2ω
=
√PSD aSD 0
1ω
∑Mi=1
√PSRi
√PRiD√
PSRi+N0aSRi aRiD
1ω
√PSD aSD
[x1
x2
]+
nD,1
nD,2ω
E[nD,2 n∗D,2 |HR→D ] = N0 (1 +
M∑
i=1
PRiD |aRid|2
PSRi + N0
)
︸ ︷︷ ︸ω2
= ω2
N0 (20)
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 22/51
Results: Power Allocation Optimality (1)
What if TOTAL power allocated to the relays was fixed?
For amplify and forward networks, the equivalent system equation can be shown to be:
y =
[ √PSD hSD 0
H211ω
√PSD hSD
]x + n (21)
y = H x + n (22)
IAF =1
2log2(1 +
PSD
N0
|hSD|2 +|H21|2
N0
) (23)
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 22/51
Results: Power Allocation Optimality (2)
Three cases considered, with all relays equivalent (same AVERAGE received SNR) :
Power to one relay (selection based on Average SNR).
Power distributed to all relays (space-time coding).
Power to opportunistic relay (Our Approach).
0 0.5 1 1.5 2 2.5 3 3.5 40
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
16 relays
CD
F of
Mut
ual I
nfor
mat
ion
bps/Hz
Selection one random relaySelecting all relaysOpportunistic Relaying
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 23/51
Results: Power Allocation Optimality (2)
Three cases considered, with all relays equivalent (same AVERAGE received SNR) :
Power to one relay (selection based on Average SNR).
Power distributed to all relays (space-time coding).
Power to opportunistic relay (Our Approach).
0 0.5 1 1.5 2 2.5 3 3.5 40
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
16 relays
CD
F of
Mut
ual I
nfor
mat
ion
bps/Hz
Selection one random relaySelecting all relaysOpportunistic Relaying
Single
All
Oppor
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 23/51
Results: Power Allocation Optimality (3)
1 2 3 4 5 6 7 8 9 102
2.5
3
3.5Average spectral efficiency
bps/
Hz
Number of Relays
Selection one random relaySelecting all relaysOpportunistic Relaying
Under a sum power constraint (and no beamforming capabilities) using all relays issuboptimal compared to opportunistic relaying.
Similar results for decode and forward.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 24/51
Results: Power Allocation Optimality (3)
1 2 3 4 5 6 7 8 9 102
2.5
3
3.5Average spectral efficiency
bps/
Hz
Number of Relays
Selection one random relaySelecting all relaysOpportunistic Relaying
Single
All
Oppor
Under a sum power constraint (and no beamforming capabilities) using all relays issuboptimal compared to opportunistic relaying.
Similar results for decode and forward.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 24/51
Overhead: Collision Probability (1)
best path @ kT
best path @ (k+1)T
Direct Relayed
|as,i|2 |ai,d|2
Source Destination
|as,j|2 |aj,d|2
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Policy I : hi = min{|asi|2, |aid|
2} Policy II : hi =2
1|asi|2
+ 1|aid|
2
=2 |asi|
2 |aid|2
|asi|2 + |aid|2
Ti =λ
hi(24)
Here λ has the units of time. For the discussion in this work, λ has simply values of µsecs.
hb = max{hi}, ⇐⇒ (25)
Tb = min{Ti}, i ∈ [1..M ]. (26)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 25/51
Overhead: Collision Probability (2)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Worst case scenario: the probability of having two ormore relays expire within the same interval c, out ofa collection of M relays is:
Pr(Collision) ≤ Pr(any Tj < Tb + c | j 6= b)
where Tb = min{Tj}, j ∈ [1,M ] and c > 0.
(a) No Hidden Relays : c = rmax + |nb − nj |max + ds
(b) Hidden Relays : c = rmax + |nb − nj |max + 2ds + dur + 2nmax
nj : propagation delay between relay j and destination. nmax is the maximum.
r: propagation delay between two relays. rmax is the maximum.
ds: receive-to-transmit switch time of each radio.
dur: duration of flag packet, transmitted by the "best" relay.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 26/51
Overhead: Collision Probability (2)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
Worst case scenario: the probability of having two ormore relays expire within the same interval c, out ofa collection of M relays is:
Pr(Collision) ≤ Pr(any Tj < Tb + c | j 6= b)
where Tb = min{Tj}, j ∈ [1,M ] and c > 0.
(a) No Hidden Relays : c = rmax + |nb − nj |max + ds
(b) Hidden Relays : c = rmax + |nb − nj |max + 2ds + dur + 2nmax
nj : propagation delay between relay j and destination. nmax is the maximum.
r: propagation delay between two relays. rmax is the maximum.
ds: receive-to-transmit switch time of each radio.
dur: duration of flag packet, transmitted by the "best" relay.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 26/51
Overhead: Collision Probability (3)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
If Tb = min{Tj}, j ∈ [1,M ] and Y1 < Y2 < . . . < YMthe ordered random variables {Tj} with Tb ≡ Y1, and Y2 thesecond minimum timer, then:
Pr(any Tj < Tb + c | j 6= b) ≡ Pr(Y2 < Y1 + c) (27)
Given that Yj = λ/h(j), Y1 < Y2 < . . . < YM is equivalent to1/h(1) < 1/h(2) < . . . < 1/h(M)
Pr(Y2 < Y1 + c) = Pr(1
h(2)<
1
h(1)+
c
λ) (28)
Ratio λc
needs to be as high as possible. λ and c are user controlled.
However λ needs to be kept small:
E[Tj ] = E[λ/hj ] ≥ λ/E[hj ] (29)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 27/51
Overhead: Collision Probability (3)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
If Tb = min{Tj}, j ∈ [1,M ] and Y1 < Y2 < . . . < YMthe ordered random variables {Tj} with Tb ≡ Y1, and Y2 thesecond minimum timer, then:
Pr(any Tj < Tb + c | j 6= b) ≡ Pr(Y2 < Y1 + c) (30)
Given that Yj = λ/h(j), Y1 < Y2 < . . . < YM is equivalent to1/h(1) < 1/h(2) < . . . < 1/h(M)
Pr(Y2 < Y1 + c) = Pr(1
h(2)<
1
h(1)+
c
λ) (31)
Ratio λc
needs to be as high as possible. λ and c are user controlled.
However λ needs to be kept small:
E[Tj ] = E[λ/hj ] ≥ λ/E[hj ] (32)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 27/51
Overhead: Collision Probability (3)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
If Tb = min{Tj}, j ∈ [1,M ] and Y1 < Y2 < . . . < YMthe ordered random variables {Tj} with Tb ≡ Y1, and Y2 thesecond minimum timer, then:
Pr(any Tj < Tb + c | j 6= b) ≡ Pr(Y2 < Y1 + c) (33)
Given that Yj = λ/h(j), Y1 < Y2 < . . . < YM is equivalent to1/h(1) < 1/h(2) < . . . < 1/h(M)
Pr(Y2 < Y1 + c) = Pr(1
h(2)<
1
h(1)+
c
λ) (34)
Ratio λc
needs to be as high as possible. λ and c are user controlled.
However λ needs to be kept small:
E[Tj ] = E[λ/hj ] ≥ λ/E[hj ] (35)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 27/51
Overhead: Collision Probability (3)
tL tHtC
Tb ds
dur
tb
tj
|nb-nj|
r r ds+2nb
CTS
CTS
CTS
flag packet
If Tb = min{Tj}, j ∈ [1,M ] and Y1 < Y2 < . . . < YMthe ordered random variables {Tj} with Tb ≡ Y1, and Y2 thesecond minimum timer, then:
Pr(any Tj < Tb + c | j 6= b) ≡ Pr(Y2 < Y1 + c) (36)
Given that Yj = λ/h(j), Y1 < Y2 < . . . < YM is equivalent to1/h(1) < 1/h(2) < . . . < 1/h(M)
Pr(Y2 < Y1 + c) = Pr(1
h(2)<
1
h(1)+
c
λ) (37)
Ratio λc
needs to be as high as possible. λ and c are user controlled.
However λ needs to be kept small:
E[Tj ] = E[λ/hj ] ≥ λ/E[hj ] (38)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 27/51
Overhead: Collision Probability (4)
Lemma: Given M ≥ 2 i.i.d. positive random variables T1, T2, . . . , TM , each withprobability density function f(x) and cumulative distribution function F (x), andY1 < Y2 < Y3 . . . < YM are the M ordered random variables T1, T2, . . . , TM , thenPr(Y2 < Y1 + c), where c > 0, is given by the following equations:
Pr(Y2 < Y1 + c) = 1− Ic (39)
Ic =M (M − 1)
∫ +∞
cf(y) [1− F (y)]M−2 F (y − c) dy (40)
Wireless channel statistics of h⇒ pdf f and cdf F of T = λ/h⇒ Pr(collision).
Example: for a mobility of 0− 3 km/h ⇒ maximum Doppler shift is fm = 2.5 Hz ⇒
minimum coherence time on the order of Tc ' 200 milliseconds.
For c/λ ≈ 1/200⇒ Pr(Collision) ≤ 0.6% for policy I.
For c ≈ 5µs⇒ λ ≈ 1ms ' 1100
Tc.
For c ≈ 1µs⇒ λ ≈ 200µs ' 11000
Tc.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 28/51
Overhead: Collision Probability (4)
Lemma: Given M ≥ 2 i.i.d. positive random variables T1, T2, . . . , TM , each withprobability density function f(x) and cumulative distribution function F (x), andY1 < Y2 < Y3 . . . < YM are the M ordered random variables T1, T2, . . . , TM , thenPr(Y2 < Y1 + c), where c > 0, is given by the following equations:
Pr(Y2 < Y1 + c) = 1− Ic (41)
Ic =M (M − 1)
∫ +∞
cf(y) [1− F (y)]M−2 F (y − c) dy (42)
Wireless channel statistics of h⇒ pdf f and cdf F of T = λ/h⇒ Pr(collision).
Example: for a mobility of 0− 3 km/h ⇒ maximum Doppler shift is fm = 2.5 Hz ⇒
minimum coherence time on the order of Tc ' 200 milliseconds.
For c/λ ≈ 1/200⇒ Pr(Collision) ≤ 0.6% for policy I.
For c ≈ 5µs⇒ λ ≈ 1ms ' 1100
Tc.
For c ≈ 1µs⇒ λ ≈ 200µs ' 11000
Tc.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 28/51
Rigorous analysis earns you trips around the world...
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 29/51
Overhead: Collision Probability (5)
200 220 240 260 280 300 320 340 360 380 4003
4
5
6
7
8
9
10x 10-3 Rayleigh and Ricean Fading vs lambda /c, for M=6
lambda /c
Pro
babi
lity
of C
ollis
ion
Policy II (harmonic), Rayleigh, SimulationPolicy II (harmonic), Rayleigh, AnalysisPolicy I (min), Ricean, SimulationPolicy I (min), Rayleigh, SimulationPolicy I (min), Rayleigh, Analysis
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 30/51
Overhead: Collision Probability (6)
Case 1
Case 2
Case 3
Case 4 Case 1 Case 2 Case 3 Case 4 1
2
3
4
5
6
7
8
9
10x 10-3
Pro
babi
lity
of C
ollis
ion
4 different topologies for M=6
Assymetry and collision probability
v=3,Policy II (harmonic)v=4,Policy II (harmonic)v=3,Policy I (min)v=4,Policy I (min)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 31/51
...and a Remark...
Stream I Stream II
Relay i
b = arg︸︷︷︸i
max{min{SNRsi, SNRid}} = max{SNRsid}, i ∈ [1..M ] (43)
b = arg︸︷︷︸i
max{min{SINRsi, SINRid}} = max{SINRsid}, i ∈ [1..M ] (44)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 32/51
...and a Remark...
Stream I Stream II
Relay i
b = arg︸︷︷︸i
max{min{SNRsi, SNRid}} = max{SNRsid}, i ∈ [1..M ] (45)
b = arg︸︷︷︸i
max{min{SINRsi, SINRid}} = max{SINRsid}, i ∈ [1..M ] (46)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 32/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 33/51
Implementation: Hardware
Rethinking wireless: approach needs access to physical (layer 1), link (layer 2), routing(layer 3).
COTS radios usually give limited access to all layers ⇒
We built our own radios. Simple, low cost, embedded Software Defined Radios (SDRs).
We built a room size demo.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 34/51
Implementation: Demo Setup
left view
right view
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 35/51
Implementation: Demo Setup
Receiver
Relay
Relay
Relay Relay
Relay
Relay
left view
right view
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 35/51
Implementation: Signal Structure
Direct transmission of 16 frames
Signal structure of each frame
Preamble 32 bits (on-off keying)
CTS Flag �packet 16/32 data frames
Direct and best relay transmission� (16 + 16 = 32 frames)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 36/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 37/51
Coordination, Cooperation and Time Keeping
Relays (or receiver) might be busy!
Relays (or receiver) might be in sleep mode.
Therefore, relays need to be awake on time!
Time keeping could simplify required scheduling.
Other researchers believe that time keeping is the basis of scalable communication.
We briefly present two approaches on network time keeping:
centralized
decentralized
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 38/51
Coordination, Cooperation and Time Keeping
Relays (or receiver) might be busy!
Relays (or receiver) might be in sleep mode.
Therefore, relays need to be awake on time!
Time keeping could simplify required scheduling.
Other researchers believe that time keeping is the basis of scalable communication.
We briefly present two approaches on network time keeping:
centralized
decentralized
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 38/51
Centralized Time Keeping
... ... ... ...
... ... ... ...
CLIENT SERVER
No control over the network: noisyenvironment.
No control over the time server: wouldlike to use existing infrastructure.
Three End-to-End algorithms werecompared:
Averaging (NIST).
Linear Programming (proposedbefore).
Kalman Filtering (our proposal).
Objective: estimate φ and θ, with minimum communication BW and computationrequirements.
Could we do better than simple averaging?
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 39/51
Centralized Time Keeping Results
0 20 40 60 80 100 120-600
-400
-200
0
200
400
600
800
number N of packets used in calculation
frequency offset estimate (ppm)
KalmanLinear ProgrammingAveraged Time Differences"Naive" Estimator
Proposed technique can improve accuracy (error) and precision (variance of error),compared to existing approaches.
Computation efficient (since it is recursive) -
Implemented and tested using existed NTP infrastructure.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 40/51
Centralized Time Keeping Results
0 20 40 60 80 100 120-600
-400
-200
0
200
400
600
800
number N of packets used in calculation
frequency offset estimate (ppm)
KalmanLinear ProgrammingAveraged Time Differences"Naive" Estimator
Kalman
Average
Proposed technique can improve accuracy (error) and precision (variance of error),compared to existing approaches.
Computation efficient (since it is recursive) -
Implemented and tested using existed NTP infrastructure.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 40/51
Decentralized Time Keeping
Network The network is the time server.
Only local communication.
Exchange timestamps and keep the highest(Lamport’s idea).
Redefine time as a periodic function!
The network re-calibrates periodically and au-tonomously.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 41/51
Decentralized Time Keeping Results
0 1 2 3 4 5 60
2
4
6
8
10
12
14
Network diameter (maximum number of hops)
Abs
olut
e er
ror a
nd s
tand
ard
devi
atio
n in
ms
Measured error in ms vs. diameter of the network
r1r2
Error could decrease with increasingNetwork diameter!
ε(tc) = Ci(tc)− Cj(tc) =
= ε(t0 + x) + (φi − φj) ∆t
∆t = tc − (t0 + x)
Error depends on communication BW.x =
propagationdelay+ transmissiondelay+
+ operating system delay.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 42/51
Decentralized Time Keeping Demo
Objective: play music in synchrony, display heartbeat at the edges...
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 43/51
Decentralized Time Keeping Demo (2)
This algorithm is based on oscillator’s coupling (no averaging).
Coupling among terminals with semi-periodic signal ≡ Entrainment.
It is relevant to natural phenomena of synchronization (fireflies, cardiac neurons etc.)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 44/51
Outline
Assumptions and Background
Approach
Performance
Implementation Example
Relevant Technologies
Conclusion
Acknowledgements
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 45/51
Conclusions (1)
Intelligent single antenna selection:
performs as good as ST-coding simultaneous transmissions, in terms ofdiversity-multiplexing tradeoff.
outperforms simultaneous transmissions, in terms of power allocation at therelays.
provides for significant RECEPTION energy savings (proactive character).
simplifies processing at the relays and the receiver (reduced complexity).
is (really) fast and (really) distributed - other proposals might follow.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 46/51
Conclusions (1)
Intelligent single antenna selection:
performs as good as ST-coding simultaneous transmissions, in terms ofdiversity-multiplexing tradeoff.
outperforms simultaneous transmissions, in terms of power allocation at therelays.
provides for significant RECEPTION energy savings (proactive character).
simplifies processing at the relays and the receiver (reduced complexity).
is (really) fast and (really) distributed - other proposals might follow.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 46/51
Conclusions (1)
Intelligent single antenna selection:
performs as good as ST-coding simultaneous transmissions, in terms ofdiversity-multiplexing tradeoff.
outperforms simultaneous transmissions, in terms of power allocation at therelays.
provides for significant RECEPTION energy savings (proactive character).
simplifies processing at the relays and the receiver (reduced complexity).
is (really) fast and (really) distributed - other proposals might follow.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 46/51
Conclusions (1)
Intelligent single antenna selection:
performs as good as ST-coding simultaneous transmissions, in terms ofdiversity-multiplexing tradeoff.
outperforms simultaneous transmissions, in terms of power allocation at therelays.
provides for significant RECEPTION energy savings (proactive character).
simplifies processing at the relays and the receiver (reduced complexity).
is (really) fast and (really) distributed - other proposals might follow.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 46/51
Conclusions (1)
Intelligent single antenna selection:
performs as good as ST-coding simultaneous transmissions, in terms ofdiversity-multiplexing tradeoff.
outperforms simultaneous transmissions, in terms of power allocation at therelays.
provides for significant RECEPTION energy savings (proactive character).
simplifies processing at the relays and the receiver (reduced complexity).
is (really) fast and (really) distributed - other proposals might follow.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 46/51
Conclusions (2)
This thesis suggests:
Instead of searching for ST coding for the relay channel at the physical layer,we could alternatively research smart relay selection techniques, at the routinglayer. Emphasis on instantaneous channel conditions (vs average).
Cooperative diversity is a cross-layer approach. Cross layer analysis isrequired.
Coordination and group formation are not trivial and require special attention.
Concrete centralized or decentralized network time keeping algorithms couldbe of help. Specific examples were analyzed in theory and implemented inpractice.
Formation of virtual antenna arrays can be simplified to practice, using existingRF-front ends. Proposed method is applicable today and a demonstrationexample was built.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 47/51
Conclusions (2)
This thesis suggests:
Instead of searching for ST coding for the relay channel at the physical layer,we could alternatively research smart relay selection techniques, at the routinglayer. Emphasis on instantaneous channel conditions (vs average).
Cooperative diversity is a cross-layer approach. Cross layer analysis isrequired.
Coordination and group formation are not trivial and require special attention.
Concrete centralized or decentralized network time keeping algorithms couldbe of help. Specific examples were analyzed in theory and implemented inpractice.
Formation of virtual antenna arrays can be simplified to practice, using existingRF-front ends. Proposed method is applicable today and a demonstrationexample was built.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 47/51
Conclusions (2)
This thesis suggests:
Instead of searching for ST coding for the relay channel at the physical layer,we could alternatively research smart relay selection techniques, at the routinglayer. Emphasis on instantaneous channel conditions (vs average).
Cooperative diversity is a cross-layer approach. Cross layer analysis isrequired.
Coordination and group formation are not trivial and require special attention.
Concrete centralized or decentralized network time keeping algorithms couldbe of help. Specific examples were analyzed in theory and implemented inpractice.
Formation of virtual antenna arrays can be simplified to practice, using existingRF-front ends. Proposed method is applicable today and a demonstrationexample was built.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 47/51
Conclusions (2)
This thesis suggests:
Instead of searching for ST coding for the relay channel at the physical layer,we could alternatively research smart relay selection techniques, at the routinglayer. Emphasis on instantaneous channel conditions (vs average).
Cooperative diversity is a cross-layer approach. Cross layer analysis isrequired.
Coordination and group formation are not trivial and require special attention.
Concrete centralized or decentralized network time keeping algorithms couldbe of help. Specific examples were analyzed in theory and implemented inpractice.
Formation of virtual antenna arrays can be simplified to practice, using existingRF-front ends. Proposed method is applicable today and a demonstrationexample was built.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 47/51
Conclusions (2)
This thesis suggests:
Instead of searching for ST coding for the relay channel at the physical layer,we could alternatively research smart relay selection techniques, at the routinglayer. Emphasis on instantaneous channel conditions (vs average).
Cooperative diversity is a cross-layer approach. Cross layer analysis isrequired.
Coordination and group formation are not trivial and require special attention.
Concrete centralized or decentralized network time keeping algorithms couldbe of help. Specific examples were analyzed in theory and implemented inpractice.
Formation of virtual antenna arrays can be simplified to practice, using existingRF-front ends. Proposed method is applicable today and a demonstrationexample was built.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 47/51
PapersConferencesA. Bletsas, A. Lippman, D.P. Reed, "A Simple Distributed Method for Relay Selection in Cooperative Diversity Wireless Networks, based onReciprocity and Channel Measurements", accepted for publication, IEEE 61st Semiannual Vehicular Technology Conference May 30 - June 1 2005,Stockholm, Sweden.
A. Bletsas and A. Lippman, "Spontaneous Synchronization in Multi-hop Embedded Sensor Networks: Demonstration of a Server-free Approach",Second European Workshop on Wireless Sensor Networks, January 31 - February 2 2005, Istanbul, Turkey.
A. Bletsas and A. Lippman, "Efficient Collaborative (Viral) Communication in OFDM Based WLANs", IEEE/ITS International Symposium onAdvanced Radio Technologies (ISART 2003),Institute of Standards and Technology, Boulder Colorado, March 4-7, 2003.
A. Bletsas, "Evaluation of Kalman Filtering for Network Time Keeping", IEEE International Conference on Pervasive Computing and Communications(PerCom 2003), Dallas-Fort Worth Texas, March 23-26, 2003.
B. Hubert, A. Bletsas, J. Jacobson, "Nano-Scale Structures Fabricated By All-Additive AFM -Assisted Nanoassembly", Materials Research Society(MRS), San Francisco, April 16-20, 2001.
JournalsA. Bletsas, A. Khisti, D.P. Reed, A. Lippman, "A Simple Cooperative Diversity Method based on Network Path Selection", submitted for publication,IEEE Journal on Selected Areas of Communication, special Issue on 4G, January 2005, revised April 2005.
A. Bletsas, "Evaluation of Kalman Filtering for Network Time Keeping", accepted for publication, IEEE Transactions in Ultrasonics, Ferromagneticsand Frequency Control (TUFFC).
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 48/51
Acknowledgements
"Tolerating ambiguity is a sign of maturity..."
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 49/51
Acknowledgements
"Tolerating ambiguity is a sign of maturity..."
In memory of Stephen A. Benton (1941-2003)
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 49/51
Acknowledgements
Thesis Committee: Andy Lippman, Joe Paradiso and Moe Win.
Colleagues at MIT: Ashish Khisti, Josh Lifton, David Reed and Joe Jacobson.
My UROPs: Vimal Bhalodia, Amanda Lechman and Marios Michalakis.
Colleagues outside MIT: Thucydides (Duke) Xanthopoulos.
Office-mates: Dean Christakos, Ilia Mirkin and Dimitris Vyzovitis.
Group Mates: Jamie Cooley, Jeff Hallbig, Casey Muller, Hector Yuen and the rest of the Viral Comm Gang: Arthur Petron, Kwan Hong Leeand Fulu Li.
MIT people: Judith Donath, Neil Gershenfeld’s Physics and Media Group, Hyundong Shin (LIDS).
Necsys Crew: Paula Aguilera, Steve Berezansky, Jon Ferguson, Jeannie Finks, Will Glesnes, Tom Greene, Elizabeth Harvey-Forsythe, HenryHoltzman, Jane Wojcik, Chi Yuen.
ML colleagues: Betsy Chimento, Tamara Hearn, Kevin Davis, Cornelle King, Sandy Sener and Stacie Slotnick.
My sweet aunts here at Media Lab: Polly Guggenheim and Deborah Widener.
Crazy Roomates: Maziar Tavakoli-Dastjerdi, Saul Griffith, Yael Maguire, John Maloney and Noah Vawter.
To all my friends at Crete: Yiannis, Dia, Olga, Andonis, Costas, Mihalis.
Friends here in Boston (including Greek Mafia): Thodoros Konstantakopoulos, Anna Stefanidou, Ioannis Kitsopanidis, Nikol Papadopoulou,Duke Xanthopoulos, Margarita Dekoli, Vasilis Ntziachristos, Christina Benou, Eirini Iliaki, Ioannis Kizanis, Anna Kondyli, Yiannis Zacharakis,George Themelis, Dimitris Rovas, Christina Samaraki, Georgia Konstadinopoulou, Karrie Karrahalios, Paris Smaragdis, Costas Pelekanakis,Thais Aleluia, Petros Bufunos, Thodoros Akiskalos, Christi Electris, Constadinos Caramanis, George Constadinidis, Maria-KaterinaNikolinakou, Angelina Aessopou, Wei Chai...
Brother and his family here in Boston, sister and her family in Norway, Eleftheria and her family in Thessaloniki, and parents and grandaunt inCrete...
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 50/51
Thank you!
...to Eleftheria, Thodoris, Aimilia-Anastasia, Constadinos, Christina and to my family.
aggelos@mit.edu, Ph.D. Thesis Defense, MIT June 2005. – p. 51/51
top related