To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop...

25
To Fragment or Not To Fragment: Viability of NCOFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University

Transcript of To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop...

Page 1: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

To Fragment or Not To Fragment: Viability of NC‐OFDMA in Multihop Networks

Muhammad Nazmul IslamWINLAB, Rutgers University

Page 2: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Availability of Non‐Contiguous Spectrum• Demand for wireless services is increasing rapidly

– Qualcomm predicts a 1000x increase by 2020

• FCC has opened up 300 MHz in TV bands – Plans to open up additional 500 MHz “license‐by‐rule” bands by 2020

• Any radio can use these bands if it abides by FCC rules

• If uncoordinated networks use these bands, they will adjust spectrum usage according to traffic demands– Available bands will become non‐contiguous 

• TV white space is itself non‐contiguous in nature

2

Page 3: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Case for Noncontiguous OFDMA ‐ I

3

1

2

3

A

B

C

X

• Three available channels

• Node A transmits to node C via node B.

• Node B relays node A’s data and transmits itsown data to node C. 

• Node X, an external and uncontrollableinterferer, transmits in channel 2.

2

If we use max‐min rate objective and allocate channels, node B requires two channels; node A requires one channel

Scheduling options for Node A and Node B?

Page 4: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Case for Noncontiguous OFDMA ‐ II

4

2

A

C

3

B

• Transmission in link BC suffers interference in channel 2

1 2

#1: Contiguous OFDM

X

2

A

C

B

• Spectrum fragmentation limited by number of radio front ends

1 3

2

#2: Multiple RF front ends

X

4

2

A

C

B

2

1 3

#3: Non‐Contiguous OFDM (NC‐OFDMA)

Nulled Subcarrier

X

NC‐OFDM accesses multiple fragmented spectrum chunks with single radio front end

Page 5: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

5

2

A

AP

B

2

1 3

#3: Non‐Contiguous OFDM

Nulled Subcarrier

Serial toParallel IFFT Parallel

to Serial D/A

X

X[1] X[3]X[1]

X[3]

0

x[1]x[2]x[3]

X[2]  =

NC‐OFDM accesses multiple fragmented spectrum chunks with single radio front end

• Node B places zero in channel 2 and avoids interference

• Node A, far from the interferer node X, uses channel 2.

• Both nodes use better channels.

• Node B spans three channels, instead of two.• Sampling rate increases.

Modulation

NC‐OFDM Operation

Page 6: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Benefits and Challenges of NC‐OFDMA

Benefits:• Avoids interference, incumbent users.• Uses better channels

6

Challenges:• Increases sampling rate Increases ADC & DAC power Increases amplifier power

Page 7: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Power Consumption Model 

7

• ADC and DAC power depend on the sampling rate• Other blocks deal with analog signals – power consumption does not depend on 

sampling rate We ignore programmable amplifier’s power consumption here.

.Power,Rx Power,Tx 2121 srMm

mst fppfp

m.channelin power Allotted- rateSampling- constants,,, 2121 ms pf

Page 8: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Optimization Formulation – Our Approach

Ni

isysP , min

Ni )( ,,,

isysNj

ijiRxCktiTxCkt PPPP

(N,N)(i,j) f)WNPg

(W* ijijij

02 1log

sconstraintduplex halfon,conservati flow ce,Interferen

Total System Power Minimization

Individual System Power Constraint

Capacity Constraint

i. nodeofpower System nodes. ofSet , isysPN

Formulation

Notation

ijlink at power Emittedpath,Tx si'nodeofpower Ckt , ijiTxCkt PP

ij.at Flow ij,ofgain Link Density,SpectralNoise Bandwidth, 0 ijij fgNW

Page 9: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Transmit Power Minimization ‐Waterfilling

Ni

iTxP , min

Ni ,

iTxNj

ij PP

(N,N)(i,j) f)WNPg

(W* ijijij

02 1log

sconstraintduplex halfon,conservati flow ce,Interferen

Total TrasnmitPower Minimization

Individual Transmit Power Constraint

Capacity Constraint

i. node ofpower Transmit nodes. ofSet , iTxPN

Formulation

Notation

ijlink at power EmittedijP

ij.at Flow ij,ofgain Link Density,SpectralNoise Bandwidth, 0 ijij fgNW

Page 10: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Comparison with Waterfilling ‐ I

10

• Channel width – 3 MHz. Minimum required rate – 18 Mbps.

• Waterfilling selects good channels across the whole list.

• Our approach selects two non‐contiguous good neighbors.

Point‐to‐Point Link

Page 11: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Comparison with Waterfilling ‐ II

11

• Waterfilling consumes less transmit power.

• Our approach consumes less system power.

Point‐to‐Point Link

Page 12: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

12

1

2

3

4

5

6

57 79 85 491 527 533 671

• Channel width – 6 MHz.

• Two sessions.

• Required rate – 10 Mbps.

2 5 6 17 23 24 47Index

Location(MHz)

S(1) =

D(1) = D(2) =

S(2) =

• Node 1 transmits to 3. 

• Node 2 transmits to 4.

Multi‐hop Scenario (TV Bands in Wichita, KS)

Page 13: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

13

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

‐134

‐132

‐130

‐128

Chan

nel G

ain (dB)

Link 1225617232447

Waterfilling Approach

1

2

3

4

5

6

23 24

S(1) =

D(1) = D(2) =

S(2) =

Page 14: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

14

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

‐136

‐135

‐134

‐133

‐132

Chan

nel G

ain (dB)

Link 2325617232447

Waterfilling Approach

1

2

3

4

5

6

2 17

S(1) =

D(1) = D(2) =

S(2) =

23 24

Page 15: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

15

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

‐135

‐133

‐131

‐129

Chan

nel G

ain (dB)

Link 4525617232447

Waterfilling Approach

1

2

3

4

5

6

2 17

5 17

S(1) =

D(1) = D(2) =

S(2) =

23 24

Page 16: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

16

1

2

3

4

5

6

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

2 17 2 24

5 17

S(1) =

D(1) = D(2) =

S(2) =‐136

‐134

‐132

‐130

‐128

Chan

nel G

ain (dB)

Link 5625617232447

Waterfilling Approach

23 24

Page 17: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

17

Our Approach (Low Power ADC & DAC)

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

1

2

3

4

5

6

S(1) =

D(1) = D(2) =

S(2) =

5 6

23 245 6

2

Page 18: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Comparison with Waterfilling (Low Power ADC & DAC)

18

• Our approach reduces system power consumption by 4 dB.

Page 19: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

19

Our Approach (USRP ADC & DAC)

57 79 85 491 527 533 671

2 5 6 17 23 24 47Index

Location(MHz)

1

2

3

4

5

6

17

24

2

5

S(1) =

D(1) = D(2) =

S(2) =

Page 20: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Comparison with Waterfilling (USRP ADC & DAC)

20

• USRP ADC and DAC’s power consumption curves are steeper.• Waterfilling spans more spectrum and consumes a lot of system power

• Our approach reduces system power consumption by 10 dB!

Page 21: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Conclusion

• Researchers focus on channel gains and traffic demands to determine power control, scheduling and routing variables

• Our results reveal that hardware configuration of the radio front ends, e.g., slope of ADC and DAC power consumption curves, can influence these variables.

21

Page 22: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Future Works

• Power consumption models of programmable amplifier– Non‐linear due to “bandwidth gain product”

• Investigate multi‐frond‐end radio’s power consumption– Analog power may increase since there are multiple components

– Digital power may decrease since each ADC/DAC spans narrower spectrum

22

Page 23: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Questions ?

Thank You !

23

Page 24: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

Power MinimizationTransmit Power Minimization

(Waterfilling) System Power Minimization       

(Our Approach)

Mmp

rWNgpW

m

Mm

mm

0

)1(log s.t.

pmin

02

Mmm

Mmp

rWNgpW

ff

m

Mm

mm

ss

0sConstraintSpan Spectrum

)1(log s.t.

pmin

02

2121Mm

m

Page 25: To or Not To Fragment: of NC OFDMA in Multihop · PDF fileViability of NC‐OFDMA in Multihop Networks Muhammad Nazmul Islam WINLAB, Rutgers University. Availability of Non‐Contiguous

References1. “Qualcomm data challenge,” accessed March 2013, 

http://www.qualcomm.com/media/documents/wireless‐networks‐risingmeet‐1000x‐mobile‐data‐challenge

2. C. Cordeiro, K. Challapali, D. Birru, and S. Shankar, “IEEE 802.22: the first worldwide wireless standard based on cognitive radios,” in Proc. IEEE DySPAN’2005, Nov. 2005, p. 328337.

3. “Enabling innovative small cell use in 3.5 GHZ band NPRM & order,” accessed March 2013, http://www.fcc.gov/document/enablinginnovative‐small‐cell‐use‐35‐ghz‐band‐nprm‐order

4. S. Cui, A. Goldsmith, and A. Bahai, “Energy‐constrained modulation optimization,” IEEE Transactions on Wireless Communications, vol. 4, pp. 2349 – 2360, SEP 2005

5. Y. Shi and Y. T. Hou, “Optimal power control for multi‐hop software defined radio networks,” in Proc. IEEE INFOCOM’2007, May 2007, pp. 1694–1702.

25