power control and adaptive packet scheduling for wireless...

20
29-11-2001 INFOCOM Department seminar, fall 2001 Power control and adaptive packet scheduling for wireless data Andrea Baiocchi INFOCOM Department University of Roma La Sapienza Joint work with Francesca Cuomo, Cristina Martello, Fabrizio Capriotti 29-11-2001 INFOCOM Department seminar, fall 2001 2 Overview l Context l Basic issues of mobile computing l Role of adaptive mechanism l Power control for wireless data l Power & capacity assignment adaptive to traffic load and channel quality ¨ Ad-hoc networking ¨ Cellular networking l Hints to further research activity

Upload: dangthien

Post on 18-Apr-2019

213 views

Category:

Documents


0 download

TRANSCRIPT

29-11-2001INFOCOM Department seminar, fall 2001

Power control and adaptive packet

scheduling for wireless data

Andrea Baiocchi

INFOCOM Department

University of Roma ÒLa SapienzaÓ

Joint work with Francesca Cuomo, Cristina Martello, FabrizioCapriotti

29-11-2001INFOCOM Department seminar, fall 2001

2Overview

l Context

l Basic issues of mobile computing

l Role of adaptive mechanism

l Power control for wireless data

l Power & capacity assignment adaptive to traffic load andchannel quality

È Ad-hoc networking

È Cellular networking

l Hints to further research activity

29-11-2001INFOCOM Department seminar, fall 2001

3Context

l Progetto UE - IST ÒWhyless.com - Open Mobile AccessNetworkÓ (2001-2003)

È IMST, Technical Univ Dresden, Univ. Lintfort, Lynx (Germany),Univ. RM ÒLa SapienzaÓ, Univ. PG (Italy), UCLA (USA)

l Progetto CNR-MURST ÒRete Multimediale Interattiva diAccesso allÕUtenteÓ (1998-2002)

l Progetto MURST CoFin2000 ÒReconfigurable AccessModule for Mobile Computing Applications (RAMON)(2001-2002)

l Cooperation with Coritel/Ericsson

29-11-2001INFOCOM Department seminar, fall 2001

4Major issues of mobile computing

l Capacity, QoS support, fairness

È MAC design and packet scheduling

l Reliability

È Design of coding and ARQ at data link layer and transport layer

l Energetic efficiency

È Power control, MAC and error recovery design

l Addressing, mobility management and routing

l Security

l Adaptive applications, robust session handling

29-11-2001INFOCOM Department seminar, fall 2001

5Ad-hoc networks

l Emphasis on distributed algorithms for dynamic adaptiveradio capacity sharing of radio capacity

È Limited signalling and processing capabilities

È Local knowledge of the environment

l Routing for a time-varying topology

l Specific applications: WLAN, sensor networks, ÒpervasivecomputingÓ, smart spaces

l Some available and developing technologies: Bluetooth,IEEE 802.11b, ETSI HiperLAN, UltraWideBand radio

29-11-2001INFOCOM Department seminar, fall 2001

6Role of adaptive scheduling

l Adaptive with respect to

È Traffic load (e.g. measured in terms of backlogged RLC/MACpackets, or averaged/max packet delay)

È Trasmission quality (e.g.measured by SINR, bit/block error ratio,received power level, overall interference)

È Energy consumption (e.g. measured by the amount of batteryenergy required per correclty delivered bit)

l There are different QoS requirements, e.g.

È LDD = Low Delay Data (e.g. transactions)

È LCD = Long Constrained Delay (e.g., WWW)

È UDD = Unconstrained Delay Data (e.g., e-mail, news)

29-11-2001INFOCOM Department seminar, fall 2001

7An example of adaptive scheduling

0 50 100 150 2000.5

0.6

0.7

0.8

0.9

1

1.1

0 50 100 150 2000

0.2

0.4

0.6

0.8

1

0 50 100 150 2000

0.5

1

1.5

2

2.5

3

Log-normal attenuation(mean=0 dB, stand. dev.=3 dB)

Required power to obtaintarget SNR (or best achievableSNR) normalized with respectto the maximum allowed power

SNR normalized with respectto target SNR value

29-11-2001INFOCOM Department seminar, fall 2001

8What can be gained

Mean delay

0.4 0.5 0.6 0.7 0.8 0.9 10

5

10

15

20

25

30

(T+C)*

T+CT

Offered load

Hit probability (%)

0.4 0.5 0.6 0.7 0.8 0.9 10

20

40

60

80

100

(T+C)*

T T+C

Offered load

Resource utilization (%)

0.4 0.5 0.6 0.7 0.8 0.9 10

20

40

60

80

100

(T+C)*

T+C

T

Offered load

T = Traffic information only based algorithm

T+C=Joint traffic and channel information based algorithm

(T+C)*=As above plus perfect knoledge of future channel state

Two state Markov channel model 0/1 (10/40 TI)

29-11-2001INFOCOM Department seminar, fall 2001

9Power sensitive networks

l Two paradigms

È Ad-hoc networking

È Cellular networking

l Conceptual modeling of a wireless access as a collectionof interfering links (concurrent single-hop transmissions)sharing a channel

l Local decisions

È Power adjustment -> autonomous probe

È Interference measurement -> collective reaction

l QoS (throughput, packet delay) monotonously increasingwith SINR

29-11-2001INFOCOM Department seminar, fall 2001

10Power Control (1/5)

RP g

P gi

i ii

j jij i ii=

+≥

≠∑ ηγ

l Classic approach (Foschini and Miljanic [Ô93])

l The SIR requirements of N mutually interfering links can beexpressed as

l In matrix form, the SIR requirements are

l F is nonnegative and irreducible; hence its maximummodulus eigenvalue λ is real, positive and simple.

p I F u( )− ≥

u g f g g i j Ni i i ii ji i ji ii= = = …γ η γ, , , , ,1

29-11-2001INFOCOM Department seminar, fall 2001

11Power Control (2/5)

l The existence of a feasible p is equivalent to λ<1; in thatcase (IÐF)Ð1 exists and it is nonnegative componentwise

l The vector p*= (IÐF)Ð1u is the Pareto optimal powerassignment, i.e. for any other feasible p, p ³ p*

l The iteration p(k+1)=p(k)F+u, k³0, converges to p*geometrically fast, with rate λ.

l This property carries over to the case of asynchronouspower adaptations, provided there exists a finite integer ssuch that each terminal updates its power at least onceevery s power updates in the system (Mitra [Ô93])

P k P k d k i j f u i U k

P k i U k

i j jij i i

i

( ) ( ( , , )) ( );

( ) ( )

+ = − + ∈

= ∉≠∑1

29-11-2001INFOCOM Department seminar, fall 2001

12Power Control (3/5)

l Distributed Power Control (DPC) algorithm

l SIR can fluctuate below γ during their evolution (new links)

l Active Link Protection (ALP)-DPC algorithm (Bambos[Ô95,Õ98]): A(k) set of active links (Ri(k)³γ i), B(k) set ofinactive links (Ri(k)<γi), δ a constant greater than 1

l Nice properties of ALP-DPC

È SIR protection of active links

È Bounded power overshoot

È SIR improvements of new links

P k

R kP k k i Ni

i

ii( )

( )( ), , , ,+ = ≥ = …1 0 1

γ

P k

R kP k i A k P k P k i B ki

i

ii i i( )

( )( ), ( ) ( ) ( ), ( )+ = ∈ + = ∈1 1

γ δ δ

29-11-2001INFOCOM Department seminar, fall 2001

13Power Control (4/5)

l Power explosion problem if links cannot be accommodatedat SIR thresholds δγi

l Two corrective actions

È Forced Drop Out under bounded power of the transmitter

È Voluntary Drop Out, after time out of the number of unsuccessfulupdates to become active

l Fast nonintrusive channel probing

È Two probes at power levels P(0) and δP(0) allow gauging theexpected steady-state SIR, given at time 0 all active links have SIRin [γ i,δ γi] (stabilized system)

l Critical point: mobility timescale >> power control timescale >> transmission timescale

29-11-2001INFOCOM Department seminar, fall 2001

14Power Control (5/5)

l Multichannel case (choose best channel)

l Minimum power routing in multihop network

l Power sensitive scheduling

È Minimize the average spent power given the backgroundinterference environment (correlated!), under an average packetdelay constraint

l Power sensitive error recovery

È Give up retransmission if the channel is so bad that the enerfyconsumption is not worth the transmission attempt

l Game theoretic approach, Shah [Ô98], Utility Function,Goodman and Mandayam [Ô00]

29-11-2001INFOCOM Department seminar, fall 2001

15Distributed power & capacity

assignment

l UWB terminals carrying Best Effort (BE) traffic andReservation Based (RB) traffic

l The distributed nature of the algorithms requires both localmeasurements, to assess the interference and noise beforeand after the onset of new links, and signalling, to realizethe admission rules

Access Point

INTERNET

Reference scenario

UWB access network

29-11-2001INFOCOM Department seminar, fall 2001

16Background

SNRP g

R T P g

i Nii ii

i i f k kik k i

N=

+

= …

= ≠∑η σ2

1

1

,

, ,

l UWB link model: non dispersive AWGN with MUI (Win,Scholtz [Ô97, Ô98, Ô00]); SINR of link between I-th Tx and Rx

l Constraints on link performance and parameters

l Reference: Baiocchi et alii [Ô01]

SNR i N

P P Ri i

i i

≥ = …< ≤ >

γ 1

0 0

, ,

max

29-11-2001INFOCOM Department seminar, fall 2001

17Problem statement: RB traffic

given find such that

P g R T P g R i N

P P

i ii i i f k kik k i

N

i i i

i

r 0 p>

− ≥ = …

< ≤

= ≠∑

,

, ,,

γ σ η γ2

1

1

0 max

l In matrix form

l The matrix A=Tfσ2GRΓDÐ1 is nonnegative and irreducible;let its (real) max modulus eigenvalue be λ. A solution existsiff λ<1 (sufficient condition: A diagonally dominant).

p I GR D hR D−( ) ≥− −Tf σ2 1 1Γ Γ

29-11-2001INFOCOM Department seminar, fall 2001

18Practical approach

l A safety margin is introduced on target SINR

l A new link can be activated if a feasible power (² Pmax)exists such that all active links and the new one canmaintain their SINR targets, with safety margins ³1

l with

ξ β

γ η γ η000

0 0 0 0

00

0 0 0 0

1=+( ) +( )

≥min , ,

g P

R U

g P

R Umax allowed

P g

T P gf k kk

N0 00

02

01

0 0

η σξ γ

+≥

=∑

29-11-2001INFOCOM Department seminar, fall 2001

19Problem statement: BE traffic

l The constraints can be always met since the bit rates canbe arbitrarily small (best effort traffic!)

l An optimization problem can be defined

maximize

under the constraints

H R C P SNR

P x Q x C p p p p p

SNRa

RR P P

i e ii

N

e

ii

ii i i

( , ) ( )

( ) , ( ) log ( ) ( )log ( )

, ,

r p = ⋅ ( )= ( ) = + + − −

= ≥ ≤ ≤

=

∑1

2 21 1 1

0 0γ maax i N= …1, ,

29-11-2001INFOCOM Department seminar, fall 2001

20BE traffic optimization

l For any fixed p, we have H(r,p) ² H(p) for all feasible r with

Th

rou

gh

pu

t

Bit rate; R

R*

mpa/γ

HC P u

u

P g

T P g

P Pe i

i

i ii

i f k kik k i

Ni

N

i( )( (max , ))

max , ,

*

*

,

p =+

≤ ≤

= ≠

= ∑∑ γ

γ η σ2

1

1

0 max

R C(Pe(a/R))

29-11-2001INFOCOM Department seminar, fall 2001

21Practical approach

l H(p) is convex with respect to each variable Pi in [0,Pmax]

l A sub-optimal solution can be found with a step-by-stepapproach: a new link is activated if the net increase of H(p)is positive (rate of the new link minus rate decrease ofdisturbed links). It turns out that the admission rule is (γi=γ)

H H P P P H P P H P P P

H P P P

N N N

NN

( ) ( , , , ) max ( , , , ), ( , , , )

max ( , , , ),

p = … ≤ … … ≤ …

1 2 2 2

0 11 2

0 max

max max maxε

ε ε ε

c T I u g gi f i= σ γ20 0 00max , *

with

c R

I c Ri i

i ii

N

+<

=∑

01

1max

29-11-2001INFOCOM Department seminar, fall 2001

22

Gain Costs

( ) ( ) oMAXmax RH,HmaxH 0=

Admission rules for BE traffic

l Either a new communication is forbidden or, if allowed, itmust transmit at peak power

Derive the sustainable rate (Ro) as afunction of the measured interference andof the transmitting power

Verify if the rate Ro can be sustained witha benefit for the overall throughput

T

R

H R R

I

I c Ri

i

i ii

N

= ++=

∑001

29-11-2001INFOCOM Department seminar, fall 2001

23Simulation model

l Area of 110x110 meters where N terminals are randomlydistributed (with a minimum distance of 1 meter).

l Each terminal manages RB and BE queues

l Backoff Procedure. Once a burst becomes a head of line, anumber n of transmission attempts take place. If all of themare unsuccessful, the burst is discarded.

Burst arrival: Poisson processBurst address: randomBurst size: Pareto distribution

UWB TxRB queues

BE queue

MAC trafficserver

TO-3 TO-2 TO-1

29-11-2001INFOCOM Department seminar, fall 2001

24Individual throughput

0 5 10 15 20 25 300.65

0.7

0.75

0.8

0.85

0.9

0.95

1

Burst arrival rate, λ

no BP

no AC,no BP

Persistent BP

BP (n=5)

Su

ccessfu

l T

ran

sm

issio

n P

rob

ab

ilit

y

Mean burst size = 200,000 byte

B = 80 mean burst size

# of terminals = 50

Only BE traffic BP = Backoff Procedure

29-11-2001INFOCOM Department seminar, fall 2001

25System throughput

50 60 70 80 90 100 110 120 130 140 1500

5

10

15

20

25

no BP

BP (n=5)

PBP

no AC, no BP

Overa

ll a

vera

ge s

usta

ined

rate

(M

bit

/s)

# of terminals

Mean burst size = 200,000 byteλ per terminal = 2.14 burst/s

B = 80 mean burst sizeOnly BE traffic

29-11-2001INFOCOM Department seminar, fall 2001

26Emitted power (BE+QoS traffic)

1 2 3 4 5 6 7 8 9 100.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

CLASS 1:0.125 Mbit/s

CLASS 2:0.25 Mbit/s

CLASS 4: 0.5 Mbit/s

CLASS 3:0.375 Mbit/s

Avera

ge t

x p

ow

er

(mW

)

Pmax = 2 mWλ = 70 burst/s

# of terminals = 50RB traffic fraction = 15%B = 80 mean burst size

Mean burst size = 200,000 byte

SNR margin (dB)

29-11-2001INFOCOM Department seminar, fall 2001

27Fairness vs throughput

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

1

2

3

4

5

6

Probability of a successful burst transmission

Mean burst size = 200,000 byteB = 80 mean burst sizeOverall λ = 150 burst/s

# of terminals = 60Only BE traffic

Averagesuccess

prob.=0.48Overall

averagerate=14

Mbit/s

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 10

2

4

6

8

10

12

14Averagesuccess

prob.=0.70Overall

averagerate=3.68

Mbit/s

Mean burst size = 200,000 byteB = 80 mean burst sizeOverall l = 150 burst/s # of terminals = 60Only BE traffic

Probability of a successful burst transmission

Number of terminals

29-11-2001INFOCOM Department seminar, fall 2001

28Centralized power & capacity

assignment: bibliographic hints

l SINR based power assignment with multicode asynchronousDS-CDMA: Liu et alii [Ô96]

l Evaluation of the capacity of single-code variable spreadingfactor and multicode DS-CDMA systems: Lee et alii [Ô99]

l Fair queueing in packet wireless networks: e.g. Stamoulisand Giannakis [Ô00], Bharghavan et alii [Ô99], Lu et alii [Ô99]

È Weights are defined for each user; the obtained service rate isproportional to the ratio between oneÕs weight and the sum of theweights of the active users (GPS)

È Practical packet-by-packet versions of GPS -> WFQ

È Revisiting fairness concept required in wireless access with variablequality radio channels and possibly intefering links

29-11-2001INFOCOM Department seminar, fall 2001

29Centralized power & capacity

assignment

l Discrete timesetting, demandassignment MAC

l Constraints on linkSINR and maximumallowed peak power

Capacity/Power Assignment

Resource Utilization

Resource Request

time

TI # t +1 TI # t +2TI # t-1 TI # t

Ci(t+2), Pi(t+2)

BASESTATION

MSi

Qi(t+1)

Q1(t-1)QM(t-1)

MSM

QM(t-1)MS1

Q1(t-1)

Capacity/Power Assignment

Resource Utilization

Resource Request

Capacity/Power Assignment

Resource Utilization

Resource Request

time

TI # t +1 TI # t +2TI # t-1 TI # t

Ci(t+2), Pi(t+2)

BASESTATION

MSi

Qi(t+1)

Q1(t-1)QM(t-1)

MSM

QM(t-1)MS1

Q1(t-1)

time

TI # t +1 TI # t +2TI # t-1 TI # t TI # t +1 TI # t +2TI # t-1 TI # t

Ci(t+2), Pi(t+2)

BASESTATION

MSi

Qi(t+1)

Q1(t-1)QM(t-1)

MSM

QM(t-1)MS1

Q1(t-1)

SINRP

P Ci

i i

ij j j jjj i

M

i

=+

=≠

α

β α η1

29-11-2001INFOCOM Department seminar, fall 2001

30Unified approach

l A matrix of traffic demand is defined based on two ÒdimensionsÓ

È Traffic Component (TC)

È Channel Component (CC)

Rq1

Rq2

Rq3

Rq4

Rq5

CCTransmission quality

TC

QA QL

Qu

eu

e l

en

gth

Backlo

gg

ed

packets

ag

e

QA Ð Queue AgeQL Ð Queue Length

Scheduling service priority

Rq5

29-11-2001INFOCOM Department seminar, fall 2001

31TDMA (1/2)

¥ Radio interface model

¥ CC: three regions are defined

Ð R1: SINR=γT with

Ð R2: γth≤ SINR<γT with P=Pmax/CTOT

Ð R3: SINR <γth even with P=Pmax /CTOT

¥ TC: QA (Queue Age) is used, with packet delay quantization ink levels (k=2)

The available capacityis independent of the

allocation

α η γi i T

tot i

M tot i

P

C P P

C C C C

0

1 0

≥≤

+ … + ≤ ≥

max

,

P P CT tot= ≤γ η α0 max

29-11-2001INFOCOM Department seminar, fall 2001

32TDMA (2/2)

l Priority levels:

È Max priority for R1

È Lower priority for R2

È Within R1 and R2 the service priority is defined according to the TC (age)

decreasing α

decre

asin

g a

ge

Request matrix (Ω)

Assigned capacity matrix

region 1 region 2 region 3

0,7363 0,5894 0,3686 0,2611 0,024 0,0077

42 41 16

3194

36 12 37

0,7363 0,5894 0,3686 0,2611 0,024 0,0077

42 94 37 0

31 36 0 0

CTOT=240 RPs

0.4 0.5 0.6 0.7 0.8 0.9 10

10

20

30

40

50

60

70

80

90

100

Offered Load

Me

an

De

lay (

T)

CHAOS+TDMA

CHAOSTDMA

CISTDMA

29-11-2001INFOCOM Department seminar, fall 2001

33A sample path example

100

150

10 20 30 40 50 60 700

50

Assig

ned C

apacity

CHAOS+TDMACISTDMA

10 20 30 40 50 60 700

1

2

3

4

Time (T)

αS

29-11-2001INFOCOM Department seminar, fall 2001

34CDMA (1/2)

¥ Radio interface model

¥ CC: based on path loss

¥ TC: QL (Queue Length) is used

The availablecapacity depends on

the allocationdecision

α

α ηγi i

j j jjj i

M T

i i

P

GP C

C P P

1

10

=≠

∑ +≥

max

C

C G

C

C G

G

Pi

i Ti

M

i M

i

i T i++

+

≤= ≤ ≤∑ γ γ

ηα1 1

0 1maxmax

29-11-2001INFOCOM Department seminar, fall 2001

35CDMA (2/2)

PS Ð Proportional SharingQL Ð Queue Length

decreasing α

decr

easi

ng

requ

ests

Contending Set

Total capacity = 105

8025

2015

108

0.7363 0.5894 0.3686 0.2611 0.0240 0.0077

Rqi = minQi, Clim,i

80 100 120 140 160 180 200 220 240 2600

20

40

60

80

100

120

140

160

180

200

Offered Load (RPs/TI)

Mean D

ela

y (

T)

CHAOSCDMA, QL

CHAOSCDMA, PS

CISCDMA, PS

C

PGi

i

Tlim, min ,=

αη γ

max

0

29-11-2001INFOCOM Department seminar, fall 2001

36System capacity optimization

l Heuristic algorithm in M steps

È At the k-th step a Contending Set CSk is defined including the columnsof Ω up to the k-th one

È Within CSk, capacities are assigned to the MSs according to QLpriority ( row index of the matrix Ω)

È The requests are scanned until the constraint is met or service is done

l The best CS is chosen among the explored M ones

maximize C

C C i M

C

C G

C

C G

G

P

jj

M

i

i

i Ti

M

i M

i

i T i

=

= ≤ ≤

≤ ≤ = …

++

+

1

1 1

0

0 1

1

lim,i

max

, ,

maxγ γ

ηα

29-11-2001INFOCOM Department seminar, fall 2001

37Fairness

TI

No

rma

lize

d A

ssig

ne

d B

an

dw

idth b) QL case

TI

No

rma

lize

d A

ssig

ne

d B

an

dw

idth a) PS case

29-11-2001INFOCOM Department seminar, fall 2001

38Some open research activities

l Definition and complexity evaluation of MAC signallingprocedures; case study of UMTS-TDD

l Channel probing and measurements

l QoS management

l Optimized multihop routing

l Trade-off between fairness and system capacity

l Multi-scope optimization including energetic efficiency anderror recovery

l Effect of mobility, handoff

29-11-2001INFOCOM Department seminar, fall 2001

39Riferimenti

l G.J. Foschini and Z. Miljanic: ÒA simple distributed autonomous power control algorithm and itsconvergence Ó, IEEE Trans. On Vehicular Tech., Vol. 42, no. 4, 1993.

l D. Mitra: ÒAn asynchronous distributed algorithm for power control in cellular radio systemsÓ, Proc.of 4th Winlab Workshop on Third Generation Wireless Information Networks, Rutgers Univ., 1993.

l N. Bambos, S.C. Chen and G.J. Pottie: ÒRadio link admission algorithms for wireless networks withpower control and active link quality protectionÓ, Proc. of IEEE INFOCOMÕ95, Boston, MA, USA, 1995.

l N. Bambos, ÒToward power-Sensitive Network Architectures in Wireless Communications: Concepts,Issues, and Design AspectsÓ, IEEE Personal Communications, vol. 5, no. 3, pp. 50-59, June 1998.

l V. Shah, N.B. Mandayam, and D. J. Goodman, ÒPower Control for Wireless Data Based on Utility and Pricing,Ó9th IEEE PIMRC, Sept. 1998, vol. 3, pp. 1427Ð32.

l David Goodman and Narayan Mandayam, ÒPower Control for Wireless DataÓ, IEEE PersonalCommunications , April 2000

l A. Baiocchi, F. Capriotti, F. Cuomo, C. Martello: ÒDistributed Radio Resource sharing with UWBÓ, Proc. of ISTMobile Communications Summit 2001, Barcelona, Spain, 9-12 September 2001

l A. Stamoulis and G.B. Giannakis: ÒPacket fair queueing scheduling based on multirate multipath-trnsparent CDMA for wireless networksÓ, Proc. Of INFOCOMÕ00, Tel Aviv, Israel, March 26-30 2000,Tel-Aviv, Israel, pp. 1067-1076.

l Bharghavan V., Lu S., Nandagopal T., Fair Queuing in Wireless Networks, Issues and Approaches, IEEEPersonal Communications, vol. 6, no. 1, February 1999, pp. 44-53

l Lu S., Bharghavan V., Srikant R., Fair Scheduling in Wireless Packet Networks, IEEE/ACM Trans. onNetworking, vol. 7, no. 4, August 1999, pp. 473-489.

l Lee S. J., Lee H. W., Sung D. K.: ÒCapacity of Single-Code and Multicode DS-CDMA System AccommodatingMulticlass ServicesÓ. IEEE Transactions on Vehicular Technology, vol. 48, no. 7, March 1999, pp. 376-383.

29-11-2001INFOCOM Department seminar, fall 2001

40Altri riferimenti

l Bray, J, Sturman C.F. "Bluetooth, Connect without Cables", Prentice Hall, 2001l Chan M. C., Woo T. Y. C.: ÒNext-Generation Wireless Data Services: Architecture and ExperienceÓ,

IEEE Personal Comm., February 1999, pp. 20-33..l Geng-Sheng Kuo, Po-Chan Ko and Min-Lian Kuo: "A Probabilistic Resource Estimation and Semi-

Reservation Scheme for Flow-Oriented Multimedia Wireless Networks", IEEE CommunicationsMagazine, febbraio 2001, pp. 135-141

l Ljupco Jorguseski, Erik. Fledderus, John Farserotu, Ramjee Prasad: "Radio Resource Allocation inThird-Generation Mobile Communication System", IEEE Communications Magazine, febbario 2001,pp. 117-123.

l Mattew Andrews, Krishan Kumaran, Kavita Ramanan, Alexander Stolyar, and Phil Whiting,, RajivVijayakumar : "Providing Quality of Service over a Shared Wireless Link", IEEE CommunicationsMagazine, febbraio 2001, pp.150-154.

l Wattenhofer , L. Li, P. Bahl and Y. Wang, ÒDistributed Topology Control for Power Efficient Operationin Multi-hop Wireless Ad Hoc NetworksÓ, Proc. INFOCOM 2001, pp. 1388-1397, 2001

l Mikls G., Molnr S., Fair Allocation of Elastic Traffic for a Wireless Base Station, in: Proc. IEEE GlobecomÔ99, December 5-9 1999, Rio de Janeiro, Brazil, pp. 1673-1678.

l Nandagopal T., Kim T. E., Gao X., Bharghavan V., Achieving MAC Layer Fairness in Wireless PacketNetworks, in: Proc. ACM MOBICOM 2000, August 6-11 2000, Boston, Massachussets, pp. 87-98.

l Bhagwat P., Bhattacharya P., Krishna A., Tripathi S.K., Enhancing throughput over wireless LANs usingchannel state dependent packet scheduling, in: Proc. IEEE INFOCOM Ô96, April 2-6 1996, San Francisco,California, pp. 1133-1140.

l Fragouli C., Sivaraman V., Srivastava M.B., Controlled multimedia wireless link sharing via enhanced class-based queuing with channel state dependent packet scheduling, in: Proc. INFOCOM Ô98, April 1998, SanFrancisco, California, pp. 572-580.