VoIP over a wired link

Transcription

1 VoIP over a wired link 1 1 Wireless Communicaions Phil Fleming Nework Advanced Technology Group Moorola, Inc Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

2 Acknowledgemens Rajesh Pazhyannur and Ivan Vukovic, Moorola Ilkka Norros, VTT, Finland Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

3 Ouline Par 1: VoIP over a wired link Traffic models of VoIP The hockey-sick naure of he queueing delay disribuion Large deviaions analysis ala Norros, e. al. The wo-law approximaion VoIP over wireless easer Par : Friday VoIP over wireless 3GPP EV-DO 3GPP HSDPA/HSUPA 80.16d/e Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

4 Packe Radio Access Nework Public Telephone Nework Inerne Radio Link Wireless Access Nework Backhaul Link Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

5 The Backhaul Link T1 or E1 in almos all curren cellular sysems 1.5 or.0 Mbps, resp. Fiber or xdsl will be rare over he nex five years Up o 0 kilomeers Hundreds of backhaul links per Access Nework in G sysems and possibly housands in 3G sysems. Accouns for up o 60% of recurring operaing cos Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

6 Packeized Voice in Cellular Telephony The vocoder oupus 50 voice frames per second Voice frames are carried by IP packes An 8 Kbps vocoder e.g. EVRC produces a nonempy voice payload every frame. Payloads voice + conrol bis range from 10 o 40 byes every 0 msec. Nework proocol headers add more byes. Vocoder oupu is modeled by a wo-sae Markov Chain. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

7 Queueing Model voice packes Session 1 Session Session N BTS 1.5 Mbps T1 Access Node 0 msec Voice packes Daa packes Voice buffer Daa buffer 1.5 Mbps Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

8 The Proocol Sack PPP variable 3 Byes B 3 B 6 B B B IS-634 HDLC Hdr PPP ID cudp IS 634 HdrVoice Payload CRC CRC cudp muxppp 1 Bye 1 B HDLC/PPP muxppp HDLC Hdr PPPmux ID Ln PPP ID cudppayload Ln cudp Payload Ln cudp Payload CRC Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

9 Aggregaed Voice Traffic Consider N voice sources each wih period τ. The ih source has offse value θ i beween 0 and τ. The number of bis from he j h source in he i h frame is he random variable i,j. The oal number of bis arriving a he buffer in he inerval [0, ] is hen A θ j + iτ i, j Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

10 General arrival process for aggregaed voice raffic wih random offses 1 i1 j 1 i 1 i j i 1 i+1 θ j τ θ j τ Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

11 Arrival Process of Aggregaed Voice Traffic: Offse Groups k 1 G... k i 1 } k G j k i1 } k 1 G k i } k G j } k i k i + 1 k 1 G } jτ/g τ jτ/g τ Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

12 Voice Sources We assume he sochasic processes { i,j, i0,1,, } are I.I.D. wo-sae markov chains. Le λ, σ and ρ k denoe he mean, variance and lag k correlaion of he process i,j. E[ ρ k i, j ] λ, Corr Var[ 0, j Cov Var Since { i,j } is a wo-sae Markov Chain,, k, j i, j ] σ 0, j, 0, j k, j ρ k ρ k Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

13 Simulaion Resuls 157 users over a T1 a MbpS 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E-05 Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

14 Effec of Loading number of sessions 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E E E Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

15 Effec of Correlaion one-sep correlaion shown 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E E E Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

16 Gaussian Traffic Model { Z, > 0 } is a cenered Gaussian process wih saionary incremens iff i is cenered and for any 1,, k and s he random vecor Z Z,, 1 s K Z Z k s has a mulivariae Gaussian disribuion ha is independen of s. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

17 Gaussian Traffic Model Arrival raffic number of bis in 0,] is modeled by A m + Z where m is he mean rae Z is a cenered Gaussian process wih saionary incremens Denoe v Var A E Z Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

18 Queue Lengh The queue lengh process wih server rae c is V sup A s sup Z s A Z s s c c s m s The ail disribuion of he queue lengh process is P[ V > x] P[sup Z Z c m s x] s > s Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

19 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor Queue Lengh Approximaion + + Φ > > > > > * * exp * * ] * ] [ sup ] [ sup ] [sup ] [ * 0 u v m u c x u v m u c x x m u c Z P x m u c Z P x s m c Z Z P x s m c Z Z P x P V u u u s s s s

20 Queue Lengh Approximaion The quaniy u* can be found as he value of ha minimizes x + c m v Ref: Ilkka Norros, Mos probable pah echniques for Gaussian queueing sysems, Nework 00, Pisa, Ialy Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

21 Scaling Facor A scaling facor is required o approximae P[ Q > 0]. For he case of k offse groups, we use Ρ[ Ν N λ / k, σ / k > µ / k ] For random offses use 0.5 lim n Ρ[ Ν Nλ / n, σ / n > µ / n ] Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

22 The Variance of A Define a simplified arrival process, A 0, for a single normalized voice source as A 0 i + I{ u< i 1 } + 1 where E i 0, Var i 1 and Cov i, j ρ i-j For N real voice sources, we have A D Nλ + Nσ A 0 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

23 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor The Variance of A + + < + < i i i i u i i u i i i E i E I I E A E } { 1 1 } { 1 0 ρ

24 The Large Deviaions Approximaion 157 users over Mbps 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E E E-07 LD approx simulaion 1.0E-08 Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

25 Why he hockeysick? 0.00 x 0. x + c m v x 0.4 x Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

26 Inuiion The delay of a packe ha arrives during a busy period ha has lased less han 0 msec will no depend on he correlaion srucure of he arrival process. Shor busy periods are common and lead o shor delays The delay of a packe ha arrives during a busy period ha has lased longer han 0 msec will depend on he correlaion srucure. Long busy periods are somewha rare and lead he longer delays. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

27 Validaion 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E LD 150 LD 155 LD 158 LD 160 LD E-06 Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

28 Heavy Traffic Analysis The sysem evoluion is given by Q Q0 + A c 1 { Q s > 0} 0 ds We consider a scaled version of Q Q n Y n n Assume Y n 0Q0A00 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

29 Heavy Traffic Analysis Consider a sequence of sysems indexed by n in which he mean arrival rae in he nh sysem is Nλ n The scaled queue lengh can hen be wrien ~ A n Yn + n Nλn µ + Λn + O n n 1/ Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

30 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor Heavy Traffic Analysis ˆ sup max0, where ˆ hen we assume If ˆ 1 ~ where ~ 1 1/ s N B N?? N B N Y Y N N n B N B N n n A n O? N n n n A Y s n n k k n n n µ λ σ µ λ σ µ λ µ λ σ ρ σ µ λ So Y is Refleced Brownian Moion

31 Saionary Densiy From his analysis one concludes ha Y is a Refleced Brownian Moion wih Drif and has a saionary densiy, f, given by f y µ Nλ Nσˆ exp y µ Nσ ˆ Nλ The exponen is he same as in he Gaussian LD approximaion for sufficienly large y and for small y. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

32 Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor Waiing Time Approximaion The heavy raffic analysis leads us o an approximaion which is he sum of wo scaled exponenial disribuions. > + Ρ > Ρ > ]exp / /, / [ ˆ ]exp ˆ, [ ] [ σ λ µ µ µ σ λ σ λ µ µ µ σ λ N N k k N k N N N N N W P k N N

33 Validaion 157 users over Mbps 1.0E Prob[ Delay > x ] 1.0E E-0 1.0E E E E E-07 LD approx simulaion wolaw 1.0E-08 Delay msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

34 Validaion 1.0E E Delay msec. Prob[Delay>x] 1.0E-0 1.0E E E-05 4 Offse Groups 8 Offse Groups 16 Offse Groups Uncorrelaed Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

35 Delay in VoIP over 80.16e Forward Link Prob[Delay > x] Delay msec. 00 user/secor 0 user/secor Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

36 Ongoing work Models of framed voice+daa e.g., I. Norris approach Nework sizing ools for sysem engineers Scheduling and admission conrol implicaions Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

37 Impac of he Number of Offse Groups on Waiing Times ProbDelay>x Wireless Communicaions Delay msec,x For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

38 Packeized Voice in Cellular Telephony numerical example: EVRC Le E[ Var[ Corr[ before proocol ] 16 bis ] 74 bis, ] ρ, j Typically, 0.90 ρ 0.99 i i i number bis in he i i i i+ j overhead 0 is a wo -sae markov chain j h voice frame Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

39 Correlaion Srucure of Vocoder Oupu sae markov geomeric.9 rbab3 ramb3 geomeric Lag Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

40 Voice Arrival Process We approximae he aggregae voice process {Z i, i1, } as a runcaed Gaussian process a discree Ornsein- Uhlenbeck process ~ ~ N µ + ρ Z µ, σ 1 ρ i+ 1 i Z Time msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor

41 Traffic Shaping Voice raffic is shaped in 3G sysems by dividing voice flows ino offse groups Noe: raffic wihin an offse group is correlaed raffic beween offse groups is independen offse group Time msec. Wireless Communicaions For he personal use of he paricipans in he IMA summer program on wireless communicaions. May no be reproduced or disribued in whole or in par wihou wrien consen of he auhor