Cristina Comaniciu and Narayan Mandayam

W INLAB

Rutgers University

DS-CDMA, bandwidth W

Voice rates R v

Data rates Rd

Intolerant to delays

Tolerant to moderate delays

Slotted reverse link

Can tolerate occasional errors

Error intolerant

Normalized Load

time (packet transmission scale)

residual capacity

ON/OFF voice activity model

Normalized Load

time (call arrival scale)

Resources used by voice users

Resources available for data users1 1

ll CDMA Systems are interference limited

1Outage Probability %1≤

Minimum data ratethroughput for data users

Average data ratethroughput for data users

System capacitymaximization

n Imperfections in estimating theresidual capacity

n Imperfections in scheduling thedesired number of data users

n Imperfections in power control

Goal: Elim inate first listed cause foroutage!

n Define outage condition (maps FERrequirements)

n Derive ideal residual capacity expression

n Obtain an estimate for the number of data userswhich are allowed to transmit in the next time slot

n Employ an access procedure to allow the exactpredeterm ined number of data users to transmitin the next time slot

Perfect Power Control - Power Control Feasibility condition:

S(n) = normalized load in the nth time slot;

v(n) and d(n)= # of active voice and data users;

γ i = target SIR for voice/data;

Outage: when

( ) ( ) ( )d−≤+= 1

dv and

anv

nS

1+=

iii R

Wa

g 001.0=d

( ) 1>nS

( ) ( )( )1ˆ −= nvfuncnd

n d(n) = max # of data users in the system s.t.

n Problem: v(n) - random variable d(n) - randomvariable only an estimatecan be derived

n - integer number only data users

can be scheduled for the next time slot

d−≤ 1)(nS

⇒⇒

⇒)(ˆ nd ( ) nd̂

( ) ( ) ( )nvaa

andv

dd −−= d1

NormalizedLoad

residual capacity

1

n Accuracy of approximation:

– sampling frequency fS

– step of modulation ∆∆

n Two types of distortions:

– granular noise

– slope overload distortion

Condition for no slope

overload distortion:

sfts •∆≤•

)(

( )nsns ≤)(ˆ

Steps of MDM algorithm: - first time slot: measure number of active voice usersand initialize

- for : if , then set

if , then set

if , then set

( ) ( )11ˆtt dd =

1>n ( ) ( )11ˆ −>− ndnd tt ( ) ( ) ∆−−= 1ˆˆ ndnd tt

( ) ( )11ˆ −<− ndnd tt ( ) ( ) ∆+−= 1ˆˆ ndnd tt

( ) ( )11ˆ −=− ndnd tt ( ) ( )1ˆˆ −= ndnd tt

Introduce guard margin = step of the modulation (∆)− the function to be approximated becomes:

( ) ( ) ∆−= ndndt

n no slope overload condition

n Max. # of data users allowable in the system a t each timeslot

n cumulative voice activity process - discrete Markov chain(the slot duration very small such that at most change inthe number of active voice users can occur)

n choose

( ) ( ) ( )nvaa

andv

dd −−= d1

1±

v

d

aa

=∆

sfts •∆≤•

)(

n MDM-S (Modified Delta Modulation with Scheduled Access)

– exactly users are scheduled to transm it in thenext time slot in a round-robin fashion

– no outage due to the access control

n MDM-R (Modified Delta Modulation w ith Random Access)

– access probability (ML estimate) - max. prob. thatexactly users will transm it in the next time s lot

– a probability tolerance value is defined to allow the finetuning of system performance

( ) nd̂

( ) nd̂

n Fixed # of adm itted data calls (Kd):

n SMS: Poisson arrival for data with rate :

– rate of adm itted data traffic:

( )( )[ ]

d

p

K

tndnp

−=

ˆ

( )( )[ ]

r

ptndnp

−=

ˆ

( ) ( ) ( )[ ] pa tndnpn −== ˆrr

r

Outage Condition: limited interference

Z(n) = load; and voice and data activity indicators

, = SIRs for voice/data;

= maximum allowed received PSD to background noise PSD

( ) ( ) ( ){ } ( ) ( )( )

01.0var

1,|1)(),(| ≤

−−=−>=

ZZE

QndnvZPP ndnvouth

h

( ) ( ) ( ) ( )hezen −≤+= ∑∑==

1 1

1

00

di

K

ii

d

vi

K

ii

v

dv

nG

nG

nZ

vie d

ie

1

h

( )nin ( )niz

n Two access control schemes are proposed and theperformance is compared with earlier work (W INLAB TR -127, TR-143)

u Both schemes give better performance

u MDM-R - very efficient implementation

u MDM-S - guarantees zero outage for the perfect powercontrol case

n Two data models are considered in the analysis:

u Non-transparent data service (ftp, mail, store andforward facsim ile) - W INLAB TR- 172

u Short Message Service (Poisson arrivals for data; onlyopen loop power control for data) - W INLAB TR inpreparation