1. DPRMA (Distributed Packet Reservation Multiple Access)
(수)
김 희 준

2. Contents Abstract Introduction Principle of DPRMA
Approximate Performance Analysis
Numerical Examples and Discussion
Conclusion

3. Abstract
Apply MAC scheme like TDMA for mobile ad hoc networks with emphasis on voice application support
Major Effects
simple slot reservation mechanism for voice traffic without relying on any central entity
simple solution for the hidden and exposed terminal problems uniquely present in wireless ad hoc environments
Performance test
investigated by analysis and computer simulations in comparison with IEEE
The results show that D-PRMA is much more suitable than IEEE for voice application

4. Introduction Only focuses on MAC schemes based on channel in time
Existing difficult issue
To design MAC scheme to support real-time applications
No fixed central entities can be used by the MAC layer in MANETs to coordinate communications
High dynamics of network topology caused by terminal mobility
real-time applications have requirements on QoS
the MAC scheme should be simple for implementation because terminals in such networks are portable and battery-operated personal devices
Only focuses on MAC schemes based on channel in time
No condition constantly frequency, frequency hopping

5. Introduction No central entities in mobile ad hoc environments
Aspects of Unslotted MAC scheme
No useful “Jamming” mechanism is Like The MAC in HIPERLAN
Overhead problem that the reservation in MACA/PR is maintained by asking all neighbors to exchange their reservation tables
Aspects of slotted MAC scheme
they can avoid difficulties in synchronization
Apply GPS problem (at providing global clock)
The same effort can also be found in code division multiple access (CDMA)-based third-generation cellular systems

6. Introduction Slotted-channel-based MAC schemes
a successful contention process (FPRP / E-TDMA)
a long access delay for real-time applications if a slot is reserved by a terminal at the “talkspurt” level
In the TDMA/FDD-based scheme
a slot is reserved for a voice terminal until the end of a call
PRMA
a centralized and slotted MAC scheme
providing a mechanism for slot reservation at the “talkspurt” level for voice and data applications
with a base station as the fixed entity for the MAC operation
So, Author discuss a simple extension of PRMA, termed “distributed PRMA” (D-PRMA)
with emphasis on voice application support in mobile ad hoc environments

7. PRINCIPLE OF D-PRMA PRINCIPLE OF D-PRMA Slot Reservation Scheme
Solution for the Hidden and Exposed Terminal Problems

8. PRINCIPLE OF D-PRMA Notations D-PRMA characteristics
N: Total number of terminals in the system.
F: Frame length in time units.
O : Number of slots in one frame.
m : Number of minislots in the payload of a slot.
p : Contention probability.
D-PRMA characteristics
TDMA-based scheme
uniformly attaches such fields to each slot
tries to simplify the solution for the hidden and exposed terminal problems
To facilitate a terminal to locate its reserved slot in the subsequent frames
improve channel utilization( used several minislot for contention)

9. PRINCIPLE OF D-PRMA Minislot
RTS/BI and CTS/BI
used by a terminal to both reserve a slot and prevent hidden terminals from colliding with transmission in the respective slots
if a terminal wins the contention through the first minislot of a slot
the extra minislots of this slot will be granted to the terminal as the payload
the same slot in each subsequent frame can be reserved

10. Slot Reservation Scheme
Reservation process is similar to the RTS/CTS used in IEEE
sender detects the channel idle at the beginning of a minislot
some part of RTS/BI of each minislot is dedicated to channel sensing

11. Slot Reservation Scheme
Guarantee voice traffic
Define rule to prioritize voice terminals
voice terminals start the contention from minislot 0 with probability p=1 (data terminals p < 1)
Give same probability(p<1) through m extra minislots contention
to avoid unnecessary slot reservation
the winner of a voice terminal can reserve the same slot in each subsequent frame until the end of the packet transmission
data terminal can only use one slot

12. Solution for the Hidden and Exposed Terminal Problems
consider the following two cases
When a terminal wins the contention in minislot 0, how to prevent other terminals from using any of the extra minislots in the same slot for contention?
How to prevent a terminal from contending for a reserved slot in each subsequent frame?

13. Solution for the Hidden and Exposed Terminal Problems
For case 1
use of RTS/CTS-like dialogue a part of solution
MACA consider for the same problems
a winner through minislot 0 will transmit immediately from minislot 1 of the same slot
the neighbors of the sender will detect a busy channel before trying to send an RTS
CTS/BI can be used
a terminal that receives RTS destined to it to transmit the respective CTS
all terminals hearing the CTS sent by the receiver
not allowed to transmit during the remaining period of the same slot to avoid the hidden terminal problem
Still transmit to avoid the exposed terminal problem
other terminals only hearing the RTS but not the CTS

14. Solution for the Hidden and Exposed Terminal Problems
For case 1
to avoid the exposed terminal problem
duplex communication, where a sender may also be a receiver simultaneously and vice versa
the transmission of the sender’s neighbors should not be allowed either
a terminal hearing the RTS but not the CTS
not transmit anything during the remaining period of the same slot to avoid collision with the sender’s receiving
If either the RTS and/or the CTS collide
the extra m minislots in the same slot can be still used for contention

15. Solution for the Hidden and Exposed Terminal Problems
For case 2
define that the receiver of a reserved slot will send a busy indication (BI) immediately
through the RTS/BI of minislot 0 of the same slot in each subsequent frame without channel sensing, and so will the sender through the CTS/BI
Letting the receiver transmit a BI signal first
also tries to avoid the hidden terminal problem
since not every neighbor of the receiver can hear from the sender while all neighbors of the sender can hear from the sender
A terminal hearing a BI signal
not contend for the slot in the current frame

16. Approximate Performance Analysis
Voice Traffic Model
Analysis of System State Distribution
Calculation of pi,j
Calculation of pdrop

17. Approximate Performance Analysis
analyze the performance in an one-hop environment where all terminals can hear each other
About Voice terminal
only voice terminals can start contention from minislot 0
the bandwidth to be used by data terminals is mainly that which is not being used by voice terminals
Voice packet dropping probability (pdrop )
voice packet will be dropped if it is queued beyond a threshold
Generally, should be less than 10-2 for an acceptable voice communication quality
Leftover bandwidth for data traffic(Lband)
left over by voice terminals can be used for data applications

18. Voice Traffic Model Voice terminal
generates a pattern of talkspurt and silence periods as classified by its voice activity detector
A terminal’s vocoder digitizes talkspurts into packets and suppresses silence periods
digitized packets have a fixed length
Model for voice traffic described Markov process
exponentially distributed talkspurt periods / silence periods

19. Voice Traffic Model Eqn. about periods Author said
Talkspurt periods ends within τ period
t1 = length of talkspurt
slience periods ends within τ period
t2 = length of silence
Author said
Applying Two equations, can calculate p0 and p1 by setting τ equal to F

20. Analysis of System State Distribution
durations of the talkspurt and silence periods are much longer than the length of a frame
assume that terminal state transitions between “talkspurt” and “silence” occur only at a frame’s boundary
Variable are defined to characterize system states observed at beginning and end of a frame
R(R-) : Number of terminals in “reservation” state
C(C-) : Number of terminals in “contention” state
S(S-) : Number of terminals in “silence” state
The system state

21. Analysis of System State Distribution
Finite state space
Modeled as as Markov process {Zi}
probability of the system in steady-state Zi
is the dimension of the system state space
Denote k as number of terminals with reservation
Its maximum is min(N,O)
Maximum number of contending terminals is N-k
N-k+1 is the maximum number of system states with respect to the number of contending terminals (+1 for zero contending terminal state)
where
and

22. Analysis of System State Distribution
Let pi,j the probability for a transition from system states to zi to zj
denote the one step transition probability
then, can have form with
and
Expectation Value

23. Calculation of pi,j
the “talkspurt–silence” transition is independent of the contention process
a contention process in the frame
numbers of terminals in the “reservation” state ( ) and in the “contention” state ( ) at the end of that frame
and
where sc,r is the number of terminals that have successfully made reservations in the frame
numbers of terminals in the “reservation” state ( ) and in the “contention” state ( ) at the beginning of the next frame
where dc, dr, ds denote numbers of terminals that have departed from states “contention” and “reservation” as well as “silence” at the frame’s boundary
Thus, rj and cj for state Zj that system at the beginning of the next frame

24. Calculation of pi,j
pi,j and can be calculated from the distributions of Dr, Ds,Dc and Sc,r
Where are random variable dr, ds, dc and sc,r
Let
Where x is the number of terminals successfully making reservation in a frame in the case of e free slot and c contending terminals available at the beginning of that frame aa
T(c) the probability for a successful transmission of RTS/CTS in an available slot
c contending terminals available at the beginning of that slot
Q(c) denote the probability of a successful transmission of CTS among c contending terminals with probability p through one of the m extra minislots of a free slot

25. Calculation of pi,j dd d The one-step state transmission probability
Where c=ci and e=O-ri for state Zi d
The one-step state transmission probability

26. Calculation of pdrop d The average number of packets dropped per frame
Where S’c,r is the random variable for the number of terminals
Terminals that obtain reservations in minislot 0 of a frame
computed as the ratio of the average number of voice packets dropped in a frame to the average number of packets generated per frame
In the design, the frame length can be set to the queuing delay threshold for voice packets
The average number of packets dropped per frame
equal to the average number of contending terminals
at the beginning of a frame minus
the number of terminals that obtain reservation through minislot 0 in a frame

27. Calculation of Lband A slot can be used
by data terminals if and only if no voice terminal has reserved or contend for this slot
E[Sa] is the average number of free slots
Slots available for contention in a frame
E[Svc] is the average number of voice terminals
Terminals start their contentions per frame

28. Numerical Examples and Discussion
a voice terminal in “talkspurt” generates exactly one packet per frame and each payload of a slot carries one packet, the above parameters should satisfy
rs = source rate of voice traffic
h = physical and MAC layer headers for each digitized packet
rs x F = amount of source information per packet
rs x F + h = total packet length
To, Tg = durations of minislot 0 and a slot guardband
rc = channel transmission rate

29. Numerical Examples and Discussion
Rx/Tx and Processing overhead set to zero
Nv,Nd = The number of voice terminals and data terminals are denoted
Pv, Pd = the contending probability of voice and data terminals are denoted

30. Analytical Results Versus Simulation Results

31. Performance of D-PRMA
average delay experienced by voice packets with D-PRMA
All Delay is shorter than a frame duration of 16ms
-> longer than 16 ms are dropped by the MAC layer
Pv has almost no effect on delay but Nv
-> lets voice terminals start contention from minislot 0 of
a free slot and the probability of such a successful
contention is high with the configuration given by
Table I
Author can get the probability of such unsuccessful
contention Pf,0(Nv)
-> Calculate value is Pf,0(10)= and
Pf,0(20)=
* Nv is larger, then Pf,0(Nv) is low

32. Performance of D-PRMA Figure 3 Figure 4
Pdrop generally increases with Nv
A little different results for Pv increases ( )
Pdrop generally increases with Nv
big different results for Pv increases ( )
The reason is probabilistic contention in the m extra minislot is
not so important as that in minislot 0 with prabability 1
If collision in minislot 0 occurs, Pv becomes important in determining the success of contention in the m extra minislots

33. Performance of D-PRMA Figure 5 Figure 6
Lband generally decreases with Nv
A little different results for Pv increases ( )
Lband generally decreases with Nv
big different results for Pv increases ( )

34. Performance of D-PRMA Tendency of Pdrop and Lband versus Pv
Figure 7
Figure 8
Tendency of Pdrop and Lband versus Pv
Pv around 0.5 is suitable for most case of Nv
The following simulation, Pv is set to 0.6

35. Comparison With IEEE
Simulations with OPNET (between D-PRMA and IEEE )
Data packets arrive at each data terminal according to a Poisson process with mean arrival rate

36. The channel efficiency
Ld (data traffic load)
Delay of voice packets
Nv =16 and Nv=24 (IEEE )
Nv =24 (D-PRMA)
Value over 10-2
Nv Should be controlled under about 15 to have Pdrop < 10-2
Data traffic support
IEEE performs better
Nv =16 and Nv =24 (D-PRMA)
Especially longer
The channel efficiency
the ratio of the time used to transmit user packets to a given time period
D-PRMA is lower than that given by IEEE especially in the case of high data traffic load

37. Conclusion
Pros
D-PRMA more suitable than IEEE for voice traffic while the latter is better for data traffic
Cons
deficiency of D-PRMA for data traffic
can be resolved by introducing a piggyback reservation scheme for data traffic
a proper call admission scheme to control the number of data and voice terminals
requires a voice terminal to contend for every talkspurt for high channel utilization
The channel efficiency of D-PRMA
further improved by maximizing the use of minislots for packet transmission