Title: October 16, 2003
1Lecture 6
2 -
-
- Last Lecture
- slotted Aloha for simple collision channels
- Collision Channels
- If two or more packets are simultaneously
transmitted, they will collide and they are all
lost. - Better signal processing at the physical layer
permits a better packet capture for the channel. - MPR Multi-Packet Reception Capability
-
3- Example of MPR
- Beamforming - Limits the interference
(collision) zone - CDMA
- Multiuser detection
- Coding - dirty paper coding
- How to combine Aloha with beamforming?
- Earlier work Zander,1990 has analyzed the
performance of a - joint beamforming and Aloha protocol,which
uses - - directional antennas for transmission
- - omni-dir antennas for reception
4System Model
- Large random network
-
- Generated by a two-dimensional Poisson Process,
- with average number of terminals/unit area
- The maximum range of a node r 1
- Expected number of nodes within this range is M
- M a measure of the connectivity in the network
- Assumption
- Network in high load state.
- All nodes are backlogged and have packets to
transmit.
5System Model cont.
Transmit a packet to a distant range - use
multiple hops
P
r
f
Final destination
O
?
? forward projector of the packet toward its
final destination
6- Slotted Aloha
- - Transmission probability in subsequent slots
is p -
-
- Beamforming Transmission Antenna
- - Horizontal beamwidth is
area
G is horizontal antenna gain
- Path loss propagation model
- propagation exponent is
- model
- Target SIR for correct reception is
7System Model - Performance measure
- Normalized expected forward projector
-
- - Expected projector in meters
(distance units) - Z dimensionless measure
- Z expected projector in the number of
hopped-over terminals - (average distance between two nearest terminals
) - successful transmission only node O transmits
- node P does not transmit
- no node in collision region
transmit
8gt
For dense networks gt Collision region
for this case close to a circle with radius
gt
Maximization, with respect to M, then p
Note on collision region determined as the
region of interferers for which SIR at P
?0
9gt
Optimum choices for M, p
- Even for moderate gains, it results
very good improvement Ex. G30,
beamwidth gt
10-fold improvement
10Multi-Packet Reception
- The concept of multi-packet reception is more
general and - include other particular areas as well
- CDMA, Multi-user detection etc.
- The multi-packet reception capability of a node
can be described - by the receiver MPR matrix.
-
- MPR matrix contains the reception
probabilities
P k packets are received n packets are
transmitted in the neighborhood
MPR matrix
11Multi-Packet Reception cont.
- MPR matrix include the effects of -
Noise - Type of modulation - signal processing
(beamforming,CDMA, MUD,etc.) - The MPR matrix
of conventional collision channels
12Slotted Aloha for Ad-hoc network with MPR
capability
Slotted Aloha for multi-hop networks is harder
to analyze, but there has been work to
characterize the performance for regular
networks. Note MPR matrix for node
MPR matrix for network Slotted Aloha for Regular
Network - Manhattan Network - All nodes are
located on a regular grid
13- Performance of slotted Aloha
- in Manhattan networks with MPR capability
Assumption - Every node in the network is
backlogged and the probability of
transmission is p - Traffic in the network
is uniform - Average path length
hops The max achievable end-to-end throughput
for slotted Aloha is
- Expected number of correctly received
packets given that k packets are
transmitted
- MPR matrix for the network
It is usually not trivial to decide the MPR
matrix,especially in Ad-hoc networks. (usually
the network MPR matrix node MPR matrix)
14Instability in Slotted Aloha
- Observations
- Slotted Aloha very simple scheme requires very
little overhead and coordination. - (only slot synchronization required)
- One problem Instability
- If the backlog is large beyond unstable
limit it tends to become larger and larger. - Solution for stabilizing Aloha
- - Change transmission probability ,
according to the estimated backlog. - the number of backlogged nodes
- - assume all arrivals are immediately
backlogged. - the estimated number of backlogged
nodes
Pseudo-Bayesian Algorithm
Assumption number of nodes attempting
transmission Poisson with rate
15Instability in Slotted Aloha cont.
How to estimate ? - use the feedback
information on collision status.
For perfect estimation of
It can be shown to be stable for
Pseudo-Bayesian Algorithm
16 Alternate solution for the instability - TDMA
scheduling users.
User1 User2 User3 User4 User5
- Aloha achieves lower delays when arrival rates
are low - TDMA very large delays with large number of
users - Delay for Aloha independent of the number of
users
- TDMA part of the reservation access schemes
family. - Other conflict resolution schemes
have been proposed and analyzed in the
literature. - Another multiple access technique
which we will discuss shortly is CDMA.
17Power Control
- In general, we decide if a packet can be decoded
correctly based on its received SIR. - - Improving SIR improves the chances of
correct reception. - - One important technique is power control.
- Power Control
- Select your power level that you exactly meet
your target SIR -
battery drain
- - If SIR gt , use too much power
- interference with others
- - If SIR lt , packets cannot be received
correctly
18Power Control cont.
- Assume that Q transmitters use the same channel
- They have power
- The expression for SIR at receiver is
-
the power at the transmitter
- link gain
- noise power at receiver i
19Power Control cont.
- Transmitter i is supported if
- - target SIR
- gt
power to select if all other powers are kept fixed
- Denote
- In a system has to be hold for all
20Power Control cont.
- both users can be supported
- only user1 can be supported
- only user2 can be supported
- none can be supported
21Power Control - Feasibility
- How many users you can support to maximize
capacity,while maintaining SIR requirement? - Feasibility conditions
- For Q users
-
22Power Control Feasibility cont.
- Def. The target SIR is said to
achievable, if there exists a non-negative power
vector so that holds for all . - The target SIR is achievable if the dominant
(largest) eigenvalue of matrix H, ( ) is
less or equal to one.
Is achievable only when noise is zero
Power control feasibility condition
23Power Control SIR Balancing
- Another approach to power control
- - Maximize the minimum SIR in all links SIR
Balancing Problem - Solution
- - Power control is achieved by making every
transmitters received - SIR balanced (equalized), while keeping
the balanced SIR as high - as possible.
- - For the noiseless case , find the
maximum achievable - Denote
The solution (with equality) for the power vector
is the eigenvector corresponding to the
eigenvalue .
24Power Control SIR Balancing cont.
- Potential problems with SIR balancing
- - Implement SIR balancing resulting less
than desired - target SIR for all users.
- -gt All links drop below the threshold and
become useless. - Solution
- - Shut off some users,at least
temporarily,and improve the - connections of the remaining users.
-
- A question
- How to maximize the number of users in the
system? - - not trivial -gt several removal algorithm s
25Power Control Stepwise Removal Algorithm
- Stepwise Removal Algorithm (SRA)
- Step1 Determine
- if , use
- else
-
- Step2 Remove terminal k
-
- After removal,
matrix - With go to Step 1.
- It can be shown that in the optimal power
allocation that maximizes the number of users,
some links are completely shut off and the other
users have SIR balanced links.
26Distributed Power Control (DPC)
- One problem with above discussion
- - need to know the link gain matrix (gains for
all users) - - impractical
- -gt need distributed power control
- Distributed Power Control
- - A distributed iterative power control
algorithm based on local measurements.
- received SIR at iteration n
- transmission power of transmitter i at
iteration n
27Distributed Power Control (DPC) cont.
- DPC requires only local measurements on SIR at
receivers - DPC is the same as the Jacobs relaxation method
in numerical linear algorithm - For these distributed iterative algorithm, the
speed of convergence is very important, because - - The assumption is that the system is static
(link gains do not change) until the power
control algorithm converges.
28Unconstrained Second Order Power Control (USOPC)
- Unconstrained Second Order Power Control (USOPC)
-
- It can be shown that USOPC converges much faster
than DPC.
- non-increasing sequence
29Standard Interference Function
- To support the design of various power control
algorithm,a general framework for proving the
convergence of iterative distributed power
control algorithms was proposed. - -Standard Interference Function
- A power control algorithm iterative
-
-
- I Interference function define the power
vector of the next iteration - Def. Assuming positive receiver noise, an
interference function I is called standard if
it satisfies all non-negative power vectors - 1) Positivity
- 2) Monotonicity
- 3) Scalability
Component wise
30Standard Interference Function cont.
- Proposition If the system is feasible, the
sequence of power vectors from the standard
interference function will converge to the
minimum power solution vector , starting with
any non-negative power vector, and the rate of
convergence is geometric. - Observation on convergence
- - Starting with all zero power vector, the
sequence - is monotonically increasing.
- - If , then the
sequence - is monotonically decreasing.
- - Staring with any initial power vector
, the sequence - converges geometrically to the
fixed point .
31Standard Interference Function cont.
- Observation
- - When approaching capacity,
- power level increases. power
warfare effect - DPC belongs to the class of standard
interference function - USOPC may not meet the first condition
(Positivity) - The standard interference function is a
sufficient condition but not necessary.
Monitor the capacity Admission control If
power warfare, reject the new users