Electrical Engineering and ComputerScience
Cross-Layer Designand Analysis of
Wireless NetworksWayne Stark
Achilleas Anastasopoulos, Shihyu Chang, Hua Wang
University of Michigan
Electrical Engineering and ComputerScience
Outline
• Introduction• Network and Physical Layer Design• MAC and Physical Layer Design
Electrical Engineering and ComputerScience
Layered Approach
Presentation Layer
Session Layer
Transport Layer
Network Layer
Data Link Layer
Physical Layer
Application Layer
Electrical Engineering and ComputerScience
Why cross-layer design?
• Significant performance advantages(e.g. 10 dB in certain situations).
• Forces designers to consider otherlayers.
• Layers are coupled.
Electrical Engineering and ComputerScience
What causes coupling?
• Energy constraints.• Delay constraints.• …
Electrical Engineering and ComputerScience
Why not cross-layer design?
• Difficulty.• Lack of insight into design.• Generally requires near brute-force
simulation/optimization if several layersare considered simultaneously.
Electrical Engineering and ComputerScience
Two cross layer problems• Problem 1: Network routing algorithm: For
fixed total energy maximize the normalizedthroughput between source and destinationwhile accounting for amplifier characteristics,physical layer performance and processingenergy at receiver.
• Problem 2: Determine the tradeoff betweenenergy and delay in wireless networks takinginto account the MAC and physical layers.
Electrical Engineering and ComputerScience
Simplified Network Model
S N1 N2 N3 N4 N5 N6 N7 DS N1 N2 N3 N4 N5 N6 DS N1 N2 N3 N4 N5 DS N1 N2 N3 N4 DS D
d
S N1 D
x
S N1 N2 N3 D
xS N1 N2 D
x
Electrical Engineering and ComputerScience
Assumptions/Notation• Total energy available for all the nodes in the
linear network =B (joules).• Independent errors at different nodes.• Energy Er for processing each packet at a
receiver.• Number of hops=k.• Packet duration =Tp.• Code rate =R (bits/channel use).• PDC=f(Pin).
Electrical Engineering and ComputerScience
Performance Measure• Expected number of successfully
received bits per unit bandwidth andtime and energy.
Energy used per hop(transmission and reception)
Number of packets that can be transmittedend-to-end.
Number of HopsPacketSuccessProbability
Electrical Engineering and ComputerScience
Main Result (large d)
• Functional form of throughput independent of– Error Control Coding Scheme– Modulation– Channel (Fading, Propagation Characteristics)– Amplifier Characteristics
• Specific constant δ depends on all of the above.
Electrical Engineering and ComputerScience
Conclusion: First Problem• Optimum rate in AWGN close to 1.• Uncoded better than rate 1/2 coded at optimum
distance but requires higher density of nodes.• Amplifier operating point is not an extreme point of
amplifier characteristics.• For other channels (e.g. faded channels) optimum
rate will likely decrease.• This problem encompasses physical layer and
network layer issues.
Electrical Engineering and ComputerScience
Second Problem
• Determine the tradeoff between energyand delay in wireless networks takinginto account the MAC and physicallayers.
Electrical Engineering and ComputerScience
ARQ Protocol
Tx Rx
Data: K Parity: N-K
ACK/NACK
error-free channel
noisy channel
Electrical Engineering and ComputerScience
Goal• For a fixed number of information bits,
K, determine the optimal number ofcoded bits, N, to minimize the delay.
• Note: The N that minimizes the delayalso minimizes the energy.
Electrical Engineering and ComputerScience
Packet Error Probability Bounds
For an additive white Gaussian noise channel
Electrical Engineering and ComputerScience
Notes• Turbo codes and LDPC codes
can achieve better than thecutoff rate.
• Convolutional codes are far fromcutoff rate for large block length.
• Reed-Solomon codes have nearexponential dependence on N
Electrical Engineering and ComputerScience
Main ResultFor large K (compared to 1) at theoptimum packet length (N*) the resultingerror probability is a constant.
Electrical Engineering and ComputerScience
Comments on Result• This result is independent of the channel model
and modulation technique (e.g. coherent,noncoherent, faded) except that the channel ismemoryless.
• The resulting minimum average energy anddelay depend on the above characteristics.
• Result implies that larger payloads (K) shouldtry to achieve a smaller error probability.
Electrical Engineering and ComputerScience
Extension to Include MAC Layer
A B C
Node A wants to transmit a message to Node B. Node Cwants to transmit a message to Node D. Withoutcoordination Node C’s signal will interfere with A’stransmission to Node B. Node C might start it’s transmissionafter A has already begun transmitting because C can nothear Node A’s signal. This is the hidden node problem.
D
Electrical Engineering and ComputerScience
RTS/CTS Mechanism• A transmits to B an RTS (request-to-
send) packet.• If B successfully decodes the RTS
packet then B transmits a CTS (clear-to-send) packet indicating the upcomingtransmission of data from A to B.
• Both A and C hear the CTS and now Aknows that it is clear to send a packet toB.
Electrical Engineering and ComputerScience
Problem• Determine the delay vs. energy for
different number of users.• For fixed data length, RTS, CTS,ACK
lengths determine optimal packet sizesNDATA, NRTS, NCTS, NACK.
• Similar approximations for large K canbe obtained for optimum Pe,RTS, Pe,CTS,Pe,ACK
Electrical Engineering and ComputerScience
Result• We have developed an analytical framework to
evaluate the joint distribution of energy anddelay of the RTS/CTS protocol in a noisychannel.
• Similar approximations to single user case.• Assumptions
– All n users have packets ready (heavy-loadassumption).
– All users can hear all other users.– Memoryless channel.– No multiuser reception/detection capability.– Only transmit energy is considered.
Electrical Engineering and ComputerScience
Interpretation
For short packets the fractional overheadto access the channel becomes larger.
Electrical Engineering and ComputerScience
“Basic Protocol”
• Eliminate RTS/CTS• Listen before send.• If collision of data packet then wait a
random (exponential) backoff timebefore retransmission.
Electrical Engineering and ComputerScience
Comparison with “Basic Protocol”
RTS/CTS Better
Basic Better Basic
Better
RTS/CTS Better
Electrical Engineering and ComputerScience
Interpretation of Results• For a larger number of users there is a lower
threshold for switching between the basicprotocol and RTS/CTS protocol
• For larger energy per coded bit, the transmissionrate becomes larger. The larger rate implies ashorter time to transmit a given number of bits.A shorter duration for transmission of the datapacket increases the relative burden needed totransmit the RTS/CTS packets. So the thresholdof packet length where RTS/CTS is betterbecomes larger.
Electrical Engineering and ComputerScience
Conclusion• Have shown certain invariants (optimum
distance, optimum error probability).• By considering a couple layers joint
design/optimization and analysis is possible.• Insight into performance analysis can be
obtained.• Still need to consider many other factors
(power control, data rate control, multiple-access capability of modulation and coding).
• There are many open and interestingproblems in cross-layer design.