Resource Allocation for E-healthcare Applications

1

Resource Allocation for E-healthcare Applications

Qinghua Shen

content

• Intro: e-healthcare system • Research issues• Preliminary results• conclusion

2

3

Intro: e-healthcare system

Wban: Remote monitoring

Wbans: Hospital information collection

Computing: Medical information processing

Body channel

Limited sensor energy

Mobilitydistribut

ed

Emergency traffic

support

Randomness of the

requests

content

• Intro: e-healthcare system • Research issues• Preliminary results• conclusion

4

5

Single sensor WBAN scheduling

• single sensor application Network model

one PDA and one sensor with Pmax

Time is partitioned into slots with length Ta pilot of duration αT required for transmission.Two decisions made by the sensor at each time slot

• sleep decision s(i) • Transmission decision b(i)

Traffic and Channel Model A(i): a maximum Amax and Dmax

h(i) : pathloss in power, bounded by minimum hmin and maximum hmax

i.i.d, stationary and ergodic

Energy Cost Model

Queue Update

Listening Transmission

6

Power vs. Delay trade-off

• Energy Efficient Approaches

• Delay requirements

Opportunistic Transmission exploiting channel dynamics

Sleep Scheduling Originate from sensor networks, reduce idle listening

Worst case delay Guarantee Dmax deterministic delay requirement

Average sense delay little’s law

7

Power vs. Delay trade-off

• Relationship between Energy and Delay single link Power-Rate relationship

Shannon capacity formulation A practical approximation --monomial function The average power consumption

Service rate delay relationship Queue: service process bµ(n) and the arrival rate A(n), service process is

determined by transmission policy Q(n)=Q(n-1)+A(n) - bµ(n)

Queue of a system is related to the delay • Average Delay• Worst Case Delay Qmax doesn’t guarantee a Dmax

8

Power vs. Delay trade-off

• Problem Formulation Power vs. Delay I (average sense delay [1]) Optimization Objective for V>0, the goal is to find the policy µ to minimize

Define the minimal average power can be achieved as the power needed to serve average arrival rate with no delay consideration, denoted by , it’s the solution of the following problem with a policy Ψ(H) .

minimize EP(H, Ψ(H))

subject to: E (Ψ(H)) A The policy for no delay consideration doesn’t need to take current queue state into decision making.

lower bound is proofed[1] and a drift policy achieves it

[1] R. Berry and R. Gallager, “Communication over fading channels with delay constraints,” IEEE Trans. Information Theory, vol. 48, no. 5, pp. 1135–1149, 2002.

9

Power vs. Delay trade-off

• Problem Formulation Power vs. Delay II (Worst Case Delay ) BT problem: B unit of traffic needed to transmitted by the deadline T

Continuous case, Markov Channel, monomial power rate function [2] formulation and solution• system updating equation• cost function and cost-to-go function• solve the Hamilton–Jacobi–Bellman equation backwards to obtain the

optimal control policy

Discrete case, i.i.d channel [3]• monomial: optimal policy• Shannon: no closed form

scheduling policy characteristics• More opportunistically when deadline is far away• less opportunistically when queue length is large

Transmission policy

[2] Murtaza Zafer and Eytan Modiano, Optimal Rate Control for Delay-Constrained Data Transmission over a Wireless Channel. IEEE Transactions on information theory, Vol. 54, No. 9, Sept. 2008.[3] J. Lee and N. Jindal, “Energy-efficient scheduling of delay constrained traffic over fading channels,” IEEE Trans. Wireless Communications, vol. 8, no. 4, pp. 1866–1875, 2009.

10

Single sensor WBAN scheduling

• Problem formulation 1) Lyapunov optimization theory[4] adopted

why not DP: • Curse of dimensionality

characteristics of Lyapunov optimization• decomposes a time average objective into objectives for each time slot• capture the trade-off between different system performance metrics

2) Original Problem goal: average power consumption constraints: bounded delay, feasible rate

[4] M. Neely, “Stochastic network optimization with application to communication and queueing systems,” Synthesis Lectures on Communication Networks, vol. 3, no. 1, pp. 1–211, 2010.

11

Single sensor WBAN scheduling

• Problem formulation 3) Worst-Case Delay Constraint Transform[5]

Why? No direct link between maximum delay and maximum queue a virtual queue Z(t) with a virtual arrival rate

Z(t) updates:

Lemma: Suppose system is controlled so that Z(i) Zmax, Q(i) Qmax, for all i, for some positive constants Zmax, Qmax. Then all data in queue is transmitted with a maximum delay Dmax:

Transform: from bounded delay to bounded queue length

[5] M. Neely, A. Tehrani, and A. Dimakis, “Efficient algorithms for renewable energy allocation to delay tolerant consumers,” in Proc. IEEE SmartGridComm’ 10, pp. 549–554, 2010.

12

Single sensor WBAN scheduling

• Problem formulation 4) Transform using Lyapunov optimization

why? Objectives for each time slot with illustration of the trade-off

a. quadratic form Lyapunov function

b. one-step Laypunov drift

c. upper bound of the drift

d. upper bound of the drift plus a weighted cost function

New objectives: min

Logic of minimization• minimizing the upper bound of the drift controls the delay• minimizing cost function is to minimize the energy consumption

Weighted cost function

13

Single sensor WBAN scheduling

• Problem formulation Final problem

Objectives: average of all possible states for each time nonlinear

Control variables: two decision variables one binary one integer

14

Single sensor WBAN scheduling

• Algorithm design two step algorithm

sleep scheduling

Where , and is the expectation of minimum of

Opportunistic Transmission

maximal available transmission amount given current channel

15

Single sensor WBAN scheduling

• Performance Analysis delay performance

Algorithm designed doesn’t guarantee non-positive drift define two conditions

Theorem 1. If above conditions hold, then deterministic upper bounds exist for actual queue and virtual queue as follows:

necessary for worst case delay guarantee

Recall lemma

Worst cast delay increase within

Theorem 2. Given the minimal power consumption P* that the system can achieve, the average power consumption of our proposed algorithm Pave satisfies: Pave P* + C/V , where C is a constant, at the cost of a worst-case delay increaseswithin O(V ).

Stationary randomize policy

power consumption performance

16

Single sensor WBAN scheduling

• simulation setup Body channel

model suggested by IEEE 802.15 task group 6 under the frequency band 2.4GHz

Wake up ratio: the fraction of time slots in which the sensor wakes up among the number of total time slots

Parameters' Value

17

Single sensor WBAN scheduling

• Simulation results

• Delay growth can be bounded a linear function of weighting factor

• Larger weighting factor, poorer delay

• Data accumulation: for potential better channel• Flat cliff: not in a very good channel condition • Sharp cliff: in a good channel condition

18

Single sensor WBAN scheduling

• Simulation results

• The gap between power consumption of our algorithm and the optimal one can be bounded by a function of the inverse of weighting factor

• Smaller wakeup ratio, less power consumption

• Larger virtual arrival rate, smaller delay• Larger virtual arrival rate, larger wakeup ratio

19

• A scheduling policy for single sensor WBAN application Address the energy delay trade-off problem for WBAN

limited transmission power random traffic and channel worst case delay guarantee

Propose a scheduling policy for the problem Utilize both sleep and opportunistic transmission for

energy saving Achieve worst case delay Show trade-off between power consumption and delay

Conclusion and Future works