Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
CS 8292 : Advanced Topics in Convex Optimization and its Applications Fall 2010
Outline
Wireless Power Control by LP Wireless Power Control by GP Extensions to Multiple Antennas by SDP
http://news.cnet.com/830113579 31002802637.html
3
System Considerations
How to solve optimally nonconvex power control problems? How many ways to characterize optimality? How to design distributed power control algorithms with fast convergence and good performance guarantees? How fast is fast? Can we leverage existing technology? What is the industry impact?
System Model
Interference channel with single-user decoding: Treat interference as additive Gaussian noise Control interference and meet objective using power control
Performance Metrics
Signal-to-Interference Ratio: SIRl(p) =
j=l
Gllpl Glj pj + nl
with Glj the channel gains from transmitter j to receiver l and nl the additive white Gaussian noise (AWGN) power at receiver l Attainable data rate (nats per channel use) is a function of SIR, e.g., Shannon capacity formula rl = log(1 + SIRl) Power constraints 1 p P
6
Interference Parameters
Let F be a nonnegative matrix with entries: Fij = and v= Let B be: n1 n2 nL , ,..., G11 G22 GLL B = F + (1/P )v1 . . if i = j Gij Gii , if i = j 0,
l l.
diag()v.
Fixed-point algorithm: p(k + 1) = diag()Fp(k) + diag()v Distributed Power Control (DPC) algorithm (more illuminating form): l pl(k + 1) = pl(k) l. SIRl(p(k)) Geometric convergence to (I diag()F)1diag()v if and only if (diag()F) < 1
9
Dualities
GP duality: Application to energy-robustness tradeo
minimize epl l subject to log(l/SIRl( )) 0 l, p variables: pl l.
(5)
LP Duality: Application to Beamforming uplink-downlink duality Perron-Frobenius Duality (later on max-min weighted SIR): Application to distributed fast algorithm
10
xmt =
l=1
bl u l
where bl is the transmitted data signal for the lth user Received signal at the lth user: rcvl = bluhl + l
j=l
bj u hl + n l . j
11
Downlink SIR dependent on power p and transmit beamformers U = [u1 . . . uL] SIRl(p, U) = |hul|2 l |huj |2 j=l l + vl .
12
Hl = hlh l
minimize l Tr(Ul ) subject to l(Tr(HlUj ) + vl) Tr(HlUl) 0 l, Ul 0, Rank(Ul) = 1 l variables: Ul, l. Relax the problem by removing the rank-one constraint leads to semidenite program (SDP)
13
There exists a rank-one solution for Ul for all l of the SDP Many other extensions including second order cone program (SOCP) preprocessing Stanford CVX, SeDuMi toolbox for SDP State of the art of SDP & SOCP solvers for large (but not huge) problems is quite satisfactory
1. E. Visotsky & U. Madhow, Optimum beamforming using transmit antenna arrays, Proc. of IEEE VTC, 1999 2. M. Chiang, P. Hande, T. Lan & C. W. Tan, Power control in wireless cellular networks, NOW Foundations & Trends in Networking, 2008
14
pl P .
l
Can reformulate as a GP and solve using interior point algorithm. Can reuse the existing DPC algorithm in any way? For distributed solution and its complete analytical solution, see lectures later
15
Summary
A seemingly hard problem can be solved by exploiting the problem structure How many possible ways to solve it? And which is the best way?1
Reading assignment: G. J. Foschini and Z. Miljanic, A Simple Distributed Autonomous Power Control Algorithm and its Convergence, IEEE Transactions on Vehicular Technology, Vol. 42, No. 4, Nov. 1993. M. Chiang, P. Hande, T. Lan and C. W. Tan, Power Control in Wireless Cellular Networks, Foundations and Trends in Networking, 2008.
1