Daniel Lee Muhammad Naeem Chingyu Hsu

Similar documents
w (1) ˆx w (1) x (1) /ρ and w (2) ˆx w (2) x (2) /ρ.

Definitions and Theorems. where x are the decision variables. c, b, and a are constant coefficients.

ACO Comprehensive Exam 9 October 2007 Student code A. 1. Graph Theory

Markov Decision Processes

Optimization Methods MIT 2.098/6.255/ Final exam

Graph Theory and Optimization Integer Linear Programming

Optimization Methods: Linear Programming Applications Assignment Problem 1. Module 4 Lecture Notes 3. Assignment Problem

Solutions for the Exam 9 January 2012

Math 155 (Lecture 3)

HOMEWORK #10 SOLUTIONS

Robust Resource Allocation in Parallel and Distributed Computing Systems (tentative)

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

Lecture 6 Chi Square Distribution (χ 2 ) and Least Squares Fitting

Riemann Sums y = f (x)

Integer Programming (IP)

Linear regression. Daniel Hsu (COMS 4771) (y i x T i β)2 2πσ. 2 2σ 2. 1 n. (x T i β y i ) 2. 1 ˆβ arg min. β R n d

Statistical Pattern Recognition

Rank Modulation with Multiplicity

Dynamic Programming. Sequence Of Decisions

Dynamic Programming. Sequence Of Decisions. 0/1 Knapsack Problem. Sequence Of Decisions

Introduction to Optimization Techniques. How to Solve Equations

Sample Size Determination (Two or More Samples)

THE SOLUTION OF NONLINEAR EQUATIONS f( x ) = 0.

Chapter 13, Part A Analysis of Variance and Experimental Design

Spectral Graph Theory and its Applications. Lillian Dai Oct. 20, 2004

Output Analysis (2, Chapters 10 &11 Law)

Analysis of Supportable Rates in Symmetric Blocking Wavelength Routers

CS284A: Representations and Algorithms in Molecular Biology

On comparison of different approaches to the stability radius calculation. Olga Karelkina

Approximate Dynamic Programming by Linear Programming for Stochastic Scheduling

Problem Set 2 Solutions

OPTIMAL ALGORITHMS -- SUPPLEMENTAL NOTES

1.225J J (ESD 205) Transportation Flow Systems

Lecture 20. Brief Review of Gram-Schmidt and Gauss s Algorithm

Optimally Sparse SVMs


ROLL CUTTING PROBLEMS UNDER STOCHASTIC DEMAND

Classification of problem & problem solving strategies. classification of time complexities (linear, logarithmic etc)

Multi-robot monitoring in dynamic environments with guaranteed currency of observations

Different kinds of Mathematical Induction

IP Reference guide for integer programming formulations.

Introduction to Optimization Techniques

Exercise 4.3 Use the Continuity Theorem to prove the Cramér-Wold Theorem, Theorem. (1) φ a X(1).

Week 5-6: The Binomial Coefficients

1 Duality revisited. AM 221: Advanced Optimization Spring 2016

L = n i, i=1. dp p n 1

B Supplemental Notes 2 Hypergeometric, Binomial, Poisson and Multinomial Random Variables and Borel Sets

Simulation of Discrete Event Systems

Model of Computation and Runtime Analysis

CSE 4095/5095 Topics in Big Data Analytics Spring 2017; Homework 1 Solutions

Introduction to Machine Learning DIS10

A collocation method for singular integral equations with cosecant kernel via Semi-trigonometric interpolation

M.Jayalakshmi and P. Pandian Department of Mathematics, School of Advanced Sciences, VIT University, Vellore-14, India.

TEACHER CERTIFICATION STUDY GUIDE

THE ASYMPTOTIC COMPLEXITY OF MATRIX REDUCTION OVER FINITE FIELDS

IIT JAM Mathematical Statistics (MS) 2006 SECTION A

Mixed Acceptance Sampling Plans for Multiple Products Indexed by Cost of Inspection

subject to A 1 x + A 2 y b x j 0, j = 1,,n 1 y j = 0 or 1, j = 1,,n 2

The picture in figure 1.1 helps us to see that the area represents the distance traveled. Figure 1: Area represents distance travelled

On Random Line Segments in the Unit Square

Divide & Conquer. Divide-and-conquer algorithms. Conventional product of polynomials. Conventional product of polynomials.

CSE 21 Mathematics for

First, note that the LS residuals are orthogonal to the regressors. X Xb X y = 0 ( normal equations ; (k 1) ) So,

Run-length & Entropy Coding. Redundancy Removal. Sampling. Quantization. Perform inverse operations at the receiver EEE

Lecture 2 February 8, 2016

1 Hash tables. 1.1 Implementation

A NEW APPROACH TO SOLVE AN UNBALANCED ASSIGNMENT PROBLEM

Singular Continuous Measures by Michael Pejic 5/14/10

Mechatronics. Time Response & Frequency Response 2 nd -Order Dynamic System 2-Pole, Low-Pass, Active Filter

6. Uniform distribution mod 1

Ma 4121: Introduction to Lebesgue Integration Solutions to Homework Assignment 5

Integrable Functions. { f n } is called a determining sequence for f. If f is integrable with respect to, then f d does exist as a finite real number

A) is empty. B) is a finite set. C) can be a countably infinite set. D) can be an uncountable set.

18.01 Calculus Jason Starr Fall 2005

4.1 Sigma Notation and Riemann Sums

6 Integers Modulo n. integer k can be written as k = qn + r, with q,r, 0 r b. So any integer.

Minimum Dominating Set Approach to Analysis and Control of Biological Networks

Some special clique problems

Design and Analysis of Algorithms

Fall 2013 MTH431/531 Real analysis Section Notes

Probability and statistics: basic terms

The Poisson Process *

Model of Computation and Runtime Analysis

15.083J/6.859J Integer Optimization. Lecture 3: Methods to enhance formulations

Lecture 2 Clustering Part II

IN many applications, the average delay packets experience. Delay-Minimal Transmission for Average Power Constrained Multi-Access Communications

Scheduling under Uncertainty using MILP Sensitivity Analysis

Polynomial identity testing and global minimum cut

Sequences, Mathematical Induction, and Recursion. CSE 2353 Discrete Computational Structures Spring 2018

ECE 564/645 - Digital Communication Systems (Spring 2014) Final Exam Friday, May 2nd, 8:00-10:00am, Marston 220

Lecture 15: Learning Theory: Concentration Inequalities

TRANSPORTATION AND ASSIGNMENT PROBLEMS

Mathematical Statistics - MS

Chapter 11 Output Analysis for a Single Model. Banks, Carson, Nelson & Nicol Discrete-Event System Simulation

Linear Programming and the Simplex Method

Lecture 7: Properties of Random Samples

Asymptotic Coupling and Its Applications in Information Theory

Generating subtour elimination constraints for the Traveling Salesman Problem

Commutativity in Permutation Groups

4.1 SIGMA NOTATION AND RIEMANN SUMS

Transcription:

omplexity Aalysis of Optimal Statioary all Admissio Policy ad Fixed Set Partitioig Policy for OVSF-DMA ellular Systems Daiel Lee Muhammad Naeem higyu Hsu

Backgroud Presetatio Outlie System Model all Admissio otrol Schemes omplexity Of Proposed Schemes Numerical Results Summary

Backgroud I Widebad DMA orthogoal variable spreadig factor (OVSF) are codes for providig differet chaels for differet users with differet data rates

[ ] [ ] [ ] [ ] [ ] [ ] 4 [ ] 4 [ ] [ ] 3 4 [ ] 4 4 [ ] [ ] k k k k k k Well kow OVSF code geeratio Nodes with the depth l from the root of the tree have code words of code legth l.

Example 4R R R

M Resource costrait: j 0 j k j k k0 Example of capacity regio max3

System Model Radom arrivals of call requests i each class ca be modeled by a stochastic process (Poisso process). For class j Average arrival rate is λ j Average call duratio μ j. I respose to each arrival of call request the etwork operator must make a call admissio cotrol (A) decisio (whether to accept or reject the call request) ad the decisio should be made based o the resource costrait. The objective ca be to maximize average throughput or miimize blockig probability etc.

System Model We allow code re-assigmet

Mai poit The greedy all Admissio Policy is ot ecessarily optimal.

Optimal policy The system ca be modeled by a Markov decisio process so a optimal policy ca be obtaied by optimizig a Markov Decisio Process. Liear Programmig Policy iteratio etc. { M ( ) } i k0 k k M k i 0 i ki Z

Previous Results: optimal policy A optimal policy ca have sigificatly better performace tha the greedy policy Accordig to small-scale cases Large-scale: large large umber of classes etc. omputatioal complexity of liear programmig (e.g. umber of variables) { M ( ) } i k0 k k M k i 0 i ki Z

Previous results: suboptimal policy Eve a fixed set policy ca have sigificatly better performace tha the greedy policy.

Optimal Fixed Set Partitioig OVSF codes are partitioed ito mutually exclusive groups of codes ad uiquely assigs a group to each service class. G j the umber of codes i the subset assiged with class j where j 0 M. Oce vector G (G 0 G. G M- ) is determied the actual partitio of the set is trivial. The umbers of codes i the subsets G 0 G G M- satisfy M j j 0 Average Through put of this scheme M ( ) ( k )( ) T G R P λ µ G F k 0 k k k j where P k ( λ k µ k) Gk 0( λ k µ k)!!

Optimal Fixed Set Partitioig Greedy withi fixed sets

omplexity Aalysis M j 0 ( G G G ) Size of the feasible set: the umber of vectors that satisfy costrait j G j 0 M

ombiatorial Aalysis ad Recursio G M ca have ay value from 0 / M. if G M is x M. The G M ca be 0 3 (/ M- -x M ) ad so o. The relatio betwee adjacet classes has a recursio Γm ( v) ( ) Let be the umber of o-egative itegral vectors G G G that ca occupy the space take by v codes 0 M of largest codes i.e. satisfyig costrait G v. The ( ) v x 0 ( ) ( v) v ( ) v ( ) Γ v Γ v x Γ i m m i 0 m Γ 0 03... m i m i 0 i

( v) ( v) ( v) Fuctios Γ Γ... Γ ca be evaluated from the recursio. 0 M ( G G G ) The umber of vectors that satisfy costrait 0 M ( ) M j M G j 0 j ΓM is.

Numerical Results

Numerical Results

Optimizatio of Markove Decisio process through LP I the case of LP its complexity is closely related to the size of state space { M ( ) } i k0 k k M k i 0 i ki Z

Let χ m (v) be the umber of o-egative iteger vectors (k 0 k k m ) that satisfy the costrait m i m k 0 i v The umber of vectors (k 0 k k m- ) that satisfy this costrait is χ m- ((v-x m )). Therefore we have the followig recursio: i ( ) ( ) m i m m k i 0 i v xm vx m χ m v ( v) ( i) i χ 0 m

χ m (v) ad Γ m (v) have a idetical recursio The size of the state space {( ) M } i k0 k k M k i 0 i ki Z M ( M) ( M ) M χ Γ

Numerical Results

Fixed Set Partitioig (FSP) omplexity The complexity of FSP operatio is characterized as O() lookups DA-A The complexity of DA-A is also O() I summary i both FSP ad DA the computatioal complexity of olie operatios is O() per call arrival.

Thak You.