I + HH H N 0 M T H = UΣV H = [U 1 U 2 ] 0 0 E S. X if X 0 0 if X < 0 (X) + = = M T 1 + N 0. r p + 1

Similar documents
Homework 5 Solutions. Problem 1

ECE559VV Project Report

U.C. Berkeley CS294: Beyond Worst-Case Analysis Luca Trevisan September 5, 2017

Lecture 3. Ax x i a i. i i

MIMO Systems and Channel Capacity

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Lecture 10 Support Vector Machines II

Department of Statistics University of Toronto STA305H1S / 1004 HS Design and Analysis of Experiments Term Test - Winter Solution

An Interactive Optimisation Tool for Allocation Problems

The Concept of Beamforming

2016 Wiley. Study Session 2: Ethical and Professional Standards Application

j) = 1 (note sigma notation) ii. Continuous random variable (e.g. Normal distribution) 1. density function: f ( x) 0 and f ( x) dx = 1

Calculation of time complexity (3%)

princeton univ. F 13 cos 521: Advanced Algorithm Design Lecture 3: Large deviations bounds and applications Lecturer: Sanjeev Arora

Chapter Newton s Method

Modeling and Simulation NETW 707

UNIVERSITY OF TORONTO Faculty of Arts and Science. December 2005 Examinations STA437H1F/STA1005HF. Duration - 3 hours

/ n ) are compared. The logic is: if the two

Errors for Linear Systems

The Gaussian classifier. Nuno Vasconcelos ECE Department, UCSD

Temperature. Chapter Heat Engine

Randomness and Computation

Flexible Allocation of Capacity in Multi-Cell CDMA Networks

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Antenna Combining for the MIMO Downlink Channel

Chapter 15 - Multiple Regression

STATISTICS QUESTIONS. Step by Step Solutions.

Chapter 5 Multilevel Models

However, since P is a symmetric idempotent matrix, of P are either 0 or 1 [Eigen-values

Here is the rationale: If X and y have a strong positive relationship to one another, then ( x x) will tend to be positive when ( y y)

Norms, Condition Numbers, Eigenvalues and Eigenvectors

The Minimum Universal Cost Flow in an Infeasible Flow Network

Lecture 4: November 17, Part 1 Single Buffer Management

One-sided finite-difference approximations suitable for use with Richardson extrapolation

Stanford University CS359G: Graph Partitioning and Expanders Handout 4 Luca Trevisan January 13, 2011

Pulse Coded Modulation

Midterm Examination. Regression and Forecasting Models

Durban Watson for Testing the Lack-of-Fit of Polynomial Regression Models without Replications

x = x 1 + :::+ x K and the nput covarance matrces are of the form ± = E[x x y ]. 3.2 Dualty Next, we ntroduce the concept of dualty wth the followng t

CS : Algorithms and Uncertainty Lecture 17 Date: October 26, 2016

Chapter 12 Analysis of Covariance

The strict priority scheduler

Topic 7: Analysis of Variance

Lecture Notes on Linear Regression

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Topic- 11 The Analysis of Variance

Notes on Frequency Estimation in Data Streams

COMPARISON OF SOME RELIABILITY CHARACTERISTICS BETWEEN REDUNDANT SYSTEMS REQUIRING SUPPORTING UNITS FOR THEIR OPERATIONS

How Strong Are Weak Patents? Joseph Farrell and Carl Shapiro. Supplementary Material Licensing Probabilistic Patents to Cournot Oligopolists *

See Book Chapter 11 2 nd Edition (Chapter 10 1 st Edition)

Predictive Analytics : QM901.1x Prof U Dinesh Kumar, IIMB. All Rights Reserved, Indian Institute of Management Bangalore

Statistics for Economics & Business

Numerical Heat and Mass Transfer

Multipath richness a measure of MIMO capacity in an environment

Simulated Power of the Discrete Cramér-von Mises Goodness-of-Fit Tests

Matrix Approximation via Sampling, Subspace Embedding. 1 Solving Linear Systems Using SVD

First day August 1, Problems and Solutions

LINEAR REGRESSION MODELS W4315

JAB Chain. Long-tail claims development. ASTIN - September 2005 B.Verdier A. Klinger

Lecture 4: Universal Hash Functions/Streaming Cont d

Chapter 11: I = 2 samples independent samples paired samples Chapter 12: I 3 samples of equal size J one-way layout two-way layout

Power Allocation for Distributed BLUE Estimation with Full and Limited Feedback of CSI

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Finite Element Modelling of truss/cable structures

Some modelling aspects for the Matlab implementation of MMA

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system

ρ some λ THE INVERSE POWER METHOD (or INVERSE ITERATION) , for , or (more usually) to

Lecture 2: Prelude to the big shrink

Chapter 8 Indicator Variables

Prof. Dr. I. Nasser Phys 630, T Aug-15 One_dimensional_Ising_Model

2 Finite difference basics

Chapter 14 Simple Linear Regression

Dr. Shalabh Department of Mathematics and Statistics Indian Institute of Technology Kanpur

LINEAR REGRESSION ANALYSIS. MODULE IX Lecture Multicollinearity

Limited Dependent Variables

Topic 23 - Randomized Complete Block Designs (RCBD)

Lecture Space-Bounded Derandomization

x = , so that calculated

Statistics for Business and Economics

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

LINEAR REGRESSION ANALYSIS. MODULE VIII Lecture Indicator Variables

is the calculated value of the dependent variable at point i. The best parameters have values that minimize the squares of the errors

Chapter 13: Multiple Regression

Composite Hypotheses testing

Outline. Communication. Bellman Ford Algorithm. Bellman Ford Example. Bellman Ford Shortest Path [1]

princeton univ. F 17 cos 521: Advanced Algorithm Design Lecture 7: LP Duality Lecturer: Matt Weinberg

Negative Binomial Regression

A Robust Method for Calculating the Correlation Coefficient

Research Article Robustness Maximization of Parallel Multichannel Systems

Error Probability for M Signals

Introduction to Analysis of Variance (ANOVA) Part 1

Chapter 3 Describing Data Using Numerical Measures

4 Analysis of Variance (ANOVA) 5 ANOVA. 5.1 Introduction. 5.2 Fixed Effects ANOVA

Assignment 5. Simulation for Logistics. Monti, N.E. Yunita, T.

This model contains two bonds per unit cell (one along the x-direction and the other along y). So we can rewrite the Hamiltonian as:

Statistical Evaluation of WATFLOOD

4DVAR, according to the name, is a four-dimensional variational method.

Hashing. Alexandra Stefan

Continuous vs. Discrete Goods

Transcription:

Homework 4

Problem Capacty wth CSI only at Recever: C = log det I + E )) s HH H N M T R SS = I) SVD of the Channel Matrx: H = UΣV H = [U 1 U ] [ Σr ] [V 1 V ] H Capacty wth CSI at both transmtter and Recever: ) r C = log 1 + σ E s γ opt N M T Waterfllng: γ opt = X) + = r γ opt = M T µ M ) T N E S σ + { X f X f X < Start wth teraton counter p=1 Calculate the constant µ µ = M T r p + 1 [ 1 + N E S r p+1 ] 1 Usng µ calculate the power n the th sub-channel from, γ = µ M ) T N E S σ = 1, r p + 1) σ If the energy n the channel wth lowest gan s negatve, dscard t by settng γ opt r p+1 = Repeat the procedure by settng, p = p + 1 The optmal power allocaton strategy s found when all the allocated power s non-negatve 1. ρ = 1 db = E s N = 1 H= [ 1 1 1 1 ] = [ 1 1 1 1 ] [ ] [ 1 1 ] H 3

σ 1 =, σ = Snce both the sngular values are equal, we can allocate equal power to the two modes We can see ths even from the waterfllng procedure Start wth teraton counter p=1 Calculate the constant µ µ = M T r p + 1 [ 1 + N E S r p+1 ] [ 1 = 1 + 1 1 σ ] 1 σ Usng µ calculate the power n the t h sub-channel from, γ = µ M ) T N E S σ = 1, r p + 1) 11 γ 1 = 1 ) = 1 = γ Snce both γ 1, γ 1 are not negatve,γ opt 1 = 1, γ opt = 1 = 11 1 a) Capacty wth CSI at both transmtter and Recever: ) ) r C = log 1 + σ E s γ opt = log 1 + σ 1γ opt = log N M T 11) = 6.9189 b) Capacty wth CSI only at Recever: C = log det I + E )) s HH H = 6.9189 N M T Comment: Snce our optmum power allocaton strategy wth CSI at transmtter gave R ss = I whch s what we use for CSI unknown case), channel knowledge at the transmtter doesn t help n ths case.. H= [ 1 1 1 1 ] = σ 1 = 4, σ = [ 1 1 1 1 ] [ ] [ 1 1 1 1 H s rank defcent and so t has only one sngular value. We should use only one mode and put all power nto t. Waterfllng procedure s not requred n ths case. a) Capacty wth CSI at both transmtter and Recever: ) ) r C = log 1 + σ E s γ opt = log 1 + σ 1γ opt = log N M T 41) = 5.3576 ] H 4

b) Capacty wth CSI only at Recever: C = log det I + E )) s HH H = 4.393 N M T Comments: Capacty values wth or wthout CSI at transmtter) are lower for ths channel compared to the prevous channel, ths can be attrbuted to the rank defcent nature of ths channel. The capacty values wth and wthout CSI are dfferent n ths case, For the prevous channel they are same). Ths s due to the fact that we avoded puttng energy nto the mode that s n the nullspace, whch we can t do f there s no CSI. As a consequence, the capacty wthout CSI s less than wth CSI. Problem 3 15 Capacty Vs. SNR 1 Capacty b/s/hz) 5 5 1 15 5 Fgure 1: Capacty wth dfferent antenna confguratons At hgher SNR s, Capacty ncreases lnearly wth the rank of channel matrx. The performance wth s better than. Ths s due to the array gan seen n case. Better performance of can be explaned by the addtonal degrees of freedom avalable. All these plots assume that transmtter doesn t have CSI. 5

Capacty wth CSI only at Recever: C = log det I + ρ )) HH H M T let, r = mn {M T, M R } Usng a low SNR approxmaton, outage formulaton s gven by, { } ρ P H F C = x M T { H P F C } = x M T M R ρm R Correspondngly for, { P h C } = x ρ Based on the tals of the dstrbutons of H F M T M R and h we can see that to have the same outage x), we need to ntegrate across more regon n case compared to. Ths leads to the followng nequalty, C C M R The above nequalty s notced to be vald from smulatons also. A smlar analyss can be carred out for hgh SNR regme, usng the followng approxmaton. C = log det I + ρ )) HH H M T Some observatons from plots: = r log ρ + r log λ For small ɛ s, and at hgh SNR s, the outage capacty rato between and s more than the rato at hgher outage values. Whch means for smaller outage values havng more antennas at the transmtter or recever) s more meanngful than at hgher ɛ s. At moderate to hgh ɛ s, outage capacty follows the pattern of ergodc capacty, whch s a lnear growth n mn {M T, M R } At very hgh ɛ s 99%), outage capacty has nterestng behavor. At moderate to hgh SNR s lnear growth n mn {M T, M R } s possble. But at very low SNR s, performs better than all other schemes. has the worst performance of all, performs slghtly better than. Ths can be explaned n terms of power dstrbuton at the transmtter. At low SNR s t s better to use only one channel, snce we don t have channel knowledge f we put energy n both modes the chance of t beng n outage s pretty hgh. Ths could be one reason for better performance of compared to. performs slghtly better than because of the recevng antennas. P. 187-189 Tse and Vshwanath) have an analytcal treatment of outage capacty for, and cases. 6

Ergodc Capacty: MATLAB code - Slght modfcaton for outage code % mt, Number of transmttng antennas % mr, Number of transmttng antennas % ITER, number of trals % SNRdB, Range of % C, varable for capacty of a system % C, varable for capacty of a system % C, varable for capacty of a system % C, varable for capacty of a system % h, random channel for wth zero mean unt varance) % h, random channel for % h, random channel for % h, random channel for clc; close all; clear all; mt = ; mr = ; ITER = 1; SNRdB = [:5]; SNR = 1.ˆSNRdB/1); C = zeros1,lengthsnr)); C = zeros1,lengthsnr)); C = zeros1,lengthsnr)); C = zeros1,lengthsnr)); for te = 1:ITER h = randn +j*randn)/sqrt); h = randnmr,1)+j*randnmr,1))/sqrt); h = randn1,mt)+j*randn1,mt))/sqrt); h = randnmr,mt)+j*randnmr,mt))/sqrt); for K = 1:lengthSNR) C K) = C K) + log1+ SNRK)*normh )ˆ); C K) = C K) + log1+ SNRK)*normh )ˆ); C K) = C K) + log1+ SNRK)*normh )ˆ/mT); C K) = C K) + logabsdeteyemr)+snrk)*h *h /mt))); end end C = C /ITER; C = C /ITER; C = C /ITER; C = C /ITER; plotsnrdb,c, r -.,SNRdB,C, b - o,snrdb,c, m,snrdb,c, k - * ) legend,,,,) xlabel ) ylabel Capacty b/s/hz) ) ttle Capacty Vs. SNR ) grd; 7

1 9 8 7 1% Outage Capacty Capacty b/s/hz) 6 5 4 3 1 5 1 15 5 1 1 1 % Outage Capacty Capacty b/s/hz) 8 6 4 5 1 15 5 18 16 99 % Outage Capacty Capacty b/s/hz) 14 1 1 8 6 4 5 1 15 5 Fgure : Outage Capacty: Increases lnearly wth the rank of channel matrx at hgh SNR s 8

.6.5 1% Outage Capacty at low SNR s Capacty b/s/hz).4.3..1 18 16 14 1 1.14.1.1 1 % Outage Capacty at Low SNR s Capacty b/s/hz).8.6.4. 18 16 14 1 1.8.7.6 99 % Outage Capacty at low SNR s Capacty b/s/hz).5.4.3..1 18 16 14 1 1 Fgure 3: Outage Capacty at low SNR s 9