Sampling Procedure of the Sum of two Binary Markov Process Realizations

Similar documents
V.Abramov - FURTHER ANALYSIS OF CONFIDENCE INTERVALS FOR LARGE CLIENT/SERVER COMPUTER NETWORKS

Solution in semi infinite diffusion couples (error function analysis)

HEAT CONDUCTION PROBLEM IN A TWO-LAYERED HOLLOW CYLINDER BY USING THE GREEN S FUNCTION METHOD

On One Analytic Method of. Constructing Program Controls

Performance Analysis for a Network having Standby Redundant Unit with Waiting in Repair

Lecture 2 M/G/1 queues. M/G/1-queue

Variants of Pegasos. December 11, 2009

This document is downloaded from DR-NTU, Nanyang Technological University Library, Singapore.

Relative controllability of nonlinear systems with delays in control

Approximate Analytic Solution of (2+1) - Dimensional Zakharov-Kuznetsov(Zk) Equations Using Homotopy

[ ] 2. [ ]3 + (Δx i + Δx i 1 ) / 2. Δx i-1 Δx i Δx i+1. TPG4160 Reservoir Simulation 2018 Lecture note 3. page 1 of 5

ESTIMATIONS OF RESIDUAL LIFETIME OF ALTERNATING PROCESS. COMMON APPROACH TO ESTIMATIONS OF RESIDUAL LIFETIME

Existence and Uniqueness Results for Random Impulsive Integro-Differential Equation

John Geweke a and Gianni Amisano b a Departments of Economics and Statistics, University of Iowa, USA b European Central Bank, Frankfurt, Germany

Implementation of Quantized State Systems in MATLAB/Simulink

( t) Outline of program: BGC1: Survival and event history analysis Oslo, March-May Recapitulation. The additive regression model

( ) () we define the interaction representation by the unitary transformation () = ()

Robustness Experiments with Two Variance Components

Let s treat the problem of the response of a system to an applied external force. Again,

Cubic Bezier Homotopy Function for Solving Exponential Equations

First-order piecewise-linear dynamic circuits

GENERATING CERTAIN QUINTIC IRREDUCIBLE POLYNOMIALS OVER FINITE FIELDS. Youngwoo Ahn and Kitae Kim

In the complete model, these slopes are ANALYSIS OF VARIANCE FOR THE COMPLETE TWO-WAY MODEL. (! i+1 -! i ) + [(!") i+1,q - [(!

5th International Conference on Advanced Design and Manufacturing Engineering (ICADME 2015)

Lecture 18: The Laplace Transform (See Sections and 14.7 in Boas)

On computing differential transform of nonlinear non-autonomous functions and its applications

J i-1 i. J i i+1. Numerical integration of the diffusion equation (I) Finite difference method. Spatial Discretization. Internal nodes.

Notes on the stability of dynamic systems and the use of Eigen Values.

THE PREDICTION OF COMPETITIVE ENVIRONMENT IN BUSINESS

CH.3. COMPATIBILITY EQUATIONS. Continuum Mechanics Course (MMC) - ETSECCPB - UPC

Part II CONTINUOUS TIME STOCHASTIC PROCESSES

EP2200 Queuing theory and teletraffic systems. 3rd lecture Markov chains Birth-death process - Poisson process. Viktoria Fodor KTH EES

Polymerization Technology Laboratory Course

Linear Response Theory: The connection between QFT and experiments

UNIVERSITAT AUTÒNOMA DE BARCELONA MARCH 2017 EXAMINATION

Time-interval analysis of β decay. V. Horvat and J. C. Hardy

Dynamic Team Decision Theory. EECS 558 Project Shrutivandana Sharma and David Shuman December 10, 2005

Density Matrix Description of NMR BCMB/CHEM 8190

DYNAMIC ECONOMETRIC MODELS Vol. 8 Nicolaus Copernicus University Toruń 2008

Comparison of Differences between Power Means 1

TSS = SST + SSE An orthogonal partition of the total SS

Density Matrix Description of NMR BCMB/CHEM 8190

Chapter 6 DETECTION AND ESTIMATION: Model of digital communication system. Fundamental issues in digital communications are

Including the ordinary differential of distance with time as velocity makes a system of ordinary differential equations.

2/20/2013. EE 101 Midterm 2 Review

Bernoulli process with 282 ky periodicity is detected in the R-N reversals of the earth s magnetic field

Track Properities of Normal Chain

e-journal Reliability: Theory& Applications No 2 (Vol.2) Vyacheslav Abramov

Outline. Probabilistic Model Learning. Probabilistic Model Learning. Probabilistic Model for Time-series Data: Hidden Markov Model

Bayesian Inference of the GARCH model with Rational Errors

Survival Analysis and Reliability. A Note on the Mean Residual Life Function of a Parallel System

Comb Filters. Comb Filters

P R = P 0. The system is shown on the next figure:

Graduate Macroeconomics 2 Problem set 5. - Solutions

3. OVERVIEW OF NUMERICAL METHODS

Mechanics Physics 151

Mechanics Physics 151

New M-Estimator Objective Function. in Simultaneous Equations Model. (A Comparative Study)

Fall 2010 Graduate Course on Dynamic Learning

Networked Estimation with an Area-Triggered Transmission Method

SELFSIMILAR PROCESSES WITH STATIONARY INCREMENTS IN THE SECOND WIENER CHAOS

Ordinary Differential Equations in Neuroscience with Matlab examples. Aim 1- Gain understanding of how to set up and solve ODE s

A New Generalized Gronwall-Bellman Type Inequality

FTCS Solution to the Heat Equation

Control Systems. Mathematical Modeling of Control Systems.

CS434a/541a: Pattern Recognition Prof. Olga Veksler. Lecture 4

A GENERAL FRAMEWORK FOR CONTINUOUS TIME POWER CONTROL IN TIME VARYING LONG TERM FADING WIRELESS NETWORKS

Pavel Azizurovich Rahman Ufa State Petroleum Technological University, Kosmonavtov St., 1, Ufa, Russian Federation

January Examinations 2012

WiH Wei He

RELATIONSHIP BETWEEN VOLATILITY AND TRADING VOLUME: THE CASE OF HSI STOCK RETURNS DATA

Advanced Machine Learning & Perception

FI 3103 Quantum Physics

Online Appendix for. Strategic safety stocks in supply chains with evolving forecasts

M. Y. Adamu Mathematical Sciences Programme, AbubakarTafawaBalewa University, Bauchi, Nigeria

Chapter Lagrangian Interpolation

A Deterministic Algorithm for Summarizing Asynchronous Streams over a Sliding Window

Example: MOSFET Amplifier Distortion

CHAPTER 10: LINEAR DISCRIMINATION

SOME NOISELESS CODING THEOREMS OF INACCURACY MEASURE OF ORDER α AND TYPE β

Genetic Algorithm in Parameter Estimation of Nonlinear Dynamic Systems

Journal of Theoretical and Applied Information Technology.

12d Model. Civil and Surveying Software. Drainage Analysis Module Detention/Retention Basins. Owen Thornton BE (Mech), 12d Model Programmer

Improvement in Estimating Population Mean using Two Auxiliary Variables in Two-Phase Sampling

A Simulation Based Optimal Control System For Water Resources

Video-Based Face Recognition Using Adaptive Hidden Markov Models

Improvement in Estimating Population Mean using Two Auxiliary Variables in Two-Phase Sampling

Chapter 2 Linear dynamic analysis of a structural system

Energy Storage Devices

Chapter 6: AC Circuits

Scattering at an Interface: Oblique Incidence

OPTIMAL EXPERIMENTAL DESIGN WITH INVERSE PROBLEMS

Inverse Joint Moments of Multivariate. Random Variables

The Analysis of the Thickness-predictive Model Based on the SVM Xiu-ming Zhao1,a,Yan Wang2,band Zhimin Bi3,c

The Finite Element Method for the Analysis of Non-Linear and Dynamic Systems

Li An-Ping. Beijing , P.R.China

Advanced time-series analysis (University of Lund, Economic History Department)

F-Tests and Analysis of Variance (ANOVA) in the Simple Linear Regression Model. 1. Introduction

Real-time Quantum State Estimation Based on Continuous Weak Measurement and Compressed Sensing

Math 128b Project. Jude Yuen

Transcription:

Samplng Procedure of he Sum of wo Bnary Markov Process Realzaons YURY GORITSKIY Dep. of Mahemacal Modelng of Moscow Power Insue (Techncal Unversy), Moscow, RUSSIA, E-mal: gorsky@yandex.ru VLADIMIR KAZAKOV Dep. of Telecommuncaons, ESIME-Zacaenco, SEPI, Naonal Polyechnc Insue of Mexco, MÉXICO E-mal: vkaz4@homal.com, Absrac: The sum of wo bnary Markov processes s consdered. Resul of he sum can have four or hree saes. In he frs case here s a Markov process, n he second case he process s non Markov. We nvesgae samplng procedure of any realzaon of he sum process. The nvesgaon mehod akes no accoun he probably of he sae omsson and probably denses funcons of sayng mes n each sae. In resul, we oban he algorhm for choosng of samplng nervals for each sae. They are dfferen. One non rval example s gven. Key-Words: - Samplng Algorhm, Bnary Markov Process, Error reconsrucon.. Inroducon The problem of samplng - reconsrucon procedure (SRP) of realzaons of random processes s exensvely examned n leraure. Majory of publcaons are devoed o he SRP nvesgaon of realzaons of connuous processes. SRP of realzaons of dsconnuous processes s weakly consdered. Here we menon some papers conneced wh he dscussed problem [, ]. In [] samplng of realzaons of he Bnary Markov Process (BMP) s nvesgaed n order o descrbe new Bnary processes. In [] SRP of realzaons of BMP s analyzed. Bu hs analyss s based on he covarance approach. There are some auhors publcaons [ -5] devoed o SRP of dsconnuous process realzaons. Invesgaons n [4, 5] are based on he Condonal mean Rule. In resuls we obaned he sascal descrpon of samplng procedure and o esmae he qualy of he reconsrucon. I s necessary o menon ha here s a grea dfference beween he SRP mehod for connuous [6 ] and dsconnuous [4, 5, ] process realzaon. In he presen paper we consder SRP of realzaons of dsconnuous process whch s he sum ς ( ) of wo ndependen BMP processes. Such a problem s neresng n some conrol and elemery sysems. There are wo dfferen cases: ) he process ς ( ) has four dfferen saes; ) he process ς ( ) has hree dfferen saes. The frs varan can be consdered by he Markov mehod suggesed n [4, 5]. The nvesgaon of he second case s more dffcul n comparson of he frs one because he process under consderaon becomes non Markov. Here a new aspec of SRP problem s occurred. In he frs me, we need o nvesgae he deermnaon of he nsan momen of he nex sample, when he locaons of prevous samples are known. Ths problem mus be solved akng no accoun he gven probably of a sae omsson. Ths SRP problem s nvesgaed n deal. Fnally, here s one non rval example. ISBN: 978--684-69-5 7

. The General Varan ξ ( ), ξ ( ) are ndependen BMP wh Le correspondng saes: {x, x } и {y, y }, x < x, y < y, and wh ranson denses (or wh he nenses) {λ, λ } and {µ, µ }. These denses have a known sense, for nsance, λ : P{x x, } = λ + o( ) ec. Le us nroduce a wo dmensonal Markov process { ( ), ( ) } ξ ξ wh four saes. The process graph s presened n Fg.. Here one can see saes wh all possble ransons and correspondng denses. We consder a sum ς ( ) = ξ ( ) + ξ ( ). Fg.. The graph of he wo dmensonal process { ξ ( ), ξ ( ) } The process ς ( ) s observable. We have o oban a samplng algorhm for any realzaon of hs process wh he prncpal condon: he probably of a sae omsson mus be no more of a gven value γ. Le us consder wo dfferen cases. The frs case When (x, y ) λ λ (x, y ) µ µ µ µ λ (x, y ) (x, y ) λ x x - x y y - y ; () he process ζ() has four saes z, z, z, z : z = x + y, z = x + y, z = x + y, z = x + y Snce z z, akng no accoun () we have z - z = ς s (x - x ) (y - y ). Then he process Markov. The graph of hs process s he same as presens n Fg.. In hs case he probably densy funcon (pdf) of sayng me ν n he saes (,,,) E α wh parameers: η E(α = λ + µ ), η E(α = λ + µ ), z = s exponenal η E(α = λ + µ ), η E(α = λ + µ ) () The ranson probables are deermned by he followng formulas: µ λ P =, P =, µ + λ µ + λ P = λ µ, P = λ + µ λ + µ () In () we gve some requred expressons only. In order o descrbe SRP of such ype of he process, s necessary o use he mehod [4, 5]. The second case When x x - x = y y - y ; (4) he process ζ() has hree saes z, u, z : z = x + y, u = x + y,= x + y, z = x + y The sae u s formed by he unon of wo saes z and z of he process n he frs case. The graph of he dscussed process s shown n Fg.. z u z Fg.. The graph of he process ζ() n he second case The pdf of me sayng η and η n he saes z and z are he same as n (),.e. hey are exponenal ISBN: 978--684-69-5 8

wh he parameers α and α correspondngly. The suaon wh pdf of sayng me η u n he sae u s anoher. The pdf of he me ηu depends on he prevous sae z (=; ). If he realzaon comes no he sae u from z, hen he pdf wll be E( α ), (=; ). Then pdf of ηu s deermned by he formula of he oal probably. I means ha pdf s no exponenal. In hs case s necessary o apply analyss whch s vald for non Markov processes, or for an arbrary pdf.. The Esmaon of he Samplng Inerval Le us nroduce some desgnaons:, j, k he pas, he presence, and he fuure saes. ζ j a random nsan me of an oupu from j. Ths momen has s begnnng from me of he las sample of he sae. θ - an nerval from L unl a curren presen me. η k a sayng me n he nex sae k. ζ θj a remander of wang me for oupu from j f ζ j > θ. The samplng nerval T j can be chosen akng no accoun he requred demand for he probably of he omsson of he nex sae k: max P{ζ θj + η k < T j } = F Σ (T j ) =γ, (5) k where F Σ (;j,k) s he probably dsrbuon funcon of he sum Σ ζ θj + η k. a) If a curren sae j s equal o z or z (hey have he exponenal fdp), hen he remander of wang saes ζ θj, j =, have he same pdf and hey do no depend on θ. There are f j () = E(, α j ), j = ;. The sayng me ηk = ηu depends on j. Is pdf s deermned by he combnaon of wo exponenal pdf, correspondng o saes z и z. Pdf for η u s descrbed by he expresson: f u ( j) = p jse( ; αs ), j =,. (6) The probably dsrbuon funcon of he sum Σ can be smply found by he convoluon of wo exponenal dsrbuons. Le us desgnae he dsrbuon funcon of he sum of wo ndependen random varables wh he exponenal pdf by F(; α, β) (here α and β are parameers of pdf). Then he dsrbuon funcon of he sum n (5) can wre n he form: F Σ (; j, u) = Σ, P js F(; α j, α s ), j =, b) Le us assume he curren sae j s he sae u. We deermne pdf of he ranson momen τ zu from z no u. Takng no accoun resuls [4, 5] and he expresson (6) one can oban he formula for pdf w u () as a lneal combnaon of wo runcaed (by he value Т ) exponenal pdf E T (, µ): w u () = pset ( ; µ s ), =,, (7) µ s = α - α s, (8) µ µ T µ e ( e ), µ, where E T (, µ) = (9) T, µ = On he base of w u () one can deermned he condonal mean and varance. They are he prncpal characerscs of SRP. We deermne he samplng nerval followng n. (a), f a curren sae j s equal o z or z. In he case he curren sae j s he sae u he dsrbuon of he remander ζ θj = ζ θu of he oupu wang me depends on he pas (z или z ) and on he pas me θ. The pdf h u ( ) of he oupu momen ζ u (ζ u = τ u + η u ) from he sae u can be calculaed by he negral akng no accoun (b). We have: mn(, T ) h u ( ) = fu ( x ) wu ( x) dx = α q sαse s, < <. () ISBN: 978--684-69-5 9

We do no wre he expressons for he coeffcens q s because hey have a cumbersome form. The pdf of he random varable ζ θu (f ζ u > θ) can be wre usng () by he formula h θu ( ) =c h u (θ + ), < < () where c s a normalzed consan. We noe ha he dsrbuon funcon F Σ (T;u,k,θ ) depends on he pas sae, me θ and he nex sae k. Ths funcon s deermned by he negral: F Σ (; u,k,θ ) = P{ζ θu + η k < } = = P{ ηk < x} hθu ( x ) dx () We do no gve he resul of he negraon because s raher complex. The fnal expresson for he samplng nerval T u T u (θ,) can be obaned by a numercal soluon of he equaon (5). Ths nerval depends on he level γ, he prevous sae and me θ. The pdf of he jump momen τ uk from u no k s deermned by he formula: α θ µ w uk () = = с qsαse s e sk, () s = < < T u, T u T u (θ, ). µ sk = α s - α k,,k =,, On he base of () one can oban he condonal mean of he jump momen (hs s he esmaon of he jump momen) and he condonal varance (hs s he qualy of he esmaon). 4. Example In order o llusrae he nfluence of sayng nervals on he SRP, would be beer o choose dfferen values of average nervals of nal flows (or denses). Namely, we pu: T = 5.6, T = 6., T =, T =8. ffu(, ) u z fu(, ).5..5 4 Fg.. Graphs of he condonal pdf f ( ν u z) for wo values z (=; ). Takng no accoun he dependence of sayng nervals ν u from dfferen saes z (=; ), we mark he average nervals n he sae u n such manner: T z. Then he average nervals of Tu( z ) and u( ) he process ς ( ) are characerzed by he followng values: T = 5.6, T ( z ) = 8.6, u u T z = 7.4, T =8. In Fg., graphs of he condonal pdf f ( u z) ν for wo values z (=; ). The connuous lne s corresponded o he sae z and he doed lne o he sae z. Le us choose he requred probably of a sae omsson as γ =.. The nvesgaed realzaon ς ( ) s he consequence of saes z =, u=.5, z =, u=.5, z =. We pu he nal value ς = =. The concree correspondng sayng mes are equal: 5.; 5;.; 5.. Calculaons of he samplng nervals and are characerzed by he followng resuls: ) T( z ) =.; T z =.. ) T( u z ) =.,.5,.8,.,.,.5,.6; T( u z ) =.4,.5,.6. ISBN: 978--684-69-5

4 ξ ( ) ζ() ξ n 4 5, n Fg. 4. One realzaon of he sum process and he se of samples In Fg 4, one realzaon of he process ς( ) s presened. Besdes hs, here are some samples desgnaed by crosses. As one can see, he samplng nervals T z and T z are consan and hey don changed n me because boh pdf of he sayng mes are exponenal. Bu he nervals are dfferen, because hey depend on pdf of sayng mes n consdered and fuure saes. The nervals n he sae u are dfferen as well and hey are changed n me. However hs dependence s raher small. Ths effec s deermned by prevous sae z or z and by he condonal pdf of he sayng mes. 5. Conclusons The deermnaon of he requred samplng nerval s a very mporan procedure n he SRP sascal descrpon of he random processes wh jumps. References: [] B.J. Redman, D.G. Lampard. Sochasc Samplng of a Bnary Random Process. IEEE Transacons on Crcu Theory. March 96, pp. - 4. [] B.Lacaze. Perodc b-samplng of saonary processes. Sgnal Processng, V. 68 (998), pp.8-9. [] V. Kazakov. Paen No 75645 USSR, (98);. Paen No 5, USSR, (98). [4] V.Kazakov, Y. Gorsky, Samplng- Reconsrucon of Bnary Markov Process.- Radelecroncs and Communcaon Sysems, Vol. 49, No, 6, pp. 6-. [5] Y. A. Gorsky, V.A.Kazakov. Samplng and Reconsrucon of Markov Processes wh lmed se of saes. Journal of Compuer and Sysem Scences Inernaonal., No, pp. 6-. [6] V. Kazakov. The samplng-reconsrucon Procedure wh a Lmed Number of Samples of Sochasc Processes and Felds on he Bass of he Condonal Mean Rule, Elecromagnec Waves and Elecronc Sysems. Vol., # -, 5, pp. 98-6. [7] V. Kazakov, D. Rodrguez. Samplng Reconsrucon Procedure of Gaussan Processes wh Jer Characerzed by Bea Dsrbuon. IEEE Transacons on Insrumenaon and Measuremen. Vol. 56, 7, #5, Ocober, pp. 84-84. [8] V.Kazakov. The reconsrucon of Gaussan processes realzaons wh an arbrary se of samples. Recen Researches n Telecommuncaons, Informacs, Elecroncs & Sgnal Processng, -h WSEAS Inernaonal Conference on Sgnal Processng, (SIP ), Lanzaroe, Canary Islands, Span, May 7-9,, pp.-5 [9] V.Kazakov, Y.Olvera. Samplng-reconsrucon procedures of non Gaussan processes by wo algorhms. Inernaonal Journal of Crcus, Sysems and Sgnal Processng, (Norh Alanc Unversy Unon - NAUN), Issue 5, Vol. 5,, pp. 557-564. [] V. Kazakov. General problems of he Samplng- Reconsrucon Procedures of Random Process Realzaons. Recen Researches n Appled Informacs & Remoe Sensng. Proceedngs of he h WSEAS Inernaonal Conference on Appled Compuer Scence (ASC-). Penang, Malaysa, Ocober -5,, pp. 98. ISBN: 978--684-69-5