Recursive Kalman-type optimal estimation and detection of hidden Markov chains

نویسندگان

  • Enzo Baccarelli
  • Roberto Cusani
چکیده

Efficient algorithms for computing the ‘a posterior? probabilities (APPs) of discrete-index finite-state hidden Markov sequences are proposed. They are obtained by reducing the APPs computation to the optimal nonlinear minimum mean square error (MMSE) estimation of the noisily observed sequences of the indicator functions associated with the chain states. Following an innovations approach, finite-dimensional and recursive Kalman-like ‘filter’ and ‘smoothers’ for the Markov chain state sequence are thus obtained, and exact expressions of their MSE performance are given. The filtered and smoothed state estimates coincide with the corresponding APP sequences. Finite-dimensional MMSE nonlinear filter and smoothers are also given for the so-called ‘number of jumps’ and for the ‘occupation time’ processes associated with the Markov state sequence. Zusammenfassung Es werden effiziente Algorithmen zur Berechnung der ‘a posteriori’ Wahrscheinlichkeiten (APPs) von verdeckten Markoff-Folgen mit diskretem Index und endlichen Zustlnden vorgestellt. Sie werden hergeleitet, indem die Berechnung der APPs auf die im Sinne des kleinsten mittleren Fehlerquadrates (MMSE) optimale nichtlineare Schatzung der verrauscht beobachteten Folgen von Indikatorfunktionen zurlckgeftihrt wird, die mit den Kettenzustanden verkniipft sind. In Anlehung an den Innovationsansatz werden fiir die Zustandsfolge der MarkotIkette endlichdimensionale, rekursive ‘Filter’ und ‘Glatter’ vom Kalmantyp hergeleitet und exakte Ausdriicke fur ihren mittleren quadratischen Fehler angegeben. Die gefilterten und geglitteten Zustandsschatzwerte stimmen mit den entsprechenden APP Folgen iiberein. Es werden such endlichdimensionale, nichtlineare MMSE Filter und Glitter fur die Prozesse der sogenannten ‘Sprunganzahl’ und der sogenannten ‘Besetzungszeit’ angegeben, die mit der Zustandsfolge der MarkotIkette verbunden sind. Rbumk On propose ici des algorithmes pour calculer les probabilites ‘a posterior? (PAP) de sequences de Markov Cachees a Ctat fini et a index discret. Ces algorithmes sont obtenus en reduisant le calcul des PAP a l’estimation optimale de l’erreur quadratique moyenne minimale (EQMM) nonlineaire des sequences bruitees observees des fonctions d’indicateur associees aux etats de la chaine. On obtient alors, par l’approche des innovations, un ‘filtre’ et des ‘lisseurs’ rtcursifs et de dimension finie pour la sequence d’etat de Markov tres proches de ceux de Kalman, et l’expression *Corresponding author. Tel.: + 39 6 44585 859; fax: + 39 6 4873300; e-mail: [email protected]. 0165-1684/96/$15.00 6 1996 Elsevier Science B.V. All rights reserved PII SO 165-l 684(96)00030-8 56 E. Baccarelli, R. Cusani / Signal Processing 51 (1996) 55-64 exacte de leurs performances EQM est do&e. Les estimees filtries et lissees comcident avec les sequences PAP correspondantes. Un filtre et des lisseurs nonlineaires de dimension finie et a EQMM sont Cgalement don& pour les processus que l’on appelle ‘le nombre de sauts’ et ‘le temps d’occupation’ associis a la sequence d&tat de Markov.

برای دانلود رایگان متن کامل این مقاله و بیش از 32 میلیون مقاله دیگر ابتدا ثبت نام کنید

ثبت نام

اگر عضو سایت هستید لطفا وارد حساب کاربری خود شوید

منابع مشابه

Empirical Bayes Estimation in Nonstationary Markov chains

Estimation procedures for nonstationary Markov chains appear to be relatively sparse. This work introduces empirical  Bayes estimators  for the transition probability  matrix of a finite nonstationary  Markov chain. The data are assumed to be of  a panel study type in which each data set consists of a sequence of observations on N>=2 independent and identically dis...

متن کامل

Recursive Estimation of Gauss-Markov Random Fields Indexed over 1-D Space

This paper presents optimal recursive estimators for vector valued Gauss-Markov random fields taking values in R and indexed by (intervals of) R or Z. These 1-D fields are often called reciprocal processes. They correspond to two point boundary value fields, i.e., they have boundary conditions given at the end points of the indexing interval. To obtain the recursive estimators, we derive two cl...

متن کامل

Particle Filtering in Pairwise and Triplet Markov Chains

The estimation of an unobservable process x from an observed process y is often performed in the framework of Hidden Markov Models (HMM). In the linear Gaussian case, the classical recursive solution is given by the Kalman filter. On the other hand, particle filters provide approximate solutions in more complex situations. In this paper, we propose two successive generalizations of the classica...

متن کامل

Intrusion Detection Using Evolutionary Hidden Markov Model

Intrusion detection systems are responsible for diagnosing and detecting any unauthorized use of the system, exploitation or destruction, which is able to prevent cyber-attacks using the network package analysis. one of the major challenges in the use of these tools is lack of educational patterns of attacks on the part of the engine analysis; engine failure that caused the complete training,  ...

متن کامل

Optimal HMM

In this paper conditional hidden Markov model (HMM) lters and conditional Kalman lters (KF) are coupled together to improve demodulation of di erential encoded signals in noisy fading channels. We present an indicator matrix representation for di erential encoded signals and the optimal HMM lter for demodulation. The lter requires O(N) calculations per time iteration, where N is the number of m...

متن کامل

ذخیره در منابع من


  با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید

عنوان ژورنال:
  • Signal Processing

دوره 51  شماره 

صفحات  -

تاریخ انتشار 1996