Genişletilmiş kalman süzgeci ile pasif akustik hedef hareket analizi

Başlık çevirisi mevcut değil.

Tez No: 39490
Yazar: KEMAL EVCİOĞLU
Danışmanlar: PROF.DR. A. HAMDİ KAYRAN
Tez Türü: Yüksek Lisans
Konular: Elektrik ve Elektronik Mühendisliği, Electrical and Electronics Engineering
Anahtar Kelimeler: Belirtilmemiş.
Yıl: 1994
Dil: Türkçe
Üniversite: İstanbul Teknik Üniversitesi
Enstitü: Fen Bilimleri Enstitüsü
Ana Bilim Dalı: Belirtilmemiş.
Bilim Dalı: Belirtilmemiş.
Sayfa Sayısı: 158

Özet

ÖZET Bu araştırmada bir denizaltının ürettiği gürültüleri pasif konumda gözleyerek denizaltının hareketine ait bilgilerin kestirimi yapılmaktadır. Yapılan analiz literatürde Pasif Akustik Hedef Hareket Analizi olarak bilinmektedir. Bir hedefin izlenebilmesi için mesafe bilgisinin hassas olarak elde edilebilmesi gerekir. Hassas mesafe bilgisi aktif gönderme yaparak elde edilebilir. Ancak Aktif gönderme yapan gemi, kendisiyle ilgili bilgileri de göndermektedir. Bu tezde gönderme yapılmadan mesafe bilgisinin otomatik olarak elde edilmesi amaçlanmaktadır. Otomatik analizde adaptif özellik taşıyan Genişletilmiş Kalman Süzgeci kullanılmaktadır. Genişletilmiş Kalman Süzgeci'yle yapılan analizde sistem modeli iki altmodelle ele alınmaktadır. Sistem modelinin durum uzay altmodeli doğrusaldır. Bu model hedef hareketlerini temsil eden modeldir. Hedeften gözlenen işaretlerle de gözlem altmodeli oluşturulmaktadır. Hedef ve kendi gemimiz arasındaki göreceli hareket nedeniyle hedeften gözlenen işaretlerin dengeli frekans bileşenlerinde kayma oluşur.Bu oluşum dopler oluşumu olarak bilinir. Gözlem altmodeli dopler kaymalı işaretler ve bu işaretlerin algılandığı kerteriz gözlemleriyle oluşturulmuştur. Gözlem altmodeli doğrusal değildir. Genişletilmiş Kalman Süzgeci'yle yapılan analizde gözlem altmodeline doğrusallaştırma uygulanmaktadır. Genişletilmiş Kalman Süzgeci'nde gözlem gürültüsü olarak tanımlanan ölçüm hatalarının sıfir ortalamalı beyaz Gauss gürültüsü olduğu varsayılmaktadır. Tezde gerçek yaşama uygun senaryolarla MATLAB 4.0 FOR WINDOWS programı kullanılarak analiz yapılmaktadır. Senaryolarda kullanılan teknik, literatürde MONTE CARLO ANALİZİ olarak bilinmektedir. Sonuç olarak çeşitli senaryolarla süzgeç performansı incelendiğinde kullanılan tekniğin hedef hareketlerinin kestiriminde başarılı olduğu görülmektedir. Başlangıç kestiriminin duyarlı yapılması ve gözlenebilirliğin sağlanması koşuluyla süzgeç dengeli kestirimler üretmektedir. vıı

Özet (Çeviri)

SUMMARY Target Motion Analysis is to determine the parameters related to the motion of a target using active or passive observations. These observations can be done by using radar, sonar or any other sensor. Active measurements carry precise knowledge about the range of a target which is the most important factor in a tracking process. An unavoidable disadvantage in active operation is that of disclosing the own ship data. In this case the target to be detected can listen our information and develop a sudden attack. The target can realize a passive TMA by passive observations. By passive measurements it is not possible to get a correct range information of the target. What is precise to get in a passive operation is bearing information. The great advantage in passive analysis is that the operation is secret as expected. This advantage brings strategic superiority in tactical situations in case you obtain range information somehow. Acoustic signals coming from the target as the target moves carries information about all the parameters related to the target motion including range information. Unfortunately these signals bear also noise that hides the intelligent parameters.What must be done in this case is to exploit Linear Estimation Theory. Acoustic signals are born and propagate in a nonlinear time and space variant ambient. This important fact must be considered in modelling. That is Linear Estimation Theory cannot be used directly. Some approximation must be done to the real model. Approximations must be performed without losing much information. TMA operations can be realized manually or automatically. Old TMA's were manual. After the adaptive filtering techniques have been developed it's become easy to filter out the noise while estimating the necessary information as powerful as possible. In this thesis TMA is performed using passive observations with only one sensor. The sensor is a sonar which has the necessary signal to noise ratio to detect the signals that carries information about the parameters of a target Although most of the previous passive TMA studies advise using more than one sensor this work searchs an economic but an efficient algorithm using only one sensor. vniThe passive analysis area is underwater. In a nonlinear ambient passive acoustic bearing and passive doppler shifted noisy observations are used with a famous adaptive filter. The filter is Extended Kalman Filter (EKF) which is an extended version of Kalman Filter. Extended Kalman Filter serves to analyse nonlinear observations in an ongoing estimation and tracking process. Extended Kalman Filter offers the possibility of doing recursive estimations. The estimation process is automatically carried out.EKF gives possibility to do a real time analysis in case the algorithm is applied to the real world phenomena. So the passive tracking technic used in this thesis is automated with EKF. Not all the algorithms used in passive acoustic TMA are automatic and recursive. In this work cartesian coordinates are used for the passive target motion analysis. The model is being established in a plane. So depth dimension is abandoned. Most of the previous acoustic passive TMA studies were devoted to the Bearings Only Analysis. Passive bearing observations were used to estimate the range of the target. As this work uses EKF let's talk about the passive automated algorithms using EKF. Bearings Only Analysis with Extended Kalman Filter were not succesful at the beginning. There are many reasons for the filter not to track efficiently.These problems were isolated in later studies and extra powerful methods for EKF were developed for tracking only with bearings observations. Although some accessories imported to EKF using bearing only model,inherent problems created obstacles in estimating range information. Solution uniqeness problem is the reason. There are infinite number of range solutions that gives the same bearing. So it is necessary to initialize the filter with a precise range information.With a precise knowledge the filter estimates efficiently in a linear target motion. Unfortunately in some cases the filter diverges. When the target executes a turn filter statistics change and the filter can't follow maneuver. Another important concept in an adaptive estimation is the observability of the system. When the observable conditions are lost the filter becomes instable. Observability must be preserved during the estimation process. Observability is a general condition that must always be available. It is strictly related to the model of the filter and the filter stability. The bearings only TMA technic needs own ship maneuver to preserve observability. rxAs a result the TMA model is very important. A model must reflect the real phenomena as far as possible. It isn't only the bearing observation that is used in estimating the position information of a target. There are many methods that exploit the characteristic properties of a signal for estimating the parameters of a target that relate to the motion. Generally intensity, frequency, signal time delay and bearing information are used for this purpose. Most of the researchers prefer to use more than one information relating to the motion of a target.But unfortunately it is not possible to use all kinds of information at the same time. There occurs technical problems. So a suitable combination is preferred. In this thesis passive doppler frequency information is used together with the bearing information.Received signals coming from a bearing are examined also in frequency domain. A real time frequency analyzer is assumed and a real time analysis is done. The frequency model in this thesis is a new model. Basically the idea is that; An underwater target creates useful noise that reflects the finger print. It is assumed that this fingerprint has a bandwidth that has a center frequency. By a frequency analyser in the sonar system that is capable to detect the signals The noisy frequency measurements are performed in real time. The operator examines the noisy signals from its bearing. As the target moves the bearing information changes together with the frequency information. Frequency information has an important phenomena. The stable frequency components shift during the relative motion of the target and own ship. This phenomena is known as the doppler effect. Also the center frequency of the finger print of the target shifts.For there are many nonlinear effects in the acoustic observation area the operator cannot detect the center frequency but can guess it. He can import this guess as an initial guess in the estimation process. In a planar motion doppler shifted center frequency model together with the zero mean Additive White Gaussian Measurement error frequency and the received signal bearing model are used. In the model the bearing and the frequency measurement errors are modeled as AWGN.They are zero mean and a known variance. The measurement errors must be AWGN for the Extended Kalman Filter to provide optimal estimations. EKF is a suboptimal filter. The reason comes from the linearization process. The linearization must approximate the real nonlinear model.In this thesis the linearization process uses Taylor approximation.The nonlinear frequency and the bearing model are opened to the series around the best possible estimate.The first derivative of the nonlinear observation model is used.This is called the first order approximation. The state space model used in EKF is linear.Besides the linear motion of a target the random changes that occur during the costant speed-constant route motion are taken into account by modelling these changes as random impulses. The target maneuver can be examined by the statistical changes. An adaptive gate is established in this thesis to track a target that executes a turning maneuver. The algorithm used in this thesis needs a succesful initialization. Once the operator starts the algorithm no further interference is needed. At the beginning of the thesis an introduction to TMA is done. Chapter 1 is dedicated to the explanation of TMA and the contents of the thesis. In Chapter 2 the TMA methods are introduced to give a previous knowledge to the reader.This introduction allows the reader to compare these methods with the method used in this thesis. Chapter 3 is devoted to the introduction of hydrodynamics. Sound, sonar concept and noise phenomena in underwater acoustics are given briefly. In Chapter 4 the system model to be used in passive acoustic target motion analysis with EKF is established.The system model is obtained in two steps.First the observation model and later the state space model are built. In Chapter 5 Kalman Filter, In Chapter 6 Extended Kalman Filter are introduced to the reader. In Chapter 7 the analysis is performed and the algorithm used in the program is obtained. In Chapter 8 the simulations are presented. The results obtained in scenarios are examined. There are 4 scenarios presented in this chapter. The scenarios simulate own ship and the target motion.The analysis used in the scenarios is known as Monte Carlo Analysis.Random AWGN noise is injected to the bearing and frequency observations obtained from the target as the target moves as if the real data about the motion of the target and own ship are known during the whole tracking process. The simulation program is MATLAB 4.0 FOR WINDOWS.The first step in the scenario is to create the real target observation data according to the established scenario.The second step is carried out by the program. In this step zero mean and known variance AWGN is injected. The noise here represents the frequency and doppler measurement errors. In the third step adaptive filtering is performed with EKF.Once the program is initialized by the operator the algorithm automatically XIcarries on the analysis. The last step in the program is to plot the estimated position data of the target. Submarine as a target executes a turn in the first scenario.The filter senses the turn maneuver. The second scenario is adjusted so as to collapse the bearing observability. The third scenario is adjusted so as to maintain observability both in frequency and the bearing observations. In the fourth scenario the system is completely unobservable. Under this condition the filter collapses.For healing the system observability is reestablished. As a result the filter becomes stable. EKF used in this thesis provides succesful and stable estimates with good initial conditions. As the initialization performance decreases the estimation performance decreases. As mentioned previously a new doppler model related to the relative motion of the target and the observer is presented.The bearing model is as known. Simulations that are carried out using only frequency observation give also succesful results in case observability is maintained. An interesting result is obtained with the center frequency estimation. The filter always converges. The convergence velocity is related to the initial covariance matrix. The reason why the filter provides wonderful frequency estimates can be grasped by looking at the model. Although the model is nonlinear the filter sees the center frequency in linearity during the relative motion. The most important problem comes inherently from the solution uniqueness of the model. Both the frequency and bearing have infinitely many solutions.There are many parameters that creates the same bearing and the same frequency. The model problem must be solved. To my opinion EKF is not responsible for the inefficient estimations but the model chosen and the linearization process.If a powerful model and linearization are used better results can be hoped provided that the observability condition is maintained. Finally, The algorithm presented in this thesis has a relative success among other estimation algorithms.Further study must be done as always. xn

Benzer Tezler

Tez No
318819
Comparison of emitter localization methods with a moving platform in three dimensions
Hareketli tek platform ile üç boyutlu hedef konum belirleme tekniklerinin karşılaştırılması
BURCU TUFAN
Yüksek Lisans
İngilizce
2012
Elektrik ve Elektronik Mühendisliği Orta Doğu Teknik Üniversitesi
Elektrik-Elektronik Mühendisliği Ana Bilim Dalı
PROF. DR. TEMEL ENGİN TUNCER
Tez No
554417
Tactical missile guidance and homing loop applications
Taktiksel füze güdümü ve hedef izleme döngüsü uygulamaları
ERDEM TURAN
Yüksek Lisans
İngilizce
2019
Savunma ve Savunma Teknolojileri İstanbul Teknik Üniversitesi
Savunma Teknolojileri Ana Bilim Dalı
PROF. DR. CENGİZ HACIZADE
Tez No
199498
Genişletilmiş kalman süzgeci ile gezgin robot konumunun belirlenmesi
Mobile robot localization with extended kalman filter
KORAY HASAN HATİPOĞLU
Yüksek Lisans
Türkçe
2007
Elektrik ve Elektronik Mühendisliği Başkent Üniversitesi
Elektrik-Elektronik Mühendisliği Ana Bilim Dalı
Y.DOÇ.DR. HAMİT ERDEM
Tez No
251642
Tekerlek yol sürtünme katsayısı için gözetleyici tasarımı
Observer design for tire road friction
MÜMİN AKTAŞ
Yüksek Lisans
Türkçe
2008
Elektrik ve Elektronik Mühendisliği İstanbul Teknik Üniversitesi
Mekatronik Mühendisliği Ana Bilim Dalı
PROF. DR. LEVENT GÜVENÇ
Tez No
318967
Short range thrusting projectile tracking
Kısa menzilli itkiye sahip mermilerin takibi
OZAN ÖZGÜN BİLGİN
Yüksek Lisans
İngilizce
2012
Elektrik ve Elektronik Mühendisliği Orta Doğu Teknik Üniversitesi
Elektrik-Elektronik Mühendisliği Ana Bilim Dalı
PROF. DR. MÜBECCEL DEMİREKLER

Geri Dön