SlideShare a Scribd company logo
Av-738
Adaptive Filter Theory
Lecture 4- Optimum Filtering
[Kalman Filters]
Dr. Bilal A. Siddiqui
Air University (PAC Campus)
Spring 2018
Kalman Filters
• Kalman filter is both an optimal and recursive (adaptive filter)
• It has a unique mathematical formulation (i.e. state space concepts)
• It is solved recursively. It is updated using previous state estimates and
new data (innovation), therefore no need for storage.
• It is mostly implemented in the discrete form, and easily implementable
on digital computers
• It gives a unifying framework for RLS filters.
• Developed by Rudolf Kalman in 1969 []
Time Domain Description of Dynamical
Sytems
• One way to describe a linear dynamical system’s evolution is in the
frequency (or transform) domain by means of transfer functions.
• Another equivalent way is to define it in time-domain.
• If the system is linear, it can be described in State-Space platform.
• A set of variables that can uniquely define the time evolution of a
dynamical system is called the state-vector 𝑋. Elements of that vector are
the system’s states 𝑥.
• In many cases, the states themselves are not measured directly. Some
combination of states may be measured. These are measurements 𝑦
mostly contaminated by measurement noise.
• The state space model of the process is a linear combination of inputs and
previous values of states. In other words, states have memories (or
integrations)
Linear Dynamical
System
• v1 is process “noise” which is a measure of how inaccurate our linear model given by
𝑥 𝑛 + 1 = 𝐹 𝑛 + 1, 𝑛 𝑥 𝑛 + 𝑣1(𝑛) is. If there are M states, 𝑂 𝐹 = 𝑀 × 𝑀 and 𝑂 𝑥 =
𝑀 × 1, 𝑂 𝑣 = 𝑀 × 1
• F is known as the state transition matrix relating states at n+1 and n. Process noise is
measured as zero mean white noise (its correlation matrix Q1 is diagonal). Let the system be
time-invariant. The 𝐹 𝑛 + 1, 𝑛 = 𝐹 for all n.
• Measured quantities are linear combinations of states 𝑦 𝑛 = 𝐶 𝑛 𝑥 𝑛 + 𝑣2 𝑛 . For N
measurements, 𝑂 𝑦 = 𝑁 × 1, 𝑂 𝐶 = N × 𝑀 and 𝑣2 is a Nx1 measurement noise vector,
which may represent sensor noise and linearization errors. Its correlation matrix Q2 is also
diagonal.
• It is assumed that process and measurement noises are uncorrelated. 𝐸 𝑣1 𝑛 𝑣2
𝐻
(𝑛) = 0
The Kalman Filtering Problem
The Innovation Process
• Let 𝑦 𝑛−1 denote all measurements uptil n-1, i.e. 𝑦 𝑛−1 = 𝑦 𝑛 − 1 … 𝑦 1
• Let 𝑦(𝑛|𝑦 𝑛−1) be the one-step prediction or best estimate (min mean square error sense) of y(n) using all
measurement made till n-1.
• We define the prediction error 𝛼 𝑛 = 𝑦 𝑛 − 𝑦(𝑛|𝑦 𝑛−1) as process “innovation”, which represents the new
information (innovation) in the data observed at time n.
• Properties:
Innovation Propagation
• Suppose we have the more generic system 𝑥 𝑘 = 𝐹𝑥 𝑘 + 𝑣1(𝑛), then we can write a recursive relation
𝑥 𝑛 + 1 = 𝐹 𝑛
𝑥 0 +
𝑖=0
𝑛−1
𝐹 𝑛−1−𝑖
𝑣1 𝑖
• multiplying v2(n) and taking expectation
𝐸 𝑥 𝑛 + 1 𝑣2
𝐻
𝑛 = 𝐹𝐸 𝑥 0 𝑣2
𝐻
𝑛 + 𝐸
𝑖=0
𝑛−1
𝐹 𝑛−1−𝑖
𝑣1 𝑖 𝑣2
𝐻
𝑛
• Since noise v2 is uncorrelated with initial state x(0) and noise v1 this means 𝐸 𝑥 𝑘 𝑣2
𝐻
𝑛 for 0 ≤ 𝑘 ≤ 𝑛.
• Since 𝑦 𝑛 = 𝐶𝑥 𝑛 + 𝑣2 𝑛 , we can also prove that 𝐸 𝑦 𝑘 𝑣2
𝐻
𝑛 = 0 for 0 ≤ 𝑘 ≤ 𝑛 − 1 , i.e. v2 is orthogonal to past measurements.
• Therefore, 𝑦 𝑛 𝑦 𝑛−1 = C 𝑥(𝑛|𝑦 𝑛−1) as v2 cannot be predicted using past measurements.
• Therefore we can write innovations as 𝛼 𝑛 = 𝑦 𝑛 − C 𝑥(𝑛|𝑦 𝑛−1)
• Substitute measurement equation,
𝛼 𝑛 = 𝐶𝑥 𝑛 + 𝑣2 𝑛 − C 𝑥 𝑛 𝑦 𝑛−1 = 𝐶𝜖 𝑛, 𝑛 − 1 + 𝑣2(𝑛)
where 𝜖 𝑛, 𝑛 − 1 = 𝑥 𝑛 − 𝑥 𝑛 𝑦 𝑛−1 is the predicted state-error vector at n using the measurements till n-1
We can prove that 𝜖 𝑛, 𝑛 − 1 is orthogonal to v1 and v2 for for all times less than n.
A more generic model
• Suppose we have a more generalized process model of a discrete-
time linear system
𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1 + 𝑤 𝑘−1
where uk is a known and deterministic input and wk is Gaussian zero-
mean white noise with covariance Qk.
• How does the mean of the state xk change with time?
𝑥 𝑘 = 𝐸 𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1
• Now, how does the covariance of xk change with time?
Covariance Propagation
• Since
• Note that innovation or state error 𝜖 𝑘 = 𝑥 𝑘 − 𝑥 𝑘 is uncorrelated with noise
wk-1 (or process noise at k-1, as we said earlier)
• This means the state covariance matrix propagates as
This is called the discrete-time Lyapunov equation or Stein equation. Also note that since x0 and wi are
Gaussian 𝑥 𝑘 = 𝑁( 𝑥 𝑘, 𝑃𝑘)
Example (Predator + Prey)
Homework 2 (a)
• Simulate this model using Matlab
• But we could also have found the steady states and covariances
directly. Prove that the expression for steady states analytically are:
Derivation of the Kalman Filter
Our goal is to estimate the state xk based on our knowledge of the system
dynamics and the availability of the noisy measurements yk. The amount of
information that is available to us for our state estimate varies depending on
the particular problem that we are trying to solve.
Posteriors and Priots
• If we have all of the measurements up to and including time k available for use in
our estimate of xk, then we can form an a posteriori estimate, which we denote
as 𝑥 𝑘
+
• If we have all of the measurements before (but not including) time k available for
use in our estimate of xk , then we can form an a postiori estimate, which we
denote as 𝑥 𝑘
−
Process of Estimation
• After we process the measurement at time (k-1), we have an
estimate Of xk−1(denoted xk−1
+
) and the covariance of that estimate
(denoted Pk−1
+
).
• When time k arrives, before we process the measurement at time k,
we compute an estimate of 𝑥 𝑘(denoted 𝑥 𝑘
−
) and
the covariance of that estimate (denoted Pk
−
).
• Then we process the measurement at time k to refine our estimate of
𝑥 𝑘. The resulting estimate is denoted 𝑥 𝑘
−
, with covariance Pk
+
.
Step 1: Time-Update of state and covariance
estimate
• We begin the estimation process with at, our best estimate of the initial state x0.
Given 𝑥0
+
, how should we compute 𝑥1
−
? We want to set 𝑥1
−
= 𝐸 𝑥1 , but we know
𝑥0
+
= 𝐸 𝑥0 . Therefore, using 𝑥 𝑘 = 𝐸 𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1
𝑥1
−
= 𝐹0 𝑥0
+
+ 𝐺0 𝑢0, or generalizing 𝑥 𝑘
−
= 𝐹𝑘−1 𝑥 𝑘−1
+
+ 𝐺 𝑘−1 𝑢 𝑘−1
• This is called the time update equation for 𝑥
• From time 𝑘 − 1 +
to time 𝑘−
, the state estimate propagates the same way that the
mean of the state propagates.
• This makes sense intuitively. We do not have any additional measurements available to
help us update our state estimate between time 𝑘 − 1 +
and time 𝑘−
, so we should
just update the state estimate based on our knowledge of the system dynamics.
• Using similar arguments, we can write for covariance
Step 2: Measurement-update of state
• Given 𝑥 𝑘
−
, how should we compute 𝑥 𝑘
+
?
• A general form of filter update is
i.e. some part of the innovation is added to the previous estimate. K is
some gain matrix, known as Kàlmàn filter gain.
• In our case, 𝑥 𝑘
+
= 𝑥 𝑘
−
+ 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘
−
• To find Kalman gain, the derivation is a bit more complex.
• We begin by noting that the state error vector propagates as
𝐸 𝜖 𝑘
+
= 𝐸 𝑥 𝑘 − 𝑥 𝑘
+
= 𝐸 𝑥 𝑘 − 𝑥 𝑘
−
− 𝐾𝑘 𝐻𝑥 𝑘 + 𝑣 𝑘 − 𝐻 𝑘 𝑥 𝑘
−
= 𝐸 𝜖 𝑘
−
− 𝐾𝑘(𝐻 𝑘 𝜖 𝑘
−
+𝑣 𝑘)
⇒ 𝐸 𝜖 𝑘
+
= 𝐼 − 𝐾𝑘 𝐻 𝑘 𝐸 𝜖 𝑘
−
− 𝐾𝑘 𝐸 𝑣 𝑘
Step 2: Measurement-update of covariance
• Estimation error covariance Pk propagates as
=
𝑃𝑘
+
= 𝐸 𝜖 𝑘
+
𝜖 𝑘
+ 𝑇
= E 𝐼 − 𝐾𝑘 𝐻 𝑘 𝐸 𝜖 𝑘
−
− 𝐾𝑘 𝐸 𝑣 𝑘 … 𝑇
= 𝐼 − 𝐾𝑘 𝐻 𝑘 E 𝜖 𝑘
−
𝜖 𝑘
−𝑇
𝐼 − 𝐾𝑘 𝐻 𝑘
𝑇
+ 𝐾𝑘 𝐸 𝑣 𝑘 𝑣 𝑘
𝑇
𝐾𝑘
𝑇
−
𝐾𝑘E 𝑣 𝑘 𝜖 𝑘
−𝑇
𝐼 − 𝐾𝑘 𝐻 𝑘
T + 𝐼 − 𝐾𝑘 𝐻 𝑘 E 𝜖 𝑘
−
𝑣 𝑘
𝑇
𝐾𝑘
𝑇
• Since estimation error at 𝑘−
is independent of measurement noise at k
𝑃𝑘
+
= 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
𝐼 − 𝐾𝑘 𝐻 𝑘
𝑇 + 𝐾𝑘 𝑅 𝑘 𝐾𝑘
𝑇
• After some elaborate steps, this can also be shown as (prove as part of HW 2)
𝑃𝑘
+
= 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
• This is the recursive formula for the covariance of estimation error. This is consistent with intuition
that as the measurement noise increases (i.e. 𝑅 𝑘 increases) the uncertainty in our estimate also
increases (i.e., 𝑃𝑘 increases).
• Note that Pk should be positive definite since it is a covariance matrix, and the form of 𝑃𝑘
+
guarantees that Pk will be positive definite, assuming that Pk-1 and Rk are positive definite.
Finding Kalman Gain
• Let’s find the optimal Kalman gain matrix by minimizing the posterior error
covariance at time k (equivalent to MSE since 𝑃𝑘
+
= 𝐸 𝜖 𝑘
+
𝜖 𝑘
+ 𝑇
).
𝜕𝑃𝑘
+
𝜕𝐾𝑘
= 0
𝜕
𝜕𝐾𝑘
𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
𝐼 − 𝐾𝑘 𝐻 𝑘
𝑇 + 𝐾𝑘 𝑅 𝑘 𝐾𝑘
𝑇
= 0
−2 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
𝐻 𝑘
𝑇
+ 2𝐾𝑘 𝑅 𝑘 = 0
This can be solved for KK as
𝐾𝑘 = 𝑃𝑘
−
𝐻 𝑘
𝑇
𝐻 𝑘 𝑃𝑘
−
𝐻 𝑘
𝑇
+ 𝑅 𝑘
−1
Using some more elaborate steps (prove as part of HW 2)
𝐾𝑘 = Pk
+
Hk
T
Rk
−1
Kalman filter summarized
• We summarize the discrete-time Kalman filter by combining the above
equations into a single algorithm.
Comments about implementation
• The 𝑃𝑘
+
= 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
𝐼 − 𝐾𝑘 𝐻 𝑘
𝑇
+ 𝐾𝑘 𝑅 𝑘 𝐾𝑘
𝑇
equation of covariance
update is more stable and robust than 𝑃𝑘
+
= 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘
−
equation, even
though the latter is less computationally intensive. The first expression
ensures Pk is symmetric positive definite, the latter does not.
• Calculation of 𝑃𝑘
−
, 𝐾𝑘and 𝑃𝑘
+
does not depend on the measurements yk ,
but depends only on the system parameters Fk, Hk, Qk and Rk. Therefore,
Kalman gain Kk can be calculated offline before the system operates and
saved in memory (if statistics Q, F and R are known). In that case, linear
time-invariant Kalman filter is like the Weiner filter and not adaptive.
• In contrast, as we will see later, the filter gain and covariance for nonlinear
systems cannot (in general) be computed offline because they depend on
the measurements. In that case, the Kalman filter has important nonlinear
variants.
Important Properties
• If wk and vk are zero-mean, uncorrelated, and white, then the Kalman
filter is the best linear solution to the above problem, i.e., the Kalman
filter is the best filter that is a linear combination of the
measurements. There may be a nonlinear filter that gives a better
solution, but the Kalman filter is the best linear filter.
• Even if the noise is not Gaussian, the Kalman filter is still the optimal
linear filter. Noise only needs to be white and zero-mean.
• If wk and vk are coloured or correlated, then the KF can be modified.
• For nonlinear systems, various formulations of nonlinear Kalman
filters approximate the solution to the linear problem we solved.
One Step KF Equations
• We will see how the a priori and a posteriori Kalman filter equations
can be combined into a single equation. This may simplify computer
implementation of the equations. We have
𝑥 𝑘+1
−
= 𝐹𝑘 𝑥 𝑘
+
+ 𝐺 𝑘 𝑢 𝑘
• Substituting 𝑥 𝑘
+
= 𝑥 𝑘
−
+ 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘
−
𝑥 𝑘
−
= 𝐹𝑘 𝑥 𝑘
−
+ 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘
−
+ 𝐺 𝑘 𝑢 𝑘
𝑥 𝑘
−
= 𝐹𝑘 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑥 𝑘
−
+ 𝐹𝑘 𝐾𝑘 𝑦 𝑘 + 𝐺 𝑘 𝑢 𝑘
• This shows that the a priori state estimate can be computed directly
from its value at the previous time step, without computing the a
posteriori state estimate in between. We can similarly show
Example
• Suppose we have a noise-free with position r , velocity v, and
constant acceleration a.
Solution
Simulation
%%
clc;clear all;close;
%Kalman filter for position measurement
%AV-738 Spr 18, Dr. Bilal A. Siddiqui
T=5;
R=30^2;
a=0.01;
F=[1 T T^2/2 %position
0 1 T %velocity
0 0 1]; %acceleration
G=0;
H=[0 0 1];% only position being measured
w=0;%process noise
x(:,1)=[0 0 a]';%initial value of state vector
x_post(:,1)=[0 0 0]';%initial estimate of state
vector
P_post(:,:,1)=1E5*eye(3);%initial estimate of error
covariance
for k=2:60
%calculate a priori covariance estimate
P_prior(:,:,k)=F*P_post(:,:,k-1)*F';
%calculate Kalman gain
K(:,:,k)=P_prior(:,:,k)*H'*(H*P_prior(:,:,k)*H'+R)^-1;
%calculate a priori estimate of state
x_prior(:,k)=F*x(:,k-1);
%system evolution
x(:,k)=F*x(:,k-1);%take measurement
y(k)=H*x(:,k)+sqrt(R)*randn;
%calculate innovation
innov(k)=y(k)-H*x_prior(:,k);
%calculate a posteriori estimate of state
x_post(:,k)=x_prior(:,k)+K(:,:,k)*innov(k);
%calculate a posteriori estimate of covariance
P_post(:,:,k)=(eye(3)- ...
K(:,:,k)*H)*P_prior(:,:,k)*+(eye(3)-K(:,:,k)*H)'+...
K(:,:,k)*R*K(:,:,k)';
end
Results
Homework 2 (c)
• First of all, we were able to reduce estimation error below the
measurement noise floor. How was that possible?
• Also, if we start measuring the acceleration directly (use a 1-g noise,
i.e.R=1) instead of the position, we will get a good estimate still, but the
error covariance will keep increasing. Why?
• In practical terms, if we inject position measurements at a slower rate,
what will happen to the error covariance, if we were taking accelerometer
measurements at a higher rate?
• Simulate the error covariance if we measure accelerometer at every
sampling instant, but receive GPS position with noise variance of 10 after
every five sampling instants.
• Comment your answer and analyze the results.
A word of caution
• The theory presented in this lecture (based on “Optimal State Estimation” b Dan
Simon, J. Wiley & Sons 2006) makes the Kalman filter an attractive choice for
state estimation.
• But when a Kalman filter is implemented on a real system it may not work, even
though the theory is correct. Main causes are (a) finite precision arithmetic and
(b) modelling errors.
• The theory presented also assumes that the system model is precisely known. It
is assumed that the F , Q, H, and R matrices are exactly known, and it is assumed
that the noise sequences {wk} and {vk} are pure white, zero-mean, and completely
uncorrelated. If any of these assumptions are violated, as they always are in real
implementations, then the Kalman filter assumptions are violated and the theory
may not work.
•
Some Remedies

More Related Content

What's hot

Kalman Filter
 Kalman Filter    Kalman Filter
Kalman Filter
Mohamed Magdy
 
Adaptive filter
Adaptive filterAdaptive filter
Adaptive filter
Vijay Kumar
 
Kalman filter for Beginners
Kalman filter for BeginnersKalman filter for Beginners
Kalman filter for Beginners
winfred lu
 
Particle filter
Particle filterParticle filter
Particle filter
Mohammad Reza Jabbari
 
Kalman Filter | Statistics
Kalman Filter | StatisticsKalman Filter | Statistics
Kalman Filter | Statistics
Transweb Global Inc
 
Ant colony opitimization numerical example
Ant colony opitimization numerical exampleAnt colony opitimization numerical example
Ant colony opitimization numerical example
Harish Kant Soni
 
14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron
Andres Mendez-Vazquez
 
Wiener Filter
Wiener FilterWiener Filter
Wiener Filter
Akshat Ratanpal
 
Report kalman filtering
Report kalman filteringReport kalman filtering
Report kalman filteringIrfan Anjum
 
Genetic Algorithm by Example
Genetic Algorithm by ExampleGenetic Algorithm by Example
Genetic Algorithm by Example
Nobal Niraula
 
Dsp U Lec04 Discrete Time Signals & Systems
Dsp U   Lec04 Discrete Time Signals & SystemsDsp U   Lec04 Discrete Time Signals & Systems
Dsp U Lec04 Discrete Time Signals & Systems
taha25
 
Practical Swarm Optimization (PSO)
Practical Swarm Optimization (PSO)Practical Swarm Optimization (PSO)
Practical Swarm Optimization (PSO)
khashayar Danesh Narooei
 
Principle of Maximum Entropy
Principle of Maximum EntropyPrinciple of Maximum Entropy
Principle of Maximum Entropy
Jiawang Liu
 
Av 738 - Adaptive Filtering Lecture 1 - Introduction
Av 738 - Adaptive Filtering Lecture 1 - IntroductionAv 738 - Adaptive Filtering Lecture 1 - Introduction
Av 738 - Adaptive Filtering Lecture 1 - Introduction
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Particle swarm optimization
Particle swarm optimizationParticle swarm optimization
Particle swarm optimization
anurag singh
 
Lecture 16 KL Transform in Image Processing
Lecture 16 KL Transform in Image ProcessingLecture 16 KL Transform in Image Processing
Lecture 16 KL Transform in Image Processing
VARUN KUMAR
 
Computing DFT using Matrix method
Computing DFT using Matrix methodComputing DFT using Matrix method
Computing DFT using Matrix method
Sarang Joshi
 
Bayesian Inference and Filtering
Bayesian Inference and FilteringBayesian Inference and Filtering
Bayesian Inference and Filtering
Engin Gul
 

What's hot (20)

Kalman Filter
 Kalman Filter    Kalman Filter
Kalman Filter
 
Kalman filter
Kalman filterKalman filter
Kalman filter
 
Kalman_filtering
Kalman_filteringKalman_filtering
Kalman_filtering
 
Adaptive filter
Adaptive filterAdaptive filter
Adaptive filter
 
Kalman filter for Beginners
Kalman filter for BeginnersKalman filter for Beginners
Kalman filter for Beginners
 
Particle filter
Particle filterParticle filter
Particle filter
 
Kalman Filter | Statistics
Kalman Filter | StatisticsKalman Filter | Statistics
Kalman Filter | Statistics
 
Ant colony opitimization numerical example
Ant colony opitimization numerical exampleAnt colony opitimization numerical example
Ant colony opitimization numerical example
 
14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron14 Machine Learning Single Layer Perceptron
14 Machine Learning Single Layer Perceptron
 
Wiener Filter
Wiener FilterWiener Filter
Wiener Filter
 
Report kalman filtering
Report kalman filteringReport kalman filtering
Report kalman filtering
 
Genetic Algorithm by Example
Genetic Algorithm by ExampleGenetic Algorithm by Example
Genetic Algorithm by Example
 
Dsp U Lec04 Discrete Time Signals & Systems
Dsp U   Lec04 Discrete Time Signals & SystemsDsp U   Lec04 Discrete Time Signals & Systems
Dsp U Lec04 Discrete Time Signals & Systems
 
Practical Swarm Optimization (PSO)
Practical Swarm Optimization (PSO)Practical Swarm Optimization (PSO)
Practical Swarm Optimization (PSO)
 
Principle of Maximum Entropy
Principle of Maximum EntropyPrinciple of Maximum Entropy
Principle of Maximum Entropy
 
Av 738 - Adaptive Filtering Lecture 1 - Introduction
Av 738 - Adaptive Filtering Lecture 1 - IntroductionAv 738 - Adaptive Filtering Lecture 1 - Introduction
Av 738 - Adaptive Filtering Lecture 1 - Introduction
 
Particle swarm optimization
Particle swarm optimizationParticle swarm optimization
Particle swarm optimization
 
Lecture 16 KL Transform in Image Processing
Lecture 16 KL Transform in Image ProcessingLecture 16 KL Transform in Image Processing
Lecture 16 KL Transform in Image Processing
 
Computing DFT using Matrix method
Computing DFT using Matrix methodComputing DFT using Matrix method
Computing DFT using Matrix method
 
Bayesian Inference and Filtering
Bayesian Inference and FilteringBayesian Inference and Filtering
Bayesian Inference and Filtering
 

Similar to Av 738 - Adaptive Filtering - Kalman Filters

lecture_18-19_state_observer_design.pptx
lecture_18-19_state_observer_design.pptxlecture_18-19_state_observer_design.pptx
lecture_18-19_state_observer_design.pptx
AnshulShekhar3
 
Av 738-Adaptive Filters - Extended Kalman Filter
Av 738-Adaptive Filters - Extended Kalman FilterAv 738-Adaptive Filters - Extended Kalman Filter
Av 738-Adaptive Filters - Extended Kalman Filter
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Week_10.2.pdf
Week_10.2.pdfWeek_10.2.pdf
Week_10.2.pdf
Mir Shah
 
Intro to Quant Trading Strategies (Lecture 6 of 10)
Intro to Quant Trading Strategies (Lecture 6 of 10)Intro to Quant Trading Strategies (Lecture 6 of 10)
Intro to Quant Trading Strategies (Lecture 6 of 10)
Adrian Aley
 
Kalman filter demonstration
Kalman filter demonstrationKalman filter demonstration
Kalman filter demonstration
Ahmed Yahia Kallel
 
Av 738- Adaptive Filtering - Background Material
Av 738- Adaptive Filtering - Background MaterialAv 738- Adaptive Filtering - Background Material
Av 738- Adaptive Filtering - Background Material
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Stat 2153 Introduction to Queiueng Theory
Stat 2153 Introduction to Queiueng TheoryStat 2153 Introduction to Queiueng Theory
Stat 2153 Introduction to Queiueng Theory
Khulna University
 
Time series Modelling Basics
Time series Modelling BasicsTime series Modelling Basics
Time series Modelling Basics
Ashutosh Kumar
 
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
AI Robotics KR
 
Kalman Filter Basic
Kalman Filter BasicKalman Filter Basic
Kalman Filter Basic
National Cheng Kung University
 
Lec1 01
Lec1 01Lec1 01
Lecture 23 24-time_response
Lecture 23 24-time_responseLecture 23 24-time_response
Lecture 23 24-time_response
Syed Ali Raza Rizvi
 
Av 738- Adaptive Filtering - Wiener Filters[wk 3]
Av 738- Adaptive Filtering - Wiener Filters[wk 3]Av 738- Adaptive Filtering - Wiener Filters[wk 3]
Av 738- Adaptive Filtering - Wiener Filters[wk 3]
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Transient and Steady State Response - Control Systems Engineering
Transient and Steady State Response - Control Systems EngineeringTransient and Steady State Response - Control Systems Engineering
Transient and Steady State Response - Control Systems Engineering
Siyum Tsega Balcha
 
Stochastic optimal control & rl
Stochastic optimal control & rlStochastic optimal control & rl
Stochastic optimal control & rl
ChoiJinwon3
 
Linear control system Open loop & Close loop Systems
Linear control system Open loop & Close loop SystemsLinear control system Open loop & Close loop Systems
Linear control system Open loop & Close loop Systems
SohaibUllah5
 
controllability-and-observability.pdf
controllability-and-observability.pdfcontrollability-and-observability.pdf
controllability-and-observability.pdf
LAbiba16
 
Servo systems
Servo systemsServo systems
Servo systems
cairo university
 
Kalman filters
Kalman filtersKalman filters
Kalman filters
AJAL A J
 

Similar to Av 738 - Adaptive Filtering - Kalman Filters (20)

lecture_18-19_state_observer_design.pptx
lecture_18-19_state_observer_design.pptxlecture_18-19_state_observer_design.pptx
lecture_18-19_state_observer_design.pptx
 
Slideshare
SlideshareSlideshare
Slideshare
 
Av 738-Adaptive Filters - Extended Kalman Filter
Av 738-Adaptive Filters - Extended Kalman FilterAv 738-Adaptive Filters - Extended Kalman Filter
Av 738-Adaptive Filters - Extended Kalman Filter
 
Week_10.2.pdf
Week_10.2.pdfWeek_10.2.pdf
Week_10.2.pdf
 
Intro to Quant Trading Strategies (Lecture 6 of 10)
Intro to Quant Trading Strategies (Lecture 6 of 10)Intro to Quant Trading Strategies (Lecture 6 of 10)
Intro to Quant Trading Strategies (Lecture 6 of 10)
 
Kalman filter demonstration
Kalman filter demonstrationKalman filter demonstration
Kalman filter demonstration
 
Av 738- Adaptive Filtering - Background Material
Av 738- Adaptive Filtering - Background MaterialAv 738- Adaptive Filtering - Background Material
Av 738- Adaptive Filtering - Background Material
 
Stat 2153 Introduction to Queiueng Theory
Stat 2153 Introduction to Queiueng TheoryStat 2153 Introduction to Queiueng Theory
Stat 2153 Introduction to Queiueng Theory
 
Time series Modelling Basics
Time series Modelling BasicsTime series Modelling Basics
Time series Modelling Basics
 
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]Sensor Fusion Study - Ch5. The discrete-time Kalman filter  [박정은]
Sensor Fusion Study - Ch5. The discrete-time Kalman filter [박정은]
 
Kalman Filter Basic
Kalman Filter BasicKalman Filter Basic
Kalman Filter Basic
 
Lec1 01
Lec1 01Lec1 01
Lec1 01
 
Lecture 23 24-time_response
Lecture 23 24-time_responseLecture 23 24-time_response
Lecture 23 24-time_response
 
Av 738- Adaptive Filtering - Wiener Filters[wk 3]
Av 738- Adaptive Filtering - Wiener Filters[wk 3]Av 738- Adaptive Filtering - Wiener Filters[wk 3]
Av 738- Adaptive Filtering - Wiener Filters[wk 3]
 
Transient and Steady State Response - Control Systems Engineering
Transient and Steady State Response - Control Systems EngineeringTransient and Steady State Response - Control Systems Engineering
Transient and Steady State Response - Control Systems Engineering
 
Stochastic optimal control & rl
Stochastic optimal control & rlStochastic optimal control & rl
Stochastic optimal control & rl
 
Linear control system Open loop & Close loop Systems
Linear control system Open loop & Close loop SystemsLinear control system Open loop & Close loop Systems
Linear control system Open loop & Close loop Systems
 
controllability-and-observability.pdf
controllability-and-observability.pdfcontrollability-and-observability.pdf
controllability-and-observability.pdf
 
Servo systems
Servo systemsServo systems
Servo systems
 
Kalman filters
Kalman filtersKalman filters
Kalman filters
 

More from Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS

Me314 week09-root locusanalysis
Me314 week09-root locusanalysisMe314 week09-root locusanalysis
Me314 week09-root locusanalysis
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Me314 week08-stability and steady state errors
Me314 week08-stability and steady state errorsMe314 week08-stability and steady state errors
Me314 week08-stability and steady state errors
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Me314 week 06-07-Time Response
Me314 week 06-07-Time ResponseMe314 week 06-07-Time Response
Me314 week 06-07-Time Response
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Me314 week05a-block diagreduction
Me314 week05a-block diagreductionMe314 week05a-block diagreduction
Me314 week05a-block diagreduction
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME-314- Control Engineering - Week 03-04
ME-314- Control Engineering - Week 03-04ME-314- Control Engineering - Week 03-04
ME-314- Control Engineering - Week 03-04
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME-314- Control Engineering - Week 01
ME-314- Control Engineering - Week 01ME-314- Control Engineering - Week 01
ME-314- Control Engineering - Week 01
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Marketing Presentation of Mechanical Engineering @ DSU for High School Students
Marketing Presentation of Mechanical Engineering @ DSU for High School StudentsMarketing Presentation of Mechanical Engineering @ DSU for High School Students
Marketing Presentation of Mechanical Engineering @ DSU for High School Students
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
2 Day Workshop on Digital Datcom and Simulink
2 Day Workshop on Digital Datcom and Simulink2 Day Workshop on Digital Datcom and Simulink
2 Day Workshop on Digital Datcom and Simulink
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
Why Choose Graduate Studies at the University you work in?
Why Choose Graduate Studies at the University you work in?Why Choose Graduate Studies at the University you work in?
Why Choose Graduate Studies at the University you work in?
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME 312 Mechanical Machine Design - Introduction [Week 1]
ME 312 Mechanical Machine Design - Introduction [Week 1]ME 312 Mechanical Machine Design - Introduction [Week 1]
ME 312 Mechanical Machine Design - Introduction [Week 1]
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME438 Aerodynamics (week 12)
ME438 Aerodynamics (week 12)ME438 Aerodynamics (week 12)
ME438 Aerodynamics (week 12)
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME438 Aerodynamics (week 11)
ME438 Aerodynamics (week 11)ME438 Aerodynamics (week 11)
ME438 Aerodynamics (week 11)
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME-438 Aerodynamics (week 10)
ME-438 Aerodynamics (week 10)ME-438 Aerodynamics (week 10)
ME-438 Aerodynamics (week 10)
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME-438 Aerodynamics (week 9)
ME-438 Aerodynamics (week 9)ME-438 Aerodynamics (week 9)
ME-438 Aerodynamics (week 9)
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 
ME438 Aerodynamics (week 8)
ME438 Aerodynamics (week 8)ME438 Aerodynamics (week 8)
ME438 Aerodynamics (week 8)
Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS
 

More from Dr. Bilal Siddiqui, C.Eng., MIMechE, FRAeS (20)

Me314 week09-root locusanalysis
Me314 week09-root locusanalysisMe314 week09-root locusanalysis
Me314 week09-root locusanalysis
 
Me314 week08-stability and steady state errors
Me314 week08-stability and steady state errorsMe314 week08-stability and steady state errors
Me314 week08-stability and steady state errors
 
Me314 week 06-07-Time Response
Me314 week 06-07-Time ResponseMe314 week 06-07-Time Response
Me314 week 06-07-Time Response
 
Me314 week05a-block diagreduction
Me314 week05a-block diagreductionMe314 week05a-block diagreduction
Me314 week05a-block diagreduction
 
ME-314- Control Engineering - Week 03-04
ME-314- Control Engineering - Week 03-04ME-314- Control Engineering - Week 03-04
ME-314- Control Engineering - Week 03-04
 
ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02ME-314- Control Engineering - Week 02
ME-314- Control Engineering - Week 02
 
ME-314- Control Engineering - Week 01
ME-314- Control Engineering - Week 01ME-314- Control Engineering - Week 01
ME-314- Control Engineering - Week 01
 
Marketing Presentation of Mechanical Engineering @ DSU for High School Students
Marketing Presentation of Mechanical Engineering @ DSU for High School StudentsMarketing Presentation of Mechanical Engineering @ DSU for High School Students
Marketing Presentation of Mechanical Engineering @ DSU for High School Students
 
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
"It isn't exactly Rocket Science" : The artsy science of rocket propulsion
 
2 Day Workshop on Digital Datcom and Simulink
2 Day Workshop on Digital Datcom and Simulink2 Day Workshop on Digital Datcom and Simulink
2 Day Workshop on Digital Datcom and Simulink
 
Why Choose Graduate Studies at the University you work in?
Why Choose Graduate Studies at the University you work in?Why Choose Graduate Studies at the University you work in?
Why Choose Graduate Studies at the University you work in?
 
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
ME 312 Mechanical Machine Design [Screws, Bolts, Nuts]
 
ME 312 Mechanical Machine Design - Introduction [Week 1]
ME 312 Mechanical Machine Design - Introduction [Week 1]ME 312 Mechanical Machine Design - Introduction [Week 1]
ME 312 Mechanical Machine Design - Introduction [Week 1]
 
labview-cert
labview-certlabview-cert
labview-cert
 
WindTunnel_Certificate
WindTunnel_CertificateWindTunnel_Certificate
WindTunnel_Certificate
 
ME438 Aerodynamics (week 12)
ME438 Aerodynamics (week 12)ME438 Aerodynamics (week 12)
ME438 Aerodynamics (week 12)
 
ME438 Aerodynamics (week 11)
ME438 Aerodynamics (week 11)ME438 Aerodynamics (week 11)
ME438 Aerodynamics (week 11)
 
ME-438 Aerodynamics (week 10)
ME-438 Aerodynamics (week 10)ME-438 Aerodynamics (week 10)
ME-438 Aerodynamics (week 10)
 
ME-438 Aerodynamics (week 9)
ME-438 Aerodynamics (week 9)ME-438 Aerodynamics (week 9)
ME-438 Aerodynamics (week 9)
 
ME438 Aerodynamics (week 8)
ME438 Aerodynamics (week 8)ME438 Aerodynamics (week 8)
ME438 Aerodynamics (week 8)
 

Recently uploaded

Basic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumersBasic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumers
PedroFerreira53928
 
Thesis Statement for students diagnonsed withADHD.ppt
Thesis Statement for students diagnonsed withADHD.pptThesis Statement for students diagnonsed withADHD.ppt
Thesis Statement for students diagnonsed withADHD.ppt
EverAndrsGuerraGuerr
 
Model Attribute Check Company Auto Property
Model Attribute  Check Company Auto PropertyModel Attribute  Check Company Auto Property
Model Attribute Check Company Auto Property
Celine George
 
Template Jadual Bertugas Kelas (Boleh Edit)
Template Jadual Bertugas Kelas (Boleh Edit)Template Jadual Bertugas Kelas (Boleh Edit)
Template Jadual Bertugas Kelas (Boleh Edit)
rosedainty
 
The approach at University of Liverpool.pptx
The approach at University of Liverpool.pptxThe approach at University of Liverpool.pptx
The approach at University of Liverpool.pptx
Jisc
 
Instructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptxInstructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptx
Jheel Barad
 
Welcome to TechSoup New Member Orientation and Q&A (May 2024).pdf
Welcome to TechSoup   New Member Orientation and Q&A (May 2024).pdfWelcome to TechSoup   New Member Orientation and Q&A (May 2024).pdf
Welcome to TechSoup New Member Orientation and Q&A (May 2024).pdf
TechSoup
 
MARUTI SUZUKI- A Successful Joint Venture in India.pptx
MARUTI SUZUKI- A Successful Joint Venture in India.pptxMARUTI SUZUKI- A Successful Joint Venture in India.pptx
MARUTI SUZUKI- A Successful Joint Venture in India.pptx
bennyroshan06
 
The Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official PublicationThe Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official Publication
Delapenabediema
 
How to Break the cycle of negative Thoughts
How to Break the cycle of negative ThoughtsHow to Break the cycle of negative Thoughts
How to Break the cycle of negative Thoughts
Col Mukteshwar Prasad
 
How to Split Bills in the Odoo 17 POS Module
How to Split Bills in the Odoo 17 POS ModuleHow to Split Bills in the Odoo 17 POS Module
How to Split Bills in the Odoo 17 POS Module
Celine George
 
Language Across the Curriculm LAC B.Ed.
Language Across the  Curriculm LAC B.Ed.Language Across the  Curriculm LAC B.Ed.
Language Across the Curriculm LAC B.Ed.
Atul Kumar Singh
 
How libraries can support authors with open access requirements for UKRI fund...
How libraries can support authors with open access requirements for UKRI fund...How libraries can support authors with open access requirements for UKRI fund...
How libraries can support authors with open access requirements for UKRI fund...
Jisc
 
Supporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptxSupporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptx
Jisc
 
The geography of Taylor Swift - some ideas
The geography of Taylor Swift - some ideasThe geography of Taylor Swift - some ideas
The geography of Taylor Swift - some ideas
GeoBlogs
 
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXXPhrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
MIRIAMSALINAS13
 
Chapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptxChapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptx
Mohd Adib Abd Muin, Senior Lecturer at Universiti Utara Malaysia
 
1.4 modern child centered education - mahatma gandhi-2.pptx
1.4 modern child centered education - mahatma gandhi-2.pptx1.4 modern child centered education - mahatma gandhi-2.pptx
1.4 modern child centered education - mahatma gandhi-2.pptx
JosvitaDsouza2
 
Digital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and ResearchDigital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and Research
Vikramjit Singh
 
Unit 8 - Information and Communication Technology (Paper I).pdf
Unit 8 - Information and Communication Technology (Paper I).pdfUnit 8 - Information and Communication Technology (Paper I).pdf
Unit 8 - Information and Communication Technology (Paper I).pdf
Thiyagu K
 

Recently uploaded (20)

Basic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumersBasic phrases for greeting and assisting costumers
Basic phrases for greeting and assisting costumers
 
Thesis Statement for students diagnonsed withADHD.ppt
Thesis Statement for students diagnonsed withADHD.pptThesis Statement for students diagnonsed withADHD.ppt
Thesis Statement for students diagnonsed withADHD.ppt
 
Model Attribute Check Company Auto Property
Model Attribute  Check Company Auto PropertyModel Attribute  Check Company Auto Property
Model Attribute Check Company Auto Property
 
Template Jadual Bertugas Kelas (Boleh Edit)
Template Jadual Bertugas Kelas (Boleh Edit)Template Jadual Bertugas Kelas (Boleh Edit)
Template Jadual Bertugas Kelas (Boleh Edit)
 
The approach at University of Liverpool.pptx
The approach at University of Liverpool.pptxThe approach at University of Liverpool.pptx
The approach at University of Liverpool.pptx
 
Instructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptxInstructions for Submissions thorugh G- Classroom.pptx
Instructions for Submissions thorugh G- Classroom.pptx
 
Welcome to TechSoup New Member Orientation and Q&A (May 2024).pdf
Welcome to TechSoup   New Member Orientation and Q&A (May 2024).pdfWelcome to TechSoup   New Member Orientation and Q&A (May 2024).pdf
Welcome to TechSoup New Member Orientation and Q&A (May 2024).pdf
 
MARUTI SUZUKI- A Successful Joint Venture in India.pptx
MARUTI SUZUKI- A Successful Joint Venture in India.pptxMARUTI SUZUKI- A Successful Joint Venture in India.pptx
MARUTI SUZUKI- A Successful Joint Venture in India.pptx
 
The Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official PublicationThe Challenger.pdf DNHS Official Publication
The Challenger.pdf DNHS Official Publication
 
How to Break the cycle of negative Thoughts
How to Break the cycle of negative ThoughtsHow to Break the cycle of negative Thoughts
How to Break the cycle of negative Thoughts
 
How to Split Bills in the Odoo 17 POS Module
How to Split Bills in the Odoo 17 POS ModuleHow to Split Bills in the Odoo 17 POS Module
How to Split Bills in the Odoo 17 POS Module
 
Language Across the Curriculm LAC B.Ed.
Language Across the  Curriculm LAC B.Ed.Language Across the  Curriculm LAC B.Ed.
Language Across the Curriculm LAC B.Ed.
 
How libraries can support authors with open access requirements for UKRI fund...
How libraries can support authors with open access requirements for UKRI fund...How libraries can support authors with open access requirements for UKRI fund...
How libraries can support authors with open access requirements for UKRI fund...
 
Supporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptxSupporting (UKRI) OA monographs at Salford.pptx
Supporting (UKRI) OA monographs at Salford.pptx
 
The geography of Taylor Swift - some ideas
The geography of Taylor Swift - some ideasThe geography of Taylor Swift - some ideas
The geography of Taylor Swift - some ideas
 
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXXPhrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
Phrasal Verbs.XXXXXXXXXXXXXXXXXXXXXXXXXX
 
Chapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptxChapter 3 - Islamic Banking Products and Services.pptx
Chapter 3 - Islamic Banking Products and Services.pptx
 
1.4 modern child centered education - mahatma gandhi-2.pptx
1.4 modern child centered education - mahatma gandhi-2.pptx1.4 modern child centered education - mahatma gandhi-2.pptx
1.4 modern child centered education - mahatma gandhi-2.pptx
 
Digital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and ResearchDigital Tools and AI for Teaching Learning and Research
Digital Tools and AI for Teaching Learning and Research
 
Unit 8 - Information and Communication Technology (Paper I).pdf
Unit 8 - Information and Communication Technology (Paper I).pdfUnit 8 - Information and Communication Technology (Paper I).pdf
Unit 8 - Information and Communication Technology (Paper I).pdf
 

Av 738 - Adaptive Filtering - Kalman Filters

  • 1. Av-738 Adaptive Filter Theory Lecture 4- Optimum Filtering [Kalman Filters] Dr. Bilal A. Siddiqui Air University (PAC Campus) Spring 2018
  • 2. Kalman Filters • Kalman filter is both an optimal and recursive (adaptive filter) • It has a unique mathematical formulation (i.e. state space concepts) • It is solved recursively. It is updated using previous state estimates and new data (innovation), therefore no need for storage. • It is mostly implemented in the discrete form, and easily implementable on digital computers • It gives a unifying framework for RLS filters. • Developed by Rudolf Kalman in 1969 []
  • 3. Time Domain Description of Dynamical Sytems • One way to describe a linear dynamical system’s evolution is in the frequency (or transform) domain by means of transfer functions. • Another equivalent way is to define it in time-domain. • If the system is linear, it can be described in State-Space platform. • A set of variables that can uniquely define the time evolution of a dynamical system is called the state-vector 𝑋. Elements of that vector are the system’s states 𝑥. • In many cases, the states themselves are not measured directly. Some combination of states may be measured. These are measurements 𝑦 mostly contaminated by measurement noise. • The state space model of the process is a linear combination of inputs and previous values of states. In other words, states have memories (or integrations)
  • 4. Linear Dynamical System • v1 is process “noise” which is a measure of how inaccurate our linear model given by 𝑥 𝑛 + 1 = 𝐹 𝑛 + 1, 𝑛 𝑥 𝑛 + 𝑣1(𝑛) is. If there are M states, 𝑂 𝐹 = 𝑀 × 𝑀 and 𝑂 𝑥 = 𝑀 × 1, 𝑂 𝑣 = 𝑀 × 1 • F is known as the state transition matrix relating states at n+1 and n. Process noise is measured as zero mean white noise (its correlation matrix Q1 is diagonal). Let the system be time-invariant. The 𝐹 𝑛 + 1, 𝑛 = 𝐹 for all n. • Measured quantities are linear combinations of states 𝑦 𝑛 = 𝐶 𝑛 𝑥 𝑛 + 𝑣2 𝑛 . For N measurements, 𝑂 𝑦 = 𝑁 × 1, 𝑂 𝐶 = N × 𝑀 and 𝑣2 is a Nx1 measurement noise vector, which may represent sensor noise and linearization errors. Its correlation matrix Q2 is also diagonal. • It is assumed that process and measurement noises are uncorrelated. 𝐸 𝑣1 𝑛 𝑣2 𝐻 (𝑛) = 0
  • 6. The Innovation Process • Let 𝑦 𝑛−1 denote all measurements uptil n-1, i.e. 𝑦 𝑛−1 = 𝑦 𝑛 − 1 … 𝑦 1 • Let 𝑦(𝑛|𝑦 𝑛−1) be the one-step prediction or best estimate (min mean square error sense) of y(n) using all measurement made till n-1. • We define the prediction error 𝛼 𝑛 = 𝑦 𝑛 − 𝑦(𝑛|𝑦 𝑛−1) as process “innovation”, which represents the new information (innovation) in the data observed at time n. • Properties:
  • 7. Innovation Propagation • Suppose we have the more generic system 𝑥 𝑘 = 𝐹𝑥 𝑘 + 𝑣1(𝑛), then we can write a recursive relation 𝑥 𝑛 + 1 = 𝐹 𝑛 𝑥 0 + 𝑖=0 𝑛−1 𝐹 𝑛−1−𝑖 𝑣1 𝑖 • multiplying v2(n) and taking expectation 𝐸 𝑥 𝑛 + 1 𝑣2 𝐻 𝑛 = 𝐹𝐸 𝑥 0 𝑣2 𝐻 𝑛 + 𝐸 𝑖=0 𝑛−1 𝐹 𝑛−1−𝑖 𝑣1 𝑖 𝑣2 𝐻 𝑛 • Since noise v2 is uncorrelated with initial state x(0) and noise v1 this means 𝐸 𝑥 𝑘 𝑣2 𝐻 𝑛 for 0 ≤ 𝑘 ≤ 𝑛. • Since 𝑦 𝑛 = 𝐶𝑥 𝑛 + 𝑣2 𝑛 , we can also prove that 𝐸 𝑦 𝑘 𝑣2 𝐻 𝑛 = 0 for 0 ≤ 𝑘 ≤ 𝑛 − 1 , i.e. v2 is orthogonal to past measurements. • Therefore, 𝑦 𝑛 𝑦 𝑛−1 = C 𝑥(𝑛|𝑦 𝑛−1) as v2 cannot be predicted using past measurements. • Therefore we can write innovations as 𝛼 𝑛 = 𝑦 𝑛 − C 𝑥(𝑛|𝑦 𝑛−1) • Substitute measurement equation, 𝛼 𝑛 = 𝐶𝑥 𝑛 + 𝑣2 𝑛 − C 𝑥 𝑛 𝑦 𝑛−1 = 𝐶𝜖 𝑛, 𝑛 − 1 + 𝑣2(𝑛) where 𝜖 𝑛, 𝑛 − 1 = 𝑥 𝑛 − 𝑥 𝑛 𝑦 𝑛−1 is the predicted state-error vector at n using the measurements till n-1 We can prove that 𝜖 𝑛, 𝑛 − 1 is orthogonal to v1 and v2 for for all times less than n.
  • 8. A more generic model • Suppose we have a more generalized process model of a discrete- time linear system 𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1 + 𝑤 𝑘−1 where uk is a known and deterministic input and wk is Gaussian zero- mean white noise with covariance Qk. • How does the mean of the state xk change with time? 𝑥 𝑘 = 𝐸 𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1 • Now, how does the covariance of xk change with time?
  • 9. Covariance Propagation • Since • Note that innovation or state error 𝜖 𝑘 = 𝑥 𝑘 − 𝑥 𝑘 is uncorrelated with noise wk-1 (or process noise at k-1, as we said earlier) • This means the state covariance matrix propagates as This is called the discrete-time Lyapunov equation or Stein equation. Also note that since x0 and wi are Gaussian 𝑥 𝑘 = 𝑁( 𝑥 𝑘, 𝑃𝑘)
  • 11. Homework 2 (a) • Simulate this model using Matlab • But we could also have found the steady states and covariances directly. Prove that the expression for steady states analytically are:
  • 12. Derivation of the Kalman Filter Our goal is to estimate the state xk based on our knowledge of the system dynamics and the availability of the noisy measurements yk. The amount of information that is available to us for our state estimate varies depending on the particular problem that we are trying to solve.
  • 13. Posteriors and Priots • If we have all of the measurements up to and including time k available for use in our estimate of xk, then we can form an a posteriori estimate, which we denote as 𝑥 𝑘 + • If we have all of the measurements before (but not including) time k available for use in our estimate of xk , then we can form an a postiori estimate, which we denote as 𝑥 𝑘 −
  • 14. Process of Estimation • After we process the measurement at time (k-1), we have an estimate Of xk−1(denoted xk−1 + ) and the covariance of that estimate (denoted Pk−1 + ). • When time k arrives, before we process the measurement at time k, we compute an estimate of 𝑥 𝑘(denoted 𝑥 𝑘 − ) and the covariance of that estimate (denoted Pk − ). • Then we process the measurement at time k to refine our estimate of 𝑥 𝑘. The resulting estimate is denoted 𝑥 𝑘 − , with covariance Pk + .
  • 15. Step 1: Time-Update of state and covariance estimate • We begin the estimation process with at, our best estimate of the initial state x0. Given 𝑥0 + , how should we compute 𝑥1 − ? We want to set 𝑥1 − = 𝐸 𝑥1 , but we know 𝑥0 + = 𝐸 𝑥0 . Therefore, using 𝑥 𝑘 = 𝐸 𝑥 𝑘 = 𝐹𝑘−1 𝑥 𝑘−1 + 𝐺 𝑘−1 𝑢 𝑘−1 𝑥1 − = 𝐹0 𝑥0 + + 𝐺0 𝑢0, or generalizing 𝑥 𝑘 − = 𝐹𝑘−1 𝑥 𝑘−1 + + 𝐺 𝑘−1 𝑢 𝑘−1 • This is called the time update equation for 𝑥 • From time 𝑘 − 1 + to time 𝑘− , the state estimate propagates the same way that the mean of the state propagates. • This makes sense intuitively. We do not have any additional measurements available to help us update our state estimate between time 𝑘 − 1 + and time 𝑘− , so we should just update the state estimate based on our knowledge of the system dynamics. • Using similar arguments, we can write for covariance
  • 16. Step 2: Measurement-update of state • Given 𝑥 𝑘 − , how should we compute 𝑥 𝑘 + ? • A general form of filter update is i.e. some part of the innovation is added to the previous estimate. K is some gain matrix, known as Kàlmàn filter gain. • In our case, 𝑥 𝑘 + = 𝑥 𝑘 − + 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘 − • To find Kalman gain, the derivation is a bit more complex. • We begin by noting that the state error vector propagates as 𝐸 𝜖 𝑘 + = 𝐸 𝑥 𝑘 − 𝑥 𝑘 + = 𝐸 𝑥 𝑘 − 𝑥 𝑘 − − 𝐾𝑘 𝐻𝑥 𝑘 + 𝑣 𝑘 − 𝐻 𝑘 𝑥 𝑘 − = 𝐸 𝜖 𝑘 − − 𝐾𝑘(𝐻 𝑘 𝜖 𝑘 − +𝑣 𝑘) ⇒ 𝐸 𝜖 𝑘 + = 𝐼 − 𝐾𝑘 𝐻 𝑘 𝐸 𝜖 𝑘 − − 𝐾𝑘 𝐸 𝑣 𝑘
  • 17. Step 2: Measurement-update of covariance • Estimation error covariance Pk propagates as = 𝑃𝑘 + = 𝐸 𝜖 𝑘 + 𝜖 𝑘 + 𝑇 = E 𝐼 − 𝐾𝑘 𝐻 𝑘 𝐸 𝜖 𝑘 − − 𝐾𝑘 𝐸 𝑣 𝑘 … 𝑇 = 𝐼 − 𝐾𝑘 𝐻 𝑘 E 𝜖 𝑘 − 𝜖 𝑘 −𝑇 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑇 + 𝐾𝑘 𝐸 𝑣 𝑘 𝑣 𝑘 𝑇 𝐾𝑘 𝑇 − 𝐾𝑘E 𝑣 𝑘 𝜖 𝑘 −𝑇 𝐼 − 𝐾𝑘 𝐻 𝑘 T + 𝐼 − 𝐾𝑘 𝐻 𝑘 E 𝜖 𝑘 − 𝑣 𝑘 𝑇 𝐾𝑘 𝑇 • Since estimation error at 𝑘− is independent of measurement noise at k 𝑃𝑘 + = 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑇 + 𝐾𝑘 𝑅 𝑘 𝐾𝑘 𝑇 • After some elaborate steps, this can also be shown as (prove as part of HW 2) 𝑃𝑘 + = 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − • This is the recursive formula for the covariance of estimation error. This is consistent with intuition that as the measurement noise increases (i.e. 𝑅 𝑘 increases) the uncertainty in our estimate also increases (i.e., 𝑃𝑘 increases). • Note that Pk should be positive definite since it is a covariance matrix, and the form of 𝑃𝑘 + guarantees that Pk will be positive definite, assuming that Pk-1 and Rk are positive definite.
  • 18. Finding Kalman Gain • Let’s find the optimal Kalman gain matrix by minimizing the posterior error covariance at time k (equivalent to MSE since 𝑃𝑘 + = 𝐸 𝜖 𝑘 + 𝜖 𝑘 + 𝑇 ). 𝜕𝑃𝑘 + 𝜕𝐾𝑘 = 0 𝜕 𝜕𝐾𝑘 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑇 + 𝐾𝑘 𝑅 𝑘 𝐾𝑘 𝑇 = 0 −2 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − 𝐻 𝑘 𝑇 + 2𝐾𝑘 𝑅 𝑘 = 0 This can be solved for KK as 𝐾𝑘 = 𝑃𝑘 − 𝐻 𝑘 𝑇 𝐻 𝑘 𝑃𝑘 − 𝐻 𝑘 𝑇 + 𝑅 𝑘 −1 Using some more elaborate steps (prove as part of HW 2) 𝐾𝑘 = Pk + Hk T Rk −1
  • 19. Kalman filter summarized • We summarize the discrete-time Kalman filter by combining the above equations into a single algorithm.
  • 20. Comments about implementation • The 𝑃𝑘 + = 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑇 + 𝐾𝑘 𝑅 𝑘 𝐾𝑘 𝑇 equation of covariance update is more stable and robust than 𝑃𝑘 + = 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑃𝑘 − equation, even though the latter is less computationally intensive. The first expression ensures Pk is symmetric positive definite, the latter does not. • Calculation of 𝑃𝑘 − , 𝐾𝑘and 𝑃𝑘 + does not depend on the measurements yk , but depends only on the system parameters Fk, Hk, Qk and Rk. Therefore, Kalman gain Kk can be calculated offline before the system operates and saved in memory (if statistics Q, F and R are known). In that case, linear time-invariant Kalman filter is like the Weiner filter and not adaptive. • In contrast, as we will see later, the filter gain and covariance for nonlinear systems cannot (in general) be computed offline because they depend on the measurements. In that case, the Kalman filter has important nonlinear variants.
  • 21. Important Properties • If wk and vk are zero-mean, uncorrelated, and white, then the Kalman filter is the best linear solution to the above problem, i.e., the Kalman filter is the best filter that is a linear combination of the measurements. There may be a nonlinear filter that gives a better solution, but the Kalman filter is the best linear filter. • Even if the noise is not Gaussian, the Kalman filter is still the optimal linear filter. Noise only needs to be white and zero-mean. • If wk and vk are coloured or correlated, then the KF can be modified. • For nonlinear systems, various formulations of nonlinear Kalman filters approximate the solution to the linear problem we solved.
  • 22. One Step KF Equations • We will see how the a priori and a posteriori Kalman filter equations can be combined into a single equation. This may simplify computer implementation of the equations. We have 𝑥 𝑘+1 − = 𝐹𝑘 𝑥 𝑘 + + 𝐺 𝑘 𝑢 𝑘 • Substituting 𝑥 𝑘 + = 𝑥 𝑘 − + 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘 − 𝑥 𝑘 − = 𝐹𝑘 𝑥 𝑘 − + 𝐾𝑘 𝑦 𝑘 − 𝐻 𝑘 𝑥 𝑘 − + 𝐺 𝑘 𝑢 𝑘 𝑥 𝑘 − = 𝐹𝑘 𝐼 − 𝐾𝑘 𝐻 𝑘 𝑥 𝑘 − + 𝐹𝑘 𝐾𝑘 𝑦 𝑘 + 𝐺 𝑘 𝑢 𝑘 • This shows that the a priori state estimate can be computed directly from its value at the previous time step, without computing the a posteriori state estimate in between. We can similarly show
  • 23. Example • Suppose we have a noise-free with position r , velocity v, and constant acceleration a.
  • 25. Simulation %% clc;clear all;close; %Kalman filter for position measurement %AV-738 Spr 18, Dr. Bilal A. Siddiqui T=5; R=30^2; a=0.01; F=[1 T T^2/2 %position 0 1 T %velocity 0 0 1]; %acceleration G=0; H=[0 0 1];% only position being measured w=0;%process noise x(:,1)=[0 0 a]';%initial value of state vector x_post(:,1)=[0 0 0]';%initial estimate of state vector P_post(:,:,1)=1E5*eye(3);%initial estimate of error covariance for k=2:60 %calculate a priori covariance estimate P_prior(:,:,k)=F*P_post(:,:,k-1)*F'; %calculate Kalman gain K(:,:,k)=P_prior(:,:,k)*H'*(H*P_prior(:,:,k)*H'+R)^-1; %calculate a priori estimate of state x_prior(:,k)=F*x(:,k-1); %system evolution x(:,k)=F*x(:,k-1);%take measurement y(k)=H*x(:,k)+sqrt(R)*randn; %calculate innovation innov(k)=y(k)-H*x_prior(:,k); %calculate a posteriori estimate of state x_post(:,k)=x_prior(:,k)+K(:,:,k)*innov(k); %calculate a posteriori estimate of covariance P_post(:,:,k)=(eye(3)- ... K(:,:,k)*H)*P_prior(:,:,k)*+(eye(3)-K(:,:,k)*H)'+... K(:,:,k)*R*K(:,:,k)'; end
  • 27. Homework 2 (c) • First of all, we were able to reduce estimation error below the measurement noise floor. How was that possible? • Also, if we start measuring the acceleration directly (use a 1-g noise, i.e.R=1) instead of the position, we will get a good estimate still, but the error covariance will keep increasing. Why? • In practical terms, if we inject position measurements at a slower rate, what will happen to the error covariance, if we were taking accelerometer measurements at a higher rate? • Simulate the error covariance if we measure accelerometer at every sampling instant, but receive GPS position with noise variance of 10 after every five sampling instants. • Comment your answer and analyze the results.
  • 28. A word of caution • The theory presented in this lecture (based on “Optimal State Estimation” b Dan Simon, J. Wiley & Sons 2006) makes the Kalman filter an attractive choice for state estimation. • But when a Kalman filter is implemented on a real system it may not work, even though the theory is correct. Main causes are (a) finite precision arithmetic and (b) modelling errors. • The theory presented also assumes that the system model is precisely known. It is assumed that the F , Q, H, and R matrices are exactly known, and it is assumed that the noise sequences {wk} and {vk} are pure white, zero-mean, and completely uncorrelated. If any of these assumptions are violated, as they always are in real implementations, then the Kalman filter assumptions are violated and the theory may not work. •