α–β filter, 302
Algebraic Riccati equation, 280, 421
Answers to summary questions, 539–541
AR Model, 12
Asymptotic distributions, 99–101
Asymptotic efficiency (see Large sample properties of estimators)
Asymptotic mean, 100
Asymptotic normality (see Large sample properties of estimators)
Asymptotic unbiasedness (see Large sample properties of estimators)
Asymptotic variance, 100–101
Augmented state-variable model (see State augmentation)
Bartlett-Brillinger-Rosenblatt formulas, 478–480
Basic state-variable model, continuous-time:
defined, 414–415
properties of, 415–416, 427–428
Basic state-variable model, discrete-time (see also Not-so-basic state-variable model):
augmented, 324–325
defined, 215–216
properties of, 216–220
Basis functions, 13
Batch processing (see Least-squares estimation processing)
Bayes risk, 210
Best linear unbiased estimation, 121–131
Best linear unbiased estimator:
batch form, 124–125
comparison with maximum a posteriori estimator, 196
comparison with mean-squared estimator, 181–184
comparison with weighted least-squares estimator, 125–126
derivation from a Kalman filter, 266
properties, 126–130
random parameters, 182–184
recursive forms, 130–131
Biases (see Not-so-basic state-variable model)
Bicepstrum (see Complex bicepstram)
Binary detection:
detection rules, 201–203
elements of, 200–204
errors, 202–203
single most-likely replacement detector, 209–210
square-law detector, 203–204
Biographies (see Fisher, biography; Gauss, biography; Kalman, biography; Wiener, biography)
Bispectrum (see Polyspectra)
Block component method, 412
BLUE (see Best linear unbiased estimator)
C(q, k) formula, 481–482
Causal invertibility, 244, 252
Central limit theorems, 101–102
Colored noises (see Not-so-basic state-variable model)
Completeness, 445
Complex bicepstrum, 496–497
Complex cepstrum, 489
Computation, 7–8, 20, 37, 51, 66–67, 131, 156–157, 184, 199, 222–223, 235, 253, 294, 314, 328, 357–358, 374–375, 392–393, 408–409, 426–427, 490–491, 514
Conditional mean:
defined, 167
properties of, 169–170
Conditional mean estimator, 184–185 (see also Nonlinear estimator)
Consequence of relinearization, 387–388
Consistency (see Large sample properties of estimators)
Constrained total least squares, 38–39
Convergence:
distribution, 93
mean-square, 97
with probability, 93
relationships among different types, 94
rth mean, 93
Convolutional model, 10, 16, 21–22, 329
Correlated noises (see Not-so-basic state-variable model)
Correlation:
biased estimator, 492
unbiased estimator, 493
Correlation-based normal equations, 494
Covariance form of recursive BLUE, 130
Covariance form of recursive least-squares estimator, 63–64
Coverage, 3–6
Cramer Rao inequality:
scalar parameter, 78–82
vector of parameters, 82–84, 86–87
Cross-sectional processing (see Least-squares estimation processing)
Cumulant-based normal equations, 484–485
Cumulants (see also Higher-order statistics applied to linear systems):
defined for random processes, 454
defined for random variables, 452
domain of support, 454
estimation of, 474–475
relation to moments, 466
slices, 455
symmetries, 457
Decision space, 200
Deconvolution (see also Steady-state MVD filter):
double-stage smoothing, 311–312
higher-order statistics, 486–488
maximum-likelihood (MLD), 196–198, 340
minimum-variance (MVD), 181, 329–338
Ott and Meder PEF, 268–269
single-stage smoothing, 308–309
Detection (see Binary detection)
Dirac delta function, 416
Discretization of linear time-varying state-variable model, 371–373
Divergence phenomenon, 267–268
Double C(q, k) algorithm, 489, 495–496
Efficiency (see Small sample properties of estimators)
Eigenvalues and poles, 513
Eigenvalues and transformations of variables, 513
Equation error, 20
Estimate types (see Filtering; Prediction; Smoothing)
Estimation (see also Philosophy):
error, 19
model, 19
Estimation algorithm M-files (see M-files)
Estimation of deterministic parameters (see Best linear unbiased estimation; Least-squares estimation processing; Maximum-likelihood estimation)
Estimation of higher-order statistics (see Cumulants; Polyspectra)
Estimation problem, 3
Estimation of random parameters (see Best linear unbiased estimator; Least-squares estimator; Maximum a posteriori estimation; Mean-squared estimation)
Estimation techniques (see Best linear unbiased estimator; Least-squares estimator; Maximum a posteriori estimation; Maximum-likelihood estimators; Mean-squared estimators; State estimation)
Estimator properties (see Large sample properties of estimators; Small sample properties of estimators)
Estimator versus estimates, 20
Examples:
aerospace vehicle control, 33–35
binomial model:
likelihood ratios, 141
probability versus likelihood, 139–140
bispectra and power spectra for two spectrally equivalent systems, 461–464
C (q, k) formula, 481–482
complete sufficient statistic, 445–446, 447
consistency, 97
convergence in probability, 98
Cramer–Rao bound (Cauchy distribution), 79–80
cumulant-based normal equation, 484–485
cumulants boost signal-to-noise-ratio, 465–466
cumulants for Bernoulli–Gaussian sequence, 454–455
cumulants: what do they look like?, 457
deconvolution, 16–17
deconvolution by double-stage smoothing, 308–309
deconvolution by single-stage smoothing, 311–312
deconvolution using higher-order statistics, 486–488
Dirac delta function in a calculation, 416
divergence phenomenon, 267–268
factorization theorem to determine sufficient statistics, 438–439
Fisher information matrix for generic linear model, 84
fixed-interval smoother, 320–321
fixed-point smoother, 324
function approximation, 13–14
GM equations for MA(q) systems, 486
hypothesis testing, 143–144
identification of coefficients of a finite-difference equation, 12–13, 110–111
impulse and frequency response plots for steady-state predictors, 286–288
impulse response identification, 10–12, 110
impulse response of steady-state predictor, 284–285
instrument calibration, 33, 64, 76
Kalman filter for first-order system with correlated disturbance and measurement noises, 352–354
Kalman filter quantities for two second-order systems, 260–262
Markov model for a colored process, 350
matrix exponential calculations, 373
maximum a posteriori estimator of population mean, 194–195
maximum-likelihood deconvolution, 196–198
maximum-likelihood estimates of parameters of a missile, 408, 410
maximum-likelihood estimation using sufficient statistics, 442–443
maximum-likelihood estimation using sufficient statistics: generic linear model, 443–444
maximum-likelihood estimator of population mean and variance, 150–151
maximum-likelihood estimator of variance, 152
mean and variance computed for a first-order dynamical system, 219
minimum-variance deconvolution, 181
MVD filters for broadband and narrowband channels, 332–336, 338–339
nonlinear estimator, 187
nonlinear model, 15–16
nonlinear predictor, 187–188
Ott and Meder prediction-error filter, 269
perturbation equations for satellite dynamics, 370
prediction performance for a first-order dynamical system, 232–233
q-slice formula, 483–484
random number generator and maximum likelihood, 139, 150–151
random number generator and probability, 138
relationship of BLUE to Kalman filter, 266–267
satellite dynamics, 366–367
satellite dynamics plus unknown parameters, 392
second-order cumulant derivation, 453
sensitivity of Kalman filter, 260, 262–265
signal-to-noise ratio for a first-order dynamical system, 221–222
single-channel Kalman filter, 266
square-law detector, 203–204
state augmentation procedure for colored disturbance, 351
state augmentation procedure for colored measurement noise, 351–352
state estimation, 14–15
state estimation and MS and MAP estimation, 198–199
state estimation errors in a feedback control system:
continuous-time, 420–421
discrete-time, 270
state-variable models:
AR models, 504–505
ARMA models, 506–507
controllable canonical form, 155
discretization of second-order differential equation, 502
MA models, 503–504
parallel connection of systems, 507–508
second-order differential equation, 500, 501–502
subsystem of two complex conjugate poles, 509
transfer function for a second-order ARMA model, 511
steady-state Kalman–Bucy filter for a second-order system, 421–423
steady-state Kalman–Bucy filter in an optimal control system, 425
steady-state Kalman filter, 281–282
sufficient statistics, 437, 438
sufficient statistics for exponential families, 440, 441
SVD computation of Ls(k), 50–51
third-order cumulants for an MA(q) model, 481
unbiasedness of variance estimator, 95
unbiasedness of WLSE, 77
uniformly minimum variance unbiased estimator, 446
uniformly minimum variance unbiased estimator: generic linear model, 446–447
Volterra series representation of a nonlinear system, 17–18
Expanding memory estimator, 36
Expectation, expansion of total expectation, 117
Exponential family of distributions, 441–444
Extended Kalman filter:
application to parameter estimation, 391–392
correction equation, 389
derived, 388–389
flowchart, 390
iterated, 390–391
nonlinear discrete-time system, 393–394
prediction equation, 388
Factorization theorem, 438
Fading memory estimator, 70
Fading-memory filtering, 268
Filtered estimate, 19
Filtering (see also Kalman–Bucy filtering):
applications, 271–276
comparisons of Kalman and Wiener filters, 293
computations, 251
covariance formulation, 251
derivation of Kalman filter, 246–247
divergence phenomenon, 267–268
examples, 259–270
information formulation, 252
innovations derivation of Kalman filter, 246–247
MAP derivation of Kalman filter, 253–255
properties, 248–253
recursive filter, 248
relationship to BLUE, 266
relationship to Wiener filtering, 286, 289–291
steady-state Kalman filter, 280–285
Finite-difference equation coefficient identification (see Identification of)
Finite-memory filtering, 268
FIR model, 10
Fisher, biography, 85–86
Fisher information matrix, 83
Fisher’s information, 78
Fixed-interval smoother (see Smoothing)
Fixed-lag smoother (see Smoothing)
Fixed memory estimator, 36
Fixed-point smoother (see Smoothing)
Function approximation, 13–14
Gauss, biography, 28–29
Gauss–Markov random sequences, 212–214, 328–329
Gaussian random sequences (see Gauss–Markov random sequences)
Gaussian random variables (see also Conditional mean):
conditional density function, 166–168
joint density function, 165–166
multivariate density function, 165
properties of, 168–169
univariate density function, 165
Gaussian sum approximation, 186
Generalized GM equation, 494
Generalized regression neural network, 186
Generic linear model, 9–10
Glossary of major results, 518–523
GM equation, 485–486
Harmonic retrieval problem, 471
Higher-order moments, 452
Higher-order statistics (see Cumulants; Polyspectra)
Higher-order statistics applied to linear systems:
Bartlett-Brillinger-Rosenblatt formulas, 478–480
C (q, k) formula, 481–482
complex bicepstrum, 496–497
cumulant-based normal equations, 484–485
deconvolution, 486–488
double C(q, k) algorithm, 489, 495–496
generalized GM equation, 494
GM equation, 485–486
MA(q) model, 481
q-slice algorithm, 498
q-slice formula, 483–484
residual time-series method, 497–498
Hypotheses (see also Binary detection):
binary, 138
multiple, 138
Identifiability, 154–155
Identification of:
coefficients of a finite-difference equation, 12–13
impulse response, 10–12
Impulse response identification (see Identification of)
Information form of recursive BLUE, 130
Information form of recursive least-squares estimator, 59–60
Innovations process:
defined, 233
properties of, 233–234
Innovations system, 252–253
Instrumental variables, 116
Invariance property of maximum-likelihood estimators (see Maximum-likelihood estimators)
Iterated extended Kalman filter (see Extended Kalman filter)
Iterated least squares, 385–386
Jordan canonical form, 508–509
Kalman, biography, 244–245
Kalman–Bucy filtering:
derivation using a formal limiting procedure, 418–420
derivation when structure of filter is pre-specified, 428–431
notation and problem statement for derivation, 416–417
optimal control application, 423–425
statement of, 417
steady-state, 421
system description for derivation, 414–415
Kalman–Bucy gain matrix, 417
Kalman filter (see Filtering; Steady-state Kalman filter; Wiener filter)
Kalman filter applications (see Filtering)
Kalman filter sensitivity system, 260, 401, 403–404
Kalman filter tuning parameter, 266
Kalman gain matrix, 246
Kalman innovations system (see Innovations system)
Lagrange’s method, 131
Lagrange variation of parameter technique, 375–376
Large sample properties of estimators (see also Least-squares estimator):
asymptotic efficiency, 103–104
asymptotic normality, 101–103
asymptotic unbiasedness, 94–95
consistency, 95–99
Least-squares estimation processing:
batch, 27–37
cross-sectional, 72–73
recursive, 58–69
Least-squares estimator:
batch form, 30–31
comparison with ML estimator, 154
examples, 33–35
initialization of recursive forms, 61, 67–69
large sample properties, 115–117
multistage, 73
normalization of data, 36–37
properties, 108–117
random parameters, 182–184
recursive covariance form, 63–64
recursive information form, 59–60
scale change, 36–37
small sample properties, 109–114
SVD computation, 49–51
vector measurements, 68
Lehmann–Scheffe theorem, 446
Levenberg–Marquardt algorithm, 401–404
Levinson algorithm, 295–300
Likelihood:
compared with probability, 137–138
conditional, 194
continuous distributions, 141, 148
defined, 137–140
unconditional, 194
Likelihood ratio:
decision making, 142–144
defined for multiple hypotheses, 141–142
defined for two hypotheses, 140
Linear algebra facts, 45
Linear model (see also Generic linear model):
defined, 10
examples of (see Deconvolution; Examples, state estimation; Identification of; Nonlinear measurement model)
Linear prediction:
backward prediction error filter, 236
forward prediction error filter, 235
Kalman filter solution, 267
lattice, 237–238
reflection coefficient, 236
tapped delay line, 236
Linear projection, 190
Linear system concepts, 294–295
Linearized nonlinear system (see Nonlinear dynamical systems)
Log-likelihood:
basic state-variable model, 398–400
defined, 148
important dynamical system, 154–156
Loss function, 184
Lyapunov equation, 220
M-files:
fixed-interval smoother, 536–538
Kalman filter, 526–529
Kalman predictor, 529–532
recursive weighted least-squares estimator, 525–526
suboptimal filter, 532–534
suboptimal predictor, 534–536
MA model, 10
Markov parameters, 512
Markov sequence, 213–214
MATLAB, 8
Matrix (see also Pseudo-inverse):
fundamental, 375
gradient, 430
Hessian, 401
intensity, 366
inversion lemma, 62
nonsingular, 31
pseudo-Hessian, 403
spectral intensity, 366
state transition, 510
Matrix Riccati differential equation, 417
Matrix Riccati equation, 250
Maximum a posteriori estimation, 192–199
Maximum a posteriori estimators:
comparison with best linear unbiased estimator, 196
comparison with mean-squared estimator, 195
Gaussian linear model, 195–199
general case, 193–195
Maximum-likelihood deconvolution (see Deconvolution)
Maximum-likelihood estimation, 147–156, 441–444
Maximum-likelihood estimators:
comparison with best linear unbiased estimator, 154
comparison with least-squares estimator, 154
exponential families, 441–444
linear model, 152–154
obtaining, 148–151
properties, 151–152
Maximum-likelihood method, 148–151
Maximum-likelihood state and parameter estimation:
computing ML, 400–404
log-likelihood function for the basic state-variable model, 398–400
steady-state approximation, 404–408
Mean-squared convergence (see Convergence, mean-square)
Mean-squared estimation, 173–188
Mean-squared estimators (see also Conditional mean estimator):
comparison with best linear unbiased estimator, 181–184
comparison with maximum a posteriori estimator, 262–263
derivation, 175–176
Gaussian case, 176–177
linear and Gaussian model, 179–181
properties of, 178–179
Measurement differencing technique, 355
Measurement problem, 2
Measurement residual, 20
Median estimator, 210
Memory:
expanding, 36
fixed, 36
Message space, 200
Method of moments, 451–452
Minimal sufficient statistic, 439, 583
Minimum-variance deconvolution (see Deconvolution)
MLD (see Deconvolution)
Mode estimator, 210
Modeling (see also Philosophy):
estimation problem, 3
measurement problem, 2
representation problem, 2
validation problem, 3
Moments (see Cumulants)
Multistage least-squares (see Least-squares estimator)
Multivariate Gaussian random variables (see Gaussian random variables)
MVD (see Deconvolution)
Nominal:
measurements, 16
parameters, 16
Nonlinear dynamical systems:
discretized perturbation state-variable model, 374
linear perturbation equations, 367–370
model, 365–366
Nonlinear estimator, 185–188
Nonlinear measurement model, 15–16
Normal equations, 32, 484–485, 494
Normalization of data, 36–37
Notation, 18–20
Not-so-basic state-variable model:
biases, 346–347
colored noises, 350–354
correlated noises, 347–350
perfect measurements, 354–357
Observation space, 200
Orthogonality condition, 32
Orthogonality principle, 177–178
Parameter estimation (see Extended Kalman filter)
Parameters, 2
Perfect measurements (see Not-so-basic state-variable model)
Perturbation equations (see Nonlinear dynamical systems)
Philosophy:
estimation theory, 7
modeling, 6
Polyspectra:
bispectrum, 458
defined, 458
domain of support, 458
estimation of bispectrum, direct method, 476–477
estimation of bispectrum, indirect method, 476
symmetries, 459
trispectrum, 458
Predicted estimate, 19
Prediction (see also Linear prediction):
general, 229–233
recursive predictor, 248
single-stage, 228–229
steady-state predictor, 283
Prediction error, 20
Prediction error filter (see Deconvolution)
Prediction error filtering (see Linear prediction)
Predictor-corrector form of Kalman filter, 246–247
Probability, transformation of variables (see also Gaussian random variables), 144
Projection (see Linear projection)
Properties of best linear unbiased estimators (see Best linear unbiased estimator)
Properties of estimators (see Large sample properties of estimators; Small sample properties of estimators)
Properties of least-squares estimator (see Least-squares estimator)
Properties of maximum-likelihood estimators (see Maximum-likelihood estimators)
Properties of mean-squared estimators (see Mean-squared estimators)
q-slice algorithm (see Higher-order statistics applied to linear systems)
q-slice formula, 483–484
Random processes (see Gauss–Markov random sequences; Second-order Gauss–Markov random sequences)
Random variables (see Gaussian random variables)
Recursive calculation of state covariance matrix, 217–218
Recursive calculation of state mean vector, 217–218
Recursive processing (see Best linear unbiased estimator; Least-squares estimation processing)
Reduced-order Kalman filter, 355–357
Reduced-order state estimator, 355
Reduced-order state vector, 354
References, 542–552
Reflection seismology, 21–22
Relinearized Kalman filter, 387
Representation problem, 2
Residual time series method (see Higher-order statistics applied to linear systems)
Restricted least-squares, 43, 120
Riccati equation (see Algebraic Riccati equation; Matrix Riccati equation)
Sample mean as a recursive digital filter, 7
Scale changes:
best linear unbiased estimator, 128–130
least-squares, 36–37
Second-order Gauss–Markov random sequences, 328–329
Sensitivity of Kalman filter, 260, 262–265
Signal-plus-noise model, 11
Signals, 2
Signal-space, 200
Signal-to-noise ratio, 220–222, 266, 465
Single-channel steady-state Kalman filter, 282–286
Single most likely replacement detector (see Binary detection)
Singular-value decomposition, derivation, 45–48 (see also Least-squares estimator)
Small sample properties of estimators (see also Least-squares estimator)
efficiency, 77–85
unbiasedness, 76–77
Smoothed estimate, 19
Smoothing:
double-stage, 309–313
single-stage, 306–309, 312–313
three types, 305
Square-law detector (see Binary detection)
Stabilized form for computing P(k + 1|k + 1), 250–251
Standard form for computing P(k + 1|k + 1), 250–251
State and parameter estimation, combined (see Extended Kalman filter; Maximum-likelihood state and parameter estimation)
State augmentation, 324, 351, 391
State equation solution, 371, 509–512
State estimation (see Filtering; Prediction; Smoothing)
State transition matrix (see Matrix)
State-variable models (see also Basic state-variable model, continuous-time; Basic state-variable model, discrete-time; Not-so-basic state-variable model):
constructing state-variable representations, 503–508
miscellaneous properties, 512–514
solutions of state equations, 509–512
state, state variables, and state space, 500–503
Steady-state approximation (see Maximum-likelihood state and parameter estimation)
Steady-state filter system, 282
Steady-state Kalman–Bucy filter, 421
Steady-state Kalman filter (see also Single-channel steady-state Kalman filter), 280
Steady-state MVD filter:
defined, 332
properties of, 337–338
relationship to IIR Wiener deconvolution filter, 338, 340
Steady-state predictor system, 283
Steady-state state-covariance matrix, 220
Stochastic convergence (see Convergence)
Stochastic linear optimal output feedback regulator problem, 424
Sufficient statistics (see also Factorization theorem; Lehmann–Scheffe theorem; Minimal sufficient statistic):
complete, 447–448
defined, 437–438
exponential families of distributions, 439–441
uniformly minimum-variance unbiased estimation, 444–448
Summary questions (see also Answers to summary questions), 8, 23–24, 39–40, 53–54, 69–70, 87–88, 104, 117–118, 132, 145, 157–158, 171, 188–189, 204–205, 223–224, 238–239, 255–256, 276–277, 300, 314–315, 341, 360–361, 376–377, 394, 409–410, 431–432, 448–449, 469, 491–492, 514–515
Time-delay estimation, 472
Total least-squares, 38–39
Trispectrum (see Polyspectra)
Unbiasedness (see Small sample properties of estimators)
Uniformly minimum-variance unbiased estimation (see Sufficient statistics)
Univariate Gaussian random variables (see Gaussian random variables)
Validation problem, 3
Variance estimator, 113–114
Volterra series representation of a nonlinear system, 17–18
Weighted least-squares estimator (see Best linear unbiased estimator; Least-squares estimator)
White noise:
discrete, 214
higher-order, 455
Wiener, biography, 291–293
Wiener filter:
derivation, 289–291
recursive, 291
relation to Kalman filter, 291, 293
Wiener–Hopf equations, 290
18.218.55.14