Cognitive Dynamic Systems Perception-action Cycle, Radar and Radio
The principles of cognition are becoming increasingly important in the areas of signal processing, communications and control. In this groundbreaking book, Simon Haykin, a pioneer in the field and an award-winning researcher, educator and author, sets out the fundamental ideas of cognitive dynamic s...
Uloženo v:
| Hlavní autor: | |
|---|---|
| Médium: | E-kniha Kniha |
| Jazyk: | angličtina |
| Vydáno: |
Cambridge
Cambridge University Press
22.03.2012
|
| Vydání: | 1 |
| Témata: | |
| ISBN: | 9780521114363, 0521114365 |
| On-line přístup: | Získat plný text |
| Tagy: |
Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
|
Obsah:
- 6.13.3 Communications between scene analyzer and perceptual memory
- 6.3 Baseband model of radar signal transmission -- 6.3.1 Baseband models of the transmitted and received signals -- 6.3.2 Bank of matched fi lters and envelope detectors -- 6.3.3 State-space model of the target -- 6.3.4 Dependence of measurement noise on the transmitted signal -- 6.3.5 Closing remarks -- 6.4 System design considerations -- 6.5 Cubature Kalman filter for target-state estimation -- 6.5.1 Cubature rule of third degree -- 6.5.2 Probability-distribution flow-graph of the Bayesian filter -- 6.5.3 Time update -- 6.5.4 Measurement update -- 6.5.5 Summarizing remarks -- 6.6 Transition from perception to action -- 6.6.1 Feedback information about the target -- 6.6.2 Posterior expected error covariance matrix -- 6.7 Cost-to-go function -- 6.7.1 Cost-to-go function using mean-square error -- 6.7.2 Cost-to-go function using Shannon's entropy -- 6.7.3 Another information-theoretic viewpoint of the entropy-based cost-to-go function -- 6.8 Cyclic directed information-flow -- 6.8.1 Bottom-up transmission path -- 6.8.2 Top-down transmission path -- 6.9 Approximate dynamic programming for optimal control -- 6.9.1 Step 1: cost-to-go function for compressing information about the radar environment -- 6.9.2 Step 2: approximation in the measurement space -- 6.9.3 Special case: dynamic optimization -- 6.10 The curse-of-dimensionality problem -- 6.11 Two-dimensional grid for waveform library -- 6.12 Case study: tracking a falling object in space -- 6.12.1 Modeling the reentry problem -- 6.12.2 Radar configurations -- 6.12.3 Performance metric -- 6.12.4 Simulation results -- 6.12.5 Comments on the simulation results -- 6.13 Cognitive radar with single layer of memory -- 6.13.1 Cyclic directed information flow in cognitive radar with single layer of memory -- 6.13.2 Communication among subsystems in cognitive radar
- 4.1 Probability, conditional probability, and Bayes' rule -- 4.1.1 Conditional probability -- 4.1.2 Bayes' rule -- 4.2 Bayesian inference and importance of the posterior -- 4.2.1 Likelihood -- 4.2.2 The likelihood principle -- 4.2.3 Sufficient statistic -- 4.3 Parameter estimation and hypothesis testing: the MAP rule -- 4.3.1 Parameter estimation -- 4.3.2 Hypothesis testing -- 4.3.3 Summarizing remarks on Bayesian inference -- 4.4 State-space models -- 4.4.1 Sequential state-estimation problem -- 4.4.2 Hierarchy of state-space models -- 4.5 The Bayesian filter -- 4.5.1 Optimality of the Bayesian filter -- 4.5.2 Approximation of the Bayesian filter -- 4.6 Extended Kalman filter -- 4.6.1 Summarizing remarks on the extended Kalman filter -- 4.7 Cubature Kalman filters -- 4.7.1 Converting to spherical-radial integration -- 4.7.2 Spherical rule -- 4.7.3 Radial rule -- 4.7.4 Spherical-radial rule -- 4.7.5 Derivation of the CKF -- 4.7.6 Properties of the CKF -- 4.7.7 Summarizing remarks on the CKF -- 4.8 On the relationship between the cubature and unscented Kalman filters -- 4.8.1 Unscented Kalman filter -- 4.8.2 On the relationship between UKF and CKF -- 4.8.2.1 Theoretical considerations -- 4.8.2.2 Geometric considerations -- 4.8.2.3 Curse-of-dimensionality problem -- 4.8.3 Summarizing remarks -- 4.9 The curse of dimensionality -- 4.9.1 Case study on the curse-of-dimensionality problem -- 4.10 Recurrent multilayer perceptrons: an application for state estimation -- 4.10.1 Description of the supervised training framework using the EKF -- 4.10.2 The EKF algorithm -- 4.10.3 Decoupled EKF -- 4.10.4 Summarizing remarks on the EKF -- 4.10.5 Supervised training of neural networks using the CKF -- 4.10.6 Adaptivity considerations -- 4.11 Summary and discussion -- 4.11.1 Optimal Bayesian filter -- 4.11.2 Extended Kalman filter -- 4.11.3 Cubature Kalman filters
- 2.11.2 Generalized Hebbian algorithm -- 2.11.3 Signal-flow graph of the GHA -- 2.12 Summary and discussion -- 2.12.1 Cognition -- 2.12.2 Two different views of perception -- Notes and practical references -- 3: Power-spectrum estimation for sensing the environment -- 3.1 The power spectrum -- 3.2 Power spectrum estimation -- 3.2.1 Parametric methods -- 3.2.2 Nonparametric methods -- 3.3 Multitaper method -- 3.3.1 Attributes of multitaper spectral estimation -- 3.3.2 Multitaper spectral estimation theory -- 3.3.3 Adaptive modification of multitaper spectral estimation -- 3.3.4 Summarizing remarks on the MTM -- 3.3.5 Comparison of the MTM with other spectral estimators -- 3.4 Space-time processing -- 3.4.1 Physical interpretation of the action performed by the MTM-SVD processor -- 3.5 Time-frequency analysis -- 3.5.1 Theoretical background of nonstationarity -- 3.5.2 Spectral coherences of nonstationary processes based on the Loève transform -- 3.5.3 Two special cases of the dynamic spectrum D (t0, f ) -- 3.5.3.1 Wigner-Ville distribution -- 3.5.3.2 Cyclic power spectrum -- 3.5.4 Instrumentation for computing Loève spectral correlations -- 3.6 Cyclostationarity -- 3.6.1 Fourier framework of cyclic statistics -- 3.6.2 Instrumentation for computing the Fourier spectral correlations -- 3.6.3 Relationship between the Fourier and Loève spectral coherences -- 3.6.4 Contrasting the two theories on cyclostationarity -- 3.7 Harmonic F-test for spectral line components -- 3.7.1 Brief outline of the F-test -- 3.7.2 Point regression single-line F -test -- 3.8 Summary and discussion -- 3.8.1 The MTM for power spectrum estimation -- 3.8.2 Extensions of the MTM -- 3.8.3 Concluding remarks -- 3.8.3.1 Mathematical framework -- 3.8.3.2 Practical requirement -- Notes and practical references -- 4: Bayesian filtering for state estimation of the environment
- Cover -- Cognitive Dynamic Systems -- Title -- Copyright -- Contents -- Preface -- Acknowledgments -- 1: Introduction -- 1.1 Cognitive dynamic systems -- 1.2 The perception-action cycle -- 1.3 Cognitive dynamic wireless systems: radar and radio -- 1.3.1 Cognitive radar -- 1.3.2 Cognitive radio -- 1.4 Illustrative cognitive radar experiment -- 1.4.1 The experiment -- 1.4.2 The environment -- 1.4.3 The radar -- 1.4.4 State-space model -- 1.4.5 Simulation results -- 1.5 Principle of information preservation -- 1.5.1 Feedback information -- 1.5.2 Bayesian filtering of the measurements -- 1.5.3 Information preservation through cognition -- 1.5.4 Concluding remarks -- 1.6 Organization of the book -- Notes and practical references -- 2: The perception-action cycle -- 2.1 Perception -- 2.1.1 Functional integration-across-time property of cognition -- 2.2 Memory -- 2.2.1 Perceptual memory -- 2.2.2 Executive memory -- 2.2.3 Final reciprocal coupling to complete the cognitive information-processing cycle -- 2.2.4 Roles of memory in cognition -- 2.3 Working memory -- 2.4 Attention -- 2.4.1 Roles of attention in cognition -- 2.5 Intelligence -- 2.5.1 Efficiency of processing information -- 2.5.2 Synchronized cognitive information processing -- 2.5.3 The role of intelligence in cognition -- 2.6 Practical benefits of hierarchy in the perception-action cycle -- 2.7 Neural networks for parallel distributed cognitive information processing -- 2.7.1 Benefits of neural networks -- 2.7.2 Models of a neuron -- 2.7.3 Multilayer feedforward networks -- 2.8 Associative learning process for memory construction -- 2.8.1 Pattern association -- 2.8.2 Replicator (identity) mapping -- 2.9 Back-propagation algorithm -- 2.9.1 Summary of the back-propagation algorithm -- 2.10 Recurrent multilayer perceptrons -- 2.11 Self-organized learning -- 2.11.1 Hebb's postulate of learning
- Notes and practical references -- 5: Dynamic programming for action in the environment -- 5.1 Markov decision processes -- 5.1.1 The basic problem -- 5.2 Bellman's optimality criterion -- 5.2.1 Dynamic-programming algorithm -- 5.2.2 Bellman's optimality equation -- 5.3 Policy iteration -- 5.3.1 Formulation of the policy iteration algorithm -- 5.4 Value iteration -- 5.4.1 Formulation of the value iteration algorithm -- 5.5 Approximate dynamic programming for problems with imperfect state information -- 5.5.1 Basics of problems with imperfect state information -- 5.5.2 Reformulation of the imperfect state-information problem as a perfect state-information problem -- 5.6 Reinforcement learning viewed as approximate dynamic programming -- 5.7 Q-learning -- 5.7.1 Summarizing remarks -- 5.8 Temporal-difference learning -- 5.8.1 Multistep TD learning -- 5.8.2 Eligible traces -- 5.8.3 Two limiting cases of TD learning -- 5.8.4 Summarizing remarks -- 5.9 On the relationships between temporal-difference learning and dynamic programming -- 5.9.1 λ-return -- 5.10 Linear function approximations of dynamic programming -- 5.11 Linear GQ(λ) for predictive learning -- 5.11.1 Objective function setting the stage for approximation -- 5.11.2 The GQ(λ) algorithm -- 5.11.3 Weight-doubling trick -- 5.11.4 Eligibility traces vector -- 5.11.5 New action-state feature vector -- 5.11.6 Summarizing remarks -- 5.11.7 Practical considerations -- 5.12 Summary and discussion -- 5.12.1 Bellman's dynamic programming -- 5.12.2 Imperfect state information -- 5.12.3 Reinforcement learning -- 5.12.4 Linear GQ(k) algorithm -- 5.12.5 Greedy-GQ -- 5.12.6 New generation of approximate dynamic programming algorithms: linear GQ methods -- Notes and practical references -- 6: Cognitive radar -- 6.1 Three classes of radars defined -- 6.2 The perception-action cycle