Continuous-time to discrete-time transformation I 1 s + 1
Continuous-time to discrete-time transformation II 1 s + 1
G (s ) =
G (s ) =
1. If =??, nd an appropriate sampling period (2 points). 2. Compute G (z ) by ZOH method (4 points). 3. Compute G (z ) by zero-pole matching method (4 points).
1. If =??, nd an appropriate sampling period (2 points). 2. Compute G (z ) by impulse response method (4 points). 3. Compute G (z ) by Tustin method (4 points).
Impulse response by the correlation approach How can we nd the impulse response of a system using the correlation approach 1. when the input signal is a white noise with variance 2 ? (4 points) 2. when the input signal is not white? (6 points)
Step response by experiment How can we obtain the step response of a real system? 1. Initial conditions? (1 point) 2. How the amplitude is selected? (3 points) 3. Sampling period? (1 point) 4. How to identify a rst-order and a rst-order with delay model from the step response? (5 points)
Closed-loop identication by spectral analysis 1. Why are we interested in closed-loop identication? Give three reasons. (3 points) 2. What is the major problem? (1 points) 3. What happens when there is no disturbance and when there is no external excitation? (2 points) 4. How a good model can by obtained by spectral analysis? (4 points)
Parametric model from a nonparametric frequencydomain model 1. Dene an optimization problem?(3 points) 2. What are the problems of this optimization? (1 point) 3. How these problems can be xed (Levys method)? (4 points) 4. How Levys method can be improved? (2 points)
Pseudo Random Binary Sequence (PRBS) 1. How it is generated? (4 points) 2. Its period ? (1 points) 3. Its autocorrelation and power spectral density functions?(2 points) 4. How the richness at low frequencies can be improved? (3 points)
Closed-loop identication by parametric approach I 1. Why are we interested in closed-loop identication? Give three reasons. (3 points) 2. What are the main problems? Give two problems. (4 points) 3. How these problem can be solved? (3 points)
Autocorrelation of a colored noise n(k ) = (1 + c2 q
2
)e (k )
Impulse response by experiment How can we obtain the discrete impulse response of a real system? 1. How the amplitude is selected? (3 points) 2. How can it be computed by the numerical deconvolution?(4 points) 3. What are the main problems of the numerical deconvolution and how can it be solved? (3 points)
when e (k ) is a white noise with E {e (k )} = 0 and E {e 2 ( k ) } = 2 . 1. Compute E {n(k )}. (2 points) 2. Compute Rnn (0). (2 points) 3. Compute Rnn (1). (3 points) 4. Compute Rnn (2). (3 points)
Non-parametric model by frequency analysis Discrete-time white noise 1. Give its autocorrelation function. (2 points) 2. Give its spectral density function. (2 points) 3. How can we estimate its autocorrelation function for a nite number of data (biased and unbiased)? (4 points) 4. What is a colored noise? (2 points) 1. Explain the basis of the frequency analysis approach. (4 points) 2. How to choose the number and the range of the frequency points? (2 points) 3. How to chose the magnitude of the input signal? (2 points) 4. What are the advantages and disadvantages of the approach? (2 points)
Sampling and measurement errors in Fourier analysis 1. Explain the sampling error in the Fourier analysis approach. (3 points) Non-parametric model by Fourier analysis I Explain 5 important practical aspects in the Fourier analysis approach. (10 points) 2. How the sampling error can be reduced? (2 points) 3. Explain the measurement error in the Fourier analysis. (2 points) 4. How the measurement error can be reduced? (3 points)
Truncation error in Fourier analysis Non-parametric model by Fourier analysis II 1. Explain the basis of the Fourier analysis approach. (4 points) 2. Name three dierent source of error in Fourier analysis. (6 points) 1. Explain the truncation error in the Fourier analysis approach. (3 points) 2. How the truncation error can be reduced? (2 points) 3. What is the eect of windowing in the Fourier analysis. (5 points)
Parametric model from the impulse response Given the impulse response g (k ) for k = 0, , N 1 of Non-parametric model by spectral analysis analysis 1. Explain the basis of the spectral analysis approach. (4 points) 2. How the results can be improved by averaging? (2 points) 3. How the results can be improved by windowing? (4 points) G (z ) = b0 + b1 z 1 1 + a1 z 1 + a2 z 2 ,
1. Explain how we can compute the parameters of G (z ). (4 points) 2. How to compute b0 and b1 ? (2 points) 3. How to compute a1 and a2 ? (4 points)
Identication versus physical modeling 1. Compare physical modeling with system identication. (5 points) 2. Compare parametric identication and nonparametric identication. (5 points)
Correlation approach Starting from the convolution sum :
y (k ) =
j =0
g (j )u (k j ) + d (k )
show that Ruy (h) = g (h) Ruu (h). (10 points)
Sampling period 1. How its upper bound is computed? (2 points) 2. How its lower bound is computed? (2 points) 3. How in practice is it computed from the step response? (2 points) 4. How in practice is it computed from the frequency response? (2 points) 5. What is an anti-aliasing lter? (2 points) Input design 1. How do we choose a suciently rich input signal? (5 points) 2. How do we design a PRBS signal for identication? (5 points)
Least squares algorithm Given the equation error : (k ) = A(q 1 )y (k ) q d B (q 1 )u (k ) A(q 1 ) = 1 + a1 q 1 + + an q n with B (q 1 ) = b0 + b1 q 1 + + bm q m 1. Develop a least squares algorithm to nd the parameters of A and B . (5 points) 2. If y = [0, 1.1, 0.2, 0.1, 0.9], u = [0, 1, 1, 1, 1], n = 1, m = 0, d = 1, compute the model parameters. (5 points)
Forgetting factor Given the adaptation gain 1 1 Pk +1 = Pk + (k + 1)T (k + 1) and the parameter estimates : k +1 = k + Pk +1 (k + 1)[y (k + 1) T (k + 1) k ] 1. Explain what is and how is it chosen? (4 points) 2. Explain its eect on the parameter estimates when = 1 and < 1. (6 points)
Recursive parameter estimation algorithm Given
k
(i )y (i ) ; Pk =
k i =1
1 (i ) (i )
T
Estimation error 1. What is the parameter estimation error? (2 points) 2. Compute it for the LS algorithm if the output of the real system is presented by Y = 0 + E0 . (4 points) 3. Compute the covariance of the estimates when E0 is white? (4 points)
k = Pk
i =1
1 1. Develop a recursive equation for Pk . (4 +1 points)
k +1 . (6 2. Develop a recursive equation for points)
Instrumental Variable method Given the parameter estimation error for LS algorithm : 0 = (T )1 T E0 1. In which conditions LS estimates are unbiased? (3 points) 2. Explain the Instrumental Variable method to obtain unbiased estimates. (3 points) 3. How instrumental variables are chosen? (4 points)
Performance criteria for parametric identication 1. Give the block diagram and equation for the equation error. (2 points) 2. Write the equation error as a linear regression. (3 points) 3. Give the block diagram and equation for the output error. (2 points) 4. Give the block diagram and equation for the prediction error. (3 points)
Weighted least squares algorithm 1. Explain the weighted lest squares algorithm. (2 points) 2. Give the criterion and the parameter estimates. (3 points) 3. What is the forgetting factor. (2 points) 4. Give the weighting lter for a forgetting factor . (3 points)
Gauss-Newton algorithm 1. Explain the Gauss-Newton method for minimization of a criterion J ( ). (2 points) 2. Dene the optimization criterion for the OE method. (2 point) 3. Compute its gradient. (4 points) 4. Compute its Hessian. (2 points)
Output error method BJ method 1. Dene the BJ structure and its main assumption. (2 points) 2. Dene the ideal predictor and ideal prediction error. (3 points) 3. Dene a parameterized predictor. (3 points) 4. Dene the optimization criterion and comment on it. (2 points). 1. Dene the output error structure and its main assumption. (2 points) 2. Dene the ideal predictor and ideal prediction error. (2 points) 3. Dene a parameterized predictor. (2 points) 4. Dene the optimization criterion and comment on it. (2 points). 5. Explain the FIR structure as a special case. (2 points)
ARX method 1. Dene the ARX structure and its main assumption. (2 points) 2. Dene the ideal predictor and ideal prediction error. (2 points) 3. Dene a parameterized predictor. (2 points) 4. Write the predictor as a linear regression. (2 points) 5. Dene the optimization criterion and comment on it. (2 points). ARMAX method 1. Dene the ARMAX structure and its main assumption. (2 points) 2. Dene the ideal predictor and ideal prediction error. (3 points) 3. Dene a parameterized predictor. (3 points) 4. Dene the optimization criterion and comment on it. (2 points).
Frequency analysis of ARX method in the frequency 1. Give the expression of domain for the ARX method. (3 points) 2. What happens if G and G0 have the same structure but H and H0 do not? (2 points) 3. What happens if G and G0 do not have the same structure? (2 points) 4. In which frequency zone a better estimation is expected? (3 points)
Frequency analysis of OE method in the frequency 1. Give the expression of domain for the OE method. (3 points) 2. What happens if G and G0 have the same structure? (2 points) 3. What happens if G and G0 do not have the same structure? (2 points) 4. How can we obtain a better model in the frequency zone of interest? (3 points)
Frequency analysis of ARMAX method in the frequency 1. Give the expression of domain for the ARMAX method. (3 points) 2. What happens if G and G0 have the same structure but H and H0 do not? (2 points) 3. What happens if G and G0 do not have the same structure? (2 points) 4. In which frequency zone a better estimation is expected? (3 points)
Frequency analysis of BJ method in the frequency 1. Give the expression of domain for the BJ method. (3 points) 2. What happens if G and G0 have the same structure but H and H0 do not? (2 points) 3. What happens if G and G0 do not have the same structure? (2 points) 4. In which frequency zone a better estimation is expected? (3 points)
Prediction error method Explain the three steps of the prediction error method for parametric model identication. 1. Step 1 (6 points). 2. Step 2 (2 points). 3. Step 3 (2 points).
Closed-loop identication by parametric approach II 1. Explain the direct method. (2 points) 2. Explain the use of IV method. (2 points) 3. Explain the indirect method. (3 points) 4. Explain its properties. (3 points)
Statistical validation 1. What is the whiteness test? (3 points) 2. What is the cross-correlation test? (3 points) 3. How the condence intervals are computed? (4 points)
Model validation 1. Validation w.r.t the objective. (2 points) 2. Validation based on a new set of data. (4 points) 3. Validation by statistical method. (4 points)
Pseudo linear regression algorithm Order estimation (n, m, d ) 1. From step response? (3 points) 2. From frequency response? (2 points) 3. From zero-pole cancellation? (2 points) 4. From the evolution of the loss function? (3 points) 1. Explain this iterative method to minimize the norm of prediction error. (4 points) 2. For OE structure write the predictor as a pseudo linear regression. (2 points) 3. Explain the computations in each iteration. (2 points) 4. How this method is initialized? (2 points)
