tdehaeze/dehaeze20_optim_robus_compl_filte
1
0
Fork 0
Code Issues Pull Requests Releases Activity

Add paper's outline

Browse Source
This commit is contained in:
Thomas Dehaeze 2020-08-17 15:58:25 +02:00
parent ab16477991
commit d8296eb243
1 changed files with 27 additions and 171 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

198
paper/paper.org
View File

@ -75,200 +75,56 @@
#+end_IEEEkeywords #+end_IEEEkeywords
* Introduction * Introduction
<<sec:introduction>> <<sec:introduction>>
The basic idea of a complementary filter involves taking two or more sensors, filtering out unreliable frequencies for each sensor and combining the filtered outputs to get a better estimate throughout the entire bandwidth of the system.
To achieve this, the sensors included in the filter should complement one another by performing better over specific parts of the system bandwidth.
** Establish the importance of the research topic :ignore: * Optimal Super Sensor Noise: $\mathcal{H}_2$ Synthesis
# What are Complementary Filters <<sec:optimal_fusion>>
A set of filters is said to be complementary if the sum of their transfer functions is equal to one at all frequencies, (i.e.) its magnitude is one and its phase is zero.
# Why Design of Complementary Filter is important ** Sensor Fusion Architecture
The proper design of this particular kind of filter is of primary importance in a wide range of applications.
** Descriptions of the applications of complementary filtering :ignore: ** Super Sensor Noise
Often, multiple sensors with different noise or dynamical properties are used to measure the same physical quantity.
In such case, complementary filters can be used to merge the sensors and forms a "super sensor" that has gives a better estimate of the physical quantity over a wider bandwidth.
This is called sensor blending or sensor fusion.
# List of all the applications ** $\mathcal{H}_2$ Synthesis of Complementary Filters
# Increasing Sensor Bandwidth ** Example
This is widely used for the attitude estimation of unmanned aerial vehicles using various kind of sensors (accelerometers, gyroscopes, vision sensors, inclinometer) cite:zimmermann92_high_bandw_orien_measur_contr,corke04_inert_visual_sensin_system_small_auton_helic,min15_compl_filter_desig_angle_estim.
cite:shaw90_bandw_enhan_posit_measur_using_measur_accel Fast position measurement of flexible structure ** Robustness Problem
# Merging of different sensor types * Robust Sensor Fusion: $\mathcal{H}_\infty$ Synthesis
cite:matichard15_seism_isolat_advan_ligo (relative displacement measurement at low frequencies with inertial at high frequencies) <<sec:robust_fusion>>
# Reducing sensor noise ** Representation of Sensor Dynamical Uncertainty
cite:hua04_polyp_fir_compl_filter_contr_system
# Improving the control robustness ** Super Sensor Dynamical Uncertainty
cite:collette15_sensor_fusion_method_high_perfor
** Current research focus of the field / Literature review :ignore: ** $\mathcal{H_\infty}$ Synthesis of Complementary Filters
# Discuss the different approach to complementary filter design
The design methods for such filters goes from simple analytical formulas
# Analytical Formulas with limited number of parameter that let the designer choose the parameter value ** Example
# First Order
cite:corke04_inert_visual_sensin_system_small_auton_helic
# Second Order * Optimal and Robust Sensor Fusion: Mixed $\mathcal{H}_2/\mathcal{H}_\infty$ Synthesis
cite:min15_compl_filter_desig_angle_estim <<sec:optimal_robust_fusion>>
cite:jensen13_basic_uas
# Third Order and Higher orders ** Sensor Fusion Architecture
cite:shaw90_bandw_enhan_posit_measur_using_measur_accel
cite:zimmermann92_high_bandw_orien_measur_contr
cite:matichard15_seism_isolat_advan_ligo
cite:collette15_sensor_fusion_method_high_perfor
# FIR Filters ** Synthesis Objective
cite:hua05_low_ligo
cite:hua04_polyp_fir_compl_filter_contr_system
cite:matichard15_seism_isolat_advan_ligo
# Non-linear methods ** Mixed $\mathcal{H}_2/\mathcal{H}_\infty$ Synthesis
cite:mahony08_nonlin_compl_filter_special_orthog_group
# Time Varying ** Example
cite:pascoal99_navig_system_desig_using_time
# Alternate Formulation * Experimental Validation
cite:jensen13_basic_uas (feedback system, P, PI, classical control theory for filter design) <<sec:experimental_validation>>
cite:brown72_integ_navig_system_kalman_filter
# LMI / convex Optimization ** Experimental Setup
cite:pascoal99_navig_system_desig_using_time
# Least Square method for finding the optimal filter coefficients ** Sensor Noise and Dynamical Uncertainty
cite:min15_compl_filter_desig_angle_estim
** Describe a gap in the research :ignore: ** Mixed $\mathcal{H}_2/\mathcal{H}_\infty$ Synthesis
# There is a need to design optimal complementary filters
Although
** Describe the paper itself / the problem which is addressed :ignore: ** Super Sensor Noise and Dynamical Uncertainty
In this paper, we propose
** Introduce Each part of the paper :ignore:
The body of the paper consists of five parts followed by a conclusion.
* H-Infinity synthesis of complementary filters
<<sec:hinf_filters>>
First order complementary filters are easy to synthesize. For instance, one can use the following filters
\begin{equation}
H_H(s) = \frac{s/\w_0}{1 + s/\w_0}; \quad H_L(s) = \frac{1}{1 + s/\w_0}
\end{equation}
with $\w_0$ is the tuning parameter corresponding to the crossover frequency of the filters.
However, the manual design of higher order complementary filters is far more complex and we have to use an automatic synthesis technique.
As shown in Sec. ref:sec:trans_perf, most of the performance requirements can be expressed as upper bounds on the magnitude of the complementary filters.
Thus, the $\mathcal{H}_\infty$ framework seems adapted and we here propose a technique to synthesis complementary filters while specifying uppers bounds on their magnitudes.
** $\hinf$ problem formulation
<<sec:hinf_conf>>
In this section, we formulate the $\hinf$ problem for the synthesis of complementary filters.
The synthesis objective is to shape an high pass filter $H_H$ and a low pass filter $H_L$ while ensuring their complementary property ($H_H + H_L = 1$).
To do so, we define two weighting functions $w_L$ and $w_H$ that will respectively used to shape $H_L$ and $H_H$.
The synthesis problem is then
\begin{subnumcases}{\text{Find } H_L, H_H \text{ such that}}
H_L \text{ and } H_H \text{ are stable} \label{eq:hinf_cond_stability}\\
H_L + H_H = 1 \label{eq:hinf_cond_complementarity} \\
|w_L H_L| \le 1 \quad \forall\omega \label{eq:hinf_cond_hl} \\
|w_H H_H| \le 1 \quad \forall\omega \label{eq:hinf_cond_hh}
\end{subnumcases}
To express this synthesis problem into an $\hinf$ synthesis problem, we define the following generalized plant $P$ (also shown on Fig. ref:fig:sf_hinf_filters_plant_b):
#+NAME: eq:generalized_plant
\begin{equation}
\colvec{w\\u} = P \colvec{z_H \\ z_L \\ v}; \quad P = \begin{bmatrix} w_H & -w_H \\ 0 & w_L \\ 1 & 0 \end{bmatrix}
\end{equation}
#+name: fig:sf_hinf_filters_plant_b
#+caption: Generalized plant for the synthesis of the complementary filters
#+attr_latex: :float t :scale 1
file:./figs/sf_hinf_filters_plant_b.pdf
The $\hinf$ synthesis objective is then to design a stable filter $H_L$ (Fig. ref:fig:sf_hinf_filters_b) such that the $\mathcal{H}_\infty$ norm of the transfer function from $w$ to $[z_H, \ z_L]$ is less than $1$:
\begin{equation}
\hnorm{\begin{matrix} (1 - H_L) w_H \\ H_L w_L \end{matrix}} \le 1
\end{equation}
Which is equivalent to
#+NAME: eq:hinf_problem
\begin{equation}
\hnorm{\begin{matrix} H_H w_H \\ H_L w_L \end{matrix}} < 1 \text{ by choosing } H_H = 1 - H_L
\end{equation}
#+name: fig:sf_hinf_filters_b
#+caption: $\hinf\text{-synthesis}$ of complementary filters
#+attr_latex: :float t :scale 1
file:./figs/sf_hinf_filters_b.pdf
Performance conditions eqref:eq:hinf_cond_hl and eqref:eq:hinf_cond_hl are satisfied by eqref:eq:hinf_problem.
Complementary condition eqref:eq:hinf_cond_complementarity is satisfied by design: $H_H = 1 - H_L$ and thus $H_L + H_H = 1$.
The stability condition eqref:eq:hinf_cond_stability is guaranteed by the $H_\infty$ synthesis (*reference*).
Using this synthesis method, we are then able to shape at the same time the high pass and low pass filters while ensuring their complementary.
** Control requirements as $\mathcal{H}_\infty$ norm of complementary filters
As presented in Sec. ref:sec:trans_perf, almost all the requirements can be specified with upper bounds on the complementary filters.
However, robust performance condition eqref:eq:robust_perf_a is not.
With the $\mathcal{H}_\infty$ synthesis the condition eqref:eq:hinf_problem only ensure
\begin{align*}
\hnorm{\begin{matrix} H_H w_H \\ H_L w_L \end{matrix}} \le 1 &\Leftrightarrow \max_\omega \sqrt{|w_L H_L|^2 + |w_H H_H|^2} \le 1\\
&\Rightarrow |w_L H_L| + |w_H H_H| \le \sqrt{2} \quad \forall\omega
\end{align*}
And thus we have almost robust stability.
** Choice of the weighting functions
<<sec:hinf_weighting_func>>
We here give some advice on the choice of the weighting functions used for the synthesis of the complementary filters.
The shape should be such that the performance requirements are met as explain in Sec. ref:sec:trans_perf.
However, one should be careful when designing the complementary filters, and should only use stable and minimum phase transfer functions.
The order of the weights should stay reasonably small as this will increase the complexity of the optimization problem.
One should not forget the fundamental limitations of feedback control such that $S + T = 1$.
Similarly, we here have that $H_L + H_H = 1$ which implies that $H_L$ and $H_H$ cannot be made small at the same time.
** Trade-off between performance and robustness
** Analytical formula of complementary filters
<<sec:analytical_complementary_filters>>
To simplify the synthesis, one can use already synthesized filters
\begin{align}
H_L(s) = \frac{1}{1 + \frac{s}{\omega_0}}\\
H_H(s) = \frac{\frac{s}{\omega_0}}{1 + \frac{s}{\omega_0}}
\end{align}
\begin{align}
H_L(s) &= \frac{(1+\alpha) (\frac{s}{\omega_0})+1}{\left((\frac{s}{\omega_0})+1\right) \left((\frac{s}{\omega_0})^2 + \alpha (\frac{s}{\omega_0}) + 1\right)}\\
H_H(s) &= \frac{(\frac{s}{\omega_0})^2 \left((\frac{s}{\omega_0})+1+\alpha\right)}{\left((\frac{s}{\omega_0})+1\right) \left((\frac{s}{\omega_0})^2 + \alpha (\frac{s}{\omega_0}) + 1\right)}
\end{align}
\begin{align}
H_L(s) &= \frac{\left(1+(\alpha+1)(\beta+1)\right) (\frac{s}{\omega_0})^2 + (1+\alpha+\beta)(\frac{s}{\omega_0}) + 1}{\left(\frac{s}{\omega_0} + 1\right) \left( (\frac{s}{\omega_0})^2 + \alpha (\frac{s}{\omega_0}) + 1 \right) \left( (\frac{s}{\omega_0})^2 + \beta (\frac{s}{\omega_0}) + 1 \right)}\\
H_H(s) &= \frac{(\frac{s}{\omega_0})^3 \left( (\frac{s}{\omega_0})^2 + (1+\alpha+\beta) (\frac{s}{\omega_0}) + (1+(\alpha+1)(\beta+1)) \right)}{\left(\frac{s}{\omega_0} + 1\right) \left( (\frac{s}{\omega_0})^2 + \alpha (\frac{s}{\omega_0}) + 1 \right) \left( (\frac{s}{\omega_0})^2 + \beta (\frac{s}{\omega_0}) + 1 \right)}
\end{align}
* Discussion
* Conclusion * Conclusion
<<sec:conclusion>> <<sec:conclusion>>
* Acknowledgment * Acknowledgment