Foundations — Observability matrix — rank test

3.5.33 · D1 · Physics › Guidance, Navigation & Control (GNC) › Observability matrix — rank test

Is page pe assume kiya gaya hai ki tumne kuch nahi dekha. Hum Observability matrix — rank test ka har letter, matrix, aur operation ground up se, ek ek brick karke banate hain. Upar se neeche padho; har item sirf wahi use karta hai jo pehle aa chuka hai.

0. Woh picture jis par hum baar baar laute hain

Ek sealed box socho. Andar spinning gears hain jo tum dekh nahi sakte. Bahar ek dial pe sirf ek needle hai. Bas yahi puri kahani hai.

• Gears ki positions = state (hidden).
• Needle ki reading = output (visible).
• Woh physics jo ek gear se doosra gear chalata hai = matrix $A$.
• Woh wiring jo gears ko needle se connect karti hai = matrix $C$.

Neeche har symbol in chaar cheezon mein se ek hai, ya unhe measure karne ka tool hai.

1. Ek number line, ek vector, aur ${\mathbb{R}}^n$

Picture: $2$ slots wala vector ek flat plane mein arrow hai; pehla number kitna right, doosra kitna upar. $3$ slots wala vector us room mein arrow hai jisme tum baithe ho. $n$ slots ke liye hum draw nahi kar sakte, lekin idea same hai: $n$ independent knobs, har ek kisi number pe set.

Topic ko ye kyun chahiye: hidden state $x$ ek vector hai ${\mathbb{R}}^n$ mein. Number $n$ yeh batata hai ki kitne hidden gears hain. Sab kuch — matrices ki size, derivatives kab rokni hain — $n$ mein count hota hai.

2. Matrix ek aisi machine hai jo vector khaati hai

Picture: grid mat socho. Socho ek machine: tum ek taraf se vector daalo, aur doosri taraf se ek (shayad alag) vector nikalti hai. Numbers ki grid bas machine ki settings hai.

"Vector khaati hai" kyun? $A$ current state $x$ leta hai aur state ka rate of change batata hai — gears ek doosre ko kaise push karte hain. $C$ state $x$ leta hai aur meter reading produce karta hai. Dono aisi machines hain jo ek vector ko doosre vector mein badal deti hain.

Agar matrices nayi hain, toh continue karne se pehle State-space representation aur Rank and null space of a matrix mein real time lagao.

3. Chaar LTI matrices: $A,B,C,D$

Parent note $\dot{x}=Ax+Bu,y=Cx+Du$ se kholti hai. Yahan us line ka har symbol hai.

Picture: neeche, arrows information flow dikhate hain. $A$ state ko apne aap loop karta hai; $C$ state ko needle tak tap karta hai.

Observability ke liye sirf $A$ aur $C$ kyun matter karte hain: tum $u$ jaante ho (tumne knobs push ki) aur tum $B,D$ jaante ho (ye woh wiring hai jo tumne banayi). Toh $y$ ka woh hissa jo $u$ ki wajah se hai calculate karke subtract kiya ja sakta hai. Jo bachta hai — free response kehlaata hai — sirf $A$ (state kaise evolve hoti hai) aur $C$ (ye sensor tak kaise pahunchi) pe depend karta hai. Isliye poora rank test sirf $(A,C)$ use karta hai.

4. Transpose $(\cdot {)}^{\top }$

Picture: grid ko ek taraf khada kar do. Ek wide row jo let rahi thi woh ek tall column ban jaati hai jo khadi hai.

Topic ko ye kyun chahiye: "duality shortcut" kehta hai $(A,C)$ ki observability $(A^{\top },C^{\top })$ ki controllability ke barabar hai. Woh sentence bhi padhne ke liye tumhe pata hona chahiye ki chhota $\top$ kya karta hai. Dekho Controllability matrix — rank test.

5. Matrices ko stack karna (block matrices)

Picture: ek ke upar ek rakhi gayi bricks. Observability matrix $\mathcal{O}$ aisi bricks ka ek tower hai: $C$, phir $CA$, phir $C{A}^2$, aur aise hi. Har brick state ka ek aur "view" hai, jo ek aur derivative se milta hai.

Topic ko ye kyun chahiye: $\mathcal{O}$ yahi tower hai. Ise stacked rows ki tarah samajhna — ek mysterious single symbol ki tarah nahi — rank test ki key hai.

6. Derivative aur $\dot{y}(0)$ — hum differentiate kyun karte hain

Picture: needle dekho. Uski position $y(0)$ hai. Lekin tum usse hilte bhi dekhte ho — woh speed $\dot{y}(0)$ hai. Aur kya woh motion speed up ho rahi hai — woh $\ddot{y}(0)$ hai. Har nayi derivative ek nayi, independent information piece hai jo tum sirf usi needle ko aur dhyan se dekhke nikalte ho.

Derivative kyun, aur nahin, maano, integral ya Fourier transform? Kyunki derivative exactly woh tool hai jo poochti hai "yeh quantity abhi kaise badal rahi hai?" — aur model $\dot{x}=Ax$ batata hai ki change $Ax$ hai. Toh output ko differentiate karna ek aur $A$ ki copy multiply karta hai, jo exactly woh hai jisse hum $CA$, $C{A}^2$ tak, state mein aur gehre pahunche. Koi aur operation yeh clean ladder of $C{A}^k$ produce nahi karta.

7. Matrix exponential $e^{At}$

Picture: ek time-machine matrix. Use $x(0)$ do aur ye $x(t)$ return karta hai — $t$ seconds ki coupled spinning ke baad har gear kahan hogi.

Topic ko ye kyun chahiye: free response $y(t)=C{e}^{At}x(0)$ hai. Yahi woh equation hai jis par observability bani hai. Tumhe $e^{At}$ compute nahi karna — tumhe sirf itna jaanna hai ki ye exist karti hai, ki ye identity se start hoti hai ($e^{A\cdot 0}=I$), aur ki ise differentiate karne se ek $A$ factor neeche aata hai: $\frac{d}{dt}{e}^{At}=A{e}^{At}$. Woh aakhri fact poori $C{A}^k$ ladder ka engine hai. Dekho Kalman Filter jahan ye reconstruction practice mein use hoti hai.

8. Identity matrix $I$ aur matrix powers $A^k$

Picture: $I$ ek aaina hai jo tumhara vector wapas de deta hai bina badlaav ke. $A^2$ matlab state ko physics ke ek time-step se chalana, phir doosre se.

Topic ko ye kyun chahiye: $\mathcal{O}$ ki rows $C,CA,C{A}^2,\dots ,C{A}^{n-1}$ hain — literally $C$ times $A$ ke successive powers. Aur $A^0=I$ isliye pehli row sirf $C$ hai.

9. Rank aur null space — test ka dil

Picture: matrix ke through ek light chalaao. Rank count karta hai kitni independent shadows wo daal sakti hai. Koi bhi state jo koi shadow nahi daalti null space mein rehti hai — woh invisible hai.

Topic ko ye kyun chahiye — yahi punchline hai:

• Hum $\left[\begin{array}{c}y(0)\\ \dot{y}(0)\\ ⋮\end{array}\right]=\mathcal{O}x(0)$ ko unknown $x(0)$ ke liye solve karna chahte hain.
• Iska ek unique answer har state ke liye tab hoga jab $\mathcal{O}$ ka rank $n$ ho (full column rank), matlab koi bhi do states same output record produce nahi karti.
• Agar rank $<n$ hai, toh koi nonzero $x_0\in \ker (\mathcal{O})$ hai jo $y(t)\equiv 0$ banata hai — ek unobservable state, ek gear jo needle ko kabhi nudge nahi karta.

Dono ideas pe deep-dive: Rank and null space of a matrix.

10. Cayley–Hamilton — kyun ladder $n-1$ pe rukti hai

Picture: ladder $C,CA,C{A}^2,\dots$ eventually naye rungs dena band kar deti hai. $C{A}^{n-1}$ ke baad, agla har rung $C{A}^n,C{A}^{n+1},\dots$ un rungs ka mix hai jo tumhare paas pehle se hain. Toh tower $\mathcal{O}$ $n$ blocks pe complete hai.

Topic ko ye kyun chahiye: ye batata hai exactly kahan rokna hai. Iske bina, hum nahi jaante ki $n$ derivatives kaafi hain. Poori kahani Cayley–Hamilton theorem mein.

Prerequisite map

Equipment checklist

Khud test karo — right side cover karo aur zabar se jawab do.

Connections

• State-space representation (jahan se $A,B,C,D$ aate hain)
• Rank and null space of a matrix (core linear-algebra tool)
• Cayley–Hamilton theorem (kyun $A^{n-1}$ pe rukते hain)
• Controllability matrix — rank test (transpose-dual)
• Kalman Filter (is reconstruction ko practice mein use karta hai)
• Observability matrix — rank test (parent — agla yahan jao)