Visual walkthrough — Direction cosine matrix (DCM) — construction from Euler angles

3.5.3 · D2 · Physics › Guidance, Navigation & Control (GNC) › Direction cosine matrix (DCM) — construction from Euler angl

---

Step 1 — "Frame" kya hota hai, aur hum kya translate karne ki koshish kar rahe hain

KYA. Ek frame sirf teen arrows hain jo ek point par chipke hote hain, har ek doosre do se right angle par, har ek ki length 1. Hum do frames ek doosre ke upar draw karte hain:

• navigation frame $N$ — arrows ${\hat{n}}_1,{\hat{n}}_2,{\hat{n}}_3$ (socho: North, East, Down — world se fixed),
• body frame $B$ — arrows ${\hat{b}}_1,{\hat{b}}_2,{\hat{b}}_3$ (spacecraft par bolted, toh yeh uske saath ghoomte hain).

Chota sa hat $\hat{}$ sirf yeh matlab hai "is arrow ki length 1 hai" (ek unit vector). Subscript ($1,2,3$) teen arrows ko number karta hai.

KYUN. Space mein ek physical direction — jaise "Sun us taraf hai" — har frame mein alag numbers ki list hoti hai. DCM ka poora kaam ek list ko doosri mein convert karna hai. Converter banana se pehle humein do frames aur unke arrows ke beech ka angle dekhna hoga.

PICTURE. Figure mein body frame (magenta) navigation frame (violet) ke relative mein ghuma hua hai. Orange angle ${\theta }_{11}$ woh angle hai jo ${\hat{b}}_1$ aur ${\hat{n}}_1$ ke beech hai. Us angle par nazar rakho — yeh hamare matrix mein pehla number banta hai.

---

Step 2 — Dot product kyon ek cosine hai (apna main tool earn karna)

KYA. Humein woh tool justify karna hoga jo abhi humne use kiya: ${\hat{b}}_i\cdot {\hat{n}}_j$ cosine kyun hota hai?

YEH tool kyun, koi aur kyun nahi? Hum ek aisa single number chahte hain jo bole "do directions kitni aligned hain?" — parallel hone par bada, perpendicular hone par zero, opposing hone par negative. Dot product wahi operation hai jiska exactly yeh behaviour hai, aur unit-length arrows ke liye yeh bina kisi extra scaling ke pure cosine mein simat jaata hai. Isliye matrix cosines se bani hai, tangents ya kisi aur cheez se nahi.

PICTURE. Arrow ${\hat{b}}_i$ ko arrow ${\hat{n}}_j$ par shadow ki tarah daalo. Shadow ki length $\cos \theta$ hai. Woh shadow length hi entry hai.

---

Step 3 — Ek plane mein ek rotation (poori machine ka atom)

KYA. 3D bhool jao abhi ke liye. Frame ko angle $\theta$ se axis $3$ (vertical arrow) ke baare mein spin karo. Axis $3$ hilta nahi; sirf arrows $1$ aur $2$ apne shared plane mein swing karte hain. Hum ek fixed point ke naye coordinates $(x^{\prime },y^{\prime })$ nikalte hain.

KYUN. Poora DCM sirf is tarah ke single-axis spins se bana hai. Agar hum ek ko theek kar lein, toh axes ko relabel karke teeno mil jaate hain. Yeh atom hai; Steps 5–6 molecule assemble karte hain.

PICTURE. Ek point radius $r$ par, old $1$-axis se angle $\varphi$ par baitha hai. Hum axes ko forward $\theta$ se rotate karte hain. Naye axes ke relative mein point ab angle $\varphi -\theta$ par baitha dikhta hai (yeh $\theta$ se peeche swing karta hua lagta hai).

---

Step 4 — Spin ko matrix $R_3(\theta )$ ki tarah package karna

KYA. Do equations ko (plus untouched $3$-axis ko) ek grid mein stack karo — ek matrix. Matrix yahan sirf ek bookkeeping table hai: row $i$ batata hai ki old coordinates se new-coordinate $i$ kaise banao.

Matrix KYUN? Kyunki rotations chain karna tab matrix multiply karna ban jaata hai, aur multiplication ka ek fixed, unambiguous order hota hai. Woh order turns ka physical order encode karega (Step 5).

PICTURE. Figure colour-code karta hai ki Step 3 ka har term kahan jaata hai: $+\sin$ top-right, $-\sin$ diagonal ke neeche, aur corner mein akela $1$ us axis ke liye jiske baare mein humne spin kiya.

---

Step 5 — Hum kyun multiply karte hain, aur exactly kis order mein

KYA. 3-2-1 sequence ka matlab hai: pehle yaw $\psi$ axis $3$ ke baare mein, phir pitch $\theta$ naye axis $2$ ke baare mein, phir roll $\varphi$ sabse naye axis $1$ ke baare mein. Har turn wahan se shuru hota hai jahan pichla turn chhoda tha.

Order KYUN flip hota hai. Frames ki chain padho: $N\stackrel{R_3(\psi )}{\to }{N}^{\prime }\stackrel{R_2(\theta )}{\to }{N}^{\prime \prime }\stackrel{R_1(\varphi )}{\to }B$. Vector ke nav-coordinates ${\mathbf{v}}_N$ pehle $R_3$ se hit hote hain, phir result $R_2$ se, phir $R_1$ se. Kyunki pehle apply hone wala operation vector ke sabse paas (rightmost) hona chahiye, isliye physical order $3,2,1$ likhta hai: ${\mathbf{v}}_B=\underset{\text{last turn, leftmost}}{\underbrace{R_1(\varphi )}}\underset{\text{middle}}{\underbrace{R_2(\theta )}}\underset{\text{first turn, rightmost}}{\underbrace{R_3(\psi )}}{\mathbf{v}}_N$

PICTURE. Figure dikhata hai frame teen turns mein step karta hua, aur uske baaju mein matrix product leftward grow hota hua — pehla turn right par, last turn left par.

---

Step 6 — Multiplication block by block karna

KYA. Pehle $R_2(\theta )R_3(\psi )$ multiply karo, phir $R_1(\varphi )$ se premultiply karo. Matrix multiply = "left table ki row ko right table ke column se dot karo."

Is grouping mein KYUN. Pehle $R_2{R}_3$ karna yaw+pitch block ko isolate karta hai; phir $R_1$ ke $\varphi$ terms sirf lower two rows mein roll stir karte hain (roll row 1 ko akela chhod deta hai). Isliye final matrix ki top row mein koi $\varphi$ nahi hota — ek fact jo tum ek nazar mein check kar sakte ho.

PICTURE. Figure top row (pure pitch-yaw, orange) ko lower two rows (roll-mixed, magenta) se highlight karta hai, taaki tum dekh sako ki $\varphi$ kahan appear karta hai aur kahan nahi.

---

Step 7 — Degenerate case: pitch $=\pm 90^{\circ }$ par gimbal lock

KYA. $\theta =90^{\circ }$ set karo, toh $c\theta =0,s\theta =1$. Matrix collapse ho jaata hai aur yaw column aur roll pattern ek hi kaam karne lagte hain.

YEH KYUN toot jaata hai. Jab body $90^{\circ }$ se seedha upar pitch karta hai, toh yaw axis (originally vertical) aur roll axis (ab vertical) ek hi physical direction mein point karte hain. Kisi ko bhi ghuma lo — body usi line ke baare mein spin karta hai, toh do alag angle-pairs $(\psi ,\varphi )$ identical attitudes produce karte hain — ek poora degree of freedom gayab ho jaata hai. DCM khud ek perfectly good rotation rehta hai; yeh angle labels hain jo singular ho jaate hain.

PICTURE. Figure dikhata hai yaw aur roll arrows $\theta =90^{\circ }$ par ek vertical line mein merge hote hue — tum $\psi$ ya $\varphi$ kuch bhi twist karo aur body identically respond karta hai.

---

Step 8 — Small-angle limit: DCM ban jaata hai $I$ minus ek skew matrix

KYA. Teeno angles ko tiny hone do. Tab $\cos \approx 1$, $\sin \approx$ angle khud (radians mein), aur do chote angles ka koi bhi product negligible hai.

Yeh limit KYUN matter karti hai. Identity ke paas, attitude errors tiny hote hain. Yeh linearised form linearized error-state filter ki poori neenv hai jo real spacecraft par use hota hai.

PICTURE. Figure exact rotation ko uske straight-line small-angle approximation ke upar overlay karta hai — yeh zero ke paas chipke rehte hain aur sirf bade angles ke liye alag hote hain.

---

Ek-picture summary

Ek arc mein poori derivation: do frames → dot product cosines deta hai → ek planar spin → $R_3$ ki tarah package karo → teen spins chained (last leads) → poora $C_{BN}$ → uske do limits (gimbal lock aur small-angle).

Recall Walkthrough ki Feynman retelling

Do log ek hi paper airplane describe karte hain; ek ghoom gaya hai. "Naak udhar point kar rahi hai" translate karne ke liye hum baar baar ek simple sawaal poochte hain: tumhara arrow mera arrow ke along kitna hai? — aur jawaab hamesha ek cosine hota hai (ek arrow ki doosre par shadow). Hum pehle seekhte hain ki ek flat map ko ek angle se spin karo aur naye left/right/up numbers likho; woh chota turn-table $R_3$ hai. Kyunki spacecraft teen baar ghoomta hai — spin, nose tilt, wings roll — hum teen aisi tables stack karte hain. Pakad yeh hai: last turn multiplication ke front par jaata hai, kyunki vector pehle wale turn se milta hai. Inhe multiply karo aur bada DCM appear hota hai, pitch top-right corner mein akela hamare checkpoint ki tarah baitha hai. Pitch ko straight-up push karo aur do turns ek hi axis ke liye ladte hain (gimbal lock); teeno turns ko whispers tak shrink karo aur poori table ban jaati hai "identity minus a tiny skew nudge," jo aise hi flight computers attitude drift track karte hain.

Recall

Har DCM entry cosine kyun hoti hai? ::: Har entry do unit vectors ka dot product hai, aur unit vectors ka dot product unke beech ke angle ka cosine hota hai. 3-2-1 product mein, kaun sa physical turn sabse baayein wala matrix hai? ::: Aakhri wala (roll, $R_1$) — kyunki pehla turn vector par pehle act karna chahiye, isliye woh rightmost hota hai. $C_{13}=-\sin \theta$ mein koi roll ya yaw kyun nahi hai? ::: Roll ($R_1$) top row ko untouched chhod deta hai aur $R_3$ ka yaw us corner tak nahi pahunchta, isliye sirf pitch usse set karta hai — ek built-in sanity check. Gimbal lock par physically kya hota hai? ::: $\theta =\pm 90^{\circ }$ par yaw aur roll axes align ho jaate hain, toh sirf $\varphi -\psi$ matter karta hai aur ek degree of freedom kho jaata hai. Small-angle limit mein, turn order kyun matter karna band kar deta hai? ::: Chote angles ke products drop ho jaate hain, bacha rehta hai $I-[\mathbf{\alpha }\times ]$; first order tak rotations commute karte hain.

Connections

• 3.5.03 Direction cosine matrix (DCM) — construction from Euler angles (Hinglish)
• Quaternions — avoiding gimbal lock
• Euler angles — kinematic differential equations
• Rotation group SO(3) and orthogonal matrices
• Angular velocity and skew-symmetric matrices
• Kalman filter — linearized attitude error state