itsg.tugraz.at W I S S E N T E C H N I K L E I D E N S C H A F T Advanced satellite geodesy Matthias Ellmer, Institut für Geodäsie 28. April 16 u www.tugraz.at itsg.tugraz.at Variational equations Goal: Setup a system of linearized obs. Equations for satellite positions. l l 0 Ax e Unknown parameters x Reference (dynamic) orbit Design matrix A Observed positions or velocities r (t ), r (t ) 2 Matthias Ellmer, Institut für Geodäsie 28. April 16 r x itsg.tugraz.at Equation of motion Equation of motion (2nd order differential equation) r(t ) f (t , r,) Specific forces (Force per unit mass) - Gravity f (r ) V (r ) with GM V ( , , r ) R - Solar radiation pressure f (t , r ) E p (t ) Cr 3 - Tides - … A r rsun m r rsun Matthias Ellmer, Institut für Geodäsie 28. April 16 R n 0 r n 1 n c m 0 nm Cnm ( , ) snm S nm ( , ) itsg.tugraz.at Integration Equation of motion (2nd order differential equation) r(t ) f (t , r,) Integrate once to get the velocity t Solution is unique except 6 integration constants - Start position and velocity - Keplerian elements - Boundary positions r (t ) r0 f (t , r (t )) dt t0 Integrate twice for the position t t r (t ) r0 r0t f (t , r (t )) dt dt t0 t0 t r0 r0t (t t ) f (t , r (t )) dt t0 4 Matthias Ellmer, Institut für Geodäsie 28. April 16 r0 r0 Position has to be known to compute the position • Necessitates reference position • Iterative approach itsg.tugraz.at Integration algorithm Compute the full acceleration signal at each epoch r(t ) f (t , r,) Compute integrated position and velocities, for example with Euler step: r (t k 1 ) r (t k ) r (t k )t r (t k 1 ) r (t k ) r(t k )t (0,0) Determine inital values, for example through least-squares fit to PODs or previous iteration of dynamic orbit. r (t ) r0 r0t r (t ) r (t ) r0 r (t ) r0 r0 5 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Evaluation of force function r(t ) f (t , r (t ),) [Mayer-Gürr, 2006] → Integration has to be repeated until computed orbit does not change anymore. 6 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Result and analysis of orbit integration The smaller this difference, the better the integration algorithm. 7 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Difference between orbits after convergence The orbit still changes by ~100 µm between iterations! Spectral domain: Large power at 1/rev 8 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Encke approach Split the observed acceleration into two components: 1. An analytically computable component, originating a reference motion 𝐫𝑟 (𝑡) 2. A perturbing component, causing the perturbation 𝐬(𝑡) r(t ) g (t ) q(t ) Johann Franz Encke (1791-1865) The perturbed acceleration, position, and velocity is given by the departure from the reference motion. s(t ) r(t ) rr (t ) q(t ) t s (t ) r (t ) rr (t ) s 0 q(t ) dt t0 t s(t ) r (t ) rr (t ) s 0 s 0t (t t ) q(t ) dt t0 9 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝐬 𝑡 , 𝐬 𝑡 , and 𝐬 𝑡 are the Encke vectors. itsg.tugraz.at Choice of reference motion g (t ) 0 →Results in linear motion Position and velocity can be derived from the equinoctial elements with high precision and efficiency, as no trigonometric functions are used. In terms of Kepler elements, the equinoctial elements are given by: a=a λ=M+ω+Ω h = e sin(ω+Ω) g (t ) GM r (t ) r (t ) The choice of reference motion can reduce the power of the computed integral significantly. 10 Matthias Ellmer, Institut für Geodäsie 28. April 16 3 k = e cos(ω+Ω) p = tan(I/2) sin Ω q = tan(I/2) cos Ω →Results in Kepler motion itsg.tugraz.at Algorithm 1. 2. 3. 4. 5. Compute accelerations along reference orbit 𝐠 𝑡 . Compute perturbing acceleration 𝒒 𝑡 = 𝐫 𝑡 − 𝐠 𝑡 . Integrate perturbing acceleration to Encke vectors 𝐬 𝑡 and 𝐬 𝑡 . Add the Encke vectors to the reference motion 𝐫𝑟 𝑡 , 𝐫𝑟 𝑡 to yield 𝐫 𝑡 , 𝐫 𝑡 . Estimate initial position and velocity 𝐫0 𝑡 , 𝐫0 𝑡 to yield 𝐫 𝑡 , 𝐫 𝑡 . The reduction of the integrated acceleration leads to improvde accuracy. 11 Matthias Ellmer, Institut für Geodäsie 28. April 16 Linear Ellipses itsg.tugraz.at Optimizing the reference motion g (t ) GM r (t ) r (t ) 3 rr , 0 r0 rr , 0 r0 g (t ) GM Through smart choice of the reference ellipses, we can reduce the power of the integral further. We perform a least squares fit of the Kepler ellipses to the dynamic orbit, thus minimizing the spatial separation and the power of the integral. 12 Matthias Ellmer, Institut für Geodäsie 28. April 16 r (t ) r (t ) 3 rr , 0 r0 rr , 0 r0 itsg.tugraz.at Convergence of the best-fit orbit The fitted orbit is much closer to the true orbit • • 13 Matthias Ellmer, Institut für Geodäsie 28. April 16 The dynamic orbit converges to within ~nm. The power spectrum of the differences is white above ~ 4 cycles/revolution. itsg.tugraz.at Variational equations Goal: Setup a system of linearized obs. equations l l 0 Ax e Unknown parameters x Reference (dynamic) orbit Design matrix A Observed positions or velocities r (t ), r (t ) 14 Matthias Ellmer, Institut für Geodäsie 28. April 16 r x itsg.tugraz.at Unknown parameters 𝐱 = 𝐩, 𝐪, 𝐫𝟎 , 𝐫𝟎 𝑻 Satellite state parameters: 𝐫0 , 𝐫0 Measurement model parameters 𝐪: • Antenna offsets • Biases • … Force model parameters 𝐩: • 𝑐𝑛𝑚 , 𝑠𝑛𝑚 • Drag coefficient • • Radiation pressure coefficient … [Source: esa] 15 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Variational equation Initial values: Position and velocity at first epoch: Satellite state: Position and velocity at each epoch: 𝐫0 𝛂= 𝐫 0 𝐫 𝑡 y(𝑡) = 𝐫 𝑡 Design matrix 𝐀: Derivation of state y(𝑡) with respect to the parameters 𝐱 = 𝐩, 𝐪, 𝐫𝟎 , 𝐫𝟎 𝐫 𝑡 𝐫 𝑡 𝝏𝐫 𝑡 𝝏𝐩 = 𝝏𝐫 𝑡 𝝏𝐩 Force model parameter sensitivity matrix 𝐒𝐩 16 𝝏𝐫 𝑡 𝝏𝐪 𝝏𝐫 𝑡 𝝏𝐪 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝝏𝐫 𝑡 𝝏𝐫𝟎 Measurement model parameter sensitivity matrix 𝐒𝐪 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝐩 𝐪 𝐫𝟎 𝐫𝟎 State transition matrix 𝚽 Variational equations: y(𝑡) = 𝐒𝐩 𝐒𝐪 𝚽 Simplified: 𝐒𝐪 = 𝟎 y(𝑡) = 𝐒 𝚽 𝐩 𝛂 𝑻 𝐩 𝐪 𝛂 itsg.tugraz.at The design matrix r0 ,r0 cnm , snm , r (t0 ) S(t0 ) 3m Φ(t0 ) 36 r (t1 ) S(t1 ) 3m Φ(t1 ) 36 r (t 2 ) S(t 2 ) 3m Φ(t 2 ) 36 r (t N ) S(t N ) 3m Parameter Sensitivity Matrix Φ(t N ) 36 Simplified: 𝐒𝐪 = 𝟎 y(𝑡) = 𝐒 State Transition Matrix 17 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝚽 𝐩 𝛂 itsg.tugraz.at State transition matrix 𝚽 A change in the initial values 𝛂 = 𝐫𝟎 , 𝐫𝟎 𝑻 will affect all positions and velocities along the orbit arc. This effect is described by the state transition matrix 𝚽(𝑡) = 𝜕𝐲(𝑡) 𝜕𝛂 The change of the state vector is given by 𝐫 𝑡 𝐳 𝑡 = y (𝑡) = 𝐫 𝑡 𝝏𝐫 𝜕𝐳(𝑡) 𝝏𝐫 𝐙𝐲 = = 𝝏𝐫 𝜕𝐲(𝑡) 𝝏𝐫 𝑡 𝑡 𝑡 𝑡 𝝏𝐫 𝝏𝐫 𝝏𝐫 𝝏𝐫 𝑡 𝑡 𝑡 𝑡 Differential equation of the state transition matrix: 𝚽 = 𝐙𝐲 𝚽 18 Matthias Ellmer, Institut für Geodäsie 28. April 16 Gravity: r V V r Gravity gradient: 2V 2 x 2V r V r x2y V xz 2V xy 2V y 2 2V yz 2V xz 2V yz 2V z 2 itsg.tugraz.at Parameter sensitivity matrix 𝐒 A change in the force model parmaters 𝐩 = 𝑐𝑛𝑚 , 𝑠𝑛𝑚 , ... velocities along the orbit arc. 𝐒(𝑡) = 𝑻 will also affect all positions and 𝜕𝐲(𝑡) 𝜕𝐩 With 𝐳 𝑡 as before: 𝝏𝐫 𝑡 𝜕𝐳(𝑡) 𝝏𝐩 𝐙𝐩 = = 𝝏𝐫 𝑡 𝜕𝐩 𝝏𝐩 19 Matthias Ellmer, Institut für Geodäsie 28. April 16 Differential equation of the parameter sensitivity matrix: 𝐒 = 𝐙𝐲 𝐒 + 𝐙𝐩 itsg.tugraz.at Linear ordinary differential equation (ODE) General form 𝚽 = 𝐙𝐲 𝚽 𝐒 = 𝐙𝐲 𝐒 + 𝐙𝐩 a (t ) y (t ) b(t ) y (t ) c(t ) −𝐙𝐲 𝚽 + 𝚽 = 0 − 𝐙𝐲 𝐒 + 𝐒 = 𝐙𝐩 Solution: 1. find a solution of the homogeneous equation (t ) a (t ) y (t ) b(t ) y (t ) 0 2. Variation of the constants Insert into the differential equation (t ) c(t ) a (t ) f (t ) (t ) b(t ) f (t ) (t ) f (t ) y (t ) f (t ) (t ) (t ) y (t ) f (t ) (t ) f (t ) Solution: t c(t ) y (t ) (t ) dt C t b(t ) (t ) 0 20 (t ) b(t ) f (t ) (t ) c(t ) f (t )a (t ) (t ) b(t ) b(t ) f (t ) (t ) c(t ) c(t ) f (t ) t 1 b(t ) (t ) S(t ) Φ(t ) Φ (t )Z p dt C t t c(t ) 0 f (t ) dt b ( t ) ( t ) t0 Matthias Ellmer, Institut für Geodäsie 28. April 16 itsg.tugraz.at Finding the integration constant t 𝑡 𝚽 −1 𝐒 𝑡 = −𝚽 𝑡 𝑡′ 𝐙𝐩 ⅆ𝑡 + 𝐂 r (t ) r0 r0t (t t ) f (t , r (t )) dt t0 𝑡0 t r (t ) r0 f (t , r (t )) dt t0 1. Determine the inital values 𝚽0 and 𝐒0 : 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝚽 𝑡 = 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝝏𝐫 𝑡 𝝏𝐫𝟎 = 𝐈 𝐈⋅𝑡 𝟎 𝐈 𝚽0 = 𝐈6,6 𝝏𝐫 0 𝝏𝐩 𝐒0 = 𝝏𝐫 0 𝝏𝐩 = 𝟎6,m 2. Find the value of 𝐂 by using intial values: 𝑡 𝐒0 = −𝚽0 𝑡0 𝑡 𝚽0−1 𝐙𝐩 ⅆ𝑡 +𝐂 0 21 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝐂 = 𝟎6,m 𝚽 −1 𝑡 ′ 𝐙𝐩 ⅆ𝑡 𝐒 𝑡 = −𝚽 𝑡 𝑡0 itsg.tugraz.at Integration of parameter sensitivity matrix Design matrix A can be found by integration: The state transition matrix at all epochs: 𝚽 𝑡 = 𝐈 𝟎 𝐈⋅𝑡 𝐈 r0 ,r0 cnm , snm , Is this correct? r (t0 ) S(t0 ) 3m Φ(t0 ) 36 r (t1 ) S(t1 ) 3m Φ(t1 ) 36 r (t 2 ) S(t 2 ) 3m Φ(t 2 ) 36 𝑡 𝚽 −1 𝑡 ′ 𝐙𝐩 ⅆ𝑡 𝐒 𝑡 = −𝚽 𝑡 𝑡0 Dependance of velocity on force model parameters: 𝝏𝐫 𝑡 𝝏𝐩 𝐙𝐩 = 𝝏𝐫 𝑡 𝝏𝐩 22 𝟎3,𝑚 = 𝝏𝛻𝑉 𝑡 𝝏𝐩 Matthias Ellmer, Institut für Geodäsie 28. April 16 r (t N ) S(t N ) 3m Φ(t N ) 36 Parameter Sensitivity Matrix State Transition Matrix itsg.tugraz.at Composition of 𝚽: We recall that the state transition matrix is defined as: 𝚽 𝑡 = 𝜕𝐲 𝑡 𝜕𝛂 𝝏𝐫 𝑡 𝝏𝐫𝟎 = 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝑡 𝝏𝐫 𝑡 𝝏𝐫𝟎 𝝏𝐫 𝑡 𝝏𝐫𝟎 We split 𝚽 into two submatrices for position and velocity: 𝚽 𝑡 𝚽 𝑡 = 𝐫 𝚽𝐫 𝑡 𝑡0 𝑡 𝐟 𝑡 ′ , 𝐫 𝑡 ′ , … ⅆ𝑡′ 𝐫 𝑡 = 𝐫0 + 𝑡0 𝑡 − 𝑡′ 𝚽𝐫 𝑡 = 𝚽𝐫 𝑡 + The position appears on both sides of the equation Correct derivative for just the position using chain rule: 𝝏𝐫 𝑡 𝝏𝐫0 + 𝐫0 𝑡 = + 𝝏𝛂 𝝏𝛂 𝑡 𝑡 − 𝑡 ′ ⋅ 𝐟 𝑡 ′ , 𝐫 𝑡 ′ , … ⅆ𝑡′ 𝐫 𝑡 = 𝐫0 + 𝐫0 𝑡 + 𝑡 𝝏𝐟 𝑡 ′ , 𝐫 𝑡 ′ , … 𝑡−𝑡 ⋅ ⅆ𝑡′ 𝝏𝛂 ′ 𝑡0 𝑡 𝝏𝐫 𝑡 𝝏𝐫0 + 𝐫0 𝑡 = + ′ ′ 𝝏𝛂 𝝏𝛂 ⋅ 𝐓𝐫 𝑡 𝚽𝐫 𝑡 ⅆ𝑡′ 𝑡 𝝏𝐟 𝑡 ′ , 𝐫 𝑡 ′ , … 𝝏𝐫 𝑡 ′ 𝑡−𝑡 ⋅ ⅆ𝑡′ 𝝏𝐫 𝑡 ′ 𝝏𝛂 ′ 0 𝑡0 𝚽𝐫 𝑡 23 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝚽𝐫 𝑡 𝛁𝛁𝐕 𝑡′ ≝ 𝐓𝐫 𝑡 ′ 𝚽𝐫 𝑡′ itsg.tugraz.at Linearization of 𝚽: We can do the same for the velocity: 𝑡 𝑡 − 𝑡 ′ ⋅ 𝐓𝐫 𝑡 ′ 𝚽𝐫 𝑡 ′ ⅆ𝑡′ 𝚽𝐫 𝑡 = 𝚽𝐫 𝑡 + 𝑡0 We introduce the integral operator Combined equation system: 𝚽 = 𝚽 + 𝐊𝐓𝚽 𝑡 𝑡 − 𝑡 ′ (⋅) ⅆ𝑡′ 𝜅= 𝚽𝐫 𝚽 𝐊 𝐓𝐫 𝚽𝐫 = 𝐫 + 𝐫 𝚽𝐫 𝐊 𝐫 𝐓𝐫 𝚽𝐫 𝚽𝐫 𝑡0 which can be discretized and represented by an integration matrix 𝐊 𝐫 . 𝚽𝐫 = 𝚽𝐫 + 𝐊 𝐫 𝐓𝐫 𝚽𝐫 Φ3 N 6 24 I 33 I 33 I 33 t0 I 33 t1I 33 t n I 33 Solve for the state transition matrix: 𝚽 = 𝐈 − 𝐊𝐓 −1 𝚽 Gravity gradient: T3 N 3 N f diag r t0 Matthias Ellmer, Institut für Geodäsie 28. April 16 𝐒: f Integrate f r t1 r𝐒t N𝑡 = −𝚽 𝑡 𝑡 𝚽 −1 𝑡 ′ 𝐙𝐩 ⅆ𝑡 𝑡0 itsg.tugraz.at Variational equations cnm , snm , r0 ,r0 Goal: Setup a system of linearized obs. equations l l 0 Ax e r (t0 ) S(t0 ) 3m Φ(t0 ) 36 r (t1 ) S(t1 ) 3m Φ(t1 ) 36 r (t 2 ) S(t 2 ) 3m Φ(t 2 ) 36 Unknown parameters x Reference (dynamic) orbit Design matrix A Observed positions or velocities r x r (t ), r (t ) r (t N ) 25 Matthias Ellmer, Institut für Geodäsie 28. April 16 S(t N ) 3m Φ(t N ) 36 itsg.tugraz.at SST range rate observations 𝜌 Projection of differential velocity onto baseline. 26 Matthias Ellmer, Institut für Geodäsie 28. April 16 Observation equation can be derived from variational equations: 𝜌 = 𝐞12 , 𝐫2 − 𝐫1 𝜕𝜌 𝜕𝜌 𝜕 𝐫 𝐀= = 𝜕𝐱 𝜕 𝐫 𝜕𝐱 This can be realized as simple matrix multiplication. itsg.tugraz.at Observation equations for SST 27 Matthias Ellmer, Institut für Geodäsie 28. April 16
© Copyright 2026 Paperzz