kshana 0.19.0

Open, reproducible PNT-resilience simulator with quantum-sensor performance models
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
// SPDX-License-Identifier: AGPL-3.0-only
//! Tightly-coupled GNSS/INS navigator on the unscented Kalman core.
//!
//! Where a *loosely*-coupled filter consumes a ready-made position/velocity fix (and so
//! needs four-plus satellites to even form one), the **tightly-coupled** filter ingests the
//! raw satellite measurements — pseudorange and Doppler (range-rate) — directly. Each
//! satellite contributes its own equations, so the navigator keeps correcting with **fewer
//! than four satellites in view** and rides through partial or total GNSS outages on its
//! propagated dynamics. The pseudorange/Doppler model is nonlinear in the state (a Euclidean
//! range and a range-rate projection), so the estimator is the scaled unscented Kalman filter
//! ([`crate::fusion::ukf::Ukf`]) rather than a Jacobian-linearised EKF.
//!
//! The state is the eight-vector `[px, py, pz, vx, vy, vz, b, d]`: ECEF position (m), velocity
//! (m/s), and the receiver clock bias `b` and drift `d` carried in range units (m and m/s, i.e.
//! `c·δt` and `c·δḟ`). The process model is constant-velocity position with a random-walk clock;
//! the measurement model maps the state to each satellite's pseudorange `ρ = |p − sᵢ| + b` and
//! range-rate `ρ̇ = (p − sᵢ)·(v − ṡᵢ)/|p − sᵢ| + d`.
//!
//! Scope (honest): this is the position/velocity/clock tightly-coupled core and its
//! outage-validation scenario. Folding it into the full 15-state INS error state (attitude and
//! IMU biases of [`crate::fusion::gnss_ins_ekf`]) and driving the satellite geometry from a live
//! SGP4 constellation with broadcast iono/tropo corrections in the loop are follow-ons (see
//! `ROADMAP.md`).

use crate::forces::gravity_accel;
use crate::fusion::ukf::Ukf;
use crate::integrator::rk4_step;

/// A GNSS satellite as seen by the navigator: ECEF position and velocity (m, m/s).
#[derive(Clone, Copy, Debug)]
pub struct Sat {
    /// Satellite ECEF position (m).
    pub pos: [f64; 3],
    /// Satellite ECEF velocity (m/s).
    pub vel: [f64; 3],
}

/// Geometric range from the state's position to `sat` (m).
fn range(state: &[f64], sat: &Sat) -> f64 {
    let d = [
        state[0] - sat.pos[0],
        state[1] - sat.pos[1],
        state[2] - sat.pos[2],
    ];
    (d[0] * d[0] + d[1] * d[1] + d[2] * d[2]).sqrt()
}

/// Modelled pseudorange `ρ = |p − sᵢ| + b` (geometric range plus the clock bias in range units).
pub fn pseudorange(state: &[f64], sat: &Sat) -> f64 {
    range(state, sat) + state[6]
}

/// Modelled range-rate (Doppler) `ρ̇ = (p − sᵢ)·(v − ṡᵢ)/|p − sᵢ| + d` — the line-of-sight
/// projection of the relative velocity plus the clock drift in range-rate units.
pub fn range_rate(state: &[f64], sat: &Sat) -> f64 {
    let d = [
        state[0] - sat.pos[0],
        state[1] - sat.pos[1],
        state[2] - sat.pos[2],
    ];
    let r = (d[0] * d[0] + d[1] * d[1] + d[2] * d[2]).sqrt();
    let vrel = [
        state[3] - sat.vel[0],
        state[4] - sat.vel[1],
        state[5] - sat.vel[2],
    ];
    (d[0] * vrel[0] + d[1] * vrel[1] + d[2] * vrel[2]) / r + state[7]
}

/// A tightly-coupled GNSS/INS navigator: the eight-state UKF plus its process-noise covariance.
pub struct TightlyCoupled {
    /// The underlying unscented Kalman filter over `[px,py,pz,vx,vy,vz,b,d]`.
    pub ukf: Ukf,
    /// Additive process-noise covariance applied each `propagate` (8×8).
    pub q: Vec<Vec<f64>>,
}

impl TightlyCoupled {
    /// A navigator with the given initial state `x0`, covariance `p0`, and process noise `q`.
    /// Uses an `α = 1` sigma-point spread (`λ = 0`, points at `±√n·Lᵢ`): the conventional
    /// `α = 1e-3` collapses the points onto the mean, which loses precision against the ~1e7-m
    /// position magnitudes of an Earth-frame navigation state.
    pub fn new(x0: Vec<f64>, p0: Vec<Vec<f64>>, q: Vec<Vec<f64>>) -> Self {
        let mut ukf = Ukf::new(x0, p0);
        ukf.alpha = 1.0;
        ukf.kappa = 0.0;
        Self { ukf, q }
    }

    /// Propagate the state forward by `dt` seconds under the constant-velocity / random-walk-clock
    /// dynamics. This is the only step taken during a GNSS outage. Returns `false` (state
    /// untouched) on a non-positive-definite covariance.
    pub fn propagate(&mut self, dt: f64) -> bool {
        let f = move |s: &[f64]| {
            vec![
                s[0] + s[3] * dt,
                s[1] + s[4] * dt,
                s[2] + s[5] * dt,
                s[3],
                s[4],
                s[5],
                s[6] + s[7] * dt,
                s[7],
            ]
        };
        self.ukf.predict(f, &self.q)
    }

    /// Propagate under the **two-body + J2 gravity field** (RK4 over the position/velocity block,
    /// random-walk clock), the dynamic coast an orbital tightly-coupled navigator runs on through a
    /// GNSS outage. For an unpowered orbiting platform gravity is the whole dynamics, so this is the
    /// faithful inertial-coast model — and unlike the straight-line [`Self::propagate`] it follows
    /// the orbit's curvature, which a constant-velocity coast cannot (the curvature alone is
    /// ~`½·(v²/r)·t²` ≈ tens of km over a two-minute outage at LEO). Returns `false` on a
    /// non-positive-definite covariance.
    pub fn propagate_orbital(&mut self, dt: f64) -> bool {
        let f = move |s: &[f64]| {
            // Derivative of the 6-vector [pos, vel] under gravity: ṙ = v, v̇ = g(r).
            let deriv = |_t: f64, y: &[f64]| {
                let a = gravity_accel([y[0], y[1], y[2]]);
                vec![y[3], y[4], y[5], a[0], a[1], a[2]]
            };
            let pv = rk4_step(&deriv, 0.0, &s[..6], dt);
            vec![
                pv[0],
                pv[1],
                pv[2],
                pv[3],
                pv[4],
                pv[5],
                s[6] + s[7] * dt,
                s[7],
            ]
        };
        self.ukf.predict(f, &self.q)
    }

    /// Update with the raw satellite measurements: pseudoranges `pr` and range-rates `rr`, one per
    /// satellite in `sats`, with per-measurement standard deviations `sigma_pr` (m) and `sigma_rr`
    /// (m/s). Works for any number of satellites — including fewer than the four a snapshot fix
    /// would need. Returns `false` on a degenerate covariance.
    pub fn update_gnss(
        &mut self,
        sats: &[Sat],
        pr: &[f64],
        rr: &[f64],
        sigma_pr: f64,
        sigma_rr: f64,
    ) -> bool {
        let k = sats.len();
        let sats_owned = sats.to_vec();
        let h = move |s: &[f64]| {
            let mut z = Vec::with_capacity(2 * sats_owned.len());
            for sat in &sats_owned {
                z.push(pseudorange(s, sat));
            }
            for sat in &sats_owned {
                z.push(range_rate(s, sat));
            }
            z
        };
        let mut z = Vec::with_capacity(2 * k);
        z.extend_from_slice(pr);
        z.extend_from_slice(rr);
        let mut r = vec![vec![0.0; 2 * k]; 2 * k];
        for i in 0..k {
            r[i][i] = sigma_pr * sigma_pr;
            r[k + i][k + i] = sigma_rr * sigma_rr;
        }
        self.ukf.update(h, &z, &r)
    }

    /// Current position error magnitude (m) against a known truth position.
    pub fn position_error(&self, truth: [f64; 3]) -> f64 {
        let x = &self.ukf.x;
        let d = [x[0] - truth[0], x[1] - truth[1], x[2] - truth[2]];
        (d[0] * d[0] + d[1] * d[1] + d[2] * d[2]).sqrt()
    }
}

#[cfg(test)]
mod tests {
    use super::*;
    use rand::SeedableRng;
    use rand_chacha::ChaCha8Rng;
    use rand_distr::{Distribution, Normal};

    // A GPS-like satellite geometry at MEO radius, well spread for good DOP.
    fn constellation() -> Vec<Sat> {
        vec![
            Sat {
                pos: [2.00e7, 1.00e7, 1.50e7],
                vel: [-1500.0, 2200.0, 600.0],
            },
            Sat {
                pos: [1.50e7, -1.20e7, 1.80e7],
                vel: [1800.0, 1500.0, -700.0],
            },
            Sat {
                pos: [2.20e7, 0.50e7, -1.00e7],
                vel: [-900.0, -2000.0, 1200.0],
            },
            Sat {
                pos: [1.00e7, 1.80e7, -1.50e7],
                vel: [2100.0, -800.0, -1000.0],
            },
            Sat {
                pos: [2.50e7, -0.80e7, 0.60e7],
                vel: [-1200.0, 1700.0, 1400.0],
            },
        ]
    }

    // Truth state at integer second `t`: a constant-velocity LEO user with a drifting clock,
    // matching the filter's process model exactly so a noiseless filter converges to truth.
    fn truth_at(t: f64) -> [f64; 3] {
        [7.0e6, 7.5e3 * t, 0.0]
    }
    fn truth_state(t: f64) -> [f64; 8] {
        [7.0e6, 7.5e3 * t, 0.0, 0.0, 7.5e3, 0.0, 30.0 + 0.1 * t, 0.1]
    }

    fn measurements(t: f64, sats: &[Sat]) -> (Vec<f64>, Vec<f64>) {
        let s = truth_state(t);
        let pr = sats.iter().map(|sat| pseudorange(&s, sat)).collect();
        let rr = sats.iter().map(|sat| range_rate(&s, sat)).collect();
        (pr, rr)
    }

    fn init_navigator() -> TightlyCoupled {
        // Perturbed initial guess: ~150 m position, ~2 m/s velocity, ~8 m clock-bias offset.
        let x0 = vec![
            7.0e6 + 150.0,
            -120.0,
            90.0,
            2.0,
            7.5e3 - 1.5,
            1.0,
            38.0,
            0.15,
        ];
        // Generous initial covariance; small process noise (truth is exactly CV + RW clock).
        let p0diag = [1.0e4, 1.0e4, 1.0e4, 1.0e2, 1.0e2, 1.0e2, 1.0e4, 1.0e0];
        let qdiag = [
            1.0e-2, 1.0e-2, 1.0e-2, 1.0e-3, 1.0e-3, 1.0e-3, 1.0e-2, 1.0e-4,
        ];
        let p0 = diag(&p0diag);
        let q = diag(&qdiag);
        TightlyCoupled::new(x0, p0, q)
    }

    fn diag(d: &[f64]) -> Vec<Vec<f64>> {
        let n = d.len();
        let mut m = vec![vec![0.0; n]; n];
        for (i, &v) in d.iter().enumerate() {
            m[i][i] = v;
        }
        m
    }

    #[test]
    fn pseudorange_and_range_rate_match_geometry() {
        // Sat at (3,4,0) from the origin ⇒ geometric range 5.
        let sat = Sat {
            pos: [3.0, 4.0, 0.0],
            vel: [0.0, 0.0, 0.0],
        };
        // Zero clock: pseudorange = 5, range-rate = 0 (no motion).
        let still = [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0];
        assert!((pseudorange(&still, &sat) - 5.0).abs() < 1e-12);
        assert!(range_rate(&still, &sat).abs() < 1e-12);
        // Clock bias 10 m and drift 2 m/s add directly.
        let clocked = [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 10.0, 2.0];
        assert!((pseudorange(&clocked, &sat) - 15.0).abs() < 1e-12);
        assert!((range_rate(&clocked, &sat) - 2.0).abs() < 1e-12);
        // Moving +x at 1 m/s toward a sat on +x ⇒ closing ⇒ range-rate −1.
        let sat_x = Sat {
            pos: [10.0, 0.0, 0.0],
            vel: [0.0, 0.0, 0.0],
        };
        let moving = [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0];
        assert!((range_rate(&moving, &sat_x) - (-1.0)).abs() < 1e-12);
    }

    #[test]
    fn noiseless_full_geometry_converges_to_truth() {
        let sats = constellation();
        let mut nav = init_navigator();
        for step in 1..=60 {
            let t = step as f64;
            nav.propagate(1.0);
            let (pr, rr) = measurements(t, &sats);
            assert!(nav.update_gnss(&sats, &pr, &rr, 1.0, 0.05));
        }
        // Noiseless measurements + an exact process model + good geometry ⇒ the filter pulls onto
        // the truth. Well under a metre after a minute.
        let err = nav.position_error(truth_at(60.0));
        assert!(err < 1.0, "converged position error = {err} m");
    }

    #[test]
    fn tracks_with_fewer_than_four_satellites() {
        // Only THREE satellites — a snapshot PVT fix is under-determined, but the tightly-coupled
        // filter is still observable over time through its dynamics, so it converges.
        let sats = &constellation()[..3];
        let mut nav = init_navigator();
        for step in 1..=120 {
            let t = step as f64;
            nav.propagate(1.0);
            let (pr, rr) = measurements(t, sats);
            assert!(nav.update_gnss(sats, &pr, &rr, 1.0, 0.05));
        }
        // Converges from a ~212 m initial error to ~13 m on three satellites — looser than the
        // sub-metre five-satellite case (weaker geometry, fewer equations per epoch) but a real
        // fix where a four-unknown snapshot PVT cannot be formed at all.
        let err = nav.position_error(truth_at(120.0));
        assert!(err < 20.0, "sub-4-satellite position error = {err} m");
    }

    #[test]
    fn survives_120s_gnss_outage_within_50m() {
        let sats = constellation();
        let mut nav = init_navigator();
        let mut rng = ChaCha8Rng::seed_from_u64(0xC0FFEE);
        let n_pr = Normal::new(0.0, 1.0).unwrap(); // σ_ρ = 1 m
        let n_rr = Normal::new(0.0, 0.05).unwrap(); // σ_ρ̇ = 5 cm/s
                                                    // 60 s of noisy GNSS to converge the position and, crucially, the velocity.
        for step in 1..=60 {
            let t = step as f64;
            nav.propagate(1.0);
            let (mut pr, mut rr) = measurements(t, &sats);
            for v in pr.iter_mut() {
                *v += n_pr.sample(&mut rng);
            }
            for v in rr.iter_mut() {
                *v += n_rr.sample(&mut rng);
            }
            assert!(nav.update_gnss(&sats, &pr, &rr, 1.0, 0.05));
        }
        let converged = nav.position_error(truth_at(60.0));
        // Now a full 120-second outage: propagate only, no measurements.
        for step in 61..=180 {
            nav.propagate(1.0);
            let _ = step;
        }
        let after = nav.position_error(truth_at(180.0));
        // The acceptance criterion: position stays within 50 m through the 120-s coast — the
        // converged velocity carries the dead-reckoning.
        assert!(
            after < 50.0,
            "post-outage error = {after} m (converged was {converged} m)"
        );
    }

    // A real curving LEO orbit (two-body + J2), integrated at 1 Hz, with a drifting clock — the
    // truth for the orbital acceptance test. The filter uses the SAME gravity model, so the coast
    // through a GNSS outage follows the orbit rather than flying off on a tangent.
    fn truth_orbit(n: usize) -> Vec<[f64; 8]> {
        use crate::forces::MU_EARTH;
        let r = 7.0e6_f64;
        let v = (MU_EARTH / r).sqrt(); // circular speed at this radius
        let inc = 28.5_f64.to_radians(); // a representative inclination
                                         // Equatorial-plane position, velocity tilted into the orbit plane (|v| = circular speed).
        let mut pv = vec![r, 0.0, 0.0, 0.0, v * inc.cos(), v * inc.sin()];
        let deriv = |_t: f64, y: &[f64]| {
            let a = gravity_accel([y[0], y[1], y[2]]);
            vec![y[3], y[4], y[5], a[0], a[1], a[2]]
        };
        let mut out = Vec::with_capacity(n + 1);
        for k in 0..=n {
            let t = k as f64;
            out.push([
                pv[0],
                pv[1],
                pv[2],
                pv[3],
                pv[4],
                pv[5],
                30.0 + 0.1 * t,
                0.1,
            ]);
            pv = rk4_step(&deriv, t, &pv, 1.0);
        }
        out
    }

    #[test]
    fn force_model_coast_holds_50m_through_outage_on_a_curving_leo_pass() {
        // The milestone acceptance: a 30-minute LEO pass with a 120-second GNSS outage in it, held
        // within 50 m. Here the satellites are fixed beacons (no visibility mask); the realism is in
        // the user's curving orbit and the force-model coast.
        let sats = constellation();
        let n = 1800; // 30 minutes at 1 Hz
        let truth = truth_orbit(n);
        // Start the filter from the true initial state perturbed by ~210 m / ~2 m/s / clock offsets.
        let s0 = truth[0];
        let x0 = vec![
            s0[0] + 150.0,
            s0[1] - 120.0,
            s0[2] + 90.0,
            s0[3] + 2.0,
            s0[4] - 1.5,
            s0[5] + 1.0,
            s0[6] + 8.0,
            s0[7] + 0.05,
        ];
        let p0 = diag(&[1.0e4, 1.0e4, 1.0e4, 1.0e2, 1.0e2, 1.0e2, 1.0e4, 1.0e0]);
        let q = diag(&[
            1.0e-2, 1.0e-2, 1.0e-2, 1.0e-3, 1.0e-3, 1.0e-3, 1.0e-2, 1.0e-4,
        ]);
        let mut nav = TightlyCoupled::new(x0, p0, q);
        let mut rng = ChaCha8Rng::seed_from_u64(0x5EED_0042);
        let n_pr = Normal::new(0.0, 1.0).unwrap();
        let n_rr = Normal::new(0.0, 0.05).unwrap();

        let outage = 900..1020; // the 120-second GNSS gap
        let mut sumsq = 0.0;
        let mut count = 0.0;
        let mut max_outage_err = 0.0_f64;
        for (step, st) in truth.iter().enumerate().take(n + 1).skip(1) {
            nav.propagate_orbital(1.0);
            if !outage.contains(&step) {
                let mut pr: Vec<f64> = sats.iter().map(|s| pseudorange(st, s)).collect();
                let mut rr: Vec<f64> = sats.iter().map(|s| range_rate(st, s)).collect();
                for v in pr.iter_mut() {
                    *v += n_pr.sample(&mut rng);
                }
                for v in rr.iter_mut() {
                    *v += n_rr.sample(&mut rng);
                }
                assert!(nav.update_gnss(&sats, &pr, &rr, 1.0, 0.05));
            }
            let err = nav.position_error([st[0], st[1], st[2]]);
            if outage.contains(&step) || step == 1020 {
                max_outage_err = max_outage_err.max(err);
            }
            // Steady-state RMS over the pass, excluding the initial 60-s convergence transient.
            if step >= 60 {
                sumsq += err * err;
                count += 1.0;
            }
        }
        let rms = (sumsq / count).sqrt();
        // Both the pass RMS and the worst error through the 120-s outage stay under 50 m — which a
        // constant-velocity coast could not, since orbital curvature alone is ~½·(v²/r)·t² ≈ 58 km
        // over 120 s at this radius. The force-model coast follows the orbit instead.
        assert!(rms < 50.0, "pass RMS = {rms} m");
        assert!(
            max_outage_err < 50.0,
            "worst outage error = {max_outage_err} m"
        );
    }
}