Struct KF

pub struct KF<T, A, M>
where
    A: DimName,
    M: DimName,
    T: State,
    DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<A> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<A, A> + Allocator<<T as State>::Size, A> + Allocator<A, <T as State>::Size>,
    <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy,
    <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,
{
    pub prev_estimate: KfEstimate<T>,
    pub process_noise: Vec<SNC<A>>,
    pub ekf: bool,
    /* private fields */
}

Defines both a Classical and an Extended Kalman filter (CKF and EKF) T: Type of state A: Acceleration size (for SNC) M: Measurement size (used for the sensitivity matrix)

Fields

prev_estimate: KfEstimate<T>

The previous estimate used in the KF computations.

process_noise: Vec<SNC<A>>

A sets of process noise (usually noted Q), must be ordered chronologically

ekf: bool

Determines whether this KF should operate as a Conventional/Classical Kalman filter or an Extended Kalman Filter. Recall that one should switch to an Extended KF only once the estimate is good (i.e. after a few good measurement updates on a CKF).

Implementations

impl<T, A, M> KF<T, A, M>

where A: DimName, M: DimName, T: State, DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<A> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, M> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<A, A> + Allocator<<T as State>::Size, A> + Allocator<A, <T as State>::Size>, <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy, <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,

pub fn new(initial_estimate: KfEstimate<T>, process_noise: SNC<A>) -> Self

Initializes this KF with an initial estimate, measurement noise, and one process noise

Examples found in repository

examples/04_lro_od/main.rs (lines 252-257)

fn main() -> Result<(), Box<dyn Error>> {
    pel::init();

    // ====================== //
    // === ALMANAC SET UP === //
    // ====================== //

    // Dynamics models require planetary constants and ephemerides to be defined.
    // Let's start by grabbing those by using ANISE's MetaAlmanac.

    let data_folder: PathBuf = [env!("CARGO_MANIFEST_DIR"), "examples", "04_lro_od"]
        .iter()
        .collect();

    let meta = data_folder.join("lro-dynamics.dhall");

    // Load this ephem in the general Almanac we're using for this analysis.
    let mut almanac = MetaAlmanac::new(meta.to_string_lossy().to_string())
        .map_err(Box::new)?
        .process(true)
        .map_err(Box::new)?;

    let mut moon_pc = almanac.planetary_data.get_by_id(MOON)?;
    moon_pc.mu_km3_s2 = 4902.74987;
    almanac.planetary_data.set_by_id(MOON, moon_pc)?;

    let mut earth_pc = almanac.planetary_data.get_by_id(EARTH)?;
    earth_pc.mu_km3_s2 = 398600.436;
    almanac.planetary_data.set_by_id(EARTH, earth_pc)?;

    // Save this new kernel for reuse.
    // In an operational context, this would be part of the "Lock" process, and should not change throughout the mission.
    almanac
        .planetary_data
        .save_as(&data_folder.join("lro-specific.pca"), true)?;

    // Lock the almanac (an Arc is a read only structure).
    let almanac = Arc::new(almanac);

    // Orbit determination requires a Trajectory structure, which can be saved as parquet file.
    // In our case, the trajectory comes from the BSP file, so we need to build a Trajectory from the almanac directly.
    // To query the Almanac, we need to build the LRO frame in the J2000 orientation in our case.
    // Inspecting the LRO BSP in the ANISE GUI shows us that NASA has assigned ID -85 to LRO.
    let lro_frame = Frame::from_ephem_j2000(-85);

    // To build the trajectory we need to provide a spacecraft template.
    let sc_template = Spacecraft::builder()
        .dry_mass_kg(1018.0) // Launch masses
        .fuel_mass_kg(900.0)
        .srp(SrpConfig {
            // SRP configuration is arbitrary, but we will be estimating it anyway.
            area_m2: 3.9 * 2.7,
            cr: 0.96,
        })
        .orbit(Orbit::zero(MOON_J2000)) // Setting a zero orbit here because it's just a template
        .build();
    // Now we can build the trajectory from the BSP file.
    // We'll arbitrarily set the tracking arc to 48 hours with a one minute time step.
    let traj_as_flown = Traj::from_bsp(
        lro_frame,
        MOON_J2000,
        almanac.clone(),
        sc_template,
        5.seconds(),
        Some(Epoch::from_str("2024-01-01 00:00:00 UTC")?),
        Some(Epoch::from_str("2024-01-02 00:00:00 UTC")?),
        Aberration::LT,
        Some("LRO".to_string()),
    )?;

    println!("{traj_as_flown}");

    // ====================== //
    // === MODEL MATCHING === //
    // ====================== //

    // Set up the spacecraft dynamics.

    // Specify that the orbital dynamics must account for the graviational pull of the Earth and the Sun.
    // The gravity of the Moon will also be accounted for since the spaceraft in a lunar orbit.
    let mut orbital_dyn = OrbitalDynamics::point_masses(vec![EARTH, SUN, JUPITER_BARYCENTER]);

    // We want to include the spherical harmonics, so let's download the gravitational data from the Nyx Cloud.
    // We're using the GRAIL JGGRX model.
    let mut jggrx_meta = MetaFile {
        uri: "http://public-data.nyxspace.com/nyx/models/Luna_jggrx_1500e_sha.tab.gz".to_string(),
        crc32: Some(0x6bcacda8), // Specifying the CRC32 avoids redownloading it if it's cached.
    };
    // And let's download it if we don't have it yet.
    jggrx_meta.process(true)?;

    // Build the spherical harmonics.
    // The harmonics must be computed in the body fixed frame.
    // We're using the long term prediction of the Moon principal axes frame.
    let moon_pa_frame = MOON_PA_FRAME.with_orient(31008);
    // let moon_pa_frame = IAU_MOON_FRAME;
    let sph_harmonics = Harmonics::from_stor(
        almanac.frame_from_uid(moon_pa_frame)?,
        HarmonicsMem::from_shadr(&jggrx_meta.uri, 80, 80, true)?,
    );

    // Include the spherical harmonics into the orbital dynamics.
    orbital_dyn.accel_models.push(sph_harmonics);

    // We define the solar radiation pressure, using the default solar flux and accounting only
    // for the eclipsing caused by the Earth and Moon.
    // Note that by default, enabling the SolarPressure model will also enable the estimation of the coefficient of reflectivity.
    let srp_dyn = SolarPressure::new(vec![EARTH_J2000, MOON_J2000], almanac.clone())?;

    // Finalize setting up the dynamics, specifying the force models (orbital_dyn) separately from the
    // acceleration models (SRP in this case). Use `from_models` to specify multiple accel models.
    let dynamics = SpacecraftDynamics::from_model(orbital_dyn, srp_dyn);

    println!("{dynamics}");

    // Now we can build the propagator.
    let setup = Propagator::default_dp78(dynamics.clone());

    // For reference, let's build the trajectory with Nyx's models from that LRO state.
    let (sim_final, traj_as_sim) = setup
        .with(*traj_as_flown.first(), almanac.clone())
        .until_epoch_with_traj(traj_as_flown.last().epoch())?;

    println!("SIM INIT:  {:x}", traj_as_flown.first());
    println!("SIM FINAL: {sim_final:x}");
    // Compute RIC difference between SIM and LRO ephem
    let sim_lro_delta = sim_final
        .orbit
        .ric_difference(&traj_as_flown.last().orbit)?;
    println!("{traj_as_sim}");
    println!(
        "SIM v LRO - RIC Position (m): {:.3}",
        sim_lro_delta.radius_km * 1e3
    );
    println!(
        "SIM v LRO - RIC Velocity (m/s): {:.3}",
        sim_lro_delta.velocity_km_s * 1e3
    );

    traj_as_sim.ric_diff_to_parquet(
        &traj_as_flown,
        "./04_lro_sim_truth_error.parquet",
        ExportCfg::default(),
    )?;

    // ==================== //
    // === OD SIMULATOR === //
    // ==================== //

    // After quite some time trying to exactly match the model, we still end up with an oscillatory difference on the order of 150 meters between the propagated state
    // and the truth LRO state.

    // Therefore, we will actually run an estimation from a dispersed LRO state.
    // The sc_seed is the true LRO state from the BSP.
    let sc_seed = *traj_as_flown.first();

    // Load the Deep Space Network ground stations.
    // Nyx allows you to build these at runtime but it's pretty static so we can just load them from YAML.
    let ground_station_file: PathBuf = [
        env!("CARGO_MANIFEST_DIR"),
        "examples",
        "04_lro_od",
        "dsn-network.yaml",
    ]
    .iter()
    .collect();

    let devices = GroundStation::load_named(ground_station_file)?;

    // Typical OD software requires that you specify your own tracking schedule or you'll have overlapping measurements.
    // Nyx can build a tracking schedule for you based on the first station with access.
    let trkconfg_yaml: PathBuf = [
        env!("CARGO_MANIFEST_DIR"),
        "examples",
        "04_lro_od",
        "tracking-cfg.yaml",
    ]
    .iter()
    .collect();

    let configs: BTreeMap<String, TrkConfig> = TrkConfig::load_named(trkconfg_yaml)?;

    // Build the tracking arc simulation to generate a "standard measurement".
    let mut trk = TrackingArcSim::<Spacecraft, GroundStation>::new(
        devices.clone(),
        traj_as_flown.clone(),
        configs,
    )?;

    trk.build_schedule(almanac.clone())?;
    let arc = trk.generate_measurements(almanac.clone())?;
    // Save the simulated tracking data
    arc.to_parquet_simple("./04_lro_simulated_tracking.parquet")?;

    // We'll note that in our case, we have continuous coverage of LRO when the vehicle is not behind the Moon.
    println!("{arc}");

    // Now that we have simulated measurements, we'll run the orbit determination.

    // ===================== //
    // === OD ESTIMATION === //
    // ===================== //

    let sc = SpacecraftUncertainty::builder()
        .nominal(sc_seed)
        .frame(LocalFrame::RIC)
        .x_km(0.5)
        .y_km(0.5)
        .z_km(0.5)
        .vx_km_s(5e-3)
        .vy_km_s(5e-3)
        .vz_km_s(5e-3)
        .build();

    // Build the filter initial estimate, which we will reuse in the filter.
    let initial_estimate = sc.to_estimate()?;

    println!("== FILTER STATE ==\n{sc_seed:x}\n{initial_estimate}");

    let kf = KF::new(
        // Increase the initial covariance to account for larger deviation.
        initial_estimate,
        // Until https://github.com/nyx-space/nyx/issues/351, we need to specify the SNC in the acceleration of the Moon J2000 frame.
        SNC3::from_diagonal(10 * Unit::Minute, &[1e-12, 1e-12, 1e-12]),
    );

    // We'll set up the OD process to reject measurements whose residuals are move than 3 sigmas away from what we expect.
    let mut odp = SpacecraftODProcess::ckf(
        setup.with(initial_estimate.state().with_stm(), almanac.clone()),
        kf,
        devices,
        Some(ResidRejectCrit::default()),
        almanac.clone(),
    );

    odp.process_arc(&arc)?;

    let ric_err = traj_as_flown
        .at(odp.estimates.last().unwrap().epoch())?
        .orbit
        .ric_difference(&odp.estimates.last().unwrap().orbital_state())?;
    println!("== RIC at end ==");
    println!("RIC Position (m): {}", ric_err.radius_km * 1e3);
    println!("RIC Velocity (m/s): {}", ric_err.velocity_km_s * 1e3);

    odp.to_parquet(&arc, "./04_lro_od_results.parquet", ExportCfg::default())?;

    // In our case, we have the truth trajectory from NASA.
    // So we can compute the RIC state difference between the real LRO ephem and what we've just estimated.
    // Export the OD trajectory first.
    let od_trajectory = odp.to_traj()?;
    // Build the RIC difference.
    od_trajectory.ric_diff_to_parquet(
        &traj_as_flown,
        "./04_lro_od_truth_error.parquet",
        ExportCfg::default(),
    )?;

    Ok(())
}

pub fn with_sncs( initial_estimate: KfEstimate<T>, process_noises: Vec<SNC<A>>, ) -> Self

Initializes this KF with an initial estimate, measurement noise, and several process noise WARNING: SNCs MUST be ordered chronologically! They will be selected automatically by walking the list of SNCs backward until one can be applied!

impl<T, M> KF<T, U3, M>

where M: DimName, T: State, DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, M> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<U3, U3> + Allocator<<T as State>::Size, U3> + Allocator<U3, <T as State>::Size>, <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy, <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,

pub fn no_snc(initial_estimate: KfEstimate<T>) -> Self

Initializes this KF without SNC

Examples found in repository

examples/02_jwst_covar_monte_carlo/main.rs (line 115)

fn main() -> Result<(), Box<dyn Error>> {
    pel::init();
    // Dynamics models require planetary constants and ephemerides to be defined.
    // Let's start by grabbing those by using ANISE's latest MetaAlmanac.
    // For details, refer to https://github.com/nyx-space/anise/blob/master/data/latest.dhall.

    // Download the regularly update of the James Webb Space Telescope reconstucted (or definitive) ephemeris.
    // Refer to https://naif.jpl.nasa.gov/pub/naif/JWST/kernels/spk/aareadme.txt for details.
    let mut latest_jwst_ephem = MetaFile {
        uri: "https://naif.jpl.nasa.gov/pub/naif/JWST/kernels/spk/jwst_rec.bsp".to_string(),
        crc32: None,
    };
    latest_jwst_ephem.process(true)?;

    // Load this ephem in the general Almanac we're using for this analysis.
    let almanac = Arc::new(
        MetaAlmanac::latest()
            .map_err(Box::new)?
            .load_from_metafile(latest_jwst_ephem, true)?,
    );

    // By loading this ephemeris file in the ANISE GUI or ANISE CLI, we can find the NAIF ID of the JWST
    // in the BSP. We need this ID in order to query the ephemeris.
    const JWST_NAIF_ID: i32 = -170;
    // Let's build a frame in the J2000 orientation centered on the JWST.
    const JWST_J2000: Frame = Frame::from_ephem_j2000(JWST_NAIF_ID);

    // Since the ephemeris file is updated regularly, we'll just grab the latest state in the ephem.
    let (earliest_epoch, latest_epoch) = almanac.spk_domain(JWST_NAIF_ID)?;
    println!("JWST defined from {earliest_epoch} to {latest_epoch}");
    // Fetch the state, printing it in the Earth J2000 frame.
    let jwst_orbit = almanac.transform(JWST_J2000, EARTH_J2000, latest_epoch, None)?;
    println!("{jwst_orbit:x}");

    // Build the spacecraft
    // SRP area assumed to be the full sunshield and mass if 6200.0 kg, c.f. https://webb.nasa.gov/content/about/faqs/facts.html
    // SRP Coefficient of reflectivity assumed to be that of Kapton, i.e. 2 - 0.44 = 1.56, table 1 from https://amostech.com/TechnicalPapers/2018/Poster/Bengtson.pdf
    let jwst = Spacecraft::builder()
        .orbit(jwst_orbit)
        .srp(SrpConfig {
            area_m2: 21.197 * 14.162,
            cr: 1.56,
        })
        .dry_mass_kg(6200.0)
        .build();

    // Build up the spacecraft uncertainty builder.
    // We can use the spacecraft uncertainty structure to build this up.
    // We start by specifying the nominal state (as defined above), then the uncertainty in position and velocity
    // in the RIC frame. We could also specify the Cr, Cd, and mass uncertainties, but these aren't accounted for until
    // Nyx can also estimate the deviation of the spacecraft parameters.
    let jwst_uncertainty = SpacecraftUncertainty::builder()
        .nominal(jwst)
        .frame(LocalFrame::RIC)
        .x_km(0.5)
        .y_km(0.3)
        .z_km(1.5)
        .vx_km_s(1e-4)
        .vy_km_s(0.6e-3)
        .vz_km_s(3e-3)
        .build();

    println!("{jwst_uncertainty}");

    // Build the Kalman filter estimate.
    // Note that we could have used the KfEstimate structure directly (as seen throughout the OD integration tests)
    // but this approach requires quite a bit more boilerplate code.
    let jwst_estimate = jwst_uncertainty.to_estimate()?;

    // Set up the spacecraft dynamics.
    // We'll use the point masses of the Earth, Sun, Jupiter (barycenter, because it's in the DE440), and the Moon.
    // We'll also enable solar radiation pressure since the James Webb has a huge and highly reflective sun shield.

    let orbital_dyn = OrbitalDynamics::point_masses(vec![MOON, SUN, JUPITER_BARYCENTER]);
    let srp_dyn = SolarPressure::new(vec![EARTH_J2000, MOON_J2000], almanac.clone())?;

    // Finalize setting up the dynamics.
    let dynamics = SpacecraftDynamics::from_model(orbital_dyn, srp_dyn);

    // Build the propagator set up to use for the whole analysis.
    let setup = Propagator::default(dynamics);

    // All of the analysis will use this duration.
    let prediction_duration = 6.5 * Unit::Day;

    // === Covariance mapping ===
    // For the covariance mapping / prediction, we'll use the common orbit determination approach.
    // This is done by setting up a spacecraft OD process, and predicting for the analysis duration.

    let ckf = KF::no_snc(jwst_estimate);

    // Build the propagation instance for the OD process.
    let prop = setup.with(jwst.with_stm(), almanac.clone());
    let mut odp = SpacecraftODProcess::ckf(prop, ckf, BTreeMap::new(), None, almanac.clone());

    // Define the prediction step, i.e. how often we want to know the covariance.
    let step = 1_i64.minutes();
    // Finally, predict, and export the trajectory with covariance to a parquet file.
    odp.predict_for(step, prediction_duration)?;
    odp.to_parquet(
        &TrackingDataArc::default(),
        "./02_jwst_covar_map.parquet",
        ExportCfg::default(),
    )?;

    // === Monte Carlo framework ===
    // Nyx comes with a complete multi-threaded Monte Carlo frame. It's blazing fast.

    let my_mc = MonteCarlo::new(
        jwst, // Nominal state
        jwst_estimate.to_random_variable()?,
        "02_jwst".to_string(), // Scenario name
        None, // No specific seed specified, so one will be drawn from the computer's entropy.
    );

    let num_runs = 5_000;
    let rslts = my_mc.run_until_epoch(
        setup,
        almanac.clone(),
        jwst.epoch() + prediction_duration,
        num_runs,
    );

    assert_eq!(rslts.runs.len(), num_runs);
    // Finally, export these results, computing the eclipse percentage for all of these results.

    // For all of the resulting trajectories, we'll want to compute the percentage of penumbra and umbra.
    let eclipse_loc = EclipseLocator::cislunar(almanac.clone());
    let umbra_event = eclipse_loc.to_umbra_event();
    let penumbra_event = eclipse_loc.to_penumbra_event();

    rslts.to_parquet(
        "02_jwst_monte_carlo.parquet",
        Some(vec![&umbra_event, &penumbra_event]),
        ExportCfg::default(),
        almanac,
    )?;

    Ok(())
}

Trait Implementations

impl<T, A, M> Clone for KF<T, A, M>

where A: DimName + Clone, M: DimName + Clone, T: State + Clone, DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<A> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<A, A> + Allocator<<T as State>::Size, A> + Allocator<A, <T as State>::Size>, <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy, <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,

fn clone(&self) -> KF<T, A, M>

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<T, A, M> Debug for KF<T, A, M>

where A: DimName + Debug, M: DimName + Debug, T: State + Debug, DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<A> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<A, A> + Allocator<<T as State>::Size, A> + Allocator<A, <T as State>::Size>, <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy, <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl<T, A, M> Filter<T, A, M> for KF<T, A, M>

where A: DimName, M: DimName, T: State, DefaultAllocator: Allocator<M> + Allocator<<T as State>::Size> + Allocator<<T as State>::VecLength> + Allocator<A> + Allocator<M, M> + Allocator<M, <T as State>::Size> + Allocator<<T as State>::Size, M> + Allocator<<T as State>::Size, <T as State>::Size> + Allocator<A, A> + Allocator<<T as State>::Size, A> + Allocator<A, <T as State>::Size> + Allocator<Const<1>, M>, <DefaultAllocator as Allocator<<T as State>::Size>>::Buffer<f64>: Copy, <DefaultAllocator as Allocator<<T as State>::Size, <T as State>::Size>>::Buffer<f64>: Copy,

fn previous_estimate(&self) -> &Self::Estimate

Returns the previous estimate

fn update_h_tilde(&mut self, h_tilde: OMatrix<f64, M, <T as State>::Size>)

Update the sensitivity matrix (or “H tilde”). This function must be called prior to each call to measurement_update.

fn time_update(&mut self, nominal_state: T) -> Result<Self::Estimate, ODError>

Computes a time update/prediction (i.e. advances the filter estimate with the updated STM).

May return a FilterError if the STM was not updated.

fn measurement_update( &mut self, nominal_state: T, real_obs: &OVector<f64, M>, computed_obs: &OVector<f64, M>, r_k: OMatrix<f64, M, M>, resid_rejection: Option<ResidRejectCrit>, ) -> Result<(Self::Estimate, Residual<M>), ODError>

Computes the measurement update with a provided real observation and computed observation.

May return a FilterError if the STM or sensitivity matrices were not updated.

fn set_process_noise(&mut self, snc: SNC<A>)

Overwrites all of the process noises to the one provided

type Estimate = KfEstimate<T>

fn set_previous_estimate(&mut self, est: &Self::Estimate)

Set the previous estimate

fn is_extended(&self) -> bool

Returns whether the filter is an extended filter (e.g. EKF)

fn set_extended(&mut self, status: bool)

Sets the filter to be extended or not depending on the value of status