Merge pull request #24 from bprobert97/update-sys-interactions

Update sys interactions

Author: bprobert97

design/system_interactions.txt

@@ -0,0 +1,52 @@
+@startuml
+
+participant "Simulation" as sim
+participant "Filter" as filter
+participant "Satellite" as sat
+participant "Consensus (PoISE)" as poise
+participant "Ledger (DAG)" as dag
+
+== Initialisation & Truth Generation ==
+sim -> sim: simulate_truth_and_meas(N, steps, dt)
+note right
+  Generates truth_hist (ground truth orbits)
+  and z_hist (noisy ISL measurements)
+end note
+
+sim -> filter: Create EKF instances (initialised with truth[0])
+sim -> sat: Create Satellite nodes
+sim -> dag: Initialise DAG with PoISE
+
+== Simulation Loop ==
+/'
+The simulation generates all possible observation records for a
+maximum possible ISL range set at 5000km. This way, the ISL range
+I want to test PoISE against can be set dynamically (up to the
+5000km maximum), but also physically impossible observations
+(such as 2 satellites at opposite sides of the earth seeing each
+other) are not created as this wastes computational effort.
+This is sufficient and efficient for research and testing.
+'/
+
+loop for each timestep k
+    sim -> filter: predict() & update(z_hist[k])
+    note right: Filter estimates state based on noisy measurements
+    filter -->> sim: ObservationRecords (with NIS)
+    note right: Observation records generated for Max ISL for 5000km
+
+    sim -> sat: update_position(truth_hist[k])
+
+    group In-Range Observation
+        sim -> sat: load_sensor_data(ObservationRecord)
+        note right: Only loads observations within specified ISL range
+        sat -> dag: submit_transaction(queue)
+    end
+end
+
+== Consensus (Async) ==
+dag -> dag: listen() to queue
+dag -> poise: validate(transaction, reputation)
+poise -->> dag: validation_result
+dag -> sat: update_reputation()
+
+@enduml

images/system_interactions.png

@@ -39,6 +39,7 @@
 Re = 6378e3
 STATE_DIM = 6 # State vector dimension (position and velocity)
 POS_VEL_DIM = 3 # Position or velocity dimension
+MAX_ISL_RANGE = 5000e3 # Maximum range for ISL observation records (5000km)
 
 # ----------------------- Result Types ---------------------
 @dataclass
@@ -476,6 +477,14 @@ def _log_nis(y: np.ndarray, ekf: ExtendedKalmanFilter, N: int, k: int,
 
             xj_idx_slice = slice(STATE_DIM*j, STATE_DIM*j+STATE_DIM)
 
+            # Skip records for satellites further than 5000km apart
+            dist = np.linalg.norm(ekf.x[STATE_DIM*i:STATE_DIM*i+3] -
+                                 ekf.x[STATE_DIM*j:STATE_DIM*j+3])
+            if dist > MAX_ISL_RANGE:
+                # Advance index by 2 as every pair is range and range rate
+                idx += 2
+                continue
+
             yij = y[idx:idx+2]
 
             # Recalculate Ht and Ho for the current pair, this is necessary.
@@ -507,6 +516,7 @@ def _log_nis(y: np.ndarray, ekf: ExtendedKalmanFilter, N: int, k: int,
                     step=k, observer=i, target=j, nis=nis, dof=yij.shape[0], time = k*dt
                 )
             )
+            # Advance index by 2 as every pair is range and range rate
             idx += 2
     return obs_records
 

src/filter.py

@@ -85,8 +85,8 @@ def test_joint_ekf_update(ekf, filter_config):
     assert not np.allclose(x_prior, x_posterior)
 
     # Check the observation records
-    expected_records = filter_config.N * (filter_config.N - 1)
-    assert len(obs_records) == expected_records
+    # Not all records are created due to the 5000km distance check
+    assert 0 < len(obs_records) <= filter_config.N * (filter_config.N - 1)
     for record in obs_records:
         assert record.step == 1
         assert record.nis >= 0