Synthetic airspeed is fused for planes without airpseed sensor when dead-reckoning

[pavloblindnology]

Bug Report

Synthetic airspeed is fused for planes without airpseed sensor when dead-reckoning.
Instead, when navigating with GPS, only side-slip fusion is performed.
I know synthetic airspeed was added exactly for this reason, but it seems illogical. And seems like it can be replaced with dead-reckoning with sideslip fusion only.

Also, with GPS, wind process noise is scaled by x10 before 1st dead-reckoning. And after turning GPS back, wind process noise is not scaled.
Related issue for copters #33451 was fixed in master.

Issue details

AP 4.6 introduced new feature to use last estimated airspeed when dead-reckoning for planes.
This feature also takes action with planes without airspeed sensor, here

ardupilot/libraries/AP_NavEKF3/AP_NavEKF3_Measurements.cpp

Line 899 in 6d578b4

} else if (lastAspdEstIsValid && !windStateIsObservable) {

This synthetic airspeed is then fused here:

ardupilot/libraries/AP_NavEKF3/AP_NavEKF3_AirDataFusion.cpp

Line 162 in 6d578b4

FuseAirspeed();

Also, before 1st dead-reckoning wind process noise is scaled by x10 here:

ardupilot/libraries/AP_NavEKF3/AP_NavEKF3_core.cpp

Line 1127 in 6d578b4

if (!tasDataDelayed.allowFusion) {

After 1st dead-reckoning, wind process noise is NOT scaled by x10, bcz tasDataDelayed.allowFusion is already true.

This results in different EKF3 air data fusion behaviour:

• With GPS only side-slip fusion is performed
  • Before 1st dead-reckoning, wind process noise is scaled by x10
  • After 1st dead-reckoning, wind process noise is NOT scaled by x10
• Without GPS both side-slip and synthetic airspeed fusion is performed

To reproduce, take off, turn GPS off and wait, then turn GPS on.

The issue with x10 wind process noise scale can fixed by adding

        } else {
            tasDataToFuse = false;
            tasDataDelayed.allowFusion = false;

here

ardupilot/libraries/AP_NavEKF3/AP_NavEKF3_Measurements.cpp

Line 913 in 6d578b4

}

Regarding synthetic airspeed fusion, seems like it can be replaced with sideslip fusion only and keeping absolute position when dead-reckoning. For this a new timeout condition can be added to setAidingMode here

ardupilot/libraries/AP_NavEKF3/AP_NavEKF3_Control.cpp

Line 326 in 6d578b4

bool airSpdUsed = (imuSampleTime_ms - lastTasPassTime_ms <= minTestTime_ms);

Smth like

            // Check if sideslip data is being used
            bool sideSlipUsed = (imuSampleTime_ms - prevBetaDragStep_ms <= minTestTime_ms);
...
            // Check if attitude drift has been constrained by a measurement source
            bool attAiding = posUsed || gpsVelUsed || optFlowUsed || sideSlipUsed || airSpdUsed || dragUsed || rngBcnUsed || bodyOdmUsed;

            // Check if velocity drift has been constrained by a measurement source
            // Currently these are all the same source as will stabilise attitude because we do not currently have
            // a sensor that only observes attitude
            velAiding = posUsed || gpsVelUsed || optFlowUsed || sideSlipUsed || airSpdUsed || dragUsed || rngBcnUsed || bodyOdmUsed;

            // Store the last valid airspeed estimate
            windStateIsObservable = !inhibitWindStates && (posUsed || gpsVelUsed || optFlowUsed || rngBcnUsed || bodyOdmUsed);
            if (windStateIsObservable && (airSpdUsed || dragUsed)) {
                lastAirspeedEstimate = (stateStruct.velocity - Vector3F(stateStruct.wind_vel.x, stateStruct.wind_vel.y, 0.0F)).length();
                lastAspdEstIsValid = true;
            }

Then, synthetic airspeed will be fused only in case of airspeed sensor failure.
While only sideslip fusion will be fused in case of dead-reckoning without airspeed sensor. Quick testing shows that it behaves very similar to original.

Version
4.6.3

Platform
[ ] All
[ ] AntennaTracker
[ ] Copter
[*] Plane
[ ] Rover
[ ] Submarine

Airframe type
Plane

Hardware type
SITL

[IamPete1]

Thanks for the report! Please make a PR so we can review the proposed changes.

[pavloblindnology]

@IamPete1 #33736