|
"Even with the default settings. It will change waypoints at about 3 miles. I can't see any real reason why that
wouldn't work for you."
In answer to this question, it doesn't work and I spent a few days pondering this and observing the LNAV behavior and I need to type a lot of characters here explaining my answer because there are some principles and terms which I don't want misunderstood. I also want to make sure you understand this is not an emotional "I own and defend this with my life" nor am I asking you to defend the program. I'm simply describing what I see when I use the program and why I have trouble with it. I also take the risk in offering some suggestions near the end.
In the subsequent narrative, I pursue these four objectives:
1) Describe an ideal (turn-only) autopilot-to-aircraft model behavior.
2) Describe some terminology which I hope will make follow-on explanations clearer.
3) Describe current LNAV behavior and why it's undesired.
4) Describe a potential solution which should preserve much of the current LNAV code yet make a dramatic improvement to the turn performance.
1) Ideally, a generalized auto pilot should behave correctly on a correctly behaving aircraft. Explanation:
When it comes to turning, a correct FS flight model which represents a passenger aircraft where the passengers are not wearing G suits and helmets and the pilot is not performing snap rolls or competitive air show maneuvers, should turn smoothly and evenly to the defined bank limit with no instability at the end of the roll. To put a finer point on it, the passenger aircraft when commanded to a 90 degree turn should turn smoothly and professionally approximately 5-8 degrees per second and at the end of the roll, there should be no bouncing back and forth (no instability.)
2) An autopilot should not be trying to correct a poor flight model which when commanded to a large turn angle, turns too fast (snap rolls) and/or bounces (unstable) at the extremes of the roll. At this point I'm not concerned with advanced corrective autothrottle corrections which are more of a discussion for VNAV.
I have been using the FMS LNAV on correctly modeled aircraft which when left to themselves without LNAV and the AP heading bug is commanded by the pilot to a 90 deg turn, rolls smoothly and evenly stopping at the limits of the roll without instability until the aircraft is point in the direction of the heading bug without overshooting. This is the behavior of my aircraft without the FMS.
Now, some definitions of terms so what I say next may be understood. These may not be scientifically correct or terms used by real world autopilot engineers, but are intended for clarity by this sorry layman:
1) "Map Track": Describes the direct line between one waypoint and another as if you were drawing a line directly between locations on a flap map. This shows up as the green track line on the HSI map display.
2) "Turn Point": Describes the actual MOMENT when the AP is command to direct the aircraft into a turn. This may have been referred to as "tracking" in the "Settings" documentation.
3) "Ground Track" is the actual path the plane takes over the ground.
4) "Resolve" or "Resolution" is the PROCESS by which the autopilot corrects the plane ground track to the map track over time.
5) "Turn Waypoint" is a waypoint which is not just a pass through in a straight line, but a waypoint over which a turn is expected after which the next waypoint represents a significant change in direction or heading.
6) "Target Waypoint": Waypoint which represents the new turn result target waypoint.
Now is the time to make a observation statement I will attempt to describe in detail afterward:
The Seconds and Distance variables which have greatly confused me do so by not doing what I expected (right or wrong.) They do not ANTICIPATE a turn, but DELAY a turn. In medium to high speed aircraft, anticipation is more helpful than a delay. The faster the speed, the longer it takes to make a turn, so if accuracy is desired (efficiency may be a better desirability) the faster the aircraft, the more anticipation is needed (assuming the roll rate is constant at all speeds- which it mostly is in FS aircraft under AP control) There is less overshoot of the "Map Track" when the turn is started before the "turn waypoint" just as a race car driver will "lead" a turn by turning to the inside of the race track as soon as possible so as to keep the curve of the turn as shallow and long as possible, then moving toward the outside of the turn near the end. The current implementation of DIST and SECS "appears" to have the purpose of making the turn longer (extending the turn) at the cost of increasing track error during the near-term initial stage ( plane is not yet pointed at the next target waypoint.)
When I use LNAV to control my aircraft, here is the set up to the turn:
1) Level flight with no roll approaching a "turn waypoint"
2) The NEXT waypoint represents a 30 degree turn to the left, 100 miles away.
3) The aircraft profile has the setting: WptChgSecs=20
At turn initialization, here is what happens:
Exactly 3.0 miles prior to crossing the "turn waypoint" the first "turn point" command is sent from LNAV and sets the AP heading bug to approximately 90 degrees to the RIGHT of the current aircraft ground track. In other words, the first command of the LNAV AP is to turn the aircraft AWAY in the opposite direction of the next waypoint. At the same time, a yellow dot and attached dotted line is drawn on the ASI representing a position 20 seconds in the future (beyond the green "map track") and the dotted line's far end intersecting the target waypoint. As the aircraft begins to turn away from the desired direction of travel and crossed the "map Track" but still getting closer to the yellow dot and line, the LNAV slowly (at apprx 1 degree/second) begins to straighten out the turn by adjusting the heading bug left, so by the time the aircraft crosses the dot, there is minimal roll (straight and level.) As the plane crosses the dot and the yellow line the LNAV slowly (apprx 1 deg/second) starts turning the plane left toward the target waypoint. By this time the track error as displayed in the FMS Progress page three is significant. In the 727 it's nearly a mile (at mach 0.84) and in the Concord it's nearly 2 miles. As LNAV tries to "Resolve" this error the AP heading bug will be directed increasingly left until it's held about 60 to 90 degrees left of the direction of the target waypoint some 100 miles distant. As the aircraft turns back to the LNAV desired yellow line and the track error drops again toward zero, the AP heading bug will slowly turn right again (1 deg/second) until the bug matches the current heading of the aircraft which is now 20 degrees left of the target waypoint heading. As the track error begins to grow in the other direction (right) the LNAV will slowly try to "resolve" the new growing error again by advancing the heading bug 1 degree/second to the right and by this time the track error is nearly a mile in the other direction when the heading bug stops advancing. Each of these first few corrective resolution turns command the aircraft into its max bank angle. Successive corrections will have smaller bank angles. If I were a passenger on this aircraft, I would think the pilot was drunk. The resulting "ground track" looks like the S turns the Space Shuttle makes slowing down during re-entry interface.
The entire process described above is repeated over and over again with the track errors and bank angles slowly getting smaller. In the Concorde, the point where the heading corrections are down to two degrees is approx 50 miles after the turn process was initialized (the yellow intercept line created.) In the 727 about 18 miles are required. BTW: the B777 and B747 cruise at the same speeds as the 727-200 so their turn behaviors are similar.
If I were to attempt to write this VNAV myself, there is how I would describe a successful turn and intercept (based on commanding the AP heading bug myself):
1) Change the WptChgSecs and WptChgDist variables so they were predictive rather than delays. The "turn point" (turn initialization) or what you may have referred to in documentation as "tracking" will now trigger before the actual "turn waypoint" by the variable distance or time.
2) At the calculated "turn point" (the newly predicted ANTICIPATED point) the AP heading bug will be immediately (not gradually as it is now) positioned so it points DIRECTLY at the target waypoint (match heading bug to target direction.)
3) Throughout the initial turn bank, the AP heading bug will be corrected as necessary to keep pointing directly at the target waypoint heading.
3) When the aircraft heading exactly matches the current target waypoint heading the code will then analyze the track error and make appropriate gradual adjustments to the heading bug to resolve the ground track to the map track.
I believe this is the process the GPS NAV program uses to great success except there is no effective prediction distance other than ALL aircraft at ALL speeds are turned at 3.0 miles or less. Aircraft faster than mach 0.70 suffer from increasingly snaking turns. After all, FS is by philosophy intended to be a general aviation simulator with jet aircraft as a concession. With the introduction of turn prediction added to a refinement of the existing code to resolve the ground track to the map track, a very effective LNAV could be created which would work with ALL aircraft at ALL speeds and the purchaser would have a much more powerful tool to customize the behavior of the LNAV to the dynamics and speed of the model.
Positive results of the three step process above:
1) LNAV will take the most efficient ground track to the target.
2) LNAV will use the aircraft's maximum bank angle at the BEGINNING of the turn (instead of in the middle as it does now) to minimize the number and intensity of successive corrective banks.
3) After having immediately turned the aircraft directly at the target waypoint, the resulting track error is as small as possible in the shortest time possible.
4) The resulting corrective turns and bank angles after pointing at the target waypoint will be 90% or more smaller than they are under the current procedure.
5) The current aircraft profiles distributed with the product will not require any change.
5) Much of of the current LNAV code can be retained. I tested this myself by:
a) waiting for the LNAV turn trigger and yellow dotted line.
b) immediately disabling LNAV and use the AP heading bug to point directly at the next waypoint target.
c) when the plane was pointing directly at the waypoint target, re-enable LNAV. The result was LNAV then made small corrective banks to resolve the track.
Again I will stress, the above procedure requires the aircraft model to behave correctly as described at the beginning of this narrative where input from the pilot to the heading bug manually does not result in snap rolls or bank limit instability. If an AP is coded to try and correct bad aircraft modeling, the navigation program becomes self defeating and performs improperly and inefficiently on well made models and only behaves "good enough" but not excellent on very slow and highly maneuverable planes.
I realize this is not what you wanted from me in answering this question and it's also a long and fatiguing explanation potentially loaded with defensive triggers. My hope is it increases understanding and contributes to future development refinements. If necessary I can create a series of screenshots or Fraps animations demonstrating the observed LVAV behavior and email them to you. It's also likely my professional software tester technique is showing through where I'm used to exactly explaining what the software is observed to to and what the software should do. Since it's obvious from the displays and actions of the LNAV it's getting all the information it needs from the FS environment and the LNAV has positive control over the heading bug, I believe it can be written to do what I describe with little additional effort. Essentially, reversing two current behaviors:
1) Make the Dist and Secs variables calculate the time and distance TO the turn waypoint (instead of after as it appears to do now.)
2) Make the maximum effort to turn the aircraft at the BEGINNING of the turn rather than at the first and subsequent track corrections as it does now.
Thanks for your patience.
-Pv-
|
Login
Register
FAQ
Search
