DCS learning curve problem.

[jackmckay]

Putting aside the positive aspects of 'game' and good visual representation of 3d models of planes, cockpits and missiles the thing I'm most pissed off about DCS is learning curve problem.

 

In general airplane evolution and lifecycle, each airframe starts from base model and evolves to more and more advanced and 'better' versions of same airframe. In airframe test phase, test pilots follows predicted behavior pattern of airframe and notices critical effects that degrades safety of operation. This critical regimes are noticed and then passed to the development team which finds solutions and corrections that eliminates dangers, either by aerodynamic or system advancements in model keeping in mind all possible unpredictable situations giving airframe enough margin of safety that standard pilot will never cross in regular operations.

 

When model is safe enough, it is passed to the standard operational wings. At this process, fighter pilot learns to master the airframe pushing limits of both airframe and pilot itself. One thing is sure, airframe geometry is fixed and atmosphere properties are standard and interactive behavior of booth is set and unchangeable.

 

In DCS things are very chaotic. Learning curve in DCS is not gradual advancement but very self destructive process. In RL what is learned it stays, here in DCS what is learned is doubtful and is prone to denial. So why bother having any learning curve if model degrades instead of evolves. It should be like in RL, models should behave worst than in RL and that means no superior abilities first, only after all aspects of behavior are standardized and unchangeable advancement is possible. That how things in RL works.

 

Every RL aerodynamic model evolves to better because it is natural process of R&D teams and here in DCS you just don't know what awaits you next time you turn on DCS after update. That makes lot of time and resources spent obsolete. So should we wait for this sim to fix its models aerodynamic behavior, and I mean pure and basic aerodynamic performance and not systems functionality and interaction, and then start learning things that won't be denied or neglected in next update?

 

What do you think? :music_whistling:

[Wizard1393]

DCS is not simulating the development process of any airframes.

 

DCS is trying to simulate existing airframes flight characteristics according to known airframe / engine performance data.

 

 

Sure, it's basically the same process as you are describing, but in DCS, the goal for developers is not to improve and evolve an aircrafts performance.

Their goal is to get the flight model simulated as close and true to real life aircraft performance as possible.

 

Sometimes, an adjustment to the performance data of the flight model isn't giving the expected results in all flight regimes, whether the effective result is over performing or under performing compared to expected performance.

Therefore you will sometimes see the performance in any aircraft under development degraded after an update. Most times that's because the aircraft was over performing compared o real life in the previous version.

 

But I get what you mean, it's difficult to perfect your manouvres when aircraft performance is ever changing. If you want to be less affected by things like this, avoid purchasing modules that are till in development, aka "early access" or "beta".

Focus on modules that do not have an "early access" or "beta" label. They will generally not get any significant/very few adjustments to aircraft performance.

Edited May 16, 2017 by chrisofsweden

[jackmckay]

How come that 'known data' is so dynamic and usually degradable? I would really like to learn under performing flight model first than over performing one.

Edited May 16, 2017 by jackmckay

[shagrat]

How come that 'known data' is so dynamic and usually degradable? I would really like to learn under performing flight model first than over performing one.

Because the 'Data' is dynamic. The DCS World models weather and all aerodynamic aspects, etc.

If anything changes or is "tweaked" it is a real-time calculation with results you cannot perfectly predict, only estimate and test as much as possible.

 

Unlike other Sims that use Simplified performance tables, DCS tries hard to emulate real airplane behaviour from model and "World" interacting.

 

Any change to any of both has an effect.

Edited May 16, 2017 by shagrat

[OnlyforDCS]

What do you think? :music_whistling:

 

I think that you are going to need more cheese to go with that whine, once the Mig29 PFM is released. :thumbup:

[jackmckay]

User manual all range charts are fixed and behavior is specific for each flight envelope segment. Airframe geometry, mass and weight distribution is relatively known and stable, airfoil lift&drag charts are available and stable, acceleration limitations are also charted, engine performance/thrust versus air density are also known and available, AoA versus drag charts are also available and stable, radar and thermal cross sections too.. etc. Most of this data depend solely on geometrical aspect of airframe. How come that airframe can over perform? Is it better to lower performance if not sure about than aggregate it?

 

And what is most important thing set on single airframe should be applicable for all other airframes too reducing time needed to develop next model.

 

egg. Right now M2K can plunge at almost 2000km/h on low alt and escape anything. Nice feature but is it real? Sources say max speed is: Mach 2.2 (2,336 km/h, 1,451 mph) at high altitude/ 1,110 km/h (690 mph) at low altitude. That is at least 75% of error margin over performance. 75% better than in real life ??? How come this model can perform in that way. What if future update of this airframe neglects that 'unrealistic feature' and kicks defensive maneuver out of mirage pilot arsenal? I would be surely pissed of on that. Same cases were with Flanker max speed Cobra full stop and later infamous wing braking feature as response to that previously placed 'error'. Eagle also over performs in some segments. And what when this 'features' comes to judgment pool later. Lot of pissed pilots there and lot of time spent utilizing unrealistic feature, in general learning curve wasted. I'm talking about DCS as full feature simulator here, at least at basic and primitive level of pure aerodynamics performance.

[shagrat]

The Performance Charts and characteristic behaviour is "written" down on paper. Based on given speed, air density, wind etc. ALL values will change.

So if DCS World uses dynamic weather for example there is no "table" to look up the performance for this specific values.

FC3 SFM uses lookup tables which extrapolation with pretty unrealistic stall behaviour, etc.

 

That is why DCS level aircraft like the Mirage use PFM/EFM.

Here the developers model aerodynamic behaviour (EDIT: as Sobek pointed out. complex set of look-up-tables for different parts of the airframe), based on real-time calculation that (PFM/EFM) interacts with the data coming from the DCS World environment.

As a result the change of air temperature alters pressure changes which change the stall behaviour in real-time.

 

The challenge for a developer is, to tweak and adjust his FM algorithms to match the performance at any given point of the chart.

 

This is what happens a lot during internal tests and to a lesser extent during pre-release Beta phases...

Edited May 17, 2017 by shagrat
added details to prevent confusion

[David OC]

At this level of FM simulation, it just can take some time because most of ED's Tech is at the cutting edge.

 

It's also why ED has the best dynamic flight models in the market now, that we get to enjoy.

 

The post below I found a while ago when researching explains it very well.

 

Belsimtek/TFC Associate

 

I am neither an FM programmer nor an aerodynamics specialist in any way, but I believe blade element theory is not only perfectly applicable to ED's AFM, but is in fact only one aspect of it. In other words what we describe as AFM includes blade element theory, but also other modeling approaches that combine to make the entire aircraft model.

 

We break the aircraft down into sub-elements (including breaking the rotor blades/propeller blades/wings/stabilizers into sub-elements like the root, mid-section, trailing edge) and apply physics equations to each element in real time to determine the forces and moments acting on it at any point in time. These are summed together to produce the total velocity of the aircraft. In addition each element affects other elements of the model, such as the wings affecting the airflow over the tail section, or the propwash effects on the fuselage of the P-51, etc.

 

 

Where I would say ED's AFM takes it a step further is in applying the same principles to primary aircraft systems modeling, like the engine, fuel, hydraulics, and most recently electrics. Obviously you can increase the level of depth in real-time modeling endlessly and the limit only depends on time available and computing power, but generally I would say that our AFM models probably feature the deepest and widest real time (what we call "honest") calculated factors of such aircraft systems you would see on the PC, at least at the consumer level. To give you an idea, when you run the starter in the P-51D and see the landing gear warning lights dim, it isn't because there is a specific script for this to happen, but because you're actually pulling virtual juice from the virtual battery that is virtually running in real time and supplying the instruments and systems with virtual power through a virtual bus, to which the lights are connected. The same applies to the engine model with its main elements like cylinder pressures, temperature and lubrication factors, the various mechanical components like the RPM governor, supercharger, radiator doors, etc. And the same applies to other mechanical systems like the flight controls and their operation between the cockpit and the control surfaces, the trim mechanisms, the landing gear. Again, it's ultimately a model and cannot be completely correct or account for every physical phenomenon, but we try to go as deep as possible in terms of real-time modeling and as wide as possible in terms of covering physical effects within reasonable limits of time and money. Otherwise we could work on any one model endlessly, which is probably what some of the devs would like to do.

 

To be short, AFM is built from the ground up. We don't build behaviors to match desired performance. We build a physical model as correctly as possible and then check it against known performance characteristics. If the model is built correctly, it should match the performance characteristics closely by itself without requiring any "top-down" scripting to do so. Of course some miscalculations or oversimplifications or just plain errors happen along the way, so testing, tuning and adjusting is always necessary.

 

Here's a description of the Ka-50 model:

http://www.digitalcombatsimulator.com/en/products/black_shark/#22728

 

Similar for the Huey:

http://www.digitalcombatsimulator.com/en/products/huey/#340494

 

The Post

 

-

Edited May 16, 2017 by David OC

[myHelljumper]

egg. Right now M2K can plunge at almost 2000km/h on low alt and escape anything. Nice feature but is it real? Sources say max speed is: Mach 2.2 (2,336 km/h, 1,451 mph) at high altitude/ 1,110 km/h (690 mph) at low altitude. That is at least 75% of error margin over performance. 75% better than in real life ???

 

Last time I checked the M2000C can't into Mach 2 at low altitude....

[shonist]

egg. Right now M2K can plunge at almost 2000km/h on low alt and escape anything. Nice feature but is it real? Sources say max speed is: Mach 2.2 (2,336 km/h, 1,451 mph) at high altitude/ 1,110 km/h (690 mph) at low altitude. That is at least 75% of error margin over performance. 75% better than in real life ??? How come this model can perform in that way. What if future update of this airframe neglects that 'unrealistic feature' and kicks defensive maneuver out of mirage pilot arsenal? I would be surely pissed of on that. Same cases were with Flanker max speed Cobra full stop and later infamous wing braking feature as response to that previously placed 'error'. Eagle also over performs in some segments. And what when this 'features' comes to judgment pool later. Lot of pissed pilots there and lot of time spent utilizing unrealistic feature, in general learning curve wasted. I'm talking about DCS as full feature simulator here, at least at basic and primitive level of pure aerodynamics performance.

 

lol whot?

[OnlyforDCS]

Last time I checked the M2000C can't into Mach 2 at low altitude....

 

I think he is talking about diving (plunge?).

 

Yeah a Mach 2+ plane definitely shouldn't be able to DIVE at Mach 2!

 

/s

[Sryan]

egg. Right now M2K can plunge at almost 2000km/h on low alt and escape anything. Nice feature but is it real? Sources say max speed is: Mach 2.2 (2,336 km/h, 1,451 mph) at high altitude/ 1,110 km/h (690 mph) at low altitude. That is at least 75% of error margin over performance. 75% better than in real life ??? How come this model can perform in that way. What if future update of this airframe neglects that 'unrealistic feature' and kicks defensive maneuver out of mirage pilot arsenal?

 

The speed of sound at sea level is approximatly -depending on conditions- 340 meters per second, or about 1225km/h. Are you telling us the M-2000C cannot break the sound barrier at low altitude? Do you have any sources backing this statement up?

 

I saw the statement for that speed on wikipedia. But wikipedia is not always reliable, and they did not post a source. So I went looking for this source myself, I found this:

 

From: A Compendium of Armaments and Military Hardware by C. Chant

 

That maximum speed is only true if the M-2000C has been equiped with bombs. So if you're getting caught by Mirages carrying bombs, you can complain :thumbup:

[myHelljumper]

I think he is talking about diving (plunge?).

 

Yeah a Mach 2+ plane definitely shouldn't be able to DIVE at Mach 2!

 

/s

 

:music_whistling:

 

I would be VERY surprised if a mach 2+ plane wouldn't be able to do mach 1 at sea level without missile and tank...

 

AFAIK you can't go mach with bombs, it would damage the plane IRL.

Edited May 16, 2017 by myHelljumper

[Mars Exulte]

Because it's a game working with 0s and 1s in a 'fluid' environment that computers do not calculate naturally. It's one thing to say 'wing shape, airflow, mass, velocity blah blah it should behave like THIS'. FANTASTIC *golf clap* Now, tell a glorified calculator how to do it. Not easy. That 'blah blah' in RL is thousands of calculations, and it has to be 'overdone', i.e. described and the computer's hand held to an excessive degree, to even attempt to render something similar.

 

Basically, you're just some random bloke in here that doesn't know much about how computers and code actually function going 'y it no work? It sup so ez evn 5yo can do it'

 

Well, if you are so certain it's a simple thing to do and the devs are incompetent, then go pick up ''C++ for dummies'' and show em how it's done. They'd probably welcome the help.

[Buzzles]

Basically, you're just a bit miffed more data is made available, or flaws in the modelling are found and corrected, and that therefore means you need to relearn things?

 

For someone who has just stated a strong engineering background that's a very suprising thing that you're not a fan of iteration.

Edited May 16, 2017 by Buzzles

[jackmckay]

I can eat misjudgment in performance in manner of +/- 10% but not some drastic changes of behavior or even introduction of critical effect. If you don't set all variables on beginning and predict atmosphere and flight model interaction behavior correctly you cant even start building fancy models or interactive cockpits. No use of that if physics is not set right from start.

[David OC]

There was a guy that worked in a Patent office and come up with this theory of relativity. Yes, there are some very talented unique people out there that can build flight sims.

 

You cannot compare DCS to these civi simulations tho, have you checked how accurate the FMs are in any of them?? The FMs here are cutting edge of sim development and is why there in beta for so long, no one is trying to get this close to real world performance data in any other sim outside the military.

 

Unlike civi simulations, we push these simulations to the limit in all regions of flight, even to regions test pilots would not be game to try. Plus this combat market is way way more competitive than civi simulations which drives this need for accuracy to insanity, take one look at the P-51 threads tells this story.

 

The FMs are dynamic algorithms that are not just set in stone, if you change the weather, heat, height, pressure, weight etc these FMs will be different, thats why even the DCS simulated weather needs to be set to stardard or to what the test data avalible said the conditions were when the test took place.

 

I posted this below for you on the first page, what part of it do you not understand? It's written by a guy that works close to ED (Belsimtek)

 

I am neither an FM programmer nor an aerodynamics specialist in any way, but I believe blade element theory is not only perfectly applicable to ED's AFM, but is in fact only one aspect of it. In other words what we describe as AFM includes blade element theory, but also other modeling approaches that combine to make the entire aircraft model.

 

We break the aircraft down into sub-elements (including breaking the rotor blades/propeller blades/wings/stabilizers into sub-elements like the root, mid-section, trailing edge) and apply physics equations to each element in real time to determine the forces and moments acting on it at any point in time. These are summed together to produce the total velocity of the aircraft. In addition each element affects other elements of the model, such as the wings affecting the airflow over the tail section, or the propwash effects on the fuselage of the P-51, etc.

 

To be short, AFM is built from the ground up. We don't build behaviors to match desired performance. We build a physical model as correctly as possible and then check it against known performance characteristics. If the model is built correctly, it should match the performance characteristics closely by itself without requiring any "top-down" scripting to do so. Of course some miscalculations or oversimplifications or just plain errors happen along the way, so testing, tuning and adjusting is always necessary.

 

 

 

-

Edited May 17, 2017 by David OC

[sobek]

So if DCS World uses dynamic weather for example there is no "table" to look up the performance for this specific values.

 

Here the developers model aerodynamic behaviour based on real-time calculation that interact with the data coming from the DCS World environment.

 

Real time solution of Navier Stokes Equations on the scale that DCS needs is not possible on a current desktop computer, period.

 

DCS still has to use good old lookup tables. The big difference is that instead of using one multi dimensional table for the entire airframe like older flight model designs, there are now multiple tables for multiple airframe segments.

Edited May 17, 2017 by sobek

[myHelljumper]

OP is wrong in his example, I'm unable to go past mach 1.5 (from a mach 2+ dive) at sea level in the 2000....

Edited May 17, 2017 by myHelljumper

[shagrat]

Real time solution of Navier Stokes Equations on the scale that DCS needs is not possible on a current desktop computer, period.

 

DCS still has to use good old lookup tables. The big difference is that instead of using one multi dimensional table for the entire airframe like older flight model designs, there are now multiple tables for multiple airframe segments.

Yep, bad wording on my side. I was focused on when (dynamic) weather comes into play. The results between the whole parts providing the "airframe PFM/EFM" behaviour and the windspeed, density etc. put into the tables give constantly changing results that somehow need to match what the performance charts show in any given situation.

 

As we can imagine that requires a lot of testing and tweaking, plus any small change in the "environment engine" may require further adjustments.

 

It isn't just feeding a performance chart into a table and tell the FM "behave like this" :D

Edited May 17, 2017 by shagrat

[microvax]

Just from a programming perspective:

 

You _can_ do everything up front. But that is ridicoulously expensive. On the scale of expensive that I do not think DCS modules have that budget.

 

Apart from that IRL you get instant feedback by aerodynamics. In a simulation enviroment you have to build up a system behaving like aerodynamics. So basically, yeah its allways going to be overperforming first in some way because the limitations are more complex then the basic things.

 

So just from a implementation standpoint its probably not going to happen. And indeed the Mirage cant go 2000kph anymore sustained at SL. It could do that in the past, but thats like eons ago.

[firmek]

Just from a programming perspective:

 

You _can_ do everything up front. But that is ridicoulously expensive. On the scale of expensive that I do not think DCS modules have that budget.

 

Apart from that IRL you get instant feedback by aerodynamics. In a simulation enviroment you have to build up a system behaving like aerodynamics. So basically, yeah its allways going to be overperforming first in some way because the limitations are more complex then the basic things.

 

So just from a implementation standpoint its probably not going to happen. And indeed the Mirage cant go 2000kph anymore sustained at SL. It could do that in the past, but thats like eons ago.

+1. Let's not mix aeronautics R&D process with a software development life-cycle.

In software development a projects that woud reach a perfect solution with first release candidate version just doesn't exist (unless it's a "hello word" application). The more complex and with more unknowns, the more testing, feedback and fixing iterations will be required.

[jackmckay]

How about simplified Lattice-Bolzman instad of Navier-Stokes? And for god sake, implement inertia tensor matrix on all models, mandatory. That means no network lags and ballistic motion guaranteed. Weight distribution could also be calculated on simple level as weight of engines, airframes and external loads is known. Thrust, lift, drag and acceleration vectors are mandatory booth for flight model as for damage engine.

Sensors are blind in many aspects. Booth west/east. Gradual sensor cone/prism zones interpolated with RCS/TCS vector values and Doppler modes should give realistic sensor signature. Also ground clutter exists on some modules other not. Missile guidance, countermeasure sensitivity and failure intensity is far from real combat scenarios and industry standard.

 

General philosophy of physics should be unified and not specific for each third party team as they should be responsible for 3d model perfection, dynamic aerodynamic input data and cockpit interaction credibility. They need to comply to the general physic model.I feel the anarchy in the air.

 

MP should be in a nutshell isolated from streaming complex positional data and act as supercomputer cluster utilizing all cores available over network. The only thing that should be duplex streamed on MP server is cockpit control inputs that should give flawless results on any client if engine is set correctly.

 

Graphics is secondary and all depends on GPU but its MS DX problem anyway. Ground mesh looks like from 90s and needs procedural generator that could also be used on dynamic weather. This is what I would do if I would make a 'simulator'. I just don't have time and I'm not paid to do that. Technology is here and DCS just needs to adapt. Fast.

 

Most of my experience comes from MP and AI chasing is still no real challenge for me as I hope for many core fans of DCS too.

 

PS: Mirage thing? It plunges at near 2000 km/h, hits dense air, doesn't overheat and its hydraulic performance is still maxed and no airframe stress at all. With full missile pack. How real is that? Gas viscosity and drag force increases on high temperatures. There are more powerful planes with greater Power/Cd ratio that can't come close of 50% of that what that 'thing' can do. Its just an example of anarchy. If it's real let it be but I'm pretty sure its not.

Edited May 18, 2017 by jackmckay

[sobek]

How about simplified Lattice-Bolzman instad of Navier-Stokes? And for god sake, implement inertia tensor matrix on all models, mandatory. That means no network lags and ballistic motion guaranteed. Weight distribution could also be calculated on simple level as weight of engines, airframes and external loads is known. Thrust, lift, drag and acceleration vectors are mandatory booth for flight model as for damage engine.

Lattice-Boltzmann supposedly has problems with higher Mach numbers. You don't think that could be a problem for a flight sim? That aside, even if Lattice-Boltzmann is vastly more efficient than Navier-Stokes, doing real time CFD might still be out of the question for performance reasons. Also it is a relatively new technique, so you won't find many people that are proficient in it.

 

General philosophy of physics should be unified and not specific for each third party team as they should be responsible for 3d model perfection, dynamic aerodynamic input data and cockpit interaction credibility. They need to comply to the general physic model.I feel the anarchy in the air.

 

This is where you are really off base. Equations of motion in DCS are not up to 3rd parties and never have been.

 

MP should be in a nutshell isolated from streaming complex positional data and act as supercomputer cluster utilizing all cores available over network. The only thing that should be duplex streamed on MP server is cockpit control inputs that should give flawless results on any client if engine is set correctly.

 

Lol yeah, and if my grandma had wheels, she'd be an omnibus. This paragraph is preposterous nonsense. If that was even remotely possible, don't you think that MP games all over the place would do it?

 

This is what I would do if I would make a 'simulator'. I just don't have time and I'm not paid to do that. Technology is here and DCS just needs to adapt. Fast.

 

Actually employing technology is a bit more complicated than throwing buzz words around.

Edited May 18, 2017 by sobek

[shagrat]

Lattice-Boltzmann supposedly has problems with higher Mach numbers. You don't think that could be a problem for a flight sim? That aside, even if Lattice-Boltzmann is vastly more efficient than Navier-Stokes, doing real time CFD might still be out of the question for performance reasons. Also it is a relatively new technique, so you won't find many people that are proficient in it.

 

 

 

This is where you are really off base. Equations of motion in DCS are not up to 3rd parties and never have been.

 

 

 

Lol yeah, and if my grandma had wheels, she'd be an omnibus. This paragraph is preposterous nonsense. If that was even remotely possible, don't you think that MP games all over the place would do it?

 

 

 

Actually employing technology is a bit more complicated than throwing buzz words around.

^this...

 

I love the part about "the procedural generator"... That in fact was introduced along with EDGE and is currently used for procedural grass, clutter and more to come in the future.

 

Also the "just throw more CPU cores into Multiplayer and have better performance" is interesting. What exactly would the different cores calculate? And how do we sync the results in the simulated world?