Why do electronic AIs, such as Garmin G5 need pitot static and GPS input for most accurate attitude information?

The electronic AIs are MEMS based. Reuse of automotive grade MEMS has been very successful in experimental avionics like the Garmin G5 or Dynon. The G5 and Dynon D10A/D100 are now FAA STC! The GPS is there to ensure accurate recalibration in flight.

I have 2 Dynon D10A EFIS/HSI in my Europa. Both are connected independently to a GPS. Once, I flew off with the Dynon D10A HSI in AI mode and the AI shutdown. After power on the AI EFIS, it took him about 1 minute to recover correct attitude. During this minute, I saw first crap information, then correct horizon but opposite to the roll and then full correct AI. The flight was also far from being stable horizontal during this minute.

MEMS are solid state accelerometers – one vendor site.

These devices drift really badly and need a continuous correction signal.

There are also certification issues which bring in additional requirements. I don’t know what they are but some of the vendors struggled to meet them.

An uncertified product can be made without the extra inputs – e.g. the Sandel SG102 which I have and that has no correction inputs and works seemingly adequately just with a MEMS gyro and gravity vector sensing. The SG102 generates a certified heading and uncertified pitch+roll.

A search on the following term
background AND erection
digs out various previous threads e.g. here

I vaguely recall Garmin use GPS whereas Avidyne and Aspen use airdata. Hence e.g. if the pitot ices over on the latter two you lose attitude info!

Why do electronic AIs, such as Garmin G5 need pitot static and GPS input for most accurate attitude information?

They do not actually need it. I did a lot of reading about this before purchasing instruments for my homebuilt. From what I gathered (right or wrong) a 3D MEMS gyro consists of 3 1 axis accelerometers and 3 1 axis gyros pointing in 3 direction. They are highly sensible and pretty much useless on their own. To get useful output, some filtering is needed. Doing just basic filtering works well, but cause drifting and inaccuracies. Still, since they measure acceleration, they can be made to self erect in a 1g environment. The only thing that really work is a Kalman filter. A Kalman filter consists of a simple dynamic model of the aircraft. However, a Kalman filter is dependent on a model of the particular aircraft itself to work really good. A Kalman filter also works better the more inputs (real world measurements) you put into it. These inputs are typically control deflection, thrust, magnetic heading and sensors such as IAS, alt etc. It can also be GPS for 3D position and speed or anything that gives some real world information to the filter. Military vehicles (missiles typically, but essentially all kinds of things) have used such systems for several decades.

For GA, a general purpose AI would have to work for lots of different aircraft. But, it’s only for display purposes, and doesn’t need to be all that accurate. It’s not going to be used to guide a missile to it’s target or couple together two space ships in space. It only needs to be on par with a mechanical AI. Therefore the Kalman filter can be made very simple, and the main input (measurement) needed is the fwd velocity apparently (the IAS). But, that velocity does not need to be accurate, it is enough to simply put in a constant, like the cruise speed, and it will work just fine in all circumstances.

Why some AIs needs to be coupled to GPS or air data, I don’t know. I can only guess the Kalman filter calculations are too slow, or poorly done in some way or the other, or they are using very poor (slow) MEMS gyros. In the last 20 years, there is probably nothing that has seen more rapid development than MEMS gyros and similar related stuff. Stuff that costed €100K 10 years ago cost €100 today.

The MGL AIs do not need any additional input than it gets from the MEMS gyro, but it will use GPS and air data if they are available. I don’t know what the difference would be, but the AI can also do without the MEMS gyro altogether. It then uses GPS and air data only (but it would not be able to tell if the aircraft was upside down).

Similarly the stand alone magnetometer has a 3D accelerometer incorporated. This enables full autocalibration among other things.

My guess is we soon will see “INS” in non certified equipment. Or at least we will see systems that are able to indicate position without GPS, INS or not. Many cars have had this for a decade already. The on board “GPS” works equally well in tunnels and other places with no GPS signal thanks to INS technology.

Peter wrote:

An uncertified product can be made without the extra inputs

RC Allen has a certified glass attitude indicator that takes only power (I don’t think it has an internal GPS receiver).

Peter wrote:

I vaguely recall Garmin use GPS whereas Avidyne and Aspen use airdata.

IIRC Garmin uses all data available (meaning both GPS and air data).

LeSving wrote:

My guess is we soon will see “INS” in non certified equipment.

It’s doable, of course. Question is what use it would be. I want INS so I can continue approach (irrelevant for a non-certified suite). En-route, time scale is typically so large the drift will make it useless unless you can switch over to some other means of correcting it (like DME/DME).

How can it be a backup instrument if it doesn’t have it, rather? Lose vacuum etc, you can’t rely on a GPS GS to tell you when you’re about to stall. It’s essential.