0
nigel99

aad algorithms

Recommended Posts

The recent incident posted by Peek prompted me to do some research on
the patents published on Cypres and Vigil. The good thing about
patents is it generally shows you how the various companies do things.

Firstly it looks like Cypres have covered the market in using a
combination of temperature and pressure sensors along with an
accelerometer. This patent runs until 2018 and is pretty broad in
scope and probably prevents the competition from accessing those
without using some clever workarounds (you can almost always work
around), alternatively the competition could license the technology
from Cypres. Details of the Cypres algorithm aren't patented as far as
I can tell - they have chosen to patent the hardware implementation.
This means that the details of their algorithm remain 'secret'.

Vigil uses a successive approximation algorithm. At first glance the
calculation gets more reliable the LONGER you are in free-fall. The
parameters used to estimate the descent rate are least reliable as you
climb through the first few hundreds of feet due to how it appears to
work. Feel free to read the patent and flow charts and agree/disagree
with my assessment. I'm pretty sure reading it that the student mode
is most susceptible to false fires and the aircraft simply 'dropping'
through turbulence over a 3/8's of a second period could cause a fire.
In short I believe the Vigil is designed to be very accurate in
descent, AFTER you have climbed to a reasonable altitude. I think a
Vigil may not be very accurate following a short free-fall descent. So
on a low altitude hop and pop things could be interesting. Note that
as the Cypres algorithm is not published similar concerns may apply -
however they have an accelerometer that gives an 'independent' metric
for the algorithms, providing substantially more data for decision
making.

Having read the patents, I would strongly recommend a Cypres over a
Vigil, and am considering trading my Vigil in for a Cypres.

I'd be really interested in seeing other peoples analysis of the various patents. This is obviously my interpretation of public documents, and companies don't always do things in the way the patent is published (and they're under no obligation to do so). So for example cypres may not use an accelerometer.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
At this point I don't think the info in the patents is much like what the software inside them is like. And I can't think of much reason to have an accelerometer in an AAD. It doesn't give you that much more useful information over what a pressure sensor gives you.

But it was a good idea to move the discussion here.

Share this post


Link to post
Share on other sites
Hi Nigel,

I am not an engineer [or in any kind of technical field]. Can you describe what a successive approximation algorithm is?

Would the operation of the Vigil be harmed in any way if it armed higher? Why does it arm at the low altitude?
"What if there were no hypothetical questions?"

Share this post


Link to post
Share on other sites
The unit could be made to arm at a higher altitude as it'll be collecting data just after take off.

A successive approximation algorithm works by improving its result as more data are collected. Early in the jump it's a rougher estimate than later.

The algorithms may just need improving on the vigil to cope with unforseen changes in pressure during climb, particularly around activation height since it's unlikely that an exit will be made at below 1500 ft (particularly on the student version).

The accelerometer on the cypres will be used to augment the barometer by sensing both freefall and opening. Typically barometers are only accurate to about 1kPa (which is about 280ft change at 1000 ft altitude). I don't know if Cypress use a more sensitive barometer than those readily available commercially.

An interesting topic nonetheless.

Share this post


Link to post
Share on other sites
Yes, it is an interesting topic. Thanks for your reply, it was helpful in my understanding.

Can the Vigil collect data (so it can improve its results) while still not being allowed to fire?

Is there a reason it has such a low arming altitude (I know this has been covered before but I just can't remember).
"What if there were no hypothetical questions?"

Share this post


Link to post
Share on other sites
Quote

At this point I don't think the info in the patents is much like what the software inside them is like. And I can't think of much reason to have an accelerometer in an AAD. It doesn't give you that much more useful information over what a pressure sensor gives you.

But it was a good idea to move the discussion here.



The big plus on an accelerometer is that it gives you an independent variable to compare against. This means a pressure change with no corresponding change in acceleration is a false positive and can be discounted.

I think Adam has done a very good job of explaining other stuff. Just both aad manufacturers claim.much better resolution than the 260ft he mentions.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

The big plus on an accelerometer is that it gives you an independent variable to compare against. This means a pressure change with no corresponding change in acceleration is a false positive and can be discounted.



You can buy accelerometers with freefall detection (like the ones that go into hard drives) and I think this would be a key piece of data when deciding when to fire the AAD.

Share this post


Link to post
Share on other sites
Quote

Yes, it is an interesting topic. Thanks for your reply, it was helpful in my understanding.

Can the Vigil collect data (so it can improve its results) while still not being allowed to fire?

Is there a reason it has such a low arming altitude (I know this has been covered before but I just can't remember).



I'm always weary of trying to 'improve' on designs without knowing the intimate details. On a safety critical design and pseudo military stuff an failure modes and effects analysis will have been done. In principle it sounds easy, just don't fire when you've not gone above 1500 feet. What we forget is the algorithm is intertwined with the height calculation.

From what I've read the vigil doesn't 'arm' at any altitude as such. It simply looks for a minimum of 3 simply looks for 3 successive readings showing a descent rate that exceeds the threshold for speed, whilst below the firing altitude. The readings are taken at 1/8 second intervals. That is why I believe the plane was climbing relatively slowly when the event took place. Just a combination of factors. My guess is if we collated all the incidents they would involve student mode (predominantly), slower climbing aircraft such as a heavily laden 182 on a hot day etc.

There is stuff in the vigil user manual regarding beeps and lights at activation altitude, I'm trying to figure out if my curiosity is worth potentially causing a cutter to fire. I've got a vacuum chamber at work and would like to see the lights and hear the beeps, I can't do that in an aircraft as my aad is against my back. I'm just worried that I dump the air back into the vacuum chamber too quickly and cost myself a new cutter.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

Quote

The big plus on an accelerometer is that it gives you an independent variable to compare against. This means a pressure change with no corresponding change in acceleration is a false positive and can be discounted.



You can buy accelerometers with freefall detection (like the ones that go into hard drives) and I think this would be a key piece of data when deciding when to fire the AAD.



Buying accelerometers is the easy part.

Building the algorithm to figure out that they have accelerated into freefall, sustained freefall, are in a study state (ie not accelerating anymore, or are in and out of acceleration that would be consistant with what a freefaller would do in RW, FF, in and out of tracking, diving, funnelling, etc) and them making a decision, THAT'S the hard part.
Remster

Share this post


Link to post
Share on other sites
Quote

Quote

The big plus on an accelerometer is that it gives you an independent variable to compare against. This means a pressure change with no corresponding change in acceleration is a false positive and can be discounted.



You can buy accelerometers with freefall detection (like the ones that go into hard drives) and I think this would be a key piece of data when deciding when to fire the AAD.



Just that a quick read shows that doing so without a license from Cypres, would probably infringe their patent with the way the claims are worded. By the way a friend reminded me today of an incident in the early 90's at our club, a visiting jumpers Cypres fired as he dropped his rig on the floor after a jump.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

Can the Vigil collect data (so it can improve its results) while still not being allowed to fire?


Yes.

Quote

Is there a reason it has such a low arming altitude (I know this has been covered before but I just can't remember).



That'd be a decision by Vigil. One of the difficulties when designing high integrity safety systems (we work with detectors that conform to IEC 61508 which is a standard for high reliability & integrity systems) is that you have to design for normal operation, abnormal operation and possible fault conditions.

Determining each possible scenario and then deciding what the outcome is a lengthy task and in the Vigil's case it has to estimate what's actually happening dependent on only one or 2 inputs. Two scenarios (e.g. a low exit by a jumper vs. a rapid pressure change within the plane) could look very similar purely from a pressure perspective hence why the accelerometer input would be useful.

Sorry for the long post. Just couldn't stop typing!

Share this post


Link to post
Share on other sites
Quote

Details of the Cypres algorithm aren't patented as far as
I can tell - they have chosen to patent the hardware implementation.
This means that the details of their algorithm remain 'secret'.



That's because you can't patent an algorithm, only the hardware designed to process the algorithm.

Have you got the patent numbers?

Share this post


Link to post
Share on other sites
Quote

Quote

Details of the Cypres algorithm aren't patented as far as
I can tell - they have chosen to patent the hardware implementation.
This means that the details of their algorithm remain 'secret'.



That's because you can't patent an algorithm, only the hardware designed to process the algorithm.

Have you got the patent numbers?



Jack, I'm far from an expert on this but in practical terms you are wrong. I've personally got my name on a world wide patent filed in the US that is for an algorithm, I've also got a patent filed in the UK for an algorithm. It might be a definition thing, to me an algorithm is a method of doing something, so a software decision tree/program flow would be an algorithm. To a mathematics person it is probably different. I know you can't patent abstract ideas.

The patent number for vigil is EP1084950 just search on espacenet for it. The cypres patents are in Helmut Cloth name and I just used the name search for theirs. Cypres patents haven't always been translated from German.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Hi all,

I can offer a little insight. In my past life I designed an autopilot and this experience has given me a very healthy respect for the design challenges for AAD's.

For the autopilot I used a barometric sensor and in the first iteration I used an accelerometer. I needed both to be able to detect slow altitude changes versus something caused by turbulence. I ultimately had to change to a MEM's gyro but that is a longer story.

What I can say that is pretty easy to get a few feet of resolution with a barometric sensor, but it takes a bit of fooling around to filter out noise. For an AAD this would be a bunch of work, because it could potentially move in and out of the shadow of the skydiver's body. This coupled with having to deal with pressure change effects in an airplane along with a 101 things I haven't begun to contemplate makes the hardware and software design of an AAD "interesting" to say the least.

I know we all want our AAD's to work flawlessly in every situation, but it's a pretty tall order.

Share this post


Link to post
Share on other sites
Quote



Jack, I'm far from an expert on this but in practical terms you are wrong. I've personally got my name on a world wide patent filed in the US that is for an algorithm, I've also got a patent filed in the UK for an algorithm. It might be a definition thing, to me an algorithm is a method of doing something, so a software decision tree/program flow would be an algorithm. To a mathematics person it is probably different. I know you can't patent abstract ideas.



One thing I've learned is that patent law is very, very weird and what might squeeze past one examiner, might not squeeze past the next. But from USSC Gottschalk v. Benson (1972), European Patent Convention Art. 52(2) and others, in general terms you cannot patent an algorithm. If you have (and it is entirely possible you squeezed one past your examiner somehow), then you either have a challengable patent or your algorithm is tied to some implementation thereof.

I would also be interested to know how you managed to get a world wide patent since they don't exist. To get one you'd have to patent your design in 190+ different countries which again is possible if you were prepared to fund your patent attorneys new yacht on the back of it.

Share this post


Link to post
Share on other sites
Quote

Quote



Jack, I'm far from an expert on this but in practical terms you are wrong. I've personally got my name on a world wide patent filed in the US that is for an algorithm, I've also got a patent filed in the UK for an algorithm. It might be a definition thing, to me an algorithm is a method of doing something, so a software decision tree/program flow would be an algorithm. To a mathematics person it is probably different. I know you can't patent abstract ideas.



One thing I've learned is that patent law is very, very weird and what might squeeze past one examiner, might not squeeze past the next. But from USSC Gottschalk v. Benson (1972), European Patent Convention Art. 52(2) and others, in general terms you cannot patent an algorithm. If you have (and it is entirely possible you squeezed one past your examiner somehow), then you either have a challengable patent or your algorithm is tied to some implementation thereof.

I would also be interested to know how you managed to get a world wide patent since they don't exist. To get one you'd have to patent your design in 190+ different countries which again is possible if you were prepared to fund your patent attorneys new yacht on the back of it.



Ok I simplified by saying world wide. It was filed in the US with lots of money (spent by the client) on getting coverage in most other major jurisdictions (Europe, Japan etc) . The primary filing was done in the US.

The 2 UK patents that I have are ONLY filed and provide coverage in the UK, as my pockets weren't deep enough to do anything else.

I realise that algorithm means something different to different people. So ok what I called an algorithm is a method for determining the best routing/switching of radio packets (one of my patents).

Anyway - have a look at the Vigil patent - I would define that as a patent on their algorithm.

I'm happy to take the patent specifics to PM, if you want to clarify stuff. I don't really want to sidetrack the main point of my thread, being the implementation/trade-offs that the manufacturers have to make.

By the way. 2 weeks ago I flew from one dz that was 1200foot higher than our destination. There were 3 of us on the flight, 2 Cypres and my Vigil. We did not climb above 1000ft from our take off point. In this scenario, neither Cypres would have fired if required to do so. My Vigil would have been the only 'operable' AAD (although it was off). We didn't exit the plane, but just giving an example from personal experience.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

[Anyway - have a look at the Vigil patent - I would define that as a patent on their algorithm.



I've only read the claims quickly but I wouldn't say it patents an algorithm. In the Vigil patent, the claims are for a device that records and processes data to activate a parachute. The interesting bit is the device compares previously stored theoretical dive data to provide an estimate of remaining altitude/freefall time.

The part in any patent document that counts are the claims, the rest of the document is generally there to provide context only. Maybe Vigil have presented their algorithm in the text, I haven't read it yet, but it doesn't appear as a claim.

I'd actually say that the user manual would be a better place to find algorithm information as patents are usually difficult to read, intentionally vague and obfuscated, and often only loosely related to what the commercialized device actually does.

Share this post


Link to post
Share on other sites
Neither of the user manuals go into any detail on 'how' the devices work. I didn't see anything in the Cypres user manual on accelerometers, and in the vigil manual I saw nothing about how it initially uses 3 readings to estimate the descent rate, and then successively refines the data to improve the accuracy/reliability of the result.

The Vigil patent has a fairly detailed flow chart on how they might be doing things. Similarly there is a block diagram showing how a Cypres might be utilising hardware elements.

While the method that is described by Vigil gets more reliable with time, there is a window where it is initially quite vulnerable. IF a Vigil is working as described in that document, with no additional checks and balances, it would be wise to avoid situations where the aircraft is climbing slowly or holding below the activation altitude (2000ft for tandems). At least if the plane is ascending rapidly, any temporary blips in pressure should result in a reduced ascent rate, rather than being interpreted as a descent. As soon as you have a descent reading below activation altitude, you're completely dependent on the rate calculation.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

IF a Vigil is working as described in that document, with no additional checks and balances, it would be wise to avoid situations where the aircraft is climbing slowly or holding below the activation altitude (2000ft for tandems)..

it is a vintage piston plane. What do you expect ? a rocket rate of climb ? Plus sometimes you MUST level or descend, due to ATC/other aircraft/changing weather etc...

I have my choice of AAD, and it is of "that other brand" :)
scissors beat paper, paper beat rock, rock beat wingsuit - KarlM

Share this post


Link to post
Share on other sites
Quote

Neither of the user manuals go into any detail on 'how' the devices work



You'd have to reverse engineer the manuals and patents to dig that info out. Even then it might be impossible to figure it out since algorithms are proprietary and unpatentable so there is no incentive for manufacturers to publish that information in its entirety and every incentive to stop their competitors from finding out how they do what they do.


Quote

While the method that is described by Vigil gets more reliable with time, there is a window where it is initially quite vulnerable. IF a Vigil is working as described in that document, with no additional checks and balances, it would be wise to avoid situations where the aircraft is climbing slowly or holding below the activation altitude (2000ft for tandems). At least if the plane is ascending rapidly, any temporary blips in pressure should result in a reduced ascent rate, rather than being interpreted as a descent. As soon as you have a descent reading below activation altitude, you're completely dependent on the rate calculation.



Well Vigil's successive approximation method doesn't necessarily mean the algorithm gets more accurate as the jump progresses. It could be used for that, or it could be used to weed out unphysical firing conditions and prevent misfires, or it could do both. If Cypres don't have something similar, then they must be doing it off pressure alone and their algorithm is as robust all the way down as Vigils was when the Vigil first left the aircraft. If that's the case, chalk one up to Vigil.

My guess is that Cypres have either found this either isn't needed or they've got their own way of doing it. They probably wont tell either way.

Share this post


Link to post
Share on other sites
my non-engineer/non-computer mind is that Airtec has got some "filters" to weed out the aberrations, which seemingly AAD doesn't have.
like trying to save you underground, or saving you because you just blew the altitude by accelerating at 30G :D
scissors beat paper, paper beat rock, rock beat wingsuit - KarlM

Share this post


Link to post
Share on other sites
Quote

Quote

IF a Vigil is working as described in that document, with no additional checks and balances, it would be wise to avoid situations where the aircraft is climbing slowly or holding below the activation altitude (2000ft for tandems)..

it is a vintage piston plane. What do you expect ? a rocket rate of climb ? Plus sometimes you MUST level or descend, due to ATC/other aircraft/changing weather etc...

I have my choice of AAD, and it is of "that other brand" :)



Just stating my view that a vigil is more likely to 'misfire' in a slow climbing aircraft. Just remember that the very strength of a cypres means that in a reasonable number of scenarios, you may as well not have it, as the safety will kick in and it will not fire. Look carefully at the diagrams showing the restricted zones when jumping into different elevation dzs, and then remember the whole climb, descend and hold limitations. Not at all saying this is wrong of the device, just jumpers often don't know enough about it. For what it is worth, I'd choose Cypres if selecting an aad new off the shelf.
Experienced jumper - someone who has made mistakes more often than I have and lived.

Share this post


Link to post
Share on other sites
Quote

By the way. 2 weeks ago I flew from one dz that was 1200foot higher than our destination. There were 3 of us on the flight, 2 Cypres and my Vigil. We did not climb above 1000ft from our take off point. In this scenario, neither Cypres would have fired if required to do so. My Vigil would have been the only 'operable' AAD (although it was off). We didn't exit the plane, but just giving an example from personal experience.



This means your exit height above ground would have been 2200 feet or less. If I did this on a planned jump I'd switch my AAD off anyhow, because of the risk of a two-out.

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

0