Roger,

This is an interesting format.

I have found that DFDL newbies frequently have very difficult formats to deal with. They end up on the cutting edge of DFDL language features. Basically, one of the first things people have to explore is whether DFDL can really handle their data problems. So hard cases first.

But anyway, to your use case....

The first way I started to model this is as a vector of bits that are the physical representation, and a computed value (dfdl:inputValueCalc).

Something like this:

...assume the schema has dfdl:lengthUnits='bits'

<dfdl:defineVariable name="Range173" type="xs:double">0.4</dfdl:defineVariable>

<sequence>
   <element name="bits" minOccurs="12" maxOccurs="12">
      <sequence>
         <element name="bit" type="xs:unsignedInt" dfdl:length="1"
              dfdl:outputValueCalc="{ .... omit for now. ...}" />
      </sequence>
  </element>
  <element name="value" type="xs:double"
     dfdl:inputValueCalc="{
          ( ../bits[1]/bit * $Range173 / 2.0) +
          ( ../bits[2]/bit * $Range173 / 4.0) +
           ...
           ( ../bits[12]/bit * ... you get the idea...range divided by the right number here...)
       }" />
</sequence>

This is doable, but really isn't that great a way to model this. Lots of overhead. It has the advantage of being able to give the bits any weighting factors at all. Note that I omitted the dfdl:outputValueCalc needed for unparsing the format above. That's also doable, but a little painful.

But the right way is to model this as an unsigned integer of length 12 (bits), and then convert to a double float and multiply by the right scale factor.

For your message example labeled 173 the value of the double is

( ( x * sign) / 4096.0) * 0.4

where sign is -1 or 1 depending on the sign bit value.

That formula is easily inverted (for the unparse direction), so you end up with this schema having raw integer and value floating point elements.

<sequence>
  <element name="raw" type="xs:unsignedInt" dfdl:length="12"
      dfdl:outputValueCalc="{ (../value / 0.4) * 4096.0 * (if (../value lt 0.0) -1.0 else 1.0) }"/>
  <element name="value" type="xs:double"
     dfdl:inputValueCalc="{ ((../raw * (if (../sign eq 1) -1.0 else 1.0) ) / 4096.0) * 0.4) }"/>
</sequence>

You'd also need a dfdl:outputValueCalc on the sign element, which is elsewhere in the data, so
that it takes on 1 or 0 based on the sign of the value element when unparsing.

This is not as generic as you would like. You need one of these for every range e.g., the 0.4 above, and for every bits length (12 above), but the scheme is clear, and at runtime this would be very efficient.

If you have many of these combinations of different lengths and ranges, then I'd whip up a little DFDL schema generator that just creates all the combinations from a common template.

I hope that helps.

-Mike Beckerle


Mike Beckerle | OGF DFDL Workgroup Co-Chair | Tresys Technology | www.tresys.com
Please note: Contributions to the DFDL Workgroup's email discussions are subject to the OGF Intellectual Property Policy


On Tue, Apr 25, 2017 at 9:11 PM, Champagne, Roger <roger.champagne@etsmtl.net> wrote:
Steve, Mike,

I just shared a spreadsheet illustrating the encoding/decoding of two different ARINC429 messages, you should have received the link through Google.

The data portion of both messages is 12 bits long, but have different maximum values or ranges. Both messages admit positive and negative values.

Encoding of floats in ARINC429 follows a few simple rules:
  • a maximum value and number of bits are established for the data;
  • the most significant bit represents one half of the maximum value;
  • the value of each successive bit is one half of the previous;
  • negative values are represented in two's complement. The sign is contained elsewhere in the 32 bit word and is not included in the data shown. I am showing the binary values for both positive and negative values for each of the two messages. I produced the twos complement values by hand, having not figured out how to do that automatically in a spreadsheet (bitwise binary addition).
In an ideal world, we would like to be able to specify in DFDL the maximum value (float) and number of bits for each message, and the encoding and decoding would be performed the same way for all messages, based on these two parameters.

If anything is unclear, please let me know. I most appreciate your help with this.

Sincerely,

Roger


--------------------------------------------------------------
Roger Champagne, P.Eng., Ph. D.
Professeur agrégé, Département de génie logiciel et des TI
Associate Professor, Department of Software and IT Engineering

École de technologie supérieure
1100, rue Notre-Dame Ouest, Montréal, Québec, H3C 1K3, Canada
Bureau / Office : A-4499
Téléphone / Phone : +1 514 396 8825 -- Télécopieur / Fax : +1 514 396 8405

On 25 April 2017 at 05:46, Steve Hanson <smh@uk.ibm.com> wrote:
Roger

Please can you supply a couple of concrete examples - for each, the ARINC 429 message, what the bits mean, and what you want to appear in the infoset.

Regards
 
Steve Hanson

IBM Hybrid Integration, Hursley, UK
Architect, IBM DFDL
Co-Chair, OGF DFDL Working Group
smh@uk.ibm.com
tel:+44-1962-815848
mob:+44-7717-378890



From:        "Champagne, Roger" <roger.champagne@etsmtl.net>
To:        Steve Hanson/UK/IBM@IBMGB
Cc:        smhdfdl@gmail.com, mbeckerle.dfdl@gmail.com
Date:        23/04/2017 15:50
Subject:        Re: My understanding of binary encodings in DFDL




Steve,

Thanks for the swift reply.

I am a DFDL newbie, so not familiar with the concepts you propose. As there are hundreds of ARINC 429 messages, and each has its own range of values and number of bits, I am guessing this is a bit complicated.

Ideally, we would be able to specify the number of bits, plus one of either range of values or value of most/least significant bit, and that would be it. If this is not directly supported but can be done through other means such as expression language, that would be great.

Otherwise, this is kind of a show stopper for us. We should have caught this sooner.

Sincerely,
Roger

--------------------------------------------------------------
Roger Champagne, P.Eng., Ph. D.

Professeur agrégé, Département de génie logiciel et des TI
Associate Professor, Department of Software and IT Engineering
École de technologie supérieure
1100, rue Notre-Dame Ouest, Montréal, Québec, H3C 1K3, Canada

Bureau / Office : A-4499

Téléphone / Phone : +1 514 396 8825 -- Télécopieur / Fax : +1 514 396 8405

Web: http://etsmtl.ca/Professeurs/rchampagne
   

Le 23 avr. 2017 3:13 AM, "Steve Hanson" <smh@uk.ibm.com> a écrit :
Hi Michael

Thanks for reaching out.  Direct contact is fine but better to post to the DFDL mailing list dfdl-wg@ogf.org.

DFDL supports binary integers of arbitrary length, but not binary floats.

Could you treat the mantissa and exponent as integers inside a hidden group, then use inputValueCalc to fabricate a float using the DFDL expression language?  

Michael has more experience with calculated values and may be able to offer more advice.

Regards
 
Steve Hanson
IBM Hybrid Integration, Hursley, UK
Architect, IBM DFDL
Co-Chair, OGF DFDL Working Group
smh@uk.ibm.com
tel:+44-1962-815848
mob:+44-7717-378890



From:        "Champagne, Roger" <roger.champagne@etsmtl.net>
To:        mbeckerle.dfdl@gmail.com, smhdfdl@gmail.com
Date:        22/04/2017 22:24
Subject:        My understanding of binary encodings in DFDL



Michael, Steve,

I am not sure this is the proper way to proceed, but I figured I would give it a try. Please redirect me to the proper channel if relevant.

I am working on an avionics project where we are trying to describe ARINC429 messages in DFDL. After a couple of months of a student working on this and requesting my assistance, I realize DFDL is perhaps not a good match for our needs.

My understanding, based on section 13.8 of the specification, is that for floating-point values represented in binary, only two encodings are supported in DFDL: IEEE 754 and IBM 390 HEX. In both cases, the data must be exactly 32 bits (4 bytes) long.

Question 1: Is this assumption exact?

The reason I ask is that ARINC429 messages typically consist in a single 32 bit word. The message data is a subset of the 32 bits, and has variable length (up to 18 bits), based on the nature of the message. The way the data is encoded is different for each message, based of the range of values for that message and the number of bits it uses. We are try to represent these messages in DFDL.

Question 2: If my assumption is correct, are there any plans to support in DFDL binary encodings of arbitrary length, based on the knowledge of number of bits and range of admissible values?

I can supply more details if required, I tried to keep this initial message short.

Thanks for any light you can shed on this matter. 

Most sincerely,

Roger

--------------------------------------------------------------
Roger Champagne, P.Eng., Ph. D.
Professeur agrégé, Département de génie logiciel et des TI
Associate Professor, Department of Software and IT Engineering

École de technologie supérieure
1100, rue Notre-Dame Ouest, Montréal, Québec, H3C 1K3, Canada
Bureau / Office : A-4499
Téléphone / Phone : +1 514 396 8825 -- Télécopieur / Fax : +1 514 396 8405
Web: http://etsmtl.ca/Professeurs/rchampagne

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU

Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU