Appendix 1 APPENDIX 1. STANDARD DATA FORMAT FOR DIGITAL FLIGHT DATA
RECORDERS, DATA STREAM FORMAT, AND CORRELATION DOCUMENTATION
1. DFDR. The DFDR records flight data in a digital format. Data is normally grouped into words that are synchronized in a data stream. The data stream must be correlated to EU or to discrete states for an accident investigator to use the data. This appendix provides the specific information that operators must provide to the NTSB investigators for them to decode and convert the raw data stream recorded by the DFDR. This data format information can be contained within an electronic file or in hardcopy document format.
2. GUIDELINES FOR DOCUMENTATION. Information regarding acceptable guidelines for documentation of DFDR data content and format are also available in the FRED document, ARINC Specification 647A. The ARINC FRED specification provides a standard format for electronic documentation of DFDR data. This standard format supersedes an older electronic format, the Flight Recorder Configuration Standard, Document TP13140E. Although the FRCS documentation format is acceptable, because FRED supersedes TP13140E, the preferred electronic format is the ARINC FRED format. The FRED specification, ARINC 647A, can be obtained by contacting:
ARINC Inc.
Document Section 2551 Riva Road Annapolis, MD 21401-7465 Telephone: (410) 266-4117 Fax: (410) 266-2047 E-mail: standards@arinc.com
3. DEFINITIONS. The following definitions apply to terminology often used in the DFDR correlation documentation. These definitions are derived from ARINC Characteristic 717-10, Flight Data Acquisition and Recording System, dated April 1, 1998. An operator using another data stream format must provide definitions unique to its format in the correlation documentation.
a. DFDR Bit Number. Defines a specific bit location within a DFDR system word on the output from the DFDR. The DFDR bit number is used to locate the bits that are dedicated to a given parameter within the word. For example DFDR bit numbers 3-12 indicate bit 3 through bit 12. The lowest bit number is normally the least significant bit.
b. Lexicon. A dictionary of all mnemonic codes and the associated parameters each represents. To aid in the standardization of parameter identification, ARINC Specification 647A appendix C contains a recommended lexicon.
c. Mnemonic Code. An abbreviation of the parameter name. It is intended to be used in formats where the parameter name is too large. The mnemonic code must be unique and used to identify the parameter as described in the applicable appendix (part 121 appendix B or M, part 125 appendix E, or part 135 appendices B through F). The aircraft manufacturer normally assigns a mnemonic code to each parameter. This code is normally correlated to the parameter
Appendix 1 name in the aircraft manufacturer’s interface control document. However, an installer or an operator may assign a mnemonic code to an added parameter. The mnemonic code must uniquely identify the parameter relative to all other parameters being recorded by the DFDR.
Additional characters may be necessary to uniquely identify the parameter as installed on the aircraft.
d. Parameter Name. The name of the function being recorded. The documentation must contain a means to correlate each recorded parameter name to those in the applicable appendix (part 121 appendix B or M, part 125 appendix E, or part 135 appendices B through F). Where possible, the parameter name should correlate to the aircraft manufacturer’s or applicant’s interface control documentation. Sufficient information must be contained in the parameter name to make it unique and to convey information on its source.
e. Range. The full range of a parameter (minimum to maximum) expressed in EU. Where the parameter range accommodated in the aircraft exceeds the parameter range specified in the regulations, (e.g., pressure altitude in large aircraft) the applicant should provide for recording the range accommodated in the aircraft. Enter “NA” for discrete parameters. Range of a parameter is defined by either:
(1) The full recordable range of the parameter on the flight recorder, or
(2) The parameter’s operational range. The operational range is typically a subset of the recordable range. The inclusion of operational range is useful in analysis of the recorded data.
NOTE: The inclusion of a parameter’s recorded range does not imply that the parameter is recorded linearly on the flight recorder.
f. Signal Source. The aircraft subsystem, or the dedicated transducer or signal conditioner, used to provide the signal for the DFDR, FDAU, or DFDAU (or equivalent).
g. Superframe Cycle. A subdivision of a given word slot address in a subframe. This typically provides 16 additional addresses. A counter provides the cycle number reference. The cycle number must be documented as a parameter.
h. Word Number. The location of a 12-bit word within the subframe. Word numbering starts at 1 (the first word in the subframe).
4. DATA STREAM FORMAT.
a. Data Stream Format. The data stream format defines where an analyst must look in the DFDR output to find a selected parameter or other information. Data stream format and encoded data format characteristics must be described and accurate to the specification provided by the manufacturer. The output of the data stream format should provide clear header and parameter information from the DFDR. Most DFDR data are encoded using the ARINC Characteristic 573 or ARINC Characteristic 717. Although these ARINC data bus encoding systems do not specify the data format, they provide the framework within which the data is formatted, in a fixed-frame format.
Appendix 1
b. Frames and Subframes. Specifically, they generate 12-bit words with a word rate that may vary, i.e., 32, 64, 128, 256, 512, or 1024 (words/second). The data is organized into frames that are repeated every 4 seconds. Each frame consists of 4 subframes. Each subframe occupies 1 second in the data stream, however, not all data streams are ARINC compliant and a subframe may occupy more or less than 1 second. The first word in each subframe normally provides the frame synchronization pattern. The data stream format should enable the analyst to locate header information and parameter information in the DFDR output.
c. Header Information. For DFDRs that conform to ARINC characteristics, the following information should be stored in a header file. (For other DFDR frame structures, the number of subframes per frame must be omitted and the frame structure must be uniquely described.) The header file must include:
(1) Aircraft make and model.
(2) Aircraft serial number.
(3) DFDR make and model/part number.
(4) Number of subframes per frame, for a fixed-frame. Or a description of each of the frame formats in a multiple-frame format data stream.
d. Record Information.
(1) For a fixed-frame format:
(a) Bits in the DFDR word.
(b) Number of DFDR words in a subframe.
(c) Time duration of the subframe (seconds).
(2) For a multiple-frame format:
(a) List of parameters in the frame, in order.
(b) The number of bits used to store each parameter.
e. Parameter Information. The identification for each parameter should include:
(1) Parameter name—must be unique from all other parameters recorded.
(2) Mnemonic code—a common abbreviation for the parameter, which must be unique from the mnemonic codes of all other parameters recorded.
f. Parameter Location. Both component(s) and timing information should be provided for each parameter sample. Usually there is only one component. For samples having more than one component the components should be ordered from least significant to most significant in the
Appendix 1 data stream. Each sample location for a parameter should have the same total number of bits. The following items must be provided for a parameter sample location:
(1) Subframe number.
(2) Word number.
(3) Bit numbers.
(4) If superframe cycles are used, also provide:
(a) Cycle counter name.
(b) Cycle numbers.
5. LEXICON OF MNEMONIC CODES. If mnemonic codes are used, a lexicon of these codes must be provided.
6. EU CONVERSION. EU conversions must convert decimal counts to the parameter value measured. Where the parameter is used by the pilot in flight (e.g., airspeed, altitude, and heading), the conversion must be correlated to the value shown to the pilot. Other values must be correlated to values sensed by the aircraft. Where an EU conversion results in an interim parameter (e.g., alternating current voltage ratio No. 1, frequency, direct current voltage ratio No. 2, potentiometer, or synchro angle) the conversion formula that converts this interim parameter to the actual parameter must also be provided. Each recorded parameter must be converted to an EU. Both signage (whether or not the raw data contains a plus or minus sign and the location of the sign in the parameter bits) and raw data range must be provided. Standardized signage is provided in ARINC Specification 647A appendices E and F. Instructions must be provided to enable the accident investigator to convert the recorded data to EU An example of acceptable conversion equations are as follows:
a. Linear Equation.
1010 CNTSAAEU ×+= where: EU is the value in engineering units 0CNTSthewhenEUofnumbertheisA 100 = 1A is the slope of the line in EU per decimal count 10CNTS is the number of binary or binary coded decimal (BCD) counts converted to decimal counts
Appendix 1
b. Piece-wise Linear Equation. In some instances it may be necessary to correlate the data using multiple linear equations. For example, the data of Figure 1-1 was obtained using two linear equations as follows:
For Altitude (EU) feet500≤ For Altitude (EU) feet500> ()() 0200231.0A 002.0A10 = −= ( )() 013351.0A 29673.0A10 = −= where, ≤ indicates equal to or less than > indicates greater than
c. Polynomial Equation.
X2 10x1021010 CNTSACNTSACNTSAAEU ×+×+×+= K where, EU is the value in engineering units 10CNTS is the number of binary or BCD counts converted to decimal counts 0A is the number of EU when the 0CNTS10 = X21 A,A,A K are coefficients developed by a curve fit
d. Unique Equation. Where a unique equation applies, explicit documentation must be developed at the time of certification and approved by the FAA. This documentation may contain sufficient information to allow a tool programmer the ability to develop a conversion that uses his tool efficiently.
7. EU DATA CORRELATION. When certifying new parameter implementations, an acceptable procedure to correlate the recorded data to the data derived from the conversion equation follows:
(1) Set or read the device being measured (radio altimeter altitude for example) to a known fixed data point.
(2) Record the data in the Raw Data (EU) column of Table 1-2.
(3) Record the decimal counts from the DFDR record in the 10CNTS column of Table 1-2.
NOTE: The DFDR normally records counts in a binary format. Most equipment manufacturers provide a digital output port and test equipment to access the data. Data can be displayed in binary, octal, or hexadecimal format with octal being the most common. Convert the counts to decimals using instructions provided by the equipment manufacturer.
(4) Repeat steps (1) through (3) above until a sufficient number of data points have been collected for correlation.
Appendix 1 NOTE: The number of required data points will vary depending upon the parameter and the transducer. As a minimum, three data points must be recorded for linear equation parameters, one at the mid-point (or null point) and the others at each end point of the range for the parameter. For polynomial equation parameters, the more data points taken for correlation will produce a higher quality conversion formula. Also, the recorded data points must include one at each end of the range and sufficient intermediate points to produce an accurate conversion formula. For piece-wise linear equations, a minimum of three data points must be recorded for each linear segment of the equation.
(5) Derive the EU value for each recorded 10CNTS value using the conversion equation or method provided. Record the derived value for each data point. When necessary intermediate conversion methods may be used to improve conversion accuracy of the parameter.
(6) The tabulated raw data and the derived data must also be plotted to confirm the prediction capability of the equation (linear, piece-wise linear, polynomial or unique). The plot must cover the full operating range for the parameter. See Figure 1-1 for an example plot. In the example of Figure 1-1, the data is so closely correlated (correlation coefficient = 0.99985) that the raw data and the derived data appear to be superimposed on each other.
FIGURE 1-1. AN EXAMPLE PLOT OF THE RADIO ALTIMETER IN FEET VERSUS
DECIMAL COUNTS
8. DISCRETE DECIPHER LOGIC. The correlation documentation must contain decipher logic for discrete parameters. The decipher logic must identify the status of the discrete represented by each binary state (e.g., “0” =OFF, “1”=ON). In instances where a group of discrete states is represented by multiple binary bits, the entire discrete word must be presented in the correlation document. See Table 1-1. The word slot, subframe, and bit logic for the specific discrete codes must be identified in the documentation.
Appendix 1 TABLE 1-1. EXAMPLE OF GROUPED DISCRETE CODES DECIPHER LOGIC Word Slot Subframe Discrete Status Discrete Codes 7 6 5 4 3 2 1 A/P Mode Throttle “OFF” 1 1 0 0 0 0 0 A/P Mode Throttle “RETARD” 1 1 0 0 0 0 1 A/P Mode Throttle “CLAMP” 1 1 0 0 0 1 0 A/P Mode Throttle “SPD/MCH,ALPHA” 1 1 0 0 0 1 1 A/P Mode Throttle “SPD/MCH,FLAP” 1 1 0 0 1 0 0 A/P Mode Throttle “SPD/MCH,SLAT” 1 1 0 0 1 0 1 A/P Mode Throttle “SPEED” 1 1 0 0 1 1 0 A/P Mode Throttle “EPR LIMIT” 1 1 0 0 1 1 1 A/P Mode Throttle “SPD/EPR LIMIT” 0 1 0 0 0
9. VALIDATION OF EU CORRELATION.
a. Correlation Coefficient. Calculate the correlation coefficient between the raw data and the derived data. Table 1-2 may be used to aid this calculation. The following method may be used to calculate the correlation coefficient for a parameter that is linearly converted.
( ) ()[] ()[]() 2/12222 yyNxxN yxyxN1 Σ−Σ××Σ−Σ× Σ×Σ−×Σ×= where, r = correlation coefficient N = number of data points x = raw data (EU) y = derived value (EU) Σ indicates the sum of the values that follow (e.g., Σx equals the sum of all x values)
Appendix 1 TABLE 1-2. CORRELATION TABLE DECIMAL COUNTS TO EU Parameter No. Parameter Conversion Method □ Linear □ Polynomial □ Unique 0A Unique Equation Reference:
1A 2A 3A 4A 5A Data Point No. Raw Data (EU) 10CNTS Derived Value (EU) 1 2 3 4 5 6 7 8 9 10
Appendix 1
b. Calculate the Correlation Coefficient. Table 1-3 demonstrates the steps in solving for the correlation coefficient.
TABLE 1-3. CORRELATION COEFFICIENT CALCULATION Parameter No. ___ Parameter Name _____________________________ Mnemonic Code _____________ Parameter Word Location ___________________ Subframe _______________ Superframe Cycle ___________________ Assigned Bits (1 through 12) __________________ Range (EU) ______________________ Sign Convention ______________________________ Correlation Coefficient _________ Data Pt No.
x Raw Data (EU) y Derived Value (EU) 2x 2y y×x 1 2 3 4 5 6 7 8 9 10
N= =Σx =Σy =Σ 2x =Σ 2y () =×Σ yx
(a) () =×Σ× yxN
(b) =Σ×Σ yx
(c) Subtract (b) from (a) () =Σ×Σ−×Σ× yxyxN
Appendix 1
(d) =Σ× 2xN
(e) Square the sum of x (Σx) and subtract this square from (d) ()[ ]=Σ−Σ× 22 xxN
(f) =Σ× 2yN
(g) Square the sum of y (Σy) and subtract this square from (f) ()[ ]=Σ−Σ× 2 yyN
(h) Multiply (e) and (g) ()[ ] ( )[ ]=Σ−Σ××Σ−Σ× 2222 yyNxxN
(i) Take the square root of (h) ()[] ()[ ]=Σ−Σ××Σ−Σ× 2222 yyNxxN
(j) Divide (c) by (i) () ()[] ()[] = Σ−Σ××Σ−Σ× Σ×Σ−×Σ×= 2222 yyNxxN yxyxN1
NOTE: The correlation coefficient (r) must not be less than 0.99.
c. Non-Linear Conversions. For non-linear parameter conversions, the following method, or method in (d) below, may be used to calculate the correlation coefficient. The Sum of Least Squares is used to determine the fit of data to a function in non-linear regression. The objective in the curve fit is to minimize the sum of the squared errors.
Coefficient of Determination or the Sum of Least Squares method is performed by using the following equation:
()( ) 2 ii n 1i xyRSS ∫−Σ= = where xi and yi are known values and f(xi) is the calculated y value from the curve fit function.
Appendix 1
d. Coefficient of Determination, R2 T E2 SS SS1R −= ( ) ( ) 2 iiiE 2 iiT yˆySS,yySS −Σ=−Σ= Where SSE is the RSS in paragraph (c) above.
SST is the Total Sum of Squares. The range of R2 is from 0 to 1.
0 = no correlation and the data is correlated when R2 = 1.