[Unicode]  Technical Reports

Unicode Technical Report #16


Version 7.2
Authors V.S. Umamaheswaran (umavs@ca.ibm.com)
Date 2001-04-29
This Version http://www.unicode.org/unicode/reports/tr16/tr16-7.2
Previous Version http://www.unicode.org/unicode/reports/tr16/tr16-7.1
Latest Version http://www.unicode.org/unicode/reports/tr16


This document presents the specifications of UTF-EBCDIC - EBCDIC Friendly Unicode (or UCS) Transformation Format.


This document has been reviewed by Unicode members and other interested parties, and has been approved by the Unicode Technical Committee as a Unicode Technical Report. It is a stable document and may be used as reference material or cited as a normative reference from another document.

A Unicode Technical Report (UTR) may contain either informative material or normative specifications, or both. Each UTR may specify a base version of the Unicode Standard. In that case, conformance to the UTR requires conformance to that version or higher.

A list of current Unicode Technical Reports is found on http://www.unicode.org/unicode/reports/. For more information about versions of the Unicode Standard, see http://www.unicode.org/unicode/standard/versions/ .
Please mail corrigenda and other comments to the author(s).



The term UTF-EBCDIC stands for EBCDIC-friendly Unicode (or UCS) Transformation Format. EBCDIC, IBM's Extended Binary Coded Decimal Interchange Code, is one of the widely used 8-bit industry encodings. Detailed information on EBCDIC can be found in the IBM publication IBM Character Data Representation Architecture, Reference and Registry, SC09-2190-00, December 1996.

To address the use of Unicode character data in byte-oriented ASCII-based systems, the Unicode Standard (see section A.2 of the Unicode Standard) (also ISO/IEC 10646 -1, Amendment no. 2) has defined UTF-8. Use of UTF-8 permits existing ASCII-based systems that have hard-coded dependency on the encoding of the ASCII repertoire of characters to safely process the corresponding Unicode characters. There is a similar requirement to transform Unicode characters to a form that is safe for EBCDIC systems for the control characters and invariant characters.

This Technical Report defines the UTF-EBCDIC.

Neither UTF-EBCDIC nor its intermediate form called UTF-8-Mod in this technical report, are intended to be used in open interchange environments. It is useful in homogeneous EBCDIC systems and networks.


The UTF-EBCDIC encoding is derived from the Unicode scalar values following a two step process: These two steps are defined below.

Note: The following notation is used in this Technical Report: X'nn .. mm' represent hexadecimal values; <bb...bb> represent values in bit notation; U+abcd represents a Unicode character.


3.1 Step 1: UTF-8-Mod

The UTF-8-Mod transformation definition is modeled after the UTF-8 definition in the Unicode standard. UTF-8-Mod transforms the Unicode scalar values into I8-sequences. The Unicode characters U+0000 to U+001F (corresponding to the C0 control characters X'00' to X'1F' of ASCII), U+0020 to U+007E (the ASCII repertoire), and U+007F (the ASCII 'DEL' control character) are represented as single bytes in the I8-sequence, similar to UTF-8. In addition, U+0080 to U+009F (corresponding to the so-called C1 set of controls in ISO/IEC 6429) are also represented as single bytes (X'80' to X'9F'). Thus the 65 Unicode characters corresponding to the 65 ISO/IEC 6429 control characters and the 95 characters corresponding to the 95 ASCII graphic characters (the G0 set) are represented in the I8-sequence as single bytes.

When these I8-sequence bytes are converted to the UTF-EBCDIC form, the corresponding 65 EBCDIC control characters and 95 EBCDIC graphic characters are preserved as single bytes in the UTF-EBCDIC byte sequence. The 95 EBCDIC graphic characters include 82 invariant (occupy the same code position) characters (including SPACE) across most EBCDIC single-byte code pages and 13 variant ASCII graphic characters (occupy varying code positions). Positions assigned to EBCDIC controls, the invariant graphic characters and the variant graphics are shown in Table B.1.

Furthermore, the values X'00'...X'9F' do not appear in any byte of an I8-sequence except as the direct representation of U+0000 to U+009F. Each Unicode scalar value that is not a part of a valid surrogate pair is represented in an I8-sequence by 1, 2, 3 or 4 bytes, depending on the value. A valid surrogate pair is first converted to its corresponding Unicode scalar value, which then maps into either 4 bytes or 5 bytes, depending on the value.

The UTF-8-Mod transformation is intended to be used only as an intermediate step in arriving at UTF-EBCDIC. It is not intended to be used elsewhere.

The I8-sequence is a variable length encoding of Unicode characters as 8-bit byte sequences, where the high bits of each byte indicate which part of the sequence a byte belongs to. Table 1 shows how the bits in a Unicode scalar value (or a valid surrogate pair) are distributed among the bytes in the I8-sequence. I8-sequence corresponding to a valid surrogate pair is also shown, including the UTF-16 transformation to convert the valid pair to the corresponding Unicode scalar value.

Table 1:  I8- Sequence Bit Distribution
Unicode Scalar Value (hex) Bit pattern of Unicode Scalar Value 1st Byte 2nd Byte 3rd Byte 4th Byte 5th Byte
0 to 7F 000000000xxxxxxx
80 to 9F 00000000100xxxxx
A0 to 3FF 000000yyyyyxxxxx
400 to 3FFF 00zzzzyyyyyxxxxx
4000 to 3FFFF 0wwwzzzzzyyyyyxxxxx
40000 to 10FFFF rwwwwwzzzzzyyyyyxxxxx
Note: The UTF-8-Mod transformation is valid for UCS-4 values X'0' to X'7FFFFFFF' (the full extent of ISO/IEC 10646 coding space). Only the Unicode scalar values corresponding to the end of plane 16 -- the reach of the UTF-16 transformation -- are shown in the above table.

A valid surrogate pair -- a high surrogate from the range X'D800' to X'DBFF' followed by a low surrogate from the range X'DC00' to X'DFFF' -- must be converted to its corresponding Unicode scalar value in the range X'10000' to X'10FFFF', using the UTF-16 transformation. The following table shows the correspondence between the bit patterns of the surrogate pairs and the corresponding I8-sequence bytes.

Unicode Scalar Value (hex) Bit pattern of valid Surrogate Pairs 1st Byte 2nd Byte 3rd Byte 4th Byte 5th Byte
10000 to 3FFFF 110110uuuuwzzzzz
+ 110111yyyyyxxxxx
40000 to 10FFFF 110110uuuuwzzzzz
+ 110111yyyyyxxxxx
where a uuuu = 000pp -1, or b uuuu = qpppp -1
(to account for addition of 1000016 as in Section 3.7, Surrogates, in the Unicode Standard 2.0)
When converting Unicode values to I8-sequences, always use the shortest number of bytes that can represent these values. This preserves uniqueness of encoding. For example the Unicode value <0000000000000001> is encoded as <00000001>, not as <11000000> <10100001>. The latter is an example of an unused I8-sequence. Do not make use of these unused byte sequences for encoding any other information.

When converting from I8-sequences to Unicode scalar values, however, implementations do not need to check that the shortest number of bytes is being used, which simplifies the conversion algorithm.

3.2 Characteristics of the I8-sequence

3.3 Step 2: Byte Conversion

Characteristics of the I8- sequence The second step of UTF-EBCDIC transforms the I8-sequences, using a reversible one-to-one mapping, into the byte sequences of UTF-EBCDIC.

The 64 control characters (U+0000 to U+001F, U+0080 to U+009F), the ASCII DELETE character (U+007F), the 95 ASCII graphic characters (including the SPACE character) (U+0020 to U+007E) are mapped respecting EBCDIC conventions, as defined in IBM Character Data Representation Architecture, CDRA, with one exception -- the pairing of EBCDIC Line Feed and New Line control characters are swapped from their CDRA default pairings to ISO/IEC 6429 Line Feed (U+000A) and Next Line (U+0085) control characters (to be in line with IBM OS/390 UNIX Services, or Open MVS practice and preference, stemming from the hard-coding of X'0A' as the New Line in most ASCII-C compilers.).

The map preserves the invariance for a set of 82 graphic characters (including SPACE) (known as the IBM Syntactic Graphic Character set), and maintains consistency with the IBM MVS Open Systems Code page (CPGID 1047) for the variant characters from within the ASCII repertoire.

The remaining 96 bytes of EBCDIC 8-bit structure are allocated to X'A0' to X'FF' -- the trailing bytes and leading bytes of the I8-sequence.(from Table 1). The minimum criterion for allocation of these bytes is that it provides for a reversible map.

The trailing and leading bytes (X'A0' to X'FF' of the I8-sequence) are paired with the unassigned UTF-EBCDIC bytes in increasing order. Table 2 and Table 3 show the byte maps between the I8-sequence bytes and UTF-EBCDIC bytes in the forward and reverse directions respectively. The resulting UTF-EBCDIC multi-byte sequences will be in the same lexical (numerical) order as their corresponding Unicode scalar values (when the sequences are zero-filled to equal number of bytes and compared with each other). Please note that the UTF-EBCDIC single-byte values, however, will not be in the same order as their corresponding Unicode scalar values.

The resulting UTF-EBCDIC byte sequence can be transparently processed in most EBCDIC systems. It also retains all the characteristics (see the section 3.2 Characteristics of the I8-sequence above) of I8-sequence mentioned earlier. Since EBCDIC code page definitions have 13 variants (and only 82 invariants) the choice of the above byte map for the graphic characters has been made to accommodate the MVS Open Systems environment for standardization purposes.

Table 2: Byte map from I8- sequence to UTF-EBCDIC byte sequence

 ß High nibble (hex)                                      Low nibble (hex) Þ (all entries are in hex)
  -0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -A -B -C -D -E -F
0- 00 01 02 03 37 2D 2E 2F 16 05 15 0B 0C 0D 0E 0F
1- 10 11 12 13 3C 3D 32 26 18 19 3F 27 1C 1D 1E 1F
2- 40 5A 7F 7B 5B 6C 50 7D 4D 5D 5C 4E 6B 60 4B 61
3- F0 F1 F2 F3 F4 F5 F6 F7 F8 F9 7A 5E 4C 7E 6E 6F
4- 7C C1 C2 C3 C4 C5 C6 C7 C8 C9 D1 D2 D3 D4 D5 D6
5- D7 D8 D9 E2 E3 E4 E5 E6 E7 E8 E9 AD E0 BD 5F 6D
6- 79 81 82 83 84 85 86 87 88 89 91 92 93 94 95 96
7- 97 98 99 A2 A3 A4 A5 A6 A7 A8 A9 C0 4F D0 A1 07
8- 20 21 22 23 24 25 06 17 28 29 2A 2B 2C 09 0A 1B
9- 30 31 1A 33 34 35 36 08 38 39 3A 3B 04 14 3E FF
A- 41 42 43 44 45 46 47 48 49 4A 51 52 53 54 55 56
B- 57 58 59 62 63 64 65 66 67 68 69 6A 70 71 72 73
C- 74 75 76 77 78 80 8A 8B 8C 8D 8E 8F 90 9A 9B 9C
D- 9D 9E 9F A0 AA AB AC AE AF B0 B1 B2 B3 B4 B5 B6
Note: I8-sequence bytes C0 ... C4, and E0, and the corresponding UTF-EBCDIC bytes 74 ... 78, and B7, will not be used with the shortest number of bytes in the transformed byte sequences. The corresponding entries are shown italicized in the above table.

Table 3: Byte map from UTF-EBCDIC byte-sequence to I8-sequence

  ß High nibble (hex)                                    Low nibble (hex) Þ (all entries are in hex)
  -0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -A -B -C -D -E -F
0- 00 01 02 03 9C 09 86 7F 97 8D 8E 0B 0C 0D 0E 0F
1- 10 11 12 13 9D 0A 08 87 18 19 92 8F 1C 1D 1E 1F
2- 80 81 82 83 84 85 17 1B 88 89 8A 8B 8C 05 06 07
3- 90 91 16 93 94 95 96 04 98 99 9A 9B 14 15 9E 1A
4- 20 A0 A1 A2 A3 A4 A5 A6 A7 A8 A9 2E 3C 28 2B 7C
5- 26 AA AB AC AD AE AF B0 B1 B2 21 24 2A 29 3B 5E
6- 2D 2F B3 B4 B5 B6 B7 B8 B9 BA BB 2C 25 5F 3E 3F
7- BC BD BE BF C0 C1 C2 C3 C4 60 3A 23 40 27 3D 22
8- C5 61 62 63 64 65 66 67 68 69 C6 C7 C8 C9 CA CB
9- CC 6A 6B 6C 6D 6E 6F 70 71 72 CD CE CF D0 D1 D2
A- D3 7E 73 74 75 76 77 78 79 7A D4 D5 D6 5B D7 D8
B- D9 DA DB DC DD DE DF E0 E1 E2 E3 E4 E5 5D E6 E7
C- 7B 41 42 43 44 45 46 47 48 49 E8 E9 EA EB EC ED
D- 7D 4A 4B 4C 4D 4E 4F 50 51 52 EE EF F0 F1 F2 F3
E- 5C F4 53 54 55 56 57 58 59 5A F5 F6 F7 F8 F9 FA
F- 30 31 32 33 34 35 36 37 38 39 FB FC FD FE FF 9F
Note: I8-sequence bytes C0 ... C4, and E0, and the corresponding UTF-EBCDIC bytes 74 ... 78, and B7, will not be used with the shortest number of bytes in the transformed byte sequences. The corresponding entries are shown italicized in the above table.

3.4 Shadow Flags

In order to assist in finding out if a byte in a UTF-EBCDIC sequence is a leading byte or a trailing byte, and how many bytes in the sequence corresponding to a Unicode character, rather than looking at the byte's bit combination (after converting into its corresponding I8-sequence), or checking the I8-sequence bytes to the known ranges of leading or trailing bytes, a shadow flags table - shown in Table 4 - containing the category of the byte can be utilized. The bytes having a value of '0' in the category table are control characters, '1' are single bytes, '9' are trailing bytes and '2'... '7' indicate the number of bytes in the sequence. Even though Table 1 shows I8-sequences of only up to 5 bytes (to transform up to plane 16), the I8-sequence can contain up to 7 bytes to address all of the UCS-4 space (31-bits) in ISO/IEC 10646 standard (see Table B.2 in Annex B).
Table 4: Shadow flags associated with UTF-EBCDIC bytes
0 = Single-octet control characters 
1 = Single-octet invariant and variant graphic characters from ASCII
2 = Lead octet of a 2-octet string                            3 = Lead octet of a 3-octet string
4 = Lead octet of a 4-octet string                            5 = Lead octet of a 5-octet string
6 = Lead octet of a 6-octet string                            7 = Lead octet of a 7-octet string
9 = A trailing octet of a multi-octet string
(Underscore indicates change from previous draft of this TR)
  ß High nibble (hex)                                    Low nibble (hex) Þ (hex)
  -0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -A -B -C -D -E -F
0- 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1- 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
2- 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
3- 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
4- 1 9 9 9 9 9 9 9 9 9 9 1 1 1 1 1
5- 1 9 9 9 9 9 9 9 9 9 1 1 1 1 1 1
6- 1 1 9 9 9 9 9 9 9 9 9 1 1 1 1 1
7- 9 9 9 9 2 2 2 2 2 1 1 1 1 1 1 1
8- 2 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2
9- 2 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2
A- 2 1 1 1 1 1 1 1 1 1 2 2 2 1 2 2
B- 2 2 2 2 2 2 2 3 3 3 3 3 3 1 3 3
C- 1 1 1 1 1 1 1 1 1 1 3 3 3 3 3 3
D- 1 1 1 1 1 1 1 1 1 1 3 3 4 4 4 4
E- 1 4 1 1 1 1 1 1 1 1 4 4 4 5 5 5
F- 1 1 1 1 1 1 1 1 1 1 5 6 6 7 7 0
Note: I8-sequence bytes C0 ... C4, and E0, and the corresponding UTF-EBCDIC bytes 74 ... 78, and B7, will not be used with the shortest number of bytes in the transformed byte sequences. The corresponding shadow flag values are shown italicized in the above table.

3.5 Signature

The signature character U+FEFF (zero width no-break space) of Unicode transforms into the I8-byte sequence X'F1 BF B7 BF' which maps to X'DD 73 66 73' in UTF-EBCDIC. When this sequence is displayed (erroneously) using different single-byte EBCDIC code pages, it can be visualized as different character strings. In Latin-1 EBCDIC code page 1047 (and coincidentally also in Latin-1 code pages 500 and 37), this byte sequence appears as "ùËÃË"(small letter u with grave, capital letter E with diaeresis, capital letter A with tilde, capital letter E with diaeresis). It can appear differently with other single-byte EBCDIC code pages. As with UTF-8, the byte-swapped ("little-endian") serialized Unicode byte strings must be converted to their "big-endian" equivalents before applying the UTF-EBCDIC transformation.

3.6 Where to Use UTF-EBCDIC?

UTF-EBCDIC is intended to be used inside EBCDIC systems or in closed networks where there is a dependency on EBCDIC hard-coding assumptions. It is not meant to be used for open interchange among heterogeneous platforms using different data encodings. Due to specific requirements for ASCII encoding for line endings in some Internet protocols, UTF-EBCDIC is unsuitable for use over the Internet using such protocols. UTF-8 or UTF-16 forms should be used in open interchange.


Annex A: Intellectual Property Related

Transcript of Letter
regarding Disclosure of IBM Technology - EF-UTF
(Hard copy is on file with the Chair of UTC and the Chair of NCITS/L2)
Transcribed on 1998-07-11


International Business Machines Corporation Route 100

Somers, NY 10589

June 2, 1998

The Chair, Unicode Technical Committee

Subject: Disclosure of IBM Technology - EBCDIC-Friendly UCS Transformation Format (EF-UTF)

The attached document entitled "EBCDIC-Friendly UCS Transformation Format (EF-UTF)" contains IBM technology that has been filed for application for Canadian Patent. However, IBM believes that the technology could be beneficial to the EBCDIC community at large; allowing the community to derive the enormous benefits provided by UCS (ISO/IEC 10646 and Unicode).

This letter is to inform you that IBM is pleased to make the attached documentation, and the associated technology that has been filed for patent, freely available to anyone concerned towards making the transformation format as part of the UCS standards.



Elizabeth G. Nichols
Director of National Language Support
and Information Development



(Note: The term EF-UTF has been changed to UTF-EBCDIC at the suggestion of UTC meting 78 -- V.S. Umamaheswaran)

Annex B: Additional Information

6.1 Controls, Variants, and Invariants in EBCDIC

The positions assigned to the 65 control characters, the 82 invariant graphic characters (including SPACE) and 13 variant graphic characters among the various EBCDIC code pages in use is shown in the following table.
Table B.1: Positions of controls, variants and invariants in EBCDIC

  ß High nibble (hex)                                      Low nibble (hex) Þ
  -0 -1 -2 -3 -4 -5 -6 -7 -8 -9 -A -B -C -D -E -F
0- cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
1- cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
2- cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
3- cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc cc
4- ii                     ii ii ii ii vv
5- ii                   vv vv ii ii ii vv
6- ii ii                   ii ii ii ii ii
7-                   vv ii vv vv ii ii ii
8-   ii ii ii ii ii ii ii ii ii            
9-   ii ii ii ii ii ii ii ii ii            
A-   vv ii ii ii ii ii ii ii ii       vv    
B-                           vv    
C- vv ii ii ii ii ii ii ii ii ii            
D- vv ii ii ii ii ii ii ii ii ii            
E- vv   ii ii ii ii ii ii ii ii            
F- ii ii ii ii ii ii ii ii ii ii           cc
cc = EBCDIC control character positions; ii = EBCDIC invariants from ASCII repertoire; vv = EBCDIC variants from ASCII repertoire

6.2 A comparison of UTF-EBCDIC and UTF-8

UTF-EBCDIC is a byte-mapped version of I8-sequence. The bit patterns of UTF-EBCDIC bytes and UTF-8 therefore are different. A comparison of the bit patterns of UTF-EBCDIC is not so meaningful. However, the I8-sequence and UTF-8 sequence can be compared to understand the salient differences between the two. UTF-8-Mod being derived from UTF-8 retains all of its salient features. A comparative summary of the basic characteristics of I8-sequence and UTF-8 sequence is shown in Table B.2 below. Note that this table shows the entire 31-bit UCS-4 range in the transformation, whereas Table 1 includes only the BMP and up to plane 16 using surrogate pairs.
Table B.2: Comparison of I8-Sequence with UTF-8 Generated Byte Sequence
  I8-sequence UTF-8-sequence Remarks
No. of bytes in transformed sequence Scalar Values
Scalar Values
1 00 to 9F 00 to 7F C0, G0 and C1 in I8-sequence 
C0 and G0 in UTF-8
2 A0 to 3FF 80 to 7FF  
3 400 to 3FFF 800 to FFFF To end of first quarter of BMP in I8-sequence; 
To end of BMP in UTF-8
4 4000 to 3 FFFF 1 0000 to 1F FFFF To end of plane 3 in I8-sequence; 
To end of plane 31 in UTF-8
5 4 0000 to 3F FFFF 20 0000 to 3FF FFFF To end of plane 63 in I8-sequence
6 40 0000 to 3FF FFFF 400 0000 to 7FFF FFFF
  • To end of UCS in UTF-8
7 400 0000 to 7FFF FFFF Not used To end of UCS in I8-sequence
Trailing Bytes 32 values - X'A0' -- X'BF' 
5 v-bits per byte
64 values - X'80' -- X'BF' 
6 v-bits per byte
I8-sequence trailing byte has only five information bits per trailing byte, compared to 6 in UTF-8
Lead Bytes 
2-Byte sequence
C0 -- DF
C0 -- DF
Same in both
3-Byte sequence
E0 -- EF
E0 -- EF
Same in both
4-Byte sequence
F0 -- F7
F0 -- F7
Same in both
5-Byte sequence
F8 -- FB
F8 -- FB
Same in both
6-Byte sequence
FC and FD
FC and FD
Same in both
7-Byte sequence
FE and FF
Not used
Only used in UTF-8-Mod


U+FFFE and U+FFFF are not used for character allocation in any plane of Unicode. U+FEFF (zero width no-break space) is used as a signature for Unicode, for both UCS-2 and UTF-16 forms. U+FFFE may strongly suggest a byte-reversed Unicode string. U+FFFF is used to represent a numeric value that is guaranteed not to be a character, for uses such as the final value at the end of an index. UTF-8 also avoids the use of X'FF' and X'FE' as octets in its sequences. In I8-sequence, however, X'FE' and X'FF' may appear. The following paragraphs expand on which combinations of X'FF' and X'FE' may occur in an I8-sequence or UTF-EBCDIC sequence.
  • X'FE' X'FF', X'FF' X'FE' and X'FF' X'FF' in the I8-sequence
The X'FE' and X'FF' are lead octets of seven-byte I8-sequence (assuming values from all the planes of UCS-4). They will be surrounded (in a properly formed I8-sequence) by a value less than X'C0'. None of the sequences X'FF' X'FF', X'FE' X'FF', and X'FF' X'FE' can appear in a well-formed I8-sequence.
  • X'FE' X'FF', X'FF'X'FE' and X'FF' X'FF' in the UTF-EBCDIC sequence
The I8-sequence to UTF-EBCDIC byte mappings are: X'FE' to X'FD', and X'FF' to X'FE' (see Table 2). The values X'FE' and X'FF' can be generated in a UTF-EBCDIC byte sequence from I8-sequence values by mapping X'FF' to X'FE' and X'9F' to X'FF' from Table 2).
X'FF' is the lead byte of a seven-byte I8 sequence and must be followed by six trailing bytes in the range X'A0' to X'BF', which does not include X'9F'. So the X'FE' X'FF' sequence cannot appear in UTF-EBCDIC.

The X'9F' is assigned to the control character -- Application Program Command (APC) -- in ISO-8 C1. According to ISO/IEC 6429, the APC is followed by a parameter string using bit combinations from 0/8 to 0/13 (X'08' to X'0D') and 2/0 to 7/14 (X'20' to X'7E') and terminated by the control function String Terminator (ST) (coded at X'9C' in C1). Therefore, the sequence X'FF' X'FF', the equivalent of two APC controls without intervening parameters or ST-s, also should not appear in UTF-EBCDIC sequence. None of the valid parameter bit combinations can generate a 7-byte I8 sequence that starts with X'FF'. So the sequence X'FF' X'FE' also cannot appear in a UTF-EBCDIC sequence.

6.4 Normalization to Fixed Width

Dealing with a variable number of bytes may not be possible or desirable in some processing situations (even though proper handling of Unicode text strings will require the ability to correctly deal with combining sequences). Normalization into a form with a fixed number of bits is needed for such cases. It would always be desirable to revert to the original 16-bit form or the corresponding 32-bit form as a normalization to fixed-width data.

However, this would be possible only if processing is tolerant to native Unicode encoding. If transparency to EBCDIC invariance and controls is needed also in the normalized form, then Unicode cannot be directly used for normalization. It can be seen from Table 1 that the last code position in the BMP (U+FFFF) requires four bytes in the I8-sequence and in the corresponding UTF-EBCDIC sequence. A 32-bit integer can be used for normalization of up to four-byte UTF-EBCDIC sequences.

The maximum Unicode scalar value that a four-byte I8-sequence or UTF-EBCDIC sequence can represent is:

<11110111 10111111 10111111 10111111> (X'3FFFF')

corresponding to the end of plane 3 in group 0. Using UTF-16 to represent planes 1 to 16, the surrogate characters in the BMP can be used. By treating the surrogate characters as any other BMP characters, up to plane 16 can be encoded using the 16-bit form, and hence can be contained within the 32-bit normalized form of UTF-EBCDIC. Care has to be taken to correctly process the corresponding UTF-EBCDIC sequence corresponding to the surrogate pairs, similar to dealing with combination sequences. When it is desirable to convert valid surrogate pairs into corresponding Unicode scalar value and then apply UTF-EBCDIC, only up to plane 3 can be contained within the 32-bit normalized value. For all values beyond group 0, plane 3 of UCS, the UTF-EBCDIC will contain more than four octets. The normalization for these cases will need 64 bits (assuming nothing between 32 and 64 bits is practical).

6.5 Mapping of Bytes in Step 2

The control code position mapping used in default Unicode to EBCDIC code page mappings, follow the pairings between ISO/IEC 6429 C0, DEL and C1 sets and EBCDIC controls as defined in IBM Character Data Representation Architecture as default, and customizing to the practice of OS/390 Unix services (MVS Open). These pairings may not suit all EBCDIC environments. A well-known problem is that of mapping EBCDIC New Line to Next Line in C1 of ISO/IEC 6429 versus Line Feed in C0 was mentioned earlier. Similarly it is known that the 13 variant characters are different among the various single byte EBCDIC code pages. The well known impact of this is exemplified by the different code positions of the Square Bracket characters. Even the lowercase a to z is variant in the EBCDIC Katakana code page. A judicious one to one byte reversible map to convert only those code points with category marked as '0'or a '1' may be employed as a step 3. Such a step 3 is not considered to be part of the UTF-EBCDIC transformation defined in this technical report, and is considered as customization to suit individual environments.

Similarly the pairing of I8-sequence bytes and UTF-EBCDIC sequence bytes could be done in multiple ways. The simplest requirement on this byte-pairing is that it should be unique and reversible. The pairing adopted in this version of the UTR is based on the request from Oracle Corporation's representative Mr. Jianping Yang -- to be able to maintain the order of the UTF-EBCDIC multi-byte sequences the same as the order of the corresponding Unicode scalar values.

6.6 Ordering of UTF-EBCDIC Sequences

The mapping of the I8-bytes to UTF-EBCDIC bytes allows the multi-byte UTF-EBCDIC sequences (corresponding to a Unicode character each) to be in the same order as their corresponding Unicode scalar values. The ordering of the trailing bytes and the leading bytes in the UTF-EBCDIC sequence (from Table 4) is:
trailing bytes << Lead bytes of 2-byte-sequence << .. .. .. 
.. .. .. << Lead bytes of 7-byte sequences

The byte values within each set are ordered in increasing order. Note that the UTF-EBCDIC single-bytes do not have this property - either among themselves or between themselves and the bytes of the multi-byte UTF-EBCDIC sequences. The single-bytes are ordered according to their CP1047 order. So doing a "binary comparison" of the text would look like:

for (i = 0; i < n; ++i) {
   byte1 = source1[i];
   byte2 = source2[i];
   if (byte1 == byte2) continue; // fast path
   // check for the single bytes vs multibytes
   if (shadow[byte1] < 2) {
       if (shadow[byte2] > 2) return - 1; // single bytes less than multi
   } else {
       if (shadow[byte2] < 2) return 1; // multibyte greater than single
   // now the shadows are of the same type, so just compare the bytes
   if (byte1 < byte 2) return - 1;
   return 1;
return 0;

The resulting order is a mix of EBCDIC CP1047 order for the single bytes and Unicode order for the multi-byte UTF-EBCDIC characters.

However, if the desired order is to be the same order as Unicode scalar values for all the characters, both the single-byte and the multi-byte characters, the intermediate I8-sequence bytes should be compared. This approach also makes the comparison immune to any local customization of the mapping (see Mapping of Bytes in Step 2) and provides a consistent Unicode value order. The following is a sample for the comparison code.

for (i = 0; i < n; ++i) {
   byte1 = source1[i];
   byte2 = source2[i];
   if (byte1 == byte2) continue; // fast path
   // compare the I8-sequence counterparts
   // take advantage of the ability of I8-sequence bytes being similar
   // to UTF-8 byte to preserve the same order as Unicode scalar values
   // ebtoi8 is the reverse mapping vector from UTF-EBCDIC to I8 bytes
   if (ebtoi8[byte1] < ebtoi8[byte2]) return - 1;
   return 1;
return 0;

If the desire is to preserve the EBCDIC order for the single-bytes (the ASCII repertoire) or the traditional order of the multi-byte sequences (such as for EBCDIC-Japanese, EBCDIC-Cyrillic, EBCDIC-Arabic etc.) localization resources such as a weight look up table in locales should be employed.

6 Acknowledgments

The UTF-EBCDIC transformation was originally created and developed in the National Language Technical Centre in IBM Toronto Laboratory by Messrs. Baldev Soor, Alexis Cheng, Rick Pond, Ibrahim Meru and V.S. (Uma) Umamaheswaran. The original version has been modified based on review feedback on the previous versions of this Unicode Technical Report.


7.2 revision: text centering problem fixed in HTML for tables

The previous revision of this technical report corrected an error in the section 3.5 Signature, to read as "... this byte sequence appears as "ùËÃË" (small letter u with grave, capital letter E with diaeresis, capital letter A with tilde, capital letter E with diaeresis )".  (Thanks to Robert Rosenberg -- Bob.Rosenberg@digitscorp.com for reporting this error.)

Copyright © 1999-2001 Unicode, Inc.. All Rights Reserved. The Unicode Consortium makes no expressed or implied warranty of any kind, and assumes no liability for errors or omissions. No liability is assumed for incidental and consequential damages in connection with or arising out of the use of the information or programs contained or accompanying this technical report.