Magnetic tape having a permanent pattern of a detectable magnetic quantity is known from GB A 1331604, which is incorporated herein by reference, and GB A 2309568. Such tape is available from Thorn Secure Science Limited under the UK registered trade mark"WATERMARK"tape, and is used as an identification or authentication means on articles of value such as bank cards or credit cards.

At the present state of the art the maximum data packing densities used in volume manufacture of this tape is 39 bits per inch (1.53 bits per mm) using industry standard F2F coding. The data format used in Watermark tape requires marker portions, known as"sentinels", with a binary digit string between successive sentinels. Each sentinel comprises 10 bits, with the data between sentinels typically comprising 60 bits (12 characters each comprising 5 bits). The data between sentinels may be incrementing or non-incrementing. If the data is non-incrementing, then the data stored on the tape is periodic, having a period of 70 binary digits. At a data packing density of 1.3 bits per mm, this implies that over 53 mm of tape must be applied to an article of value in order to identify it correctly, and if part of the tape gets damaged, or the read head bounces at the edge of the label then there is no redundancy, and so the data cannot be read correctly. Thus Watermark tape is not particularly convenient for use on small articles or documents of value.

One way to reduce the pitch of data on such tape would be to decrease the length of the sentinel. In GB-A-2 021 835 a 5 bit sentinel and eight 5 bit characters was suggested. However, such a scheme has been found not to be practical because of significant interaction between the sentinel and the data, resulting in unacceptable read errors. Another drawback with this scheme is that the data can only be read in the forward direction and not in the reverse direction. To overcome this problem a sentinel having 10 bits has been employed for many years. This normally includes a long string of binary 1's, such as 00111 11101 or 1011111000. When combined with 5 bit data characters which use 4

bits to represent decimal characters 0 to 9 and the 5th bit for"odd"parity this makes sure that the sentinel cannot occur in the twelve 5 bit data characters.

According to a first aspect of the invention, there is provided a method of labelling an article as claimed in claims 1-15.

According to a second aspect of the invention, there is provided an identification means for labelling an article as claimed in claims 19-21.

According to a third aspect of the invention there is provided a method of authenticating an article as claimed in claims 16 and 17.

Embodiments of the invention will now be described, by way of example only, with reference to the accompanying diagrammatic drawings, in which Figure 1 shows an article of value labelled according to the invention, Figure 2 shows a block diagram of a method of identification according to the invention, and Figure 3 shows a block diagram of a method of authentication according to the invention.

In Figure 1, an identification means (1) labelling an article (2) is shown.

The identification means comprises a data-carrying layer (in the present example a magnetic tape laminated to the article of value) having a permanent pattern of a detectable magnetic property. The data being carried by the layer comprising a sequence of binary digits, the sequence having a period of less than 64 binary digits, and having a length greater than or equal to one period. In the example shown in Figure 1 the data comprises marker portions or sentinels (3) with four other portions or characters (4,5,6,7) located between successive marker portions. Each sentinel comprises the binary digit string, in the present example 1010000. Other binary digit strings can be used as an alternative, for example 0010100, or 0101111, etc., but 1010000 has been chosen because it is one of the strings which is least likely to be confused with other binary digit (bit) strings when read.

Each of the other characters comprises a 7 bit number. Although in the present example there are four characters, strings having 2 or more characters may be used if desired. The four character string (4,5,6,7), comprising 28 bits, is chosen to represent a given class of articles, for example a document of value such as a share certificate or gift voucher issued by a company. In the present example these four characters give 2,214,841 possible different"first"character strings (i. e."key differs"), as will be explained below. The more characters in the

initial character string, the greater the number of key differs, but the longer the initial character string the longer the final magnetic record.

The"first"character string in the present example comprises four characters. Each of the four characters is represented by a 7 bit number which is found from a predetermined look-up table such that the marker character (1010000) does not appear if the sequence of bits is read in either direction, except at a marker. The marker character is preferably not palindromic so that when the data is read in the forward or the reverse direction a different pattern is seen, such that the reading apparatus can work out which way up the character string has been presented to it in use. Each of the characters, being 7 bits in length, can take 128 potential values. The value 1010000 and its reverse 0000101 must be excluded as these would be confused with a marker. The number 0101xxx must also be excluded as it could combine with xxxx000 in an adjacent character to produce a"phantom"marker. Numbers containing long runs of zeros must also be avoided, as this can confuse F2F decoding and/or the ability of a reader to differentiate between the method of the present invention and that of normal"WATERMARK"tape. The numbers which must be excluded at characters 2,3 and 4 are shown in Table 1, whilst additional numbers which must be excluded at the first character position are shown in Table 2. The actual complete look up table used in the present embodiment is shown in Table 3. D Exclusion D Exclusion D Exclusion 0 xx (000-0000000) 26 001 (1010-000) xx 65 xx (10-10000) 01 1 xx (0000-000000) 1 32 xx (1-010000) 0 66 xx (10-10000) 10 2 x (00000-00000) 10 33 xx (1-010000) 1 67 xx (10-10000) 11 3 x (00000-00000) 11 40 xx (000-0101) 000 74 100 (1010-000) xx 4 xx (101-0000) 100 41 xx (000-0101) 001 80 (1010000) 5 (0000101) 42 xx (000-0101) 010 84 10 (10100-00) xx 6 xx (101-0000) 110 43 xx (000-0101) 011 90 101 (1010-000) xx 7 xx (101-0000) 111 44 xx (000-0101) 100 96 110 (0000-101) xx 10 000 (1010-000) xx 45 xx (000-0101) 101 97 11 (00001-01) xx 11 xx (0-000101) 1 46 xx (000-0101) 110 104 1 (101000-0) xx 16 001 (0000-101) xx 47 xx (000-0101) 111 106 110 (1010-000) xx 20 00 (10100-00) xx 48 011 (0000-101) xx 112 111 (0000-101) xx 21 001 (1010-000) xx 52 01 (10100-00) xx 116 11 (10100-00) xx 22 xx (00-00101) 10 58 011 (1010-000) 122 111 (1010-000) xx 23 xx (00-00101) lu64100 (0000-101) xx Table 1-Codewords excluded at characters 2,3 & 4 D = decimal value of the excluded codeword D Exclusion D Exclusion D Exclusion 87101(0000-101)011193101(0000-101)110181101(0000-101)0001 88101(0000-101)100094101(0000-010)111082101(0000-101)0010 89101(0000-101)100195101(0000-101)111183101(0000-101)0011 85 101 (0000-101) 0101 91 1ol (oooo-101) 1011 92101(0000-101)110086101(0000-101)0110 Table 2-Additional codewords excluded at character 1

Table 3-Binary Table n = input, un-encode value (b, D, H) = output, encoded value in binary, decimal and hex respectively. n b D H n b D H n b D H n b D H 0000000 0 0 0100000 32 20 1000000 64 40 1100000 96 60 0000001 1 1 0100001 33 21 1000001 65 41 1100001 97 61 00000102216 01000103422 1000010664246110001098 62 0000011 3 3 17 0100011 35 23 1000011 67 43 47 1100011 99 63 0000100 4 4 18 0100100 36 24 35 1000100 68 44 48 1100100 100 64 5190100101372536100010169454911001011016500001015 6200100110382637100011070465011001101026600001106 7210100111392738100011171475111001111036700001117 0 80101000402839100100072481101000104688 990101001412940100100173495211010011056910001001 A0101010422A1001010744A1101010106000101010 6A 0001011 11 B 0101011 43 2B 41 1001011 75 4B 53 1101011 107 6B 2 C0101100442C421001100764C5411011001086C12 3 D0101101452D431001101774D5511011011096D13 4 E0101110462E441001110784E56110111011014 6E 5 0001111 15 F 0101111 47 2F 45 1001111 79 4F 57 1101111 111 6F 0010000 16 10 0110000 48 30 1010000 80 50 1110000 112 70 6 11220110001493171101000181515811100011137117 7 0010010 18 12 23 0110010 50 32 72 1010010 82 52 59 1110010 114 72 8 0010011 19 13 24 0110011 51 33 73 1010011 83 53 60 1110011 115 73 140110100523410101008454111010011674001010020 152501101015335741010101855561111010111775001010121 162601101105436751010110865662111011011876001011022 0010111 23 17 27 0110111 55 37 76 1010111 87 57 63 1110111 119 77 9 0011000 24 18 28 0111000 56 38 77 1011000 88 58 64 1111000 120 78 10 19290111001573978101100189596511110011217925 1A0111010583A1011010905A11110101227A001101026 11 1B300111011593B791011011915B6611110111237B27 12 0011100 28 1C 31 0111100 60 3C 80 1011100 92 5C 67 1111100 124 7C 13 1D320111101613D811011101935D6811111011257D29 14 0011110 30 1E 33 0111110 62 3E 82 1011110 94 5E 69 1111110 126 7E 15 1F340111111633F831011111955F7011111111277F31

Binary numbers shown in bold in Table 3 are excluded for all character places. Binary numbers shown in italics in Table 3 are excluded from character position 1. It will be noted that the numbers n in table 3 do not form a monotonically increasing sequence. This is permissible as it is a look up table.

Thus any value could appear at any place in the look up table as long as the place is the same for both the encode and decode operations.

Only 84 of the possible 128 possible values available with a 7 bit string are used in the encode direction. The 44 excluded values become parity errors if found in the reverse (decode) application of the look-up table shown as Table 3.

To avoid generating phantom markers, values of n above 70 must be avoided in the character 1 position. This is done by keeping the maximum possible number used to 2,214,840 or less. The total number of possible combinations or"key differs"is (71 x 84 x 84 x 84)/19, which rounds down to 2,214,841.

Having now described in some detail what format the data carried by the identification means takes in the preferred embodiment, the method of providing the data will be explained.

First, the person desiring to label the article of value must choose a characters string from the 2,214,841 combinations possible when using 4 characters from Table 3. In the present example this will be a decimal number, which we shall call N, which lies between zero and 2,214,841. This step is shown as block 20 in Figure 2.

The steps in the method are then as follows, with steps a to h being represented in Figure 2 by blocks 21 to 28 respectively:- a). Check that N lies between zero and 2,214,841. b). Multiply N by a numerical factor, in the present example the integer 19. This factor is known as the longitudinal redundancy check factor, or LRC. c). Convert the result (i. e. 19N) to the number base of the look up table (84 in the present example) giving a 4 digit number in base 84, d). Take this 4 digit number and using the look up table of Table 3 encode the digits sequentially into characters 1-4, each character comprising a 7 bit number, e). Add the marker 101000 to one end of the data, f). Send the resulting 35 bit sequence through a circuit of known form which applies F2F encoding to the binary data, g). Embody the code as a permanent structure of a detectable magnetic property in an identification means as previously described,

h). Repeat steps f) and g) to provide a periodic sequence of binary digits in the identification means.

A specific example of the above method will now be provided. First, the identification number 1234567 is chosen (being a decimal number between 0 and 2214841). This number is then multiplied by the integer 19 to give 23456773. To convert to a base 84 number, this is divided by 84, leaving the number 279247 and the remainder 25. Dividing 279247 by 84 gives 3324, remainder 31. Dividing 3324 by 84 gives 39 remainder 48. The final remainder (which in the present example should be less than 71) is 39. The look-up table of Table 3 is then used to convert the remainders to bit patterns as follows:- 39 = 1001000 48 = 1100100 31 = 0111100 25 = 0110101.

The sentinel 1010000 is placed at one end of the bit pattern, and the bit pattern is recorded onto an identification means, such as for example magnetic tape, the final bit pattern at the end of step (e) above reading:- 1010000 1001000 1100100 0111100 0110101.

This completes the encoding part of the labelling or identification method.

The identification means is then applied to or incorporated in an article of value such as a document of value. This may be achieved by simply laminating the identification means, in the present example magnetic tape, to the article, or by a transfer method in which an adhesive layer is applied to the surface of the tape, the article affixed to this adhesive layer, and the substrate tape is removed leaving the upper layer of what was originally the tape bonded to the surface of the article of value.

In order to authenticate the article of value, the identification means applied to the article as described above must be read and decoded. This may be performed according to the following method, which is represented in Figure 3 as a flow diagram in which blocks 30 to 41 correspond to the following steps a-1 respectively:- a) Move a reader or scanning means past the identification means, or vice versa, and apply (if desired) known media authentication techniques such as those disclosed in GB 2035659 or EP 93916121, b) Apply a F2F decoding algorithm to the scanned data,

c) Test for the presence of leading or trailing zeros, or other features indicating the presence of another data format such as the usual"WATERMARK"data format, if found decode as that format. If not found, continue as described below, d) Find 35 contiguous bits which are free from read or F2F decode errors, using known error detection techniques, and/or authentication techniques, e) Check for errors in the 35 contiguous bits by storing them in a shift register such as, for example, a ring or rotating barrel shifter, and rotate them by 35 bits (in steps of 1 bit) in both directions, checking that in total in both directions the 35 bits contain exactly one marker or sentinel-any other result signifying the presence of an error, f) Arrange the digits by rotation and, if necessary, by inverting the order of the digits, such that the marker appears at the head of the character string and character 4 appears at the tail, g) Remove the bits corresponding to the marker, leaving 28 bits corresponding to data consisting of 4 characters, each character comprising 7 bits, h) Apply the look-up table of Table 3 in reverse to each of the 4 characters to give 4 numbers in base 84. If the look-up table does not contain an entry matching one or more of the numbers this indicates the presence of an error, i) Convert the number to a number base which matches the desired output, such as, for example, base 10 in the present example, j) Divide the number obtained in step i above by the LRC (19 in the present example), and reject as erroneous if the remainder is not zero. k) Compare the result with N, the number characteristic of a genuine article of value. If they are the same, then the article is genuine, 1) Signal the result of this comparison. The result signalled may comprise for example a report of the identification number, confirmation that the extracted identification number matches a predetermined number, or a report that authentication failed due to a media authentication error or due to a data format error.

In the above encode and decode sequence, it is not convenient to use a binary checksum as an error protection feature. The use of multiplication and division by the LRC factor has similar error protection properties to that of a checksum. A random error on, for example, a modulo 19 multiplication and division error protection scheme has only a 1 in 19 chance of producing a codeword which passes the test.

The reasons for choosing the LRC factor to be 19 are as follows. The LRC factor chosen should share as few common factors as possible with the number base of the look-up table (84 in the present example). Being a prime number a little above 16 makes 19 a good candidate. It would be possible to use LRC factors of 17 or 13, these would provide slightly worse error protection but numbers which are slightly easier to work with. It would also in theory be possible to use the numbers 16,15, or 14-however, this would make for rather more interaction between the LRC factor and the number base of the look up table, and hence rather more exclusions from the look-up table.

In addition to the 19 multiply/divide LRC factor, substantial error detection is achieved in the present method by use of the character look up table and by the F2F coding.

The factor of 19 provides a similar level of protection to that used in most large magnetic stripe schemes. If a new application required the error protection to be increased fifty-fold then it will be obvious to those skilled in the art that the LRC factor in the above example can be increased to (say) 1000, which will then reduce the available"key differs"to 42,081.

Although in the above example 4 characters each comprising seven bits are used, more characters or fewer characters may be employed in the scheme if desired. Each character may comprise six or more digits, preferably seven or more digits. The digits may be expressed in a predetermined number base which need not be equal to 2 as in the above example.

The use of the method described above enables an identification means or label having a length of 27 mm to be produced having 2,214,841 possible"key differs". In comparison, a standard"WATERMARK"tape record with similar key differs has at least a 45 mm length, assuming that both are encoded at 1.53 bits per mm.

It is anticipated that this invention will find wide application in areas such as:- a) Provision of sets of identical tokens, such as event tickets, gift tokens, share certificates, promissory notes, bank notes, and signatures to authenticate that a given organisation or person has stamped or sealed the document. b) Cost critical applications: an important factor in cost control is minimisation of surface area. This is also desirable for aesthetic considerations.

As a further aspect of the invention, the above method may be applied in an analogous way to the case where the successive character strings having a

sentinel or marker therebetween are not identical but increment or decrement. In this case the binary digit sequence will not in general be periodic. In this case the digit sequence will be applied to the article, for example a document, in a registered position so that the marker is in a known position on the document.

This slightly simplifies the decoding described above, which no longer requires the rotating barrel shifter. This registration process is known in the industry, although in this particular embodiment the registration can be sensed by magnetic means. More conventional optical sensing can be used at registration if suitable optical features are included during the manufacture of the magnetic foil. In general, application needing incrementing numbers are likely to need more than four of the 7 bit characters. For example, if the embodiment shown in the example is extended from 35 to 63 bits it generates 1.1 x 10""key differs", whereas"WATERMARK"would require 90 bits to encode this. Thus the present invention provides the advantage over the prior art that more"key differs"can be provided using a shorter length of tape.