2 * Provides routines for encoding and decoding the extended Golay
5 * This implementation will detect up to 4 errors in a codeword (without
6 * being able to correct them); it will correct up to 3 errors.
8 * Wireshark - Network traffic analyzer
9 * By Gerald Combs <gerald@wireshark.org>
10 * Copyright 1998 Gerald Combs
12 * SPDX-License-Identifier: GPL-2.0-or-later
21 These entries are formed from the matrix specified in H.223/B.3.2.1.3;
22 it's first transposed so we have:
24 [P1 ] [111110010010] [MC1 ]
25 [P2 ] [011111001001] [MC2 ]
26 [P3 ] [110001110110] [MC3 ]
27 [P4 ] [011000111011] [MC4 ]
28 [P5 ] [110010001111] [MPL1]
29 [P6 ] = [100111010101] [MPL2]
30 [P7 ] [101101111000] [MPL3]
31 [P8 ] [010110111100] [MPL4]
32 [P9 ] [001011011110] [MPL5]
33 [P10] [000101101111] [MPL6]
34 [P11] [111100100101] [MPL7]
35 [P12] [101011100011] [MPL8]
37 So according to the equation, P1 = MC1+MC2+MC3+MC4+MPL1+MPL4+MPL7
39 Looking down the first column, we see that if MC1 is set, we toggle bits
40 1,3,5,6,7,11,12 of the parity: in binary, 110001110101 = 0xE3A
42 Similarly, to calculate the inverse, we read across the top of the table and
43 see that P1 is affected by bits MC1,MC2,MC3,MC4,MPL1,MPL4,MPL7: in binary,
46 I've seen cunning implementations of this which only use one table. That
47 technique doesn't seem to work with these numbers though.
50 static const unsigned golay_encode_matrix
[12] = {
65 static const unsigned golay_decode_matrix
[12] = {
82 /* Function to compute the Hamming weight of a 12-bit integer */
83 static unsigned weight12(unsigned vector
)
93 /* returns the golay coding of the given 12-bit word */
94 static unsigned golay_coding(unsigned w
)
99 for( i
= 0; i
<12; i
++ ) {
101 out
^= golay_encode_matrix
[i
];
106 /* encodes a 12-bit word to a 24-bit codeword */
107 uint32_t golay_encode(unsigned w
)
109 return ((uint32_t)w
) | ((uint32_t)golay_coding(w
))<<12;
114 /* returns the golay coding of the given 12-bit word */
115 static unsigned golay_decoding(unsigned w
)
120 for( i
= 0; i
<12; i
++ ) {
122 out
^= golay_decode_matrix
[i
];
128 /* return a mask showing the bits which are in error in a received
129 * 24-bit codeword, or -1 if 4 errors were detected.
131 int32_t golay_errors(uint32_t codeword
)
133 unsigned received_data
, received_parity
;
136 unsigned inv_syndrome
= 0;
138 received_parity
= (unsigned)(codeword
>>12);
139 received_data
= (unsigned)codeword
& 0xfff;
141 /* We use the C notation ^ for XOR to represent addition modulo 2.
143 * Model the received codeword (r) as the transmitted codeword (u)
144 * plus an error vector (e).
148 * Then we calculate a syndrome (s):
150 * s = r * H, where H = [ P ], where I12 is the identity matrix
153 * (In other words, we calculate the parity check for the received
154 * data bits, and add them to the received parity bits)
157 syndrome
= received_parity
^ (golay_coding(received_data
));
158 w
= weight12(syndrome
);
161 * The properties of the golay code are such that the Hamming distance (ie,
162 * the minimum distance between codewords) is 8; that means that one bit of
163 * error in the data bits will cause 7 errors in the parity bits.
165 * In particular, if we find 3 or fewer errors in the parity bits, either:
166 * - there are no errors in the data bits, or
167 * - there are at least 5 errors in the data bits
168 * we hope for the former (we don't profess to deal with the
172 return ((int32_t) syndrome
)<<12;
175 /* the next thing to try is one error in the data bits.
176 * we try each bit in turn and see if an error in that bit would have given
177 * us anything like the parity bits we got. At this point, we tolerate two
178 * errors in the parity bits, but three or more errors would give a total
179 * error weight of 4 or more, which means it's actually uncorrectable or
180 * closer to another codeword. */
182 for( i
= 0; i
<12; i
++ ) {
183 unsigned error
= 1<<i
;
184 unsigned coding_error
= golay_encode_matrix
[i
];
185 if( weight12(syndrome
^coding_error
) <= 2 ) {
186 return (int32_t)((((uint32_t)(syndrome
^coding_error
))<<12) | (uint32_t)error
) ;
190 /* okay then, let's see whether the parity bits are error free, and all the
191 * errors are in the data bits. model this as follows:
193 * [r | pr] = [u | pu] + [e | 0]
196 * pu = H * u => u = H' * pu = H' * pr , where H' is inverse of H
198 * we already have s = H*r + pr, so pr = s - H*r = s ^ H*r
200 * = (H' * ( s ^ H*r )) ^ r
204 * Once again, we accept up to three error bits...
207 inv_syndrome
= golay_decoding(syndrome
);
208 w
= weight12(inv_syndrome
);
210 return (int32_t)inv_syndrome
;
213 /* Final shot: try with 2 errors in the data bits, and 1 in the parity
214 * bits; as before we try each of the bits in the parity in turn */
215 for( i
= 0; i
<12; i
++ ) {
216 unsigned error
= 1<<i
;
217 unsigned coding_error
= golay_decode_matrix
[i
];
218 if( weight12(inv_syndrome
^coding_error
) <= 2 ) {
219 uint32_t error_word
= ((uint32_t)(inv_syndrome
^coding_error
)) | ((uint32_t)error
)<<12;
220 return (int32_t)error_word
;
224 /* uncorrectable error */
230 /* decode a received codeword. Up to 3 errors are corrected for; 4
231 errors are detected as uncorrectable (return -1); 5 or more errors
232 cause an incorrect correction.
234 int golay_decode(uint32_t w
)
236 unsigned data
= (unsigned)w
& 0xfff;
237 int32_t errors
= golay_errors(w
);
238 unsigned data_errors
;
242 data_errors
= (unsigned)errors
& 0xfff;
243 return (int)(data
^ data_errors
);
252 * indent-tabs-mode: nil
255 * ex: set shiftwidth=4 tabstop=8 expandtab:
256 * :indentSize=4:tabSize=8:noTabs=true: