README
1COPYRIGHT
2
3This package is copyright 2006 by Phil Karn, KA9Q. It may be used
4under the terms of the GNU Lesser General Public License (LGPL). See
5the file "lesser.txt" in this package for license details.
6
7INTRODUCTION
8
9This package provides a set of functions that implement several
10popular forward error correction (FEC) algorithms and several low-level routines
11useful in modems implemented with digital signal processing (DSP).
12
13The following routines are provided:
14
151. Viterbi decoders for the following convolutional codes:
16
17r=1/2 k=7 ("Voyager" code, now a widely used industry standard)
18r=1/2 k=9 (Used on the IS-95 CDMA forward link)
19r=1/6 k=15 ("Cassini" code, used by several NASA/JPL deep space missions)
20
212. Reed-Solomon encoders and decoders for any user-specified code.
22
233. Optimized encoder and decoder for the CCSDS-standard (255,223)
24Reed-Solomon code, with and without the CCSDS-standard "dual basis"
25symbol representation.
26
274. Compute dot product between a 16-bit buffer and a set of 16-bit
28coefficients. This is the basic DSP primitive for digital filtering
29and correlation.
30
314. Compute sum of squares of a buffer of 16-bit signed integers. This is
32useful in DSP for finding the total energy in a signal.
33
345. Find peak value in a buffer of 16-bit signed integers, useful for
35scaling a signal to prevent overflow.
36
37SIMD SUPPORT
38
39This package automatically makes use of various SIMD (Single
40Instruction stream, Multiple Data stream) instruction sets, when
41available: MMX, SSE and SSE2 on the IA-32 (Intel) architecture, and
42Altivec on the PowerPC G4 and G5 used by Power Macintoshes.
43
44"Altivec" is a Motorola trademark; Apple calls it "Velocity Engine",
45and IBM calls it "VMX". Altivec is roughly comparable to SSE2 on the
46IA-32.
47
48Many of the SIMD versions run more than an order of
49magnitude faster than their portable C versions. The available SIMD
50instruction sets, if any, are determined at run time and the proper
51version of each routine is automatically selected. If no SIMD
52instructions are available, the portable C version is invoked by
53default. On targets other than IA-32 and PPC, only the portable C
54version is built.
55
56The SIMD-assisted versions generally produce the same results as the C
57versions, with a few minor exceptions. The Viterbi decoders in C have
58a very slightly greater Eb/No performance due to their use of 32-bit
59path metrics. On the other hand, the SIMD versions use the
60"saturating" arithmetic available in these instructions to avoid the
61integer wraparounds that can occur in C when argument ranges are not
62properly constrained. This applies primarily to the "dotprod" (dot
63product) function.
64
65The MMX (MultiMedia eXtensions) instruction set was introduced on
66later Pentium CPUs; it is also implemented on the Pentium II and most
67AMD CPUs starting with the K6. SSE (SIMD Streaming Extensions) was
68introduced in the Pentium III; AMD calls it "3D Now! Professional".
69Intel introduced SSE2 on the Pentium 4, and it has been picked up by
70later AMD CPUs. SSE support implies MMX support, while SSE2 support
71implies both SSE and MMX support.
72
73The latest IA-32 SIMD instruction set, SSE3 (also known as "Prescott
74New Instructions") was introduced in early 2004 with the latest
75("Prescott") revision of the Pentium 4. Relatively little was
76introduced with SSE3, and this library currently makes no use of it.
77
78See the various manual pages for details on how to use the library
79routines.
80
81Copyright 2006, Phil Karn, KA9Q
82karn@ka9q.net
83http://www.ka9q.net/
84
85This software may be used under the terms of the GNU Lesser General
86Public License (LGPL); see the file lesser.txt for details.
87
88Revision history:
89Version 1.0 released 29 May 2001
90
91Version 2.0 released 3 Dec 2001:
92Restructured to add support for shared libraries.
93
94Version 2.0.1 released 8 Dec 2001:
95Includes autoconf/configure script
96
97Version 2.0.2 released 4 Feb 2002:
98Add SIMD version override options
99Test for lack of SSE2 mnemonic support in 'as'
100Build only selected version
101
102Version 2.0.3 released 6 Feb 2002:
103Fix to parityb function in parity.h
104
105feclib version 1.0 released November 2003
106Merged SIMD-Viterbi, RS and DSP libraries
107Changed SIMD Viterbi decoder to detect SSE2/SSE/MMX at runtime rather than build time
108
109feclib version 2.0 (unreleased) Mar 2004
110General speedups and cleanups
111Switch from 4 to 8-bit input symbols on all Viterbi decoders
112Support for Altivec on PowerPC
113Support for k=15 r=1/6 Cassini/Mars Pathfinder/Mars Exploration Rover/STEREO code
114Changed license to GNU Lesser General Public License (LGPL)
115
116feclib version 2.1 June 5 2006
117Added error checking, fixed alignment bug in SSE2 versions of Viterbi decoders causing segfaults
118
119feclib version 2.1.1 June 6 2006
120Fix test/benchmark time measurement on Linux
121