Fano decoder uploaded to ftp.ucsd.edu

[Phil Karn <karn@unix.ka9q.ampr.org>]

My Fano (sequential) decoder package is now available as

ftp://ftp.ucsd.edu/hamradio/packet/tcpip/incoming/fano.tar.gz
ftp://ftp.ucsd.edu/hamradio/packet/tcpip/incoming/fano.zip

Three different rate 1/2 constraint length 32 convolutional codes are
supported. The decoder operates on 8-bit soft decisions. A routine is
provided to build the decoder metric table assuming gaussian noise at
some estimated Es/N0. By substituting different metric tables, the
decoder can be adapted to run on other kinds of channels such as the
binary symmetric channel (BSC), i.e., with hard decoding.

Using GCC 2.5.8 on a 66 Mhz 486DX2 in protected (32-bit) mode, the
decoder runs at about 300,000 branches per second.  This translates
directly to 300,000 bits/sec on a "clean" channel. Noisy packets
decode more slowly, as is characteristic for a sequential decoder. The
performance curve is quite steep, as expected for a strong, long
constraint length code.  Simulation results show that at an Eb/N0 of
4dB, only about 1.6% of the packets decode at half speed or
slower. Below about 2.5-3dB, things start to fall apart as expected.

Although the encoder/decoder itself is in portable ANSI C, the test
driver assumes a UNIX (specifically BSDI1.1) environment with
curses/termcap. Also, the synthetic gaussian noise generator uses
log() and sqrt() quite heavily. Some 386/486 math libraries don't use
the native FPU instructions for these functions, so they can be quite
slow.

My Viterbi (maximum likelihood) decoder is next.

--Phil