\begin{abstract} % COPY FROM MAIN
In ITW 2003 we presented a new class of codes called ``Signal
codes''. The codes are revisited here, with better understanding and
after much of their properties were studied in more detail. We also
realized that a better name for the codes would be ``Convolutional
Lattice Codes''.

Convolutional lattice codes are lattice codes designed directly in
the Euclidean space without any finite alphabet code construction.
As the name may suggest, a convolutional lattice codeword (or
lattice point) is generated by convolving an integer sequence,
representing the information sequence, with a fixed,
continuous-valued, finite impulse response (FIR) filter pattern. The
FIR length is small yet, as shown in the paper, by proper choice it
generates a lattice with surprisingly large minimal distance. It is
due to the signal processing interpretation of the code construction
tha we originally called this class signal codes. For practical
usage the code construction includes a shaping mechanism inspired by
pre-coding techniques such as the Tomlinson-Harashima filter.
Convolutional lattice codes can be decoded efficiently by sequential
decoding or for better performance by bi-directional sequential
decoding. Error analysis and simulation results indicate that
convolutional lattice codes with computationally reasonable decoders
can achieve low error rate within 1dB to channel capacity.

\end{abstract}