Signal Codes: Convolutional Lattice Codes

Abstract

The coded modulation scheme proposed in this paper has a simple construction: an integer sequence, representing the information, is convolved with a fixed, continuous-valued, finite impulse response (FIR) filter to generate the codeword - a lattice point. Due to power constraints, the code construction includes a shaping mechanism inspired by precoding techniques such as the Tomlinson-Harashima filter. We naturally term these codes “convolutional lattice codes” or alternatively “signal codes” due to the signal processing interpretation of the code construction. Surprisingly, properly chosen short FIR filters can generate good codes with large minimal distance. Decoding can be done efficiently by sequential decoding or for better performance by bidirectional sequential decoding. Error analysis and simulation results indicate that for the additive white Gaussian noise (AWGN) channel, convolutional lattice codes with computationally reasonable decoders can achieve low error rate close to the channel capacity.