Semidefinite Relaxation codes for the discrete Integer Least Squares problem

 
 

Algorithms used in SDR Detector

SDR Detector is based on a dual scaling interior point method (implemented in C), with several key modifications to allow warm start, early termination and efficient stepsize selection. The warm start is made possible by truncated version of the popular Sphere Decoder. The resulting interior point algorithm has a worst case polynomial complexity. The following resources provide the foundation for the implementation of SDR Detector:

  • Semidefinite Programming (SDP) Detector:
  • Sphere Decoder (SD):
    • The foundation of SD is an algorithm for computing the shortest vector in lattices, first published in U. Fincke and M. Pohst, "Improved methods for calculating vectors of short length in a lattice, including a complexity analysis,'' Mathematics of Computation, vol. 44, pp. 463-471, 1985.
    • SD was applied to maximum-likelihood detection problem in E. Viterbo and J. Boutros, "A universal lattice code decoder for fading channels,'' IEEE Transactions on Information Theory, pp. 1639-1642, 1999.

Algorithms used in PSK Detector

The core of PSK Detector is the coordinate descent method implemented over a homotopy of the relaxed feasible set. A warm start procedure and a dimension reduction technique allow to make efficient use of symbol reliabilities to deliver near optimal error probability with highly optimized running time. The following resources provide the foundation for the implementation of SDR Detector:

  • Phase-Shift-Keying (PSK) Detector:
    • PSK Detector was introduced in Z.-Q. Luo, X. Luo, and M. Kisialiou, "An efficient quasi-maximum-likelihood decoder for PSK signals," Proc. of ICASSP '03, vol. 6, pp. VI 561 - VI 564, Hong Kong, 2003.