Trellis and convolutional precoding for transmitter-based interference presubtraction

Wei Yu, David P. Varodayan, John M. Cioffi

Research output: Contribution to journalArticlepeer-review


This paper studies the combination of practical trellis and convolution codes with Tomlinson-Harashima precoding (THP) for the presubtraction of multiuser interference that is known at the transmitter but not known at the receiver. It is well known that a straightforward application of THP suffers power, modulo, and shaping losses. This paper proposes generalizations of THP that recover some of these losses. At a high signal-to-noise ratio (SNR), the precoding loss is dominated by the shaping loss, which is about 1.53 dB. To recover shaping loss, a trellis-shaping technique is developed that takes into account the knowledge of a noncausal interfering sequence, rather than just the instantaneous interference. At rates of 2 and 3 bits per transmission, trellis shaping is shown to be able to recover almost all of the 1.53-dB shaping loss. At a low SNR, the precoding loss is dominated by power and modulo losses, which can be as large as 3-4 dB. To recover these losses, a technique that incorporates partial interference presubtraction (PIP) within convolutional decoding is developed. At rates of 0.5 and 0.25 bits per transmission, PIP is able to recover 1-1.5 dB of the power loss. For intermediate SNR channels, a combination of the two schemes is shown to recover both power and shaping losses.

Original languageEnglish (US)
Pages (from-to)1220-1230
Number of pages11
JournalIEEE Transactions on Communications
Issue number7
StatePublished - Jul 2005
Externally publishedYes


  • Broadcast channel
  • Channels with noncausal side information
  • Convolutional codes
  • Dirty-paper precoding
  • Shaping codes
  • Tomlinson-Harashima precoding (THP)
  • Trellis codes
  • Trellis precoding (TP)
  • Trellis shaping (TS)
  • Vectored digital subscriber line (DSL)

ASJC Scopus subject areas

  • Electrical and Electronic Engineering


Dive into the research topics of 'Trellis and convolutional precoding for transmitter-based interference presubtraction'. Together they form a unique fingerprint.

Cite this