A Partial Compress-and-Forward Strategy for Relay-assisted Wireless Networks Based on Rateless Coding

TL;DR

The paper addresses relay-assisted communication in a diamond network with two noisy relays and no direct source-destination link. It introduces Partial Compress-and-Forward (PCF), which hybridizes CF and AF across the two relays and uses a joint LT-code based JSCC framework with a Tanner-graph joint decoder to exploit inter-relay side information. The authors derive capacity-like bounds for PCF, formulate a convex-like optimization over compression rates and time allocation (solved via projected gradient descent), and validate the approach with LT-based joint decoding in simulations. Results show that PCF can provide significant rate gains over decode-and-forward and pure CF when channel conditions are favorable, and even when only the relays connected to one side are good, highlighting its practical relevance for relay-rich networks. The work demonstrates the viability of combining rateless coding, JSCC, and mixed relay strategies to approach network capacity under realistic constraints.

Abstract

In this work, we propose a novel partial compress-and-forward (PCF) scheme for improving the maximum achievable transmission rate of a diamond relay network with two noisy relays. PCF combines conventional compress-and-forward (CF) and amplify-and-forward (AF) protocols, enabling one relay to operate alternately in the CF or the AF mode, while the other relay works purely in the CF mode. As the direct link from the source to the destination is unavailable, and there is no noiseless relay in the diamond network, messages received from both relays must act as side information for each other and must be decoded jointly. We propose a joint decoder to decode two Luby transform (LT) codes received from both relays corresponding to the same original message. Numerical results show that PCF can achieve significant performance improvements compared to decode-and-forward (DF) and pure CF protocols when at least the channels connected to one of the relays are of high quality.

Figures (8)

• Figure 1: The wireless network model in this work consists of one source. one destination, and two noisy relays. The direct link from the source to the destination is unavailable.
• Figure 2: PCF in time domain. In this figure, we assume that Relay-2 works in the CF mode only, while Relay-1 works alternately on the CF and the AF mode.
• Figure 3: The block diagram of PCF strategy in this correspondence. We use Raptor codes at the source and LT codes as JSCC at both relays.
• Figure 4: The joint admissible region of $R_\mathbf{X}$ and $R_\mathbf{Y}$ corresponds to our network defined by the Slepian-Wolf theory.
• Figure 5: The joint decoder is represented by the Tanner graph. In the graph, circles represent variable nodes and squares represent parity-check nodes. The joint decoder is the concatenation of two belief-propagation decoders.