Broadcast- and MAC-Aware Coding Strategies for Multiple User Information Embedding

Abdellatif Zaidi, Pablo Piantanida, Pierre DuhamelIEEE TSP 2007 · OpenReview

Abstract: Multiple user information embedding is concerned with embedding several messages into the same host signal. This paper presents several implementable dirty-paper-coding (DPC)-based schemes for multiple user information embedding, through emphasizing their tight relationship with conventional multiple user information theory. We first show that depending on the targeted application and on whether the different messages are asked to have different robustness and transparency requirements or not, multiple user information embedding parallels one of the well-known multiuser channels with state information available at the transmitter. The focus is on the Gaussian broadcast channel (GBC) and the Gaussian multiple access channel (GMAC). For each of these channels, two practically feasible transmission schemes are compared. The first approach consists in a straightforward—rather intuitive—superimposition of DPC schemes. The second consists in a joint design of these DPC schemes. This joint approach heavily relies on a recent work by Kim et al. in which the authors extend the single-user Costa's DPC to the multiple user case. The results in this paper extend the practical implementations quantization index modulation (QIM), distortion-compensated QIM (DC-QIM), and scalar Costa scheme (SCS) that have been originally conceived for one user to the multiple user case. After presenting the key features of the joint design within the context of structured scalar codebooks, we broaden our view to discuss the framework of more general lattice-based (vector) codebooks and show that the gap to full performance can be bridged up using finite dimensional lattice codebooks. Performance evaluations, including bit error rates (BERs) and achievable rate region curves are provided for both methods, illustrating the improvements brought by a joint design.

Introduction

Digital watermarking traditionally focuses on hiding a single message within a host signal. However, modern applications increasingly require the embedding of multiple messages with varying requirements. For instance, a medical image might necessitate a robust watermark for patient identification alongside a fragile one for tamper detection. The primary challenge in multiple user information embedding is the mutual interference between messages. Standard techniques like Quantization Index Modulation (QIM) were designed for single-user scenarios; simply superimposing them causes each watermark to treat the others as noise, severely limiting the achievable data rates and robustness.

Multi-User Communication Models

The authors bridge the gap between multi-user information theory and practical data hiding by leveraging Dirty Paper Coding (DPC). DPC demonstrates that if a transmitter has prior knowledge of the interference (the host signal), it can "pre-cancel" it to achieve the same capacity as an interference-free channel. This paper maps information embedding to two fundamental communication models:

1. Gaussian Broadcast Channel (GBC)

In the GBC scenario, a single encoder embeds multiple messages intended for different decoders. This is ideal for applications requiring different levels of robustness and transparency. For example, one decoder might extract a high-rate fragile signal, while another extracts a low-rate robust signal.

Gaussian Broadcast Channel (GBC) model for information embedding with fragile and robust decoders.

2. Gaussian Multiple Access Channel (GMAC)

The GMAC model involves multiple independent encoders embedding messages into the same host signal for a single joint decoder. This setup is relevant when different entities need to watermark the same content independently.

Gaussian Multiple Access Channel (GMAC) model featuring multiple independent encoders embedding into a single host.

Coding Strategies and "Awareness"

The core innovation is the introduction of "Awareness" in encoder design. Rather than treating concurrent watermarks as unknown noise, an "aware" encoder incorporates knowledge of other users' codebooks and power constraints.

  • Scalar Schemes: The authors extend the Scalar Costa Scheme (SCS) by jointly optimizing inflation parameters ($\alpha$) for all users.
  • Lattice-Based Coding: To approach theoretical Shannon limits, the work transitions from 1D scalar quantization to high-dimensional lattice codebooks (e.g., Hexagonal $A_2$ or 4D $D_4$ lattices). These provide superior "shaping gain" and more efficient information packing in the signal space.
  • Successive Interference Cancellation (SIC): In the MAC scenario, the decoder employs a "peeling-off" strategy, decoding messages sequentially and subtracting them to minimize interference for subsequent users.

Key Results and Insights

Performance was evaluated using synthetic Gaussian host signals and Monte Carlo simulations to generate Bit Error Rate (BER) and capacity curves. The study found that:

  • Expanded Rate Regions: Jointly designed "aware" schemes consistently outperform simple superimposition. In GBC, awareness significantly boosts the rate of the robust watermark.
  • BER Improvements: Joint designs achieve lower BER for the same Watermark-to-Noise Ratio (WNR) compared to independent schemes.
  • Lattice Gains: Moving to a $D_4$ lattice provides a measurable gain in bits per dimension, bridging the gap to ideal DPC capacity.

Limitations and Future Work

While promising, the current models rely on a "flat-host" assumption where host variance is much larger than the quantization step. Future research should address scenarios where this assumption fails. Additionally, the computational complexity of high-dimensional lattice decoding and the challenge of perfect synchronization in real-world environments (e.g., against cropping or lossy compression) remain open problems for the multi-user case.

Summary updated

Abstract

para xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> Multiple user information embedding is concerned with embedding several messages into the same host signal. This paper presents several implementable dirty-paper-coding (DPC)-based schemes for multiple user information embedding, through emphasizing their tight relationship with conventional multiple user information theory. We first show that depending on the targeted application and on whether the different messages are asked to have different robustness and transparency requirements or not, multiple user information embedding parallels one of the well-known multiuser channels with state information available at the transmitter. The focus is on the Gaussian broadcast channel (GBC) and the Gaussian multiple access channel (GMAC). For each of these channels, two practically feasible transmission schemes are compared. The first approach consists in a straightforward—rather intuitive—superimposition of DPC schemes. The second consists in a joint design of these DPC schemes. This joint approach heavily relies on a recent work by Kim <etal/> in which the authors extend the single-user Costa's DPC to the multiple user case. The results in this paper extend the practical implementations quantization index modulation (QIM), distortion-compensated QIM (DC-QIM), and scalar Costa scheme (SCS) that have been originally conceived for one user to the multiple user case. After presenting the key features of the joint design within the context of structured scalar codebooks, we broaden our view to discuss the framework of more general lattice-based (vector) codebooks and show that the gap to full performance can be bridged up using finite dimensional lattice codebooks. Performance evaluations, including bit error rates (BERs) and achievable rate region curves are provided for both methods, illustrating the improvements brought by a joint design. </para>
BibTeX
@article{zaidi2007broadcast,
  author = {Abdellatif Zaidi and Pablo Piantanida and Pierre Duhamel},
  title = {{Broadcast- and MAC-Aware Coding Strategies for Multiple User Information Embedding}},
  journal = {IEEE Trans. Signal Process.},
  year = {2007},
  url = {https://openreview.net/forum?id=OvUmMc1XiIC},
}
Figure